I

Structural
Engineering
Handbook
Other McGraw-Hill Handbooks of Interest

A val/one and Baumeister . MARKS' ST ANDARD HANDBOOK FOR MECHANICAL ENGINEERS

Brady and Clauser • MATERIALS HANDBOOK
Brater . HANDBOOK OF HYDRAULICS
Callender . TIME-SA VER STANDARDS FOR ARCHITECTURAL DESIGN DATA
Crocker and King • PIPING HANDBOOK
Croft and Summers • AMERICAN ELECTRICIANS' HANDBOOK
Davis and Sorensen • HANDBOOK OF APPLIED HYDRAULICS
DeChiara and Callender . TIME-SA VER STANDARDS FOR BUILDING TYPES
Faherty and Williamson . WOOD ENGINEERING AND CONSTRUCTION HANDBOOK
Fink and Beaty . STANDARD HANDBOOK FOR ELECTRICAL ENGINEERS
Gieck . ENGINEERING FORMULAS
Harris • HANDBOOK OF NOISE CONTROL
Harris and Crede . SHOCK AND VIBRATION HANDBOOK
Hicks . STANDARD HANDBOOK OF ENGINEERING CALCULATIONS
Higgins . HANDBOOK OF CONSTRUCTION ENGINEERING MAINTENANCE
Higgins . MAINTENANCE ENGINEERING HANDBOOK
Manas . NATIONAL PLUMBING CODE HANDBOOK
McPartland. McGRAW-HILL'S NATIONAL ELECTRICAL CODE HANDBOOK
Merritt . BUILDING DESIGN AND CONSTRUCTION HANDBOOK
Merritt . STANDARD HANDBOOK FOR CIVIL ENGINEERS
Merritt • STRUCTURAL STEEL DESIGNER'S HANDBOOK
O'Brien . CONTRACTOR'S MANAGEMENT HANDBOOK
Parmley . FIELD ENGINEER'S MANUAL
Parmley. STANDARD HANDBOOK OF FASTENING AND JOINING
Rosa/er and Rice. STANDARD HANDBOOK OF PLANT ENGINEERING
Rossnagel, MacDonald, and Higgins • HANDBOOK OF RIGGING
Smeaton • SWITCHGEAR AND CONTROL HANDBOOK
Tuma . ENGINEERING MATHEMATICS HANDBOOK
Tuma . HANDBOOK OF NUMERICAL CALCULATIONS IN ENGINEERING

For more information about other McGraw-Hill materials,

ca// 1-800-2-MCGRAWin the United States. In other
countries, call your nearest McGraw-Hill office.
Sbuctural
Engineering
Handbook
Edited by

Edwin H. Gaylord, Jr.

Professor of Civil Engineering, Emeritus
University of Illinois at Urbana-Champaign

Charles N. Gaylord
Late Professor of Civil Engineering, Emeritus
University of Virginia

Third Edition

McGraw-Hill Publishing Company

New York San Francisco Washington, D.C. Auckland Bogota
Caracas Lisbon London Madrid Mexico City Milan
Montreal New Delhi San Juan SAo Paulo Singapore
Sydney Tokyo Toronto
Library of Congress Cataloging-in-Publication Data
Structural engineering handbook / edited by Edwin H. Gaylord, Jr.,
Charles N. Gaylord.-3rd ed.
p. cm.
Includes bibliographies and index.
ISBN 0-07-023188-5
1. Structural engineering-Handbooks, manuals, etc. I.
Gaylord, Edwin Henry. II. Gaylord, Charles N.
TA635.S77 1990
624.l-dc20 89-2577
CIP

Copyright© 1990 by McGraw-Hill, Inc. All rights reserved.

Printed in the United States of America. Except as permitted
under the United States Copyright Act of 1976, no part of this
publication may be reproduced or distributed in any form or by
any means, or stored in a data base or retrieval system, without
the prior written permission of the publisher.

567890 DOCDOC 9876543

ISBN0-07-023188-5

The editors for this book were Harold B. Crawford and Susan Thomas,
the designer was Naomi Auerbach, and the production supervisor was
Suzanne W. Babeuf. It was set in Caledonia by University Graphics,
Inc.

Printed and bound by the R. R. Donnelley & Sons Company.

Information contained in this work has been obtained by

McGraw-Hill, Inc., from sources believed to be reliable. However,
neither McGraw-Hill nor its authors guarantees the accuracy or
completeness of any information published herein and neither
McGraw-Hill nor its authors shall be responsible for any errors,
omissions, or damages arising out of use of this information. This
work is published with the understanding that McGraw-Hill and
its authors are supplying information but are not attempting to
render engineering or other professional services. If such services
are required, the assistance of an appropriate professional should
be sought.
 Contents

Preface xv
Contributors xvii

Section 1 STRUCTURAL ANALYSIS 1-1

Part 1. Fundamentals and Analysis of Elastic Trusses and Frames 1-1

David A. Pecknold
Introduction: Classification of Structures; Fundamentals of Analysis: Descrip-
tion of Structural Configuration-Basic Principles-Static Determinacy and Sta-
bility-Superposition-Symmetry; Energy Principles: Principle of Virtual
Work-Principle of Minimum Potential Energy-Principle of Minimum Com-
plementary Potential Energy-The Reciprocal Theorem; Analysis of Statically
Determinate Structures: Forces: Plane and Space Trusses-Beams and
Frames-Beam Deflections; Analysis of Statically Indeterminate Structures:
Comparison of Force and Displacement Methods; Force Method: Structure
Flexibility Coefficients-Dummy Unit-Load Method Examples-Three-
Moment Equation for Continuous Beams; Displacement Method: Structure Stiff-
ness Coefficients-Member Stiffness Matrices-Member Fixed-End Forces-
Portal Frame by Displacement Method-Slope-Deflection Equations-Moment
Distribution--Matrix Formulation of the Displacement Method-Frame Ele-
ment Stiffness Properties and Equivalent Loads Using Virtual Work

Part 2. Continua: The Finite-Element Method William C. Schnobrich 1-46

Finite-Element Method: Requirements for Finite-Element Analysis-Discreti-
zation of the Structure-Guidelines for Selection of Grid-Element Models-
Plane-Stress and Plain-Strain Elements-Plane-Stress Analysis-Beam and Plate
Bending-Shells and Combined Direct Stress and Bending-Three-Dimen-
sional Problems-Element Stiffness Matrix and Nodal Force Vector-Equilib-
rium Equations for the Assemblage-Solution for the Displacements-Solut,ion
for Element Strains and Stress.

Section 2 COMPUTER APPLICATIONS IN STRUCTURAL 2-1

ENGINEERING Steven]. Fenves

The Structural Design Process-The Program-Development Process; Program-

Development Tools: Structured Programming-Programming Languages-
Flowcharting-Tools for Organizing Programs-Tools for Organizing Data-

V
vi Contents

Generic Software-Computer Hardware; Interaction with Computers: Modes

oflnteraction-Media of Interaction-Level of Interaction; Program Types and
Sources: Program Types-Sources of Programs; Typical Applications: Analy-
sis-Proportioning-Detailing-Preparation of Final Documents

Section 3 EARTHQUAKE-RESISTANT DESIGN W. ]. Hall 3-1

Response of Simple Structures to Earthquake Motions-Earthquake Motions;

Response Spectra: Elastic Systems-Design Response Spectra-Response Spec-
tra for Inelastic Systems-Multi-Degree-of-Freedom Systems; Computation of
Period of Vibration: Fundamental Mode-Higher Modes-Modal Participation
Factors-Spring Constant for Equivalent Shear Beam; Design: General Consid-
erations-Effects of Design on Behavior-Design Lateral Forces-Seismic
Forces for Overturning Moments and Shear Distribution-Damping-Gravity
Loads-Vertical and Horizontal Excitation-Unsymmetrical Structures in Tor-
sion-Curtain-Wall Buildings-Core Walls-Parts of Buildings-Effects of Soil
Conditions-Detailing and Quality Control-Cost-Lifelines

Section 4 FATIGUE, BRITTLE FRACTURE AND LAMELLAR 4-1

TEARING W. H. Munse

Fatigue of Structural Steel: Significance of Fatigue-Fatigue of Structural

Steels-Factors Affecting Fatigue Strength-Structural Members-Riveted
Connections-Bolted Connections-Welded Connections-Design for
Fatigue-Protecting Against Fatigue; Brittle Fracture of Structural Steel: Sig-
nificance of Brittle Fractures-Initiation, Propagation, and Arrest-Factors
Affecting Brittle Fracture-Fabrication-Inspection-Lamellar Tearing

Section 5 SOIL MECHANICS H. G. Larew 5-1

Nature and Causes of Soil Deformation: Settlement-Frost Movements-

Shrinkage-Subsidence-Soil Deformation-Time-dependent Deformation;
Strength Properties of Soils: Compressive Strength-Effect of Confining Pres-
sures-Transient and Repeated Loads; Factors Affecting Bearing Pressure:
Allowable Bearing Pressure-Permissible Settlement-Elastic-plastic Defor-
mation-Shear Failures-Consolidation-Time Rate of Settlement; Cuts and
Embankments: Slope Stability; Compaction and Permeability: Compaction-
Permeability; Appendix

Section 6 SOIL EXPLORATION Thomas H. Thornburn (deceased) 6-1

and James W. Mahar
General Foundation Conditions: Sources of Information-Glacial Materials-
Wind-laid Materials-Water-laid Materials-Organic Soils-Residual Soils-
Frozen Soils; Exploratory Procedures: Test Pits-Borings-Sampling-Bore-
hole Cameras-Boring Reports-Standard Penetration Tests-In Situ Strength
Tests-Correlations with Surficial Geology-Geophysical Surveys-Load
Tests-Special Observations-Preliminary Site Inspection-Fact-finding Sur-
vey-Borings
 Contents vii

Section 7 RETAINING STRUCTURES AND FOUNDATIONS 7-1

Herbert 0. Ireland
Earth Pressures: Stresses in Earth Mass-Rankine's Theory-Coulomb's The-
ory-Equivalent-Fluid Method-Trial-Wedge Method; Retaining Walls: Types
and Behavior-Determining Earth Pressure-Bases on Piles or Piers-Bear-
ing Capacity-Stability-Drainage-Other Considerations-Gravity Walls-
Semi-gravity Walls-Cantilever Walls-Counterfort Walls-Joints; Abut-
ments; -Bulkheads: Forces on Bulkheads-Penetration of Piles-Anchorage;
Footing Foundations: Footings on Clay-Footings on Sand-Footings on Silt
and Loess; Raft Foundations: Raft on Clay-Rafts on Sand; Pier Foundations:
Open Excavations-Drilled Piers-Piers on Clay-Piers on Sand-Caisson
Foundations-Foundation Requirements; Pile Foundations: Pile-driving Equip-
ment-Pile-driving Formulas-Pile Tests-Piles in Sand-Piles in Clay-Set-
tlement of Pile Foundations-Laterally Loaded Piles-Batter Piles-Lateral
Stability of Poles-Guy Anchors-Foundations Subjected to Uplift-Improve-
ment of Subsoil-Construction Problems

Section 8 DESIGN OF STEEL STRUCTURAL MEMBERS 8-1

Lee C. Lim, William J. LeMessurier, and Hans William Hagen
Types of Steel-Shapes; Tension Members: Concentrically Loaded Tension
Members-Threaded Members-Member Types and Selection-Truss Mem-
bers; Compression Members: Column Strength-Concentrically Loaded Col-
umns-Effective Length-Amplification Factors and Frame Stability-Propor-
tioning-Local Buckling-Lacing and Perforated Cover Plates-Tapered
Columns-Slender Compression Elements; Beams: Allowable Stresses-Biaxial
Bending-Shear-Deflection-Combined Bending and Compression; Plate
Girders: Web-Flanges-Lengths of Flange Plates-Lateral Buckling-
Requirements for Stiffeners-Combined Bending and Shear; Welded Connec-
tions: Welding Processes-Weld Classification-Weldability-Electrodes-
lnspection-Fillet-welded Joints-Groove-welded Joints-Concentrically
Loaded Connections-Beam Seat Connections-Stiffened Beam-Seat Connec-
tions-Framed Beam Connections-Moment-resistant Beam Connections; Riv-
eted and Bolted Connections: Rivets-High-Strength Bolts-Installation of
High-Strength Bolts-Inspection of High-Strength Bolts-Unfinished Bolts-
Turned Bolts-Ribbed Bolts-Bearing Bolts-High-Strength Tension Control
and Tension Set Bolts-Shear Connections-Eccentrically Loaded Connec-
tions, Fasteners in Shear-Eccentrically Loaded Connections, Fasteners in Ten-
sion-Flexible Beam-Seat Connections-Stiffened Beam-Seat Connections-
Framed Beam Connections-Moment-resistant Beam Connections-Pinned
Connections; Bearing Plates and Splices: Beam Bearing Plates-Column
Bases-Compression-Member Splices

Section 9 PLASTIC DESIGN OF STEEL FRAMES Lynn S. Beedle 9-1

and T. V. Galambos
Inelastic Bending-Indeterminate Structures; Analysis: Theorems-Statical
Method-Mechanism Method-Moment Check-Instantaneous Center-Dis-
tributed Loads; Deflection Analysis: Deflection at Ultimate Load-Deflection at
Working Load; Design Requirements: Specifications-Loads and Forces-Gen-
eral Design Procedure-Preliminary Design-Analysis; Secondary Design Con-
siderations: Axial Force-Lateral Bracing-Local Buckling-Shear-Frame
Instability; Connections: Corner Connections-Interior Beam-to-Column
Connections
viii Contents

Section 10 DESIGN OF COLD-FORMED STEEL STRUCTURAL 10-1

MEMBERS Roger A. LaBoube
Materials-Shapes and Uses; Design: Section Properties-Thin Compression
Elements-Uniformly Compressed Elements-Elements with Stress Gradi-
ent-Elements with an Edge Stiffener-Elements with an Intermediate
Stiffener-Members-Compression Members-Flexural Members-Beam
Webs-Combined Bending and Axial Compression-Connections-Wall
Studs-Effects of Cold Forming on Steel Properties-Tests for Special Cases-
Industrial Storage Racks-Shear Diaphragms-Folded-Plate and Shell Roofs

Section 11 DESIGN OF ALUMINUM STRUCTURAL MEMBERS 11-1

john W. Clark
Materials and Specifications: Shapes-Codes and Specifications-Applications
and Alloys; Design of Tension Members: Yielding and Fracture-Welded Ten-
sion Members; Design of Compression Members: Column Formulas-Lacing-
Local Buckling of Plates, Legs, and Flanges in Edge Compression-Stiffeners for
Flat Plates in Edge Compression-Local Buckling of Round Tubes in End Com-
pression-Welded Compression Members; Beams: Yielding and Ultimate
Strength-Lateral-Torsional Buckling-Local Buckling of Beams-Shear
Strength of Beam Webs; Plate Girders: Lateral Buckling-Design of Web-
Vertical Stiffeners-Longitudinal Stiffeners-Bearing Stiffeners; Combined
Loading: Combined Bending and Axial Load-Plate and Tube Buckling under
Combined Stress; Connections: Riveted Connections-Bolted Connections-
Welded Connections; Design for Repeated Loads

Section 12 DESIGN OF REINFORCED CONCRETE STRUCTURAL 12-1

MEMBERS William L. Gamble

Concrete-Reinforcement-Specifications, Codes, and Standards-Strength

Design and Working-Stress Design-AC! Load and Reduction Factors-Preci-
sion-Rectangular Beams-Continuity-Doubly Reinforced Beams-Tee
Beams-Special Beam Shapes-Shear and Diagonal Tension-Development
and Anchorage of Reinforcement-Splices-Bar Cutoffs and Bend Points-
Deflection-Column Design-Combined Compression and Bending-Column
Splices-Columns With Biaxial Bending-Stairs-Wall Footings-Column
Footings-Walls-Slabs-Structural Framing Systems

Section 13 DESIGN OF PRESTRESSED-CONCRETE 13-1

STRUCTURAL MEMBERS T. Y. Lin and Paul Zia

Notation; Materials: Concrete-Steel-Grouting; Methods and Systems of

Prestressing: Tensioning Methods-Pretensioning-Posttensioning Systems;
Loss of Prestress: Elastic Shortening of Concrete-Creep-Shrinkage-Relax-
ation in Steel-Slippage of Tendons during Anchoring-Friction-Effective
Prestress-Elongation of Tendons; Analysis for Flexure: Basic Concepts-
Stress in Steel-Cracking Moment-Ultimate Moment-Composite Sections;
Design for Flexure: Preliminary Design-Elastic Design-Ultimate Design-
Balanced-Load Design-Deflections; Shear, Bond, and Bearing: Principal Ten-
sion-Web Reinforcement-Prestress Transfer Bond-Anchorage; Typical Sec-
tions: Beam Sections-Span-Depth Ratios-Cable Layouts-Tendon Protection
and Spacing-Partial Prestress-Combination of Prestressed and Reinforced
Concrete; Continuous Beams: Continuous-Beam C Lines-Load-Balancing
' Met~od-Ultimate Strength of Continuous Beams; Design Examples
 Contents ix

Section 14 DESIGN OF COMPOSITE MEMBERS Ira Hooper, 14-1

Michael A. Grubb, and Ivan M. Viest
Definitions-Elastic Properties of Cross Section-Plastic Strength of Beam
Cross Section-Critical Column Stress-Combined Compression and Flex-
ure-Shear Connectors-Unsymmetrical Steel Sections-Negative-Moment
Sections-Deflections and Vibrations-Composite Columns; Building Design:
Assumptions-Design of Composite Beams-Selection of Steel Section-
Design of Shear Connectors-Design of Composite Columns; Bridge Design:
Assumptions-Design of Composite Beams-Steel-Member Selection-Design
of Shear Connectors

Section 15 MASONRY CONSTRUCTION Walter L. Dickey 15-1

Materials: Burned-Clay Units-Brick-Structural Clay Tile-Concrete Units-

Dimensions-Mortar-Grout; Nomenclature; Unreinforced Masonry; Rein-
forced Masonry: Materials-Design-Allowable Stresses-Beams-Walls-
Slender Walls-Columns-Connections-Diaphragms; Tests and Inspection:
Compressive Strength of Masonry-Field Tests; Detailing and Construction:
Detailing-Concrete Foundations-Storing-Workmanship-Shoring and Scaf-
folding- Prefabrication

Section 16 TIMBER STRUCTURES Kenneth P. Milbradt 16-1

Structural Properties of Wood: Anisotropic Nature of Wood-Elastic Con-

stants-Directional Strength Properties-Factors Affecting Strength-Work-
ing Stresses for Sawn Lumber-Glued-Laminated Lumber-Plywood; Fasten-
ers: Bolts-Split Rings and Shear Plates-Truss Plates; Beams: Flexure-
Shear-Bearing-Deflections-Lateral Stability-Continuous Spans-Pitched
and Tapered Beams; Columns: Solid Columns-Box Columns-Spaced Col-
umns-Beam Columns; Trusses: Proportions-Design of Members-Deflec-
tions-Camber-Bracing-Trussed Joists; Arches: Three-Hinged Tudor
Arch-Two-Hinged Arches; Shell Structures: Domes-Barrel Vaults-Hyper-
bolic Paraboloids

Section 17 ARCHES AND RIGID FRAMES Thomas C. Kavanagh 17-1

(deceased) and Robert C. Y. Young
Nomenclature and Classification; Analysis: Assumptions-Kern Relationships-
Finite-Element Analysis-Energy Methods; Design of Arches: General Proce-
dure-Preliminary Selection of Shape-Approximations for Special Shapes-
Intermediate Design-Approximations of Whitney Data-Final Design-
Unsymmetrical Arches-Ultimate Design of Concrete Arches; Design of
Frames: Steel Frames-Concrete Rigid-frame Bridges-Design-Arched
Bents, Continuous Arches on Elastic Piers; Special Topics: Second-Order The-
ory-Interaction of Arch and Deck-Buckling of Arches-Laterally Loaded
Arches and Frames-Skewed Barrel Arches and Rigid-Frame Slabs; Construc-
tion and Details: Concrete Arches and Frames-Steel Arches and Rigid
Frames-Economics

Section 18 STEEL AND CONCRETE BRIDGES Arthur L. Elliott 18-1

General-Aesthetics-Loads-Maximum Moments and Shear in Simple

Spans-Positive Moments in Continuous Spans-Negative Moments in Contin-
x Contents

uous Spans-Shears in Continuous Spans-Impact-Wind-Other Loads-

Design Methods; Steel Bridges: Floor Systems-Concrete Floors-Steel
Floors-Floor Beams; Bearing and Expansion Details: End Bearings-Expan-
sion Hangers-Deck Expansion Joints; Beam and Plate-Girder Bridges: Beam
Bridges-Plate-Girder Bridges-Composite Beam Bridges-Continuous
Spans-Spacing-Lateral Systems-Deflection-Welded Plate Girders-
Design of a Welded Plate Girder-Web Splices-Field Splices; Truss Bridges:
Proportions-Loads and Stresses-Truss Members-Lateral Forces; Concrete
Bridges: Camber, Plastic Flow, and Shrinkage; Slab Bridges: Simple Spans-
Continuous Spans-Design of Bents-Typical Details, Continuous Slabs; T-
heam Bridges: Economics-Design of a T-beam Bridge-Design of Substruc-
ture-Typical Details; Box-Girder Bridges: Economics-Proportions-
Design-Substructure; Prestressed-Concrete Bridges: Standard Sections-
Stresses-Path of Prestressing Force-Friction Losses-Ultimate Load-Web
Reinforcement-Uplift-Live-Load Deflection; Bridge Railings: Railing
Design-Curbs and Sidewalks-Pedestrian Railings; Administration of a Bridge
System: Maintenance and Inspection of Existing Bridges-Bridge
Administration

Section 19 STEEL-PLATE-DECK BRIDGES Roman Wolchuk 19-1

Applications-Economic Considerations-Structural Behavior-Deck Plate-

Design Stresses and Loads; Open-Rib Decks: Design of Ribs-Design of Floor
Beams-Stresses in Ribs and Floor Beams; Closed-Rib Decks: Design of Closed
Ribs; Box Girders: Analysis of Box Girders-Design in Accordance With Clas-
sical Elastic Theory; Design with Consideration of Imperfections: Effect of
Imperfections on Behavior of Steel Plating-Unstiffened Plate Panel under Axial
Compression-Stiffened Plate Panel under Axial Compression-Plate Panel
under Combined Shear and Axial Load-Load-Bearing Diaphragms; Construc-
tion Details: Deck Details, Fabrication, and Erection-Erection of Box Girders;
Wearing Surfaces: Seal and Tack Coating-Surface Courses; Railroad Bridges

Section 20 CURVED STEEL I-GIRDER BRIDGES 20-1

William N. Poellot, Jr.
Design Considerations: Bridge Geometry-Thermal Effects-Framing
Arrangements-Number of Design Girders-Connection Details-Camber;
Curved Girder Mechanics: Torsion, Warping Stresses-Radial Component of
Flange Force-V-Load Approximation-Lateral Flange Bending-Centrifugal
Force Effect; Diaphragm Forces

Section 21 BUILDINGS-GENERAL DESIGN 21-1

CONSIDERATIONS Stephen]. Y. Tang and Ian R. Chin
Planning Building Structures: Selection of Structural Scheme-Spatial Require-
ments-Wind Systems-Deflection-Structural Materials-Fire Resistance-
Deterioration-Provision for Environmental-Control Systems-Limitations of
Various Systems; Loads: Dead Load-Live Load-Snow Loads-Wind Loads-
Seismic Loads; Floor and Roof Construction: Floor and Roof Systems-Floor
Finish-Roofing; Wall Construction: Type ofWalls-Nonbearing Walls-Bear-
ing Walls-Windows; Stairs: Planning-Types-Framing-Steel Stairs-Con-
crete Stairs-Escalators; Miscellaneous Considerations: Openings and Voids-
Thermal and Seismic Movement
 Contents xi

Section 22 INDUSTRIAL BUILDINGS James M. Fisher 22-1

Design Philosophy-Identification of Client Requirements-Building Layout-

Identifying Structural Requirements-Selection of Roofing and Wall Material-
Selection of Bay Size; Framing Systems: Concrete-Framed Buildings-Wood-
Framed Buildings-Steel-Framed Buildings-Lateral-Load Systems: Materials
Handling; Crane Buildings: Fatigue-Crane Girder Design-Crane Columns

Section 23 TALL BUILDINGS Morton H. Eligator, Sing L. Chu, 23-1

and Louis A. Occhicone
Framing: Bay Sizes-Columns-Elevator Shafts-Moving Stairs-Stairwells-
Transfer Girders and Trusses; Wind Bracing: Braced Bents, Rigid Frames, Shear
Walls, and Interaction Systems-Tubular Frames, Tube Within a Tube, and
Combinations-Fixed and Partially Fixed Joints in Steel Structures-Wind-
Load Determination-Wind Deflection-Wind-Shear Dissipation-Approxi-
mate Methods of Analysis-Computer Methods-Damping Systems-Thermal
Effects

Section 24 THIN-SHELL CONCRETE STRUCTURES 24-1

David P. Billington
Introduction to Design-Simplified Analyses-Basis for the Membra"i-ie The-
ory-General Method of Analysis-Classification of Shells; Shell Walls: Behav-
ior of Cylindrical Shell Walls-Cylindrical Tanks-Hyperboloids-Membrane
Theory-Bending Theory-Examples-Behavior of Barrel-Shell Roofs-Long
Barrels-Short Barrels-Transverse Frames-Barrel-Shell Reinforcement;
Folded Plates: Analysis of Folded Plates-Prestressed Folded Plates-Contin-
uous Folded Plates; Translation Shells of Double Curvature: Membrane The-
ory-Elliptic Paraboloids-Hyperbolic Paraboloids; Dimensioning; Stability
and Safety; Construction; Appearance

Section 25 SUSPENSION ROOFS Lev Zetlin and I. Paul Lew 25-1

Examples of Suspension Structures; Design of Suspension Systems: Anchorage

Forces-Dynamic Behavior-Single Cable under Uniformly Distributed
Load-Configuration and Shapes of Suspension Structures; Double Layer of
Prestressed Cables: Damped Suspension Systems-Structural Relationships-
Notation-Preliminary Design of Double-Layer Cable System-Analysis of
Double-Layer Cable System-Behavior of Pair-Set of Cables-Application to
Preliminary Design of Cable Grids and Membranes-Load Combinations for
Selection of Cables-Types of Cables-Fittings-Membranes

Section 26 REINFORCED-CONCRETE BUNKERS AND 26-1

SILOS German Gurfinkel

Introduction-Bin Pressures-Emptying Pressures on Silo Walls, DIN 1055

Sheet 6-Emptying Pressures in Funnel-Flow Silos, ACI 313-Silage-Pres-
sures on Horizontal Bottoms and Inclined Hoppers-Earthquake Forces; Wall
Forces: Circular Silos-Rectangular and Polygonal Silos-Thermal Effects;
Design of Walls: Minimum Thickness of Circular Walls-Maximum Crack
Width-Walls in Tension-Walls in Tension and Flexure-Walls in Compres-
sion-Walls in Compression and Flexure-In-Plane Bending of Walls-Walls
xii Contents

Subjected to Thermal Stresses-Vertical Reinforcement-Details and Place-

ment of Reinforcement; Design of Bottoms: Bottom Pressure-Plane Bottoms-
Conical Hoppers-Pyramidal Hoppers-Hopper-Supporting Beams-Col-
umns-Roofs-Failures-Dust Explosions in Grain Elevators and Flour Mills;
Examples

Section 27 STEEL TANKS Robert S. Wozniak 27-1

Reservoirs: Capacity-Shell Design-Bottom Plates-Concrete Ringwall-

Roofs; Standpipes: Design-Anchorage-Foundations; Elevated Tanks:
Roofs-Bottoms-Balcony or Ring Girder-Columns-Single-Pedestal
Tanks-Foundations; Accessories; Bins: Forces-Miscellaneous Details;
Materials

Section 28 TOWERS AND TRANSMISSION POLE 28-1

STRUCTURES Alain H. Peyrot and Gene M. Wilhoite
Types of Towers-Materials-Height Limitations-Loads-Candelabra; Free-
standing Towers: Stresses-Foundations; Guyed Towers: Wind-Design of
Guys-Ice Loading-Guy Tensioning-Guy Vibration-Design of Mast-
Foundations-Erection; Transmission Towers: Types-Loads-Vibration-
Stress Analysis-Steel Tension Members-Aluminum Tension Members-Steel
Compression Members-Aluminum Compression Members-Limiting Slender-
ness Ratios-Bolts-Tower Design-Foundations-Tower Tests; Pole Struc-
tures: Design-Material-Pole Splices-Foundations

Section 29 BURIED CONDUITS Raymond]. Krizek 29-1

Types of Conduits-Analysis and Design; Loads on Conduits: Loads on Ditch

Conduits-Loads on Projecting Conduits-Loads on Conduits in Wide
Ditches-Loads on Negative Projecting and Imperfect-Ditch Conduits-Sur-
face Loads; Rigid Conduits: Supporting Strength-Bedding Classes for Trench
Conduits-Bedding Classes for Embankment Installation-Monolithic Con-
duits; Flexible Conduits: Ring Compression-Deflection-Pipe Arches-
Arches on Rigid Foundations; Pressure Conduits: Flexible Pressure Conduits-
Rigid Pressure Conduits; Modern Design Methodology: Elasticity Solution-
Finite-Element Solution-CANDE (Culvert ANalysis and DEsign); Additional
Design Considerations: Handling Criteria-Durability-Camber-Wrappings
and Coatings; Construction Considerations: Site Preparation-Bedding-
Fill Construction-Compaction Procedures-Strutting-Joints-Backpacking;
Long-Span Corrugated-Metal Conduits

Section 30 CHIMNEYS Shih-Lung Chu, Shu-Jin Fang, and Max Zar 30-1
Materials-Diameter and Heights; Design Loads: Dead Loads-Wind Loads-
Earthquake Forces-Pressure Differentials-Temperature Differentials-Nat-
ural Frequency of Vibration; Steel Stacks: Allowable Stresses-Cone-to-Cylin-
der Junction-Circumferential Stiffeners-Anchor Bolts-Base Ring for
Anchor Bolts-Guyed Stacks-Braced Stacks-Resonant Vibrations; Rein-
forced Concrete Chimneys: ACI Standard-Vibration due to Wind; Linings:
Foundations
 Contents xiii

APPENDIX A-1

Torsional Properties of Solid Cross Sections; Torsional Properties of Closed

Thin-Walled Cross Sections; Torsional Properties of Open Cross Sections; Effec-
tive-Length Coefficients for Columns; Buckling of Plates under Edge Stress;
Stiffened Beam Webs; SI Conversion Factors

Index follows the Appendix.

 Preface

This Handbook provides engineers, architects, and students of civil

engineering and architecture with an authoritative reference work on
the planning and design of a variety of engineered structures. Among
the structures covered are industrial buildings, tall buildings, bridges,
thin-shell structures, arches, suspension roofs, tanks for liquid storage,
bins and silos for granular materials, retaining walls, bulkheads, steel
transmission towers and poles, chimneys, and buried conduits. Design
in reinforced concrete, prestressed concrete, steel, composite con-
struction, wood, aluminum, and masonry are covered. Sections on soil
mechanics, soil exploration, and foundations, and a comprehensive
treatment of structural analysis, give the designer the information likely
to be needed for these phases of design. Earthquake-resistant design
and design against fatigue, brittle fracture, and lamellar tearing are
treated.
In this third edition every section has been reviewed, and many have
been revised extensively. The section on structural analysis has been
completely rewritten so as to emphasize computer-based techniques,
and the treatment of finite-element idealization of various types of
structures, including recommendations as to suitable types of elements,
has been expanded.
The section on steel structural members has been expanded to
include coverage of the new Load and Resistance Factor Design spec-
ification (LRFD). Design of composite beams and girders according to
the LRFD specification is also covered, and a detailed description and
an example of the autostress design procedure for highway bridges is
given. There is also a new section on curved steel I-girder bridges.
A discussion oflamellar tearing and suggestions on how to avoid it has
been added to the section on fatigue and brittle fracture. Coverage of
fatigue is updated, with comprehensive tables for determining the
fatigue life of various types of joints.
The section on cold-formed steel members has been rewritten so as
to conform to changes in the 1986 edition of the American Iron and
xv
xvi Preface

Steel Institute specification, which is an extensive revision of earlier

editions. The sections on structural members of reinforced concrete,
prestressed concrete, masonry, wood, and aluminum have also been
revised to conform to the latest design specifications.
The 30 sections have been written by 45 contributors. They have pre-
sented their material in ready-to-use form wherever possible. To this
end, derivations of formulas are omitted in all but a few instances and
many worked-out examples are given. Background information,
descriptive matter, and explanatory material have been condensed or
omitted. Because each section treats a subject which is broad enough
to fill a book in itself, the contributors have had to select that material
which in their judgment is likely to be most useful to the greatest num-
ber of users. However, sources of additional material are noted for most
of the topics which could not be treated in sufficient detail
Each section was edited to minimize duplication, to arrange the con-
tents of the book in a logical order, and to see that important topics
were not overlooked. The contributors' painstaking efforts, their coop-
eration in the review and editing of their work, and their patience dur-
ing the time it has taken to complete the third edition are gratefully
acknowledged.

Edwin H. Gaylord, Jr.

 Contributors

Lynn S. Beedle Director, Fritz Engineering Laboratory, Lehigh University (Plastic Design
of Steel Frames)
David P. Billington Professor of Civil Engineering, Princeton University (Thin-Shell Con-
crete Structures)
Senior Consultant, Wiss,Janney, Elstner Associates, Inc., Chicago, IL (Build-
Ian R. Chin
ings-General Design Considerations)
Sing L. Chu Senior Associate Partner, Weiskopf and Pickworth, Consulting Engineers,
New York, NY (Tall Buildings)
John W. ClarkFormerly Technical Adviser, Alcoa Laboratories, Alcoa Center, PA (Design
of Aluminum Structural Members)
Walter L. Dickey Consulting Engineer, Higgins Brick Company, Los Angeles, CA (Masonry
Construction)
Morton H. Eligator Partner, Weiskopf and Pickworth, Consulting Engineers, New York, NY
(Tall Buildings)
Arthur L. Elliott Bridge Engineer, Sacramento, CA (Bridges)
Steven J. Fenves Sun Company University Professor of Civil Engineering, Carnegie-Mel-
lon University (Computer Applications in Structural Engineering)
James M. Fisher Vice President, Computerized Structural Design, Inc., Milwaukee, WI
(Industrial Buildings)
T. V. Galambos Professor of Civil and Mineral Engineering, University of Minnesota-
Twin Cities (Plastic Design)
William L. GambleProfessor of Civil Engineering, University of Illinois at Urbana-Cham-
paign (Design of Reinforced-Concrete Structural Members)
Michael A. Grubb AISC Marketing, Inc., Pittsburgh, PA (Design of Composite Members)
German Gurfinkel Professor of Civil Engineering, University of Illinois at Urbana-Cham-
paign (Reinforced-Concrete Bunkers and Silos)
Hans William Hagen Partner, LeMessurier Consultants Inc., Consulting Engineers, Cam-
bridge, MA (Design of Steel Structural Members)
Professor and Head of Civil Engineering, University of Illinois at Urbana-
William J. Hall
Champaign (Earthquake-Resistant Design)
Ira Hooper Seelye Stevenson Value & Knecht, New York, NY (Design of Composite
Members)
Professor Emeritus of Civil Engineering, University of Illinois at
Herbert O. Ireland
Urbana-Champaign (Retaining Structures and Foundations)

xvii
xviii Contributors

Thomas C. Kavanagh (deceased) Former Vice-President, Iffiand Kavanagh Waterbury,

P.C., New York, NY (Arches and Rigid Frames)
Raymond J. Krizek Professor of Civil Engineering, Northwestern University (Buried
Conduits)
Roger A. LaBoube Associate Professor, University of Missouri-Rolla, Rolla, MO (Design
of Steel Cold-Formed Members)
H. G. Larew Professor of Civil Engineering, University of Virginia (Soil Mechanics)
William J. LeMessurierSenior Partner, LeMessurier Consultants, Inc., Consulting Engi-
neers, Cambridge, MA (Design of Steel Structural Members)
I. Paul Lew Vice President, Lev Zetlin Associates, Inc., Engineers and Designers, New
York, NY (Suspension Roofs)
Lee C. Lim President, Lim Consultants, Inc., Consulting Engineers, Cambridge, MA
(Design of Steel Structural Members)
T. Y. Lin Professor Emeritus of Civil Engineering, University of California, Berkeley
(Design of Prestressed Concrete Structural Members)
James W. Mahar Resident Engineer, Department of Civil Engineering, University of Illi-
nois at Urbana-Champaign (Soil Exploration)
Kenneth P. Milbradt Associate Professor Emeritus of Civil Engineering, Illinois Institute
of Technology (Timber Structures)
W. H. Munse Professor Emeritus of Civil Engineering, University of Illinois at Urbana-
Champaign (Fatigue, Brittle Fracture, and Lamellar Tearing)
Louis A. Occhicone Senior Associate Partner, Weiskopf and Pickworth, Consulting Engi-
neers, New York, NY (Tall Buildings)
David A. Pecknold Professor of Civil Engineering, University of Illinois at Urbana-Cham-
paign (Structural Analysis-Fundamentals and Analysis of Elastic Trusses and Frames)
Alain H. PeyrotProfessor of Civil Engineering, University of Wisconsin-Madison (Towers
and Transmission Poles)
William N. Poellot, Jr.Project Manager, HDR-Richardson Gordon, Inc., Pittsburgh, PA
(Curved Steel I-Girder Bridges)
William C. SchnobrichProfessor of Civil Engineering, University of Illinois at Urbana-
Champaign (Structural Analysis-Continua: The Finite-Element Method)
Shih-Lung Chu Associate and Head, Structural Analytical Division, Sargent and Lundy,
Engineers, Chicago, IL (Chimneys)
Shu-Jin Fang Associate and Assistant Head, Structural Analytical Division, Sargent and
Lundy, Engineers, Chicago, IL (Chimneys)
Stephen J. Y. TangProfessor Emeritus of Architecture, University of Oregon (Buildings-
General Design Considerations)
Thomas H. Thornburn (deceased) Former Consulting Soils Engineer, Las Vegas, NV (Soil
Exploration)
Ivan M. Viest Consultant, Bethlehem, PA (Design of Composite Members)
Gene M. Wilhoite Transmission Line Consultant, Chattanooga, TN (Towers and Transmis-
sion Pole Structures)
Roman Wolchuk Consulting Engineer,Jersey City, NJ (Steel-Plate-Deck Bridges)
Robert S. Wozniak Senior Engineer, Chicago Bridge and Iron Company, Oak Brook, IL
(Steel Tanks)
Robert C. Y. Young Vice-President, Iffiand Kavanagh Waterbury, P.C., New York, NY
(Arches and Rigid Frames)
 Contributors xix

Max Zar Former Partner and Manager of Structural Department, Sargent and Lundy,
Engineers, Chicago, IL (Chimneys)
Lev Zetlin President, Zetlin/Argo, New York, NY (Suspension Roofs)
Paul Zia Professor of Civil Engineering, North Carolina State University (Design of Pre-
stressed Concrete Structural Members)
 ABOUT THE EDITORS

EDWIN H. GAYLORD, Professor Emeritus of Civil

Engineering with the University of Illinois at
Urbana-Champaign, is a registered structural
engineer in Illinois.
The late CHARLES N. GAYLORD was Professor
Emeritus of Civil Engineering at the University
of Virginia.
 Section 1
Structural Analysis

Part 1. Fundamentals and Analysis of

Elastic Trusses and Frames
DAVID A. PECKNOLD
Professor of Civil Engineering, University of Illinois at Urbana-
Champaign

INTRODUCTION
In order to perform a structural analysis, the structural engineer invariably must idealize,
or model, the actual structure, to a greater or lesser degree, depending on the complexity
of the structure itself, the purpose for which the analysis is being made, and the compu-
tational tools at hand.
Computational tools which are easily available make it feasible to model structures for
analysis so that structural response can be calculated well into the nonlinear range, taking
account of inelastic material behavior and geometry changes under load. Dynamic
response to, for example, earthquake, wind, and wave loadings can be routinely deter-
mined. Such analyses can provide useful information on postelastic stress redistribution,
can identify potential failure mechanisms, and can provide better estimates of safety mar-
gins. In many cases, however, analyses are made under the assumption that the structure
behaves in a linear elastic manner, that the deflections are small, and that the loads are
applied slowly. Even if these assumptions are not all reasonable approximations in a given
case, a linear elastic static analysis is often made as a first step, since such an analysis often
provides useful information in its own right and also provides a context within which the
results of more sophisticated analyses can be evaluated and understood.
Structural modeling for analysis must be done with a clear understanding of the struc-
tural load-carrying mechanisms which are involved and an appreciation for essential
behavior versus unimportant detail; in some cases modeling is relatively routine, while in
others it may require a great deal of skill and experience.
1. Classification of Structures Structures can be classified for analytical purposes in a
number of ways. They can be regarded naturally in many cases as consisting of a number
1-1
1-2 Structural Analysis

of structural components each of which may carry a relatively simple system of forces.
This decomposition into simpler components can be done at several levels; thus a struc-
ture may be regarded as consisting of a number of subassemblages or substructures, each
of which may, in turn, consist of other, simpler substructures; until finally at the lowest
level the basic components are simple structural members or elements. This hierarchical
decomposition is useful, both for computational purposes and for conceptual, descriptive
purposes.
Structural members or elements can be classified in terms of the essential force systems
which they carry and in terms of their shapes and dimensions. One possible classification
scheme is as follows:
I. Line elements: bars, rods, columns, or cables in which the cross-sectional dimen-
sions are small compared to the length of the member, and the force system is axial, either
tension or compression
2. Surface elements (two-dimensional equivalent of line elements): panels, deep
beams, and tension and compression membranes in which the thickness is small compared
to the surface dimensions, and the force system is two-dimensional and lies in the plane
of the midsurface
3. Line elements which carry load by bending: beams or arches
4. Surface flexural elements: plates or shells
5. Hybrid elements: for example, beam columns, which are subject to both axial load
and bending
Structural components which are of general shape and are subjected to three-dimen-
sional stress systems are solids.

FUNDAMENTALS OF ANALYSIS
2. Description of Structural Configuration The state or configuration of the structure may
be defined in terms of the degrees offreedom of the structure. These are defined as a set
of independent coordinates, or pieces of geometric information, minimum in number,
required to uniquely define the shape and orientation of the structure. A rigid body in
space has six degrees of freedom, which can be taken as three cartesian components of
displacement and three rotations. A planar truss has 2j degrees of freedom: the two carte-
sian components of displacement of each joint, where j is the number of joints. A planar
frame has 3j degrees of freedom: two displacements and a rotation for each joint. In each
of these examples, the degree of freedom is a natural, physical choice used to describe
the deformed configuration. However, the degrees of freedom can be chosen in a some-
what more abstract way, which is very useful in some cases. For example, a degree of
freedom may simply be an amplitude or multiplier of a special deflected shape or config-
uration of the structure. A good example in structural dynamics is the set of natural modes
of vibration of the structure. A completely general deflected shape of the structure can
be specified as a linear combination of these mode shapes. The amplitude of each of these
mode shapes in that linear combination is a degree of freedom. A simply supported beam
provides another example. The rotations at the supports at either end of the beam can be
defined as the (two) degrees of freedom for the beam. Alternatively, the deflected shape
can be described in terms of the superposition of two special shapes: a symmetric deflec-
tion shape in which the rotations at the ends of the beam are of equal magnitude but in
opposite rotational senses, and an antisymmetric deflection shape in which the rotations
are of equal magnitude and the same rotational sense. The amount of each of these shapes
which is present in the general shape can be selected as a degree of freedom. In these
latter cases, the independent shapes (the mode shapes, or the symmetric/antisymmetric
beam deflection shapes) play the same role as the cartesian unit vectors do in the case of
the truss joint displacements.
Other Structural Variables. In the case of, for example, a planar truss under load, it is
obvious that if the two cartesian displacement components of every joint are known, then
the position of each joint in the loaded configuration is known. Thus, member lengths in
the deformed configuration can be determined, leading directly to member elongations,
or strains. From these strains and knowledge of the material behavior, the stresses, or
alternatively member axial forces, can be determined. Thus, specification of the values of
the degrees of freedom leads directly to determination of the other structural variables
of interest. In the procedure just described, the degrees of freedom are viewed as the
 Basic Principles 1-3

fundamental variables, and the strains and stresses as derived variables. However, since
unique relationships between these variables exist for elastic structures, it follows that
one may consider, for example, the member forces to be the fundamental variables, and
the joint displacements and strains to be derived variables. The choice of different fun-
damental variables leads to different general metho..J.s of structural analysis. The displace-
ment method and the force method are the two best-known such general methods. Various
hybrid methods have also been developed, in particular, for the finite-element analysis of
plates and shells.
3. Basic Principles In the solution to any structural analysis problem, the structural
variables must satisfy three basic physical laws, which are described here in the context
of a truss:
Equilibrium: The member forces (stresses) and the externally applied joint loads must
satisfy the equations of static (or dynamic) equilibrium.
Compatibility: The member elongations (strains) and the joint displacements must be
geometrically compatible; that is, structural continuity must be preserved.
Constitutive laws: The axial force (stress) and elongation (strain) of each member
must satisfy the behavior laws for the particular material of which the member or struc-
ture is constructed.
For the simple planar truss shown in Fig. 1, these relationships are as follows:
J; + f2 COS a = Px4
(1)
f 2 sin a + h = Py4
e1 = Ux4
e2 = u, 4 cos a + uy4 sin a (2)
e3 = Uy4

(3)

where fk = axial tension

ek = elongation of member k
u,i, uyi = cartesian components of displacement of joint i
Pxi> Pu, = external loads acting on joint i

Uy4,Py4
@]
Member 1 Ux4,Px4

3
i V

[gJ _l
~H~
L1 = H Cross-sectionol area of member k

L2 = JH2+ v2 Modulus of elasticity of member k

L3 = V
Fig. 1 Planar truss.
1-4 Structural Analysis

In the general case of a plane or space truss, Eqs. (1) to (3) can be written in the form
ill=p
e = Cu
w
(5)
f = Ee (6)
in which the degrees of freedom are contained in the displacement vector u, the member
forces (stresses) in f, and the member elongations (strains) in e. If the plane truss has j
joints, b members, and r ~ 3 restrained degrees of freedom, the equilibrium matrix H is
(2j - r X b), the compatibility matrix C is (b X 2j - r), and the matrix E, which contains
member stiffness information, is (b X b). Equations (4) to (6) display the general structure
of the equations which appear throughout structural analysis, from simple trusses to
finite-element models of very complex structures. In the latter case, (usually numerical)
integrations over element volumes are necessary. The force vector Hf is often called the
internal resisting force vector i. It consists simply of the member internal forces resolved
into the directions of the global degrees of freedom.
4. Static Determinacy and Stability Several important special cases can be identified from
Eqs. (4) to (6). If the equilibrium matrix H of a plane truss is square (2j - r = b) and
nonsingular, the truss is statically determinate and statically stable; the member forces f
can be determined from the equations of joint equilibrium, Eq. (4), and the member elon-
gations and joint displacements can be subsequently determined. If 2j - r > b, the truss
is unstable; there are not enough bar forces to satisfy statics for all possible loading con-
ditions. If 2j - r < b, the truss is statically indeterminate. It is possible that a truss is
statically unstable even if 2j - r ~ b. This condition corresponds to a critical geometric
arrangement of members and/or supports which produces instability. 1 Instability, as dis-
cussed here, refers to the possibility of (possibly small) rigid-body motions of the entire
structure, or of substructures within the structure, taking place, without requiring mem-
ber deformations. Such rigid-body modes are called kinematic modes. This type of insta-
bility must be distinguished from elastic instability or buckling.
If the truss is statically indeterminate, the compatibility and stress-strain relations must
be utilized in the solution. In this case, the joint displacements can be expressed as
u = K- 1p (7)
in which the matrix K = HEC is the structure stiffness matrix. The relation HT = C can
be verified for the example shown in Fig. 1 and Eqs. (1) to (3). It is true in general, as
will be shown subsequently. The matrix Eis square, symmetric, and positive definite. As
a result, the structure stiffness matrix K possesses the same properties.
Consequences of Static Indeterminacy. In statically indeterminate structures, particu-
lar systems of internal forces can exist in the absence of external loads. These systems of
internal forces are sometimes called prestress, residual stress, autostress, or selfstress. (It
is possible to stabilize structures which possess kinematic modes by means of prestress.)
The classification of a structure as statically determinate or indeterminate is not, as in the
case of some classification systems in structural analysis, merely for the purposes of anal-
ysis. The behavior of statically indeterminate structures differs fundamentally from that
of determinate structures. Some of these differences in behavior, and their consequences,
can be inferred from Eqs. (4) to (7). For example, in a statically determinate structure
undergoing small displacements, the internal forces are determined directly and uniquely
from the external loads. If the loads are zero, so are the internal forces. No states of resid-
ual stress are possible. Thus, in a statically determinate structure no internal stresses are
generated by, for example, differential temperature changes, differential support move-
ments, or a lack of fit of the structural members (i.e., differences between ideal and actual
member dimensions). Conversely, in an indeterminate structure, all these effects can
result in locked-in stresses, which can be of considerable magnitude. On the other hand,
it is evident from Eqs. (4) to (7) that the relative member stiffnesses influence the internal
force distributions in indeterminate structures. This has consequences in member pro-
portioning for design, which must usually be carried out in an iterative design/analysis
cycle. A second major consequence of this property of indeterminate structures is that,
when inelastic behavior occurs, resulting in localized stiffness changes in the structure, a
redistribution of internal stresses takes place if the structure can tolerate the associated
deformations. Alternative load paths can therefore be developed in indeterminate struc-
tures once significant inelastic action occurs. Determinate structures, in contrast, have no
capacity for significant stress redistribution, and thus do not have any extra safety margin.
 Symmetry 1-5

5. Superposition A fundamental property of linear systems is that solutions can be

superposed. That is, ifloads pi, p 2 produce displacements Ui, u 2 , respectively, then a load
p 1 + p 2 produces a displacement u 1 + u 2 • This can, in fact, be regarded as the definition
of a linear system. Therefore, the principle of superposition applies for small displace-
ments of elastic systems. This principle is used in developing many of the methods of
linear structural analysis and also as a routine calculation technique.
6. Symmetry Symmetry properties can be used as an organizing concept for under-
standing the behavior of particular classes of structures. Additionally, these properties of
structures are routinely used in analysis, in order to reduce the amount of computation
which must be done, especially in the case oflarge structures. Symmetry is often recog-
nized and utilized by the analyst when the external load also possesses some degree of
symmetry. However, although the structure must possess some particular symmetry
properties in order for these methods to be utilized, the external load need satisfy no
particular symmetry conditions.
The simple portal frame shown in Fig. 2a possesses reflection symmetry about a vertical
plane x-x'. The applied loading can be decomposed into two loading conditions, one sym-
metric with respect to the symmetry plane, and one antisymmetric with respect to this
plane.

p IX P/2 1
x P/2 P/2 1
x P/2

m m
I
I

I
+ I
I

tl
I I
X X X

P/2 1x

I
I
Ooly ,ecticol
displacement
allowed
(al

+
n'!r
p/2 Ix

I
Rotutioo
horizontalood
displacement
allowed

I I
I I
I I
x' x'
(bl
Fig. 2 Use of symmetry: (a) Decomposition ofload into symmetric and antisymmetric com-
ponents. (b) Symmetric and antisymmetric structures of reduced size.

One-half of the original structure, the reduced structure, can then be analyzed under
two different sets of boundary conditions at the plane of symmetry, as shown in Fig. 2b,
and the results superposed. Although reflection symmetry is the most commonly occur-
ring and most commonly utilized type of structural symmetry, there are many other kinds
which occur in structures. Rotational, or cyclic, symmetry, for example, describes the
synthesis of a structure by rotation of a substructure or repeating unit about an axis. Even
when advantage is not taken of symmetry in performing the structural analysis, the nec-
essary symmetry in the computed results provides an additional check on the correctness
of the results.

ENERGY PRINCIPLES
Energy principles provide alternative ways of viewing some of the basic results and meth-
ods of structural analysis. 2 They also lead to extremely powerful techniques for obtaining
approximate solutions, for example, in the finite-element analysis of plates and shells and
general two- and three-dimensional solids.
1-6 Structural Analysis

7. Principle of Virtual Work The principle of virtual work states that a body (or mechan-
ical system) is in equilibrium under the action of a system of forces if, and only if, the
virtual work done by the system of forces during an arbitrary virtual displacement is equal
to zero. This principle can be regarded as an alternative, parallel statement of the con-
ditions of equilibrium. It is not restricted to any particular material behavior assumption,
and so it is applicable to inelastic as well as elastic systems.

(al (bl
Fig. 3 Principle of virtual work for a particle. (a) Rigid particle
in two dimensions under a system of forces. (b) Resultant force
and virtual displacement.

The concepts involved in this principle can be illustrated by considering the equilib-
rium of a rigid particle in a plane (Fig. 3). The particle is considered_Jo ~ave ze!:_o dimen-
sions, i.e., it is a point in two-dimensional space. A ~stem of forces F 1 , F 2 , ••• F,. acts on
the particle. The resultant of this system of forces is R = E7_ 1 F,. The virtual displacement
iiu is defined in general as an arbitrary reversible displacement consistent with the con-
straints. The virual displacement is arbitrary in both magnitude and direction. Thus, since
the rigid particle in the plane has two degrees of freedom, there are two independent
pieces of information necessary to specify the virtual displacement. Rather than magni-
tude and direction, these could alternatively be taken as the components iil!:J and iiuy of
iiu with respect to cartesian axes (x, y). The work done by one of the forces F, were it to
move through the displacement iiu would be F, · iiu. Therefore, the virtual work iiW done
by the system of forces is iiW = E7_ 1 (F, · iiu) = R · iiu. The statement of the principle
of virtual work is that the particle is in equilibrium if, and only if, iiW = 0 for an arbitrary
iiu. The latter requirement is essential, as it precludes the special choice in which iiu is
orthogonal to R. Therefore iiW = 0 requires that R = 0 for equilibrium. The virtual work
can also be written in the form iiW = (E,'_ 1 F,,) iiu, = (E,'_ 1 F.,) iiuy, Thus iiW = 0 is
completely equivalent to E7_ 1 F,, = 0 and E7_ 1 F., = 0, the two equations of static equi-
librium in the plane. One equation of equilibrium is generated corresponding to each
degree of freedom of the system. The coefficient of each virtual displacement in the
expression for ii Wis called a generalized force.
A rigid body with finite dimensions, shown in Fig. 4, has three degrees of freedom in
two dimensions, two translations iiu,, iiuy in the plane, and a rotation ii/J, about an axis
normal to the plane through point 0. Thus the components of the virtual displacement

y y
8uyi
F;
xWUxi
8uy ly.
I I
X X
88zt/
8ux

Fig. 4 Rigid-body virtual displacements.

 Principle of Virtual Work 1-7

ou; at the point of application (x;, y;) of one of the forces F; can be expressed in terms of
the three independent quantities oux, ouy, o/Jz as
OUx; = OUx - y, · o/Jz (8)
OUy; = OUy + X; . o/J,
The virtual work is now given by the expression oW = E,'- 1 (F; · ou;), or oW =

(E,'= I Fx;) . oux + (E,'= I Fy,) . OUy + [ L'';= I (-y,Fxi + x,Fy,)] . o/J,, which leads to the three
equations of static equilibrium in the plane: E:'-iFx, = 0, E,'- 1Fy; = 0, and E,'- 1M,, = 0.
The body under consideration need not be rigid for the foregoing relations to apply.
The virtual displacement was simply chosen as if the body were rigid, but this virtual
displacement can just as well be imposed on a deformable body. Thus, imposition of rigid
body virtual displacements on a deformable body generates the overall equations of static
equilibrium.
A deformable body may also be subjected to virtual-displacement patterns which, if
they were actual displacements, would distort the body. Such virtual-displacement pat-
terns provide information about the internal forces. For example, the simple plane truss
shown in Fig. 1 has two degrees of freedom ux 4, uy 4. Let virtual displacements oux 4, ou" 4
be imposed on the structure. These joint, or nodal, virtual displacements define ·a
deflected shape of the entire structure which is a perturbation from the equilibrium posi-
tion, as illustrated in Fig. 5. Such a virtual-displacement pattern would, if it were actually
imposed on the structure, produce length changes in some or all of the structural mem-
bers. Therefore, the internal forces, which would tend to resist such changes, must con-
tribute to the virtual-work expression in this case. It is important to note that the forces
do not change during the virtual deformation of the structure; they are the forces existing
in the equilibrium position.

Equilibrium
position

Fig. 5 Virtual displacement from

equilibrium position.

Since the joints of the truss can be regarded as rigid particles, the foregoing relations
can be directly applied to them. Equations (1) are the equilibrium equations for joint 4.
Thus the virtual work can be expressed as
oW = lPx4 - u; + f2 cos a)] . 0Ux4 + [py4 - /J2 sin a + f3)] . 0Uy4 (9)
This expression can be separated into two components, the first involving the external
loads, called the external virtual work oWext, and the second involving the internal forces,
called the internal virtual work oW,n,•
0We<I = Px4 . oux4 + Py4 . 0Uy4 (10a)
oW,n, = -lf1 . 0Ux4 + fz . (oux4 cos a + 0Uy4 sin a) + f3 . OUy4]
= -lf1 · oe 1 + f 2 • oe 2 + f; · oe3] (lOb)
In the general case, Eqs. (4) to (6), the virtual work can be written as
ow = ouT . (p - Hf) (11)
and the external and internal virtual work can be written as
oWext = OUT. p (12a)
1-8 Structural Analysis

Comparison of the latter two expressions in Eq. (12a) shows that e = Hru. Since e =
Cu from Eq. (5), it follows that C = Hr. The principle of virtual work can therefore be
written in the form
oW,n, = -our · Hf
-o(H7u)7 • f
-oeT • f

Compatibility
i
OUT. p = oeT. f
t
Equilibrium
In Eq. (13), the internal force f and the external load p satisfy the equations of equilib-
rium, Eq. (4). The (virtual) joint displacements ou and the corresponding (virtual) mem-
ber elongations oe satisfy the compatibility equations, Eq. (5). Material behavior descrip-
tions are not involved in this relation, and it therefore applies to inelastic materials as
well. Equation (13) may be regarded as a general reciprocal relation which relates any
compatible set of kinematic variables (ou, oe) and any equilibrated set of static variables (p,
f). For some purposes, it is convenient to take these compatible kinematic variables (ou,
oe) as the actual displacements and elongations, and the equilibrated static variables (p,
f) as a conveniently chosen auxiliary load system. This is the basis for the unit dummy
load method (Art. 12).
8. Principle of Minimum Potential Energy The principle of minimum potential energy
treats the displacements or degrees of freedom of the structure as primary variables. It is
closely related to the displacement or stiffness methods of structural analysis. The (total)
potential energy II is equal to the sum of the (internal) strain energy U and the load poten-
tial, or external potential energy, Q. These quantities are defined as follows.
Strain Energy. The strain energy is the stored energy of deformation. For a one-
dimensional structural member with an internal force vs. elongation relation as shown in
Fig. 6, the strain energy is the work done by the force in producing the current elonga-
tion. Thus, U(e) = le f (e) de, the area under the curve up to the current elongation e.
The gradual change of force with elongation is taken into account in this calculation.
This quantity involves not only the material response but also the cross-sectional area
and length of the member. It is more convenient instead to define a strain-energy density,
i.e., strain energy per unit volume, U0 which involves only the material response and not
the dimensions of the member. Thus, in terms of stress u and strain ,, the strain-energy
density U0 (,) = l' u(E) dE. For a linear elastic material in a state of uniaxial stress U0 (,) =
½u(,), = ½E,2 where Eis the modulus of elasticity. The first-order change in strain energy
due to a change in elongation or strain is denoted oU and is given by the expression oU =
foe as shown in Fig. 6; thus the relation oU = -oW,n, between the first-order change in
strain energy and the internal virtual work.

4- _-e:::::..r,,___ .!.. 82 u = l 8f8e

2 2
@
.:::, 8u = f8e
0
E
Q)

1:
H

Elongation e

Strain energy U
Fig. 6 Strain energy.
 Principle of Minimum Potential Energy 1-9

These relations can be generalized to multiaxial states of stress, resulting in

U =½ i [uxfx + Uyfy + Txyl'xy + · · ·] dV (l4a)

= .!
2
JV
fT. u dV

oU = fvluxOfx + UyOfy + TxyO'Yxy +···I dV (l4b)

= i OfT. u dV

The expression for U given in Eq. (14a) is valid for three-dimensional linear elastic
systems. The expression for oU in Eq. (14b) is general.
Expressions for U and oU for various types of structural members are obtained by insert-
ing in Eq. (14) the appropriate assumptions regarding strain distributions in each partic-
ular case. For example, in conventional beam theory, it is assumed that plane sections
remain plane. The strains fx are therefore constrained to vary linearly through the beam
depth, so fx = fxo - z · Ky - y · Kz in which the x-axis is the longitudinal axis of the beam
and the y- and z-axes are the principal axes of the cross section, fxo is the axial strain at
the beam centroidal axis, and Ky = 1/Ry and K, = l/R. are the beam curvatures for bending
in the x-z and the x-y planes, respectively. This strain distribution describes a beam under
axial force and biaxial bending about its principal axes. In addition, transverse shear
deformations are neglected. Therefore, inserting these constraints in Eq. (14) and decom-
posing the volume integration into an integration over the cross section followed by inte-
gration along the length of the beam, the following expressions result.

(l5a)

(15b)

in which

Nxo = i Ux dA M. =- i yux dA (16)

For the truss-type structural system described by Eqs. (4) to (7), the strain energy is
given by the expression
(17)
This expression for strain energy in terms of the stiffness matrix K is general.
External Potential Energy. For conservative external loadings, a potential energy Q of
the loads can be defined such that n = -ur • p.
Potential Energy. The total potential energy II of the structural system is II = U + n.
It is regarded as a function of the displacements, and therefore of the degrees of freedom,
of the structure. The principle of minimum potential energy provides the criterion for
selecting the actual displacements of the structure (i.e., those which also satisfy equilib-
rium) from the set of all possible displacements which satisfy the constraints (boundary
conditions). It may be stated in the following way: Of all displacements which satisfy the
constraints (boundary conditions on displacements) those which also satisfy equilibrium
make the potential energy an absolute minimum.
The potential energy is

(18)
If a small displacement increment ou is taken from the equilibrium configuration u, the
potential energy changes from II to II + D.II. The change in potential energy is equal to
D.II = ouT(Ku - p) + ½ou~u (19)
= oII + ½0 2II
According to the principle, D.II must be positive since any displacement away from the
equilibrium position results in an increase in potential energy. Therefore, for an equilib-
1-10 Structural Analysis

rium position, oil = 0, which leads to the equilibrium equations of the stiffness method
Ku = p. The additional requirement for stable equilibrium is that 0211 > 0, which provides
analytical conditions for the study of structural stability.
Application of the minimum potential energy approach generates the same equations
as does the virtual-work approach. Both approaches lead to powerful approximation
methods.
9. Principle of Minimum Complementary Potential Energy The principal of minimum com-
plementary energy treats the stresses (internal forces) as primary variables. It is closely
related to the force or flexibility methods of structural analysis. It provides the criterion
for selecting the actual stresses (i.e., those which correspond to compatible strains) from
the set of all possible stresses which satisfy equilibrium and stress boundary conditions.
It may be stated in the following way: Of all stresses (internal forces) which satisfy equi-
librium and stress boundary conditions, those which also satisfy the associated strain com-
patibility conditions make the complementary potential energy an absolute minimum.
10. The Reciprocal Theorem The reciprocal theorem relates two separate loaded config-
urations of a linearly elastic structure. 1 Figure 7 shows two separate loading conditions
acting on the same linearly elastic structure. The displacements are shown in the figure
with subscripts which have the following meaning: d,1 is the displacement at point i due
to the action of load P1. The displacements are the components of the resultant structural
displacements in the directions of the loads at the same points; that is, the respective
force and displacement variables are work-conjugate.

P2
P1

.6.,,id11
--------
d21 ~/~ L"<I,,,--------L.z',
d22

Load case 1 Load case 2

Fig. 7 The reciprocal theorem.

The reciprocal theorem states that P1d12 = P2d21 . It is sometimes stated in terms of unit
loads, i.e., for the special case P 1 = P 2 = 1, where it is known as Maxwell's law. Moments
and rotations can be considered in this relationship as well as forces and displacements.
The terms which appear must be work-conjugate pairs.
The two configurations need not be restricted to single loads. The generalization to
multiple loads is usually called Betti's law. It can be stated in the following form. The
work done by the first loading system acting through the displacements produced by the
second loading is equal to the work done by the second loading system acting through
the displacements produced by the first loading. The main use of these relationships is in
demonstrating symmetry of stiffness and flexibility coefficients and in developing a
method (the Miiller-Breslau principle 2) for constructing influence lines.

ANALYSIS OF STATICALLY DETERMINATE STRUCTURES

Statically determinate models of structures are often useful. For example, a truss model
in which each member carries only axial force can capture the important aspects of struc-
tural behavior in many cases, even though the assumption of frictionless pin joints con-
necting the members is far from accurate. Further, the detailed analysis of members or
of substructures is often carried out as a final (postprocessing) step after the analysis of
the full (statically indeterminate) structure. In this case, the indeterminate structural
analysis provides sufficient information so that the detailed member substructure analysis
can proceed as if the member or substructure were statically determinate.
Forces
11. Plane and Space Trusses Ideal trusses have frictionless pin joints at which the cen-
troidal axes of connecting members meet (no eccentricities), and are subjected to loads
only at these joints. As a result, only axial force, either tension or compression, is carried
 Beams and Frames 1-11

by each member. Of these assumptions, the critical one is that loads are applied only at
the joints. Calculations with both a rigid-jointed frame model and an ideal truss model of
the same structure with loads applied only at the joints generally yield axial forces which
agree closely with each other. Additional internal forces (shears and bending moments)
exist in the frame, but these are due only to compatibility effects and are not required for
equilibrium. The additional stresses due to these bending effects are called secondary
stresses and are often in the range of 15 to 20 percent, or less, of the stresses due to the
axial forces.
Plane and space trusses are described in general by Eqs. (4) to (6). For plane trusses,
the criterion for static determinacy is that b + r = 2j, in which b = number of members,
r = number of independent reaction components, and j = number of joints. For space
trusses, the criterion for static determinacy is that b + r = 3j. The degree of static inde-
terminacy is (b + r) - 2j and (b + r) - 3j, respectively, when these indices are positive.
These criteria determine whether or not the complete set of reactions and internal forces
can be determined from the equations of statics, i.e., they relate to overall static deter-
minacy. It is possible that external reactions can be determined, or that some member
forces can be determined, even if the structure is statically indeterminate overall, as
determined by these criteria.
Certain special truss configurations, called simple trusses and compound trusses, which
are statically determinate overall, do not require a simultaneous solution of the equations
of joint equilibrium, Eq. (4). A simple truss can be defined as one which can be assembled
by starting with a simple triangular-shaped panel and then connecting new joints one by
one, using two new members for each joint. A compound truss consists of simple-truss
substructures. For these two special configurations, the reactions and member forces can
be determined by use of the method ofjoints and the method of sections. These two meth-
ods are also useful as methods for checking results of analyses of indeterminate structures.
Although statics does not provide enough information to uniquely determine member
forces in this case, spot checks should be made to ensure that equilibrium is satisfied by
the computed reactions and member forces.
12. Beams and Frames Rigid frames, or moment frames, consist of members connected
together by joints which are assumed to be capable of transferring moments between
connecting members. In some cases, more complicated structural models may be used,
in which some of the joints are assumed to be frictionless pins or in which there are other
types of force releases. Each such release limits or constrains the internal force system at
a given location in the structure and therefore provides additional information which aug-
ments the equations of statics. These additional equations are sometimes called equations
of condition.
For plane frames, the criterion for static determinacy is that 3b + r = 3j + c, in which
c = number of releases and the remaining symbols are as defined in Art. 11. For space
frames, the criterion for static determinacy is that 6b + r = 6j + c. The degree of static
indeterminacy is (3b + r) - (3j + c) and (6b + r) - (6j + c), respectively, when these
indices are positive. These criteria determine whether or not the complete set of reac-
tions and internal forces can be determined from the equations of statics; i.e., they relate
to overall static determinacy. As with trusses, it is possible that external reactions can be
determined (if r = 3 + c for plane frames or r = 6 + c for space frames) or that some
member forces can be determined, even if the structure is statically indeterminate over-
all, as determined by these criteria.
Reaction Calculations Using Virtual Work. Throughout structural analysis there are
two alternative approaches: (1) direct application of the basic principles (equilibrium,
compatibility, constitutive laws), and (2) a work-energy approach, using virtual work as
the fundamental tool. In the case of statically determinate structures, reactions can be
determined by directly using the equations of static equilibrium. When the structure is
complex, perhaps with a number of releases, freebodies of substructures as well as free-
bodies of the entire structure will have to be considered using the equilibrium approach.
Alternatively, virtual work can be used to determine external reactions by applying suit-
ably chosen rigid-body virtual displacements which violate the constraint corresponding
to the reaction under consideration, i.e., so that the appropriate reaction does some vir-
tual work. For complex, but statically determinate, structures with force releases, the
virtual-work approach is likely to be more convenient than the direct equilibrium
approach.
1-12 Left Running Head (B)

'L
H w

Fig. 8 Frame reaction using virtual

work.

Figure 8 shows a simple portal frame which is pinned at the base and which has a
moment release at one of the beam-to-column connections. In order to calculate the ver-
tical reaction R, a virtual-displacement pattern is chosen in which the right-hand support
is moved vertically. The resulting pattern is such that the angle between column and
girder at the moment release can change, but the angle between column and girder at
the rigid joint must remain a right angle. The virtual work of the reaction is R · liu, while
the virtual work of the applied distributed load is -wH · ½H/L · liu. These are both
contributions to the external virtual work liWe<t• The internal virtual work is zero in this
case, since the virtual-displacement pattern is that of a rigid body. Thus, by setting oW
= liWe,t = 0, the reaction is found to be R = wH2/2L. One of the characteristics of the
virtual-work approach is that it is often possible to avoid calculating auxiliary information
which is not of direct interest.
The analysis of a frame can be considered to consist of the determination of the external
reactions and the internal forces in each member at the joints of the frame. The subse-
quent determination of internal forces (and stresses) in a given member at a given mem-
ber cross section (or at all cross sections) can be considered to be beam analysis. The
relations between load, shear, and bending moment in the beam are determined by exam-
ining a freebody of a differential element of the beam (Fig. 9), which gives
dV
dx = p(x) (20)
dM
dx = V(x) + m(x)
where p(x) = distributed lateral load
m(x) = distributed applied moment
The sign convention is shown in Fig. 9.

p(x)dx
y

tttttttttt
M ( l cJ;,r l) M•dM

V \ . . - dx ---I V+dV

Fig. 9 Load, shear, and bending moment sign convention.

 Beam Deflections 1-13

Equations (20) provide expressions for the slopes of the shear and moment diagrams at
any axial coordinate in the beam. They can be integrated between two arbitrary cross
sections of the beam x = a and x = b to provide the following relations.

V(b) - V(a) = lb p(x) dx + ~ Pc;

(21)
M(b) - M(a) = lb V(x) dx + lb m(x) dx + ~ Mc;
which show that the change in shear between two sections is equal to the resultant exter-
nal load acting on the beam between those two sections; and the change in bending
moment between two sections is equal to the area under the shear diagram plus the resul-
tant external moment acting on the beam between those two sections.
13. Beam Deflections Methods for the calculation of beam deflections are of two gen-
eral types: they are based either on (1) the integration of the differential equation for
beam deflections (using double integration or one of the many analogs for it, such as the
moment-area method or conjugate beam); or, (2) work-energy methods, for example the
dummy unit-load method.
Differential Equation for Beam Deflections. Equations (20) relate load to shear and
shear to bending moment. Two additional differential relations which incorporate strain-
displacement compatibility and the (elastic) stress-strain relation complete the descrip-
tion of beam response. The two equations are
d0 M
ds EI
dy
- = tan(}
dx
where EI =flexural stiffness of beam
(J =angle between x-axis and tangent to beam axis
If rotations of the beam axis are small, 02 « 1, the curvature d0/ds can be approximated
as d0/dx, and the slope dy/dx = tan (J as (J so that these equations can be written
d0 = M(x)
dx EI (22)
dy = 0(x)
dx
Equations (22) are often combined into the single relation
d 2 y = M(x)
dx 2 EI
For a statically determinate beam, the bending moment can be determined by statics, and
then the deflected shape from this equation can be determined by double integration or
a number of alternative equivalent methods.
Table 1 gives some standard beam formulas, showing in most cases maximum values of
shear, moment, slope, and deflection.
Example. For the first case in Table 1, the bending moment M(x) = M0 (1 - x/L). Two successive
integrations of d 2 y/dx 2 = M(x)/EI gives

= Mo (x
2 3
x
EI 2 - + C,x + C2
)
y(x) 6L

Constants C 1 and C2 are found from the boundary conditions y(0) = 0 and y(L) = 0. Then

y(x) = ~;;2 [- (f) + (fr - (fn

2 3

When the boundary conditions are such that the beam is not statically determinate, Eqs. (20) and (22)
must be solved together rather than sequentially.
Classical Analogs for Double Integration. Equations (20) and (22) have a symmetrical
structure which suggests some of the analogs which have been developed for calculating
...
"'

TABLE 1 Formulas for Beams

Structure Shear i~ Moment (J Slope ~ Deflection

t
Simply supported beam

(i_Mo MoL Mol2

AA sA 0.= 3EI Ymox = 0.062 EI
Mo

r----_ s.=---;:- Mo
89=-
M0L
6EI ot x = 0.422L

~ X

iw
A A C sA w M _ WL WL 2 WL3·
s.= T c-4 0.=-89=16EI Ye = 48EI

---------------
a (b S _ Wb
A-L 0=Wob (L+b)
A GEIL
-15._A s..21.. M _ Wab wo 2 b 2
o - L Yo= 3EIL
~ S9=-WL0 08 =- Wob ( L + o)
GEIL

w
1! t t \t t t 11 wL wL 2
Mc= -8- 8 • = -B8 = 24EI
wL 3 5wL 4
s.=2 Ye = 384EI
~

E~w 2 7wL 3 wL 4
SA = WGL Mmox = 0.064 wL 8
• = 3GOEI Ymox = 0.00652 El

~ Se=- w3L ot x = 0.577L 88 = -3GOEI

8wL3
ot x = 0.519L

f-+- X
 TABLE 1 Formulas for Beams ( Continued)
w

A 6~
C b.B wl wL 2 5wL3 wl 4
SA =4 Mc = 12 eA =-es =192EI
~
Ye= 120EI

Fixed beam

A~
tw ~ 0
C w WL WL 3
SA= eA=es=0
~ 2 Mc = s Ye = 192EI
V "--J

Aa a
t b ~0 A
Wb 2
S =-
L3
(3a+b) M
A
=---
Wab 2
L2
Wa 3 b3
eA=es=0
~ Ss = - -
wa 2
(3b+a) Ms=-
Wba 2
L2
Ya = 3EIL3
~
"'l L3

A~ '''''c'''''
.,.,..------... ~O
SA=
wl wL2
MA= Ms=-12 eA=es=0
wL 4
2 Ye = 384EI

V ~
w 3wl wL 2 wL 4

~ ==:x::c:r:a:J1l 0 30
SA= MA= - Ymax = 0.00131 EI
A
20
.....-----..... 7wl
sS = - - Ms=-
wL 2
20
eA =es =0
at x = 0.525L
V ~ 20

A ~~~ 0 wl 5wL 2 0.7wl 4

...... ......----..... SA= 4 M
A
=M
S
=-~-
96
eA = es = o Ye= 384EI

"'
V "J
....... TABLE 1 Formulas for Beams (Continued)
0,
Structure Shear H Moment () Slope ~ Deflection t
Cantilever beam

A~o
""s MoL
-
M L2
0
0 Mo BA=EI YA -- 2EI
I I

At ~ B
WL 2 WL 3
w Me=-WL BA= - 2EI YA=m
~
w
Aiiiliiill~s
wL 2 wL 3 wL 4
Se= - wl Me =--2- BA= - 6EI YA= 8EI

~
A -=r:er:e011l wB
wl wL 2 wL 3 wL 4
Se=-2 Me=-s BA=- 24EI YA= 30EI
~

: ~ B
S __ wl wl 2 wL 3 11wl 4
e- 2 Me= --2- BA=-8EI YA= 120El
~
 TABLE 1 Fonnulas for Beams (Continued)

Propped cantilever
-
VSMo
tS A ~ B
MoL2
Ymax= - -
r---_ SA=- 3Mo
2L
Ma=-~
2
BA= MoL
4EI
27EI

f-- X
---:::::::::::J at X= L
3

A tS
!w - ~ B
Ma=- 3WL Ymox= 0.00932 WL
3
16 EI
~ SA= 5W B - WL2
16 A- 32EI
Mc= 5WL at x-0.447L
~x ":::] 32

a Wb 2
AtS ( b ~B SA= (a +2Ll
2L3
Ma=_Wab (a+E..) Wob 2 wa 2b 3
BA= - - Yo= 12EIL3 ( 3 L+a)
L2 2 4EIL
S _ Wo
a - - - ( 3 L2 -a 2)
. -------------"-I
w
2L3

4
A,! f f f f f f f f f ~ B Ymox = 0.0054 wL
S _ 3wL wL 2 B _ wL3 EI
~ A- -
8
Ma=-a A - 48EI

~x '(J at x =0.422L

Mmox=0.03wL 2 4
ALS-=r::r:r::r:IJI WB Ymox = 0.00239 wL
....-------.... SA= wL
10
at x = 0.447L
wL2
wL3
BA=--
120EI
EI

f-- X "-J Ma=

15
at X - 0.447L

w
Mmox - 0.0423 wl 2
AffID:cn-=- ~ B
SA= 11wL at x=0.329L BA- -wL3
--
Ymox = 0.00305 wL
EI
4

... 40 BOE! at x =0.402L

......' ------------- "'1
~x
Ma=- 7wL2
120
1-18 Structural Analysis

beam deflections. In particular, it should be noted that Eqs. (22) are very similar to the
equations for shear and moment, Eqs. (20), if 0 is thought of as shear, y as a bending
moment, and M/EI as a distributed load in some auxiliary, fictitious beam. This is the basis
for the conjugate beam method. 3 Other well-known methods, such as the moment-area
method 4 also use this analogy. The moment-area method is cast in a geometrical form
which emphasizes visualization of the deflected shape. It is usually stated in the form of
two moment-area principles, which specify procedures for calculating (1) changes t!i.0 in
the beam tangent angle between two beam cross sections, and (2) deflection of the beam
at a given section with respect to the tangent drawn from another section. The methods
used for calculating these quantities in the moment-area method are equivalent to Eqs.
(21) for calculating changes in shear and bending moment in the beam, using the curva-
ture M/EI as the load. These methods can be used to calculate deflections due to general
inelastic strains by substituting the actual curvature, however it is caused, for the term
M/EI.
The Dummy Unit-Load Method. Deflection calculations can also be carried out very
conveniently using work-energy methods. The dummy unit-load method is probably the
most versatile of this class of methods. Alternatives such as Castigliano's second theorem 1
are equivalent to the dummy unit-load method. The dummy unit-load method employs
the principle of virtual work, displayed in Eq. (13) for trusses. The actual strains and
displacements of the structure are selected as the compatible set of kinematic variables.
The equilibrated set of static variables is chosen as an external load of unit magnitude,
the dummy unit-load (placed on the structure so that it is work-conjugate to the desired
displacement), and its associated internal forces. Thus the expression for the external vir-
tual work reduces to our · p = 1 · !!i., where!!,. is the desired displacement. The dummy
unit-load method can then be expressed in the general form

!!. = J>T · u dV (23)

in which the dummy unit load and its associated stresses u satisfy equilibrium, and the
displacement !!,. results from the actual strains <.
The displacement which is calculated is actually a generalized displacement-any type
of deformation response (rotations, relative deflections, etc.) can be calculated by apply-
ing the appropriate work-conjugate dummy unit load. A rotation can be calculated using
the dummy unit load as a concentrated moment. Relative deflections can be calculated
using pairs of dummy unit loads.
A characteristic of the dummy unit-load method is that unwanted intermediate results
need not usually be calculated. On the other hand, a single calculation gives the gener-
alized displacement at only one location in the structure, and if a more complete descrip-
tion of the deflected shape is desired, other approaches may be preferable.
In order to carry out calculations using the dummy unit-load method, the internal vir-
tual-work expression given in general form in Eq. (23) must be made more specific.
Internal-Virtual-Work Expressions. The internal-virtual-work expressions for specific
types of structural members are developed from Eq. (23) by imposing the appropriate
kinematic assumptions. For beams subjected to axial force and biaxial bending, the result
is Eq. (15). If torsional rotations and transverse shear deformations are also included, the
internal virtual work in a single member consists of the contributions

bW; 0, =- l~o (N, 0<,o + My OKy + Mz OKz + Vy O"fy + Vz O'Yz + T bq,) dx (24)

The internal virtual work for the structure is obtained by summing member contributions.
For members of a truss, the axial force Nx is constant along the member length, and
f~-o D<xo dx = be. For a truss, therefore, the internal virtual work for the whole structure
is - ~k Nk bek as indicated in Eq. (13).
The dummy unit-load formula givea by Eq. (23) can therefore be written as

!!. = ~ [ nkek + J~:, (mK + v-y + tq,) dx] (25)

for the special case of planar bending, in which the (constant) axial force n, bending
moment m, shear force v, and torsional moment t are the internal force system associated
 Beam Deflections 1-19

with the dummy unit load; and e, K, 'Y, </) are the actual deformations causing the gener-
alized deflection A. The summation is over all members in the structure.
The deflection formula, Eq. (25), separates the total deflection into distinct compo-
nents; axial length changes, and flexural, shear, and torsional deformations. Furthermore,
the actual deformations need not be caused by loads-they may be due to, for example,
temperature changes in the structure. In the case when the deformations are produced
by loads, the actual deformations are related to the actual internal forces N, M, V, T by
NL M V T
e=- K =- 'Y = - <j)=- (26)
AE £] A,G GJ
The effective shear area A, which appears in Eq. (26) accounts for the nonuniform dis-
tribution of shear stress over the beam cross section. For rectangular cross sections, it is
often taken as A, = A/1.2. For WF sections, a good approximation is A, = Aweb· The
treatment of torsional deformations does not include warping torsion effects.
Example. To calculate the vertical deflection at midspan of the simple, statically determinate truss
shown in Fig. 10, a unit load is placed at point C as shown. For trusses, Eq. (25) reduces to

a= Lk nkek (27)

If the member elongations are due only to member internal forces, then Eq. (27) specializes further
to
"'"' NkLk
a= L., n k - - (28)
k AkEk

The calculations are summarized in Table 2. The member length changes due to the actual load are
shown in the column headed "e." The member forces nk due to the dummy unit load are shown in
the column headed "n."' The quantity nk can be interpreted as the vertical deflection of joint C due
to a unit length change of member k with all other members remaining unchanged in length. Thus if
member AF were to be elongated by, say, 0.1 in (due to any cause), then joint C would deflect verti-
cally by an amount (-0.707)(0.1) = -0.0707 in (the negative sign indicates that the deflection is in

20 K

iG
7

4(q) 10 ft

(al

Member forces nk

(bl
Fig. 10 Truss deflection by dummy unit-load method. (a) Actual load. (b)
Dummy unit load.
1-20 Structural Analysis

TABLE 2 Deflection of Truss of Fig. 10*

Member N, kips L, in A, in 2 e, in n ne, in

AB,DE +30.0 120 1.5 0.080 +0.5 +0.040

BC,CD +40.0 120 2.0 0.080 +1.0 +0.080
FG,GH -30.0 120 2.0 -0.060 -0.5 +0.030
AF,EH -42.4 169.7 2.5 -0.096 -0.707 +0.068
BG,DG -14.1 169.7 1.0 -0.080 -0.707 +0.057
BF,DH +10.0 120 1.0 0.040 +0.5 +0.020
CG + 0.0 120 1.0 0.000 +1.0 +0.000
~ne = +0.59
0
E = 30,000 ksi for all members.

the opposite direction to the unit load, i.e., upward in this case). The vertical component of the deflec-
tion of joint C is calculated as 0.59 in.
Example. The calculation of deflections due to flexure is illustrated by the simple portal frame
shown in Fig. Ila. The horizontal deflection of joint C is sought. Therefore a dummy unit load is
placed at C in the horizontal direction (Fig. llc). If the contributions of axial deformations and shear
deformations to the deflection are disregarded, Eq. (25) becomes

a=L fLk m
( J,-o ~ dx) (29)
k EI
in which M/EI are the member curvatures due to internal bending moments, and m are the bending
moments corresponding to the dummy unit load. The flexural stiffness EI is taken as constant and is
the same for all members. The curvatures are shown in Fig. llb, and the bending moments due to the
dummy unit load in Fig. llc.

r
H w
"' Pin

L
(al

wH 2 Pin
2 EI
/
Parabolic
C
7 H

D _l D

M
Curvatures EI Moments m

(bl (Cl
Fig. 11 Frame deflection by dummy unit-load method.
 TABLE 3 Values of J:L L mM dx
Linear M diagrams Porabol ic M diagrams

~ ~ ~
L
&gin Mo
M,
Ml
I Mo~ ~M,
L
ModM,
L L/2
1

L/2 L L ___j
M,
L L

ml
L
I mML
1
2 mM 0L
,;::,
1
2 mM 1 L
1
2 mUM 0 +M 1 )
2
3 mM 1L
1
3 mM 1 L
3
1
mU2M 0 -M 1 l

1 1 1 1 1 1 1
mor---_ 2 m 0 ML 3m0Mol
..., 6 m 0 M1 L
6 m0 U2M 0 +M 1 )
3 m0 M 1L
12 m0 M1 L 12
m 0 L(5M 0 -M 1 )

1 1 1 1 1 1 1
~m, 2 m 1ML
6 m 1 M0 L 3
m 1 M 1L
6 m 1 L(2M 1+M 0 ) 3 m 1M 1L
4 m 1M 1 L
4 m 1L(M 0 -M 1 )

~ [m 0 (2M 0 +M 1) ~ [ mo(5Mo-M1l
1
1 1 1 1 1
ML(m 0+m 1 ) M0U2m 0 +m 1 ) M1L(m 0 + 2m 1 ) M1Um 0 +m 1 ) M 1L(m 0 +3m 1)
... mollm, 2 6 6
+m 1 (2M 1 +M 0 l]
3 12
+ 3m 1 (M 0 -M1l]
~ L
1-22 Structural Analysis

Values of typical integrals which appear in Eq. (29) are given in Table 3. For the frame of Fig. 11,
they have the values ½,wH4 /EI for AB; wLH3/6EI for BC; and O for CD. Thus the horizontal component
of deflection of joint C arising solely from bending of the frame is Ile = (½,wH4/EI)(l + 0.8L/H).

ANALYSIS OF STATICALLY INDETERMINATE STRUCTURES

The two fundamental methods of structural analysis are the force method and the dis-
placement method. The force method is also sometimes called the compatibility method
or the flexibility method. The displacement method is also sometimes called the equilib-
rium method or the stiffness method.
14. Comparison of Force and Displacement Methods The force method and the displace-
ment method are dual methods. Each step or procedure in one has its mirror-image equiv-
alent in the other. In the force method, equilibrium is satisfied at all stages, and the gov-
erning equations of the method result from the imposition of the conditions of
compatibility. In the displacement method, compatibility is satisfied at all stages, and the
governing equations of the method result from the imposition of the equations of equilib-
rium. The force method is closely related to the principle of minimum complementary
potential energy, and the displacement method is closely related to the principle of min-
imum potential energy.
The displacement method is the basis for the overwhelming majority of production
computer codes for structural analysis, because of a number of practical computational
advantages. Nevertheless, the concepts embodied in the force method are fundamental
to a full understanding of structural theory and analysis.
The three-span continuous beam shown in Fig. 12 illustrates the fundamental concepts
of these methods.
Force Method. Consider a structure which is statically indeterminate to the nth
degree. For the beam shown in Fig. 12a, n = 2. A statically determinate stable structure,

( ( U1 U2

~ ;)
15,
A {} B
fc ¥D
15,
A B C
4}
D
R1 R2

11 II

Ir
.r 2J ~52 ~1
$)
P2

{}
R1= 0 R2 =O U1 =O U2 =O

+ +
~~
~ { {} ~}· f 11 R2=O

R2=1
}·
,,,~u,•1 '"
t)
U2=O
la

+ +
~ C (} '22~}2
k12

~{}
U1= 0
(a) (b)

Fig. 12 Comparison of (a) force and (b) displacement methods.

 Comparison of Force and Displacement Methods 1-23

the primary structure, is produced by introducing sufficient releases. The generalized

forces corresponding (in a work-conjugate sense) to these releases are the redundants.
The primary structure is now subjected to n + 1 loading cases: the first of these is the
external loading condition; each of the remaining n loading cases corresponds to a unit
value of one redundant with all other redundants set equal to zero. The generalized dis-
placements at the releases in each of these loading cases are defined as the flexibility
coefficients. The flexibility coefficent Ji) is the generalized displacement work-conjugate
to redundant R, due to a unit value ot redundant R1 with all other redundants equal to
zero.
A linear combination of then + 1 loading cases is constructed to give expressions for
the total generalized displacement at each release, as indicated in Fig. 12a. For this exam-
ple these two generalized displacements are the vertical displacements at supports C and
D, which must be zero in the indeterminate beam. The equations for this example are

f11R1 + f12R2 + D1 = D? = 0 (30)

f21R1 + f22R2 + D2 = D2 = O
Thus the equations of the force method express compatibility conditions which are sat-
isfied by choosing the appropriate combination (i.e., the values of the redundants R;) of
the n loading cases to superpose on the external load case.
In general, the equations of the force method have the form
F ·R +D= D0 (31)
or F · R = D
where F = symmetric, positive definite structure flexibility matrix
D = D0 - D · D
= generalized displacement of primary structure due to external load
D 0 = specified generalized displacement at releases (often equal to zero)
Displacement Method. Consider a structure which is kinematically indeterminate to
the nth degree. This means that there are n unknown degrees of freedom U,. Kinematic
indeterminacy in the displacement method is the counterpart to static indeterminacy in
the force method. In the beam shown in Fig. 12b, the degree of kinematic indeterminacy
(the number of degrees of freedom) is n = 2. There is, in general, no particular correla-
tion between the degree of indeterminacy of a structure and the number of degrees of
freedom which it has.
The first step in the displacement method is to produce a kinematically determinate
structure by fixing all degrees of freedom. This fixed-end structure is the entity in the
displacement method which is the dual of the primary structure in the force method. It
is usually highly statically indeterminate.
For convenience, external loads are separated into two types of loads: element loads,
which are applied directly to the members, and joint loads which act directly on the
joints. The fixed-end structure is now subjected to n + 1 load cases. The first consists of
the external member loads acting on the fully restrained structure. Each of the remaining
n loading cases corresponds to a unit value of one of the degrees of freedom with all other
degrees of freedom equal to zero. The generalized forces at the degrees of freedom in
each of these loading cases are defined as the stiffness coefficients. The stiffness coefficient
kii is the generalized force work-conjugate to the degree of freedom U, which is due to a
unit value of degree of freedom ½ with all other degrees of freedom equal to zero.
A linear combination of the n + 1 cases is constructed to give expressions for the total
generalized force at each degree of freedom. The element loads give rise to a set of gen-
eralized forces P,, often called the fixed-end forces or fixed-end moments. The general-
ized force at each degree of freedom must be equal to the specified (externally applied)
generalized force-the joint load. In the example shown in Fi!!;. 12b, the two degrees of
freedom are joint rotations at B and C. The equations are
k11 U1 + k12U2 + P1 = P/ = O (32)
k21U1 + k22U2 + P2 =Pi= 0
Thus the equations of the displacement method express equilibrium conditions which are
satisfied by choosing the appropriate combination (i.e., the values of the degrees of free-
dom U,) of the II loading cases to superpose on the external load case.
1-24 Structural Analysis

In general, the equations of the displacement method have the form

K-U+P=P 0 (33)
K ·U =P
where K = symmetric, positive definite structure stiffness matrix
P = P0 - P·P= generalized force in fixed-end structure due to element loads
po = specified (externally applied) generalized force at degrees of free-
dom, i.e., joint loads
Force Method
The steps involved in the force method are (1) selection of redundants; (2) calculation of
generalized displacements in the primary structure for the external loading D1; (3) cal-
culation of the generalized displacements (the flexibility coefficients f,) for the n unit
redundant loading cases; (4) imposition of compatibility conditions work-conjugate to the
redundants; (5) solution of the simultaneous equations for the values of the n redundants
R1; (6) back substitution for member forces and deformations.
Difficulties arise in setting up a procedure for making a standardized, systematic choice
of redundants for general structures. This has led to a strong preference for displacement
methods in production computer codes for structural analysis. Nevertheless, for specific
classes of structures, the force method can be very effective and efficient.
15. Structure Flexibility Coefficients The generalized displacements D1 and f,i of the pri-
mary structure can be calculated using any method for calculating displacements in stat-
ically determinate structures. Both direct methods and work-energy methods are avail-
able. The dummy unit-load method is illustrated in the following article.
16. Dummy Unit-Load Method Examples
Statically Indeterminate Truss. The planar truss shown in Fig. 13 is two degrees stat-
ically indeterminate (r = 3,j = 8, b = 15; b + r - 2j = 2). The redundants R1 and R2
are chosen as the internal bar forces (tension positive) in members CF and CH. Therefore,
releases are inserted in members CF and CH, consisting of infinitesimal cuts. The gener-
alized displacements which are work-conjugate to the redundants are therefore the rel-
ative displacements (i.e., opening or closing) of the cuts. Whenever internal generalized
forces (as opposed to external reactions) are chosen as redundants, the corresponding
generalized displacements are relative quantities of this type. The relative displacements
due to the external loads are D1 and 152 •
In order to calculate the relative displacement D 1 the required dummy unit-load system
is a pair of unit forces which induce bar forces n/. Thus D 1 = ~nle1, where e1 are the
member elongations in the primary structure due to the external actions. If these external
actions consist only of mechanical loads, then e1 = N1L;/A1E1, where N1 are the bar forces
due to the external loads. The same dummy unit-load system is used to calculate the rel-
ative displacementsfll andfi 2, due to the loadings R1 = 1, R2 = 0 and R1 = 0, R2 = 1,
respectively. This dummy unit-load system happens to be the same as the load system for
R1 = 1, R2 = 0. The relative displacements D2 , /2 i, and/2 2 are calculated in an analogous
way using a dummy unit-load system consisting of a pair of unit forces which induce bar
forces nf. This second dummy unit-load system happens to be the same as the load system
for R 1 = 0, R2 = 1. The formulas for the flexibility coefficients shown in Fig. 13 verify
thatfi1 > 0.f22 > 0, andfi2 = f21•
Table 4 summarizes the calculations. The redundants R1 and R2 are found from

0.01732 0.00200] [ R 1 ] 0.1224 ]

[ 0.00200 0.01 732 R2 [ 0.1224

[t] 6.34 ]
[ 6.34

The total member forces in the indeterminate truss are found as a final step from N, =N 1
+ n/R1 + nm2-
Statically Indeterminate Frame. The portal frame shown in Fig. 14 is statically inde-
terminate to the first degree. The horizontal thrust at the base of the columns is chosen
as the redundant R1 (the sign convention chosen is such that a negative sign for R1 indi-
cates a positive thrust). The dummy unit-load system which is required in order to cal-
 Dummy Unit-Load Method Examples 1-25

20 K 20 K

A~ E ] o
6~~
ft

1.. 4 ® 10 ft ..I

Dummy unit loadings

Forces n1 n2
t t t I I

- N l
D = l;n;e1
l i =l

l L; N l l L;
n1-- f = rn;n;--
A1Ei 11 i=l AiEi

/ ~
V .J/~ "-. 2 Li
n·-- f 12
N
= :En1n1- -
l 2 L; N 2 2 L;
f 22 = :En;n;--
~
1
t::. A1E1 i =l A;E; i-1 A;E;
R =0R =1
1 2
Fig. 13 Statically indeterminate truss by force method.

TABLE 4 Statically Indeterminate Truss (Fig. 13) by Force Method*

Mem. L; A; N; n{ nl e; n/e; n're; Ju !12 f22 N;

AB 120 1.5 +30.0 0,080 +30.0
DE 120 1..5 +30.0 0,080 +30.0
BC 120 2.0 +40.0 -0.707 0.080 -0.0566 0.001 +35.5
CD 120 2.0 +40.0 -0.707 0.080 -0.0566 0.001 +35.5
FG 120 2.0 -30.0 -0.707 -0.060 0.0424 0.001 -34.5
CH 120 2.0 -30.0 -0.707 -0.060 0.0424 0.0()] -34.5
AF 169.7 2.5 -42.4 -0.096 -42.4
EH 169.7 2.5 -42.4 -0.096 -42.4
BG 169.7 1.0 -14.1 +!.O -0.080 -0.0800 0.00566 -7.8
DC 169.7 1.0 -14.1 +1.0 -0.080 -0,0800 0.00566 -7.8
BF 120 1.0 +10.0 -0.707 0.040 -0.0283 0.002 +5.5
DH 120 1.0 +10.0 -0.707 0.040 -0.0283 0.002 +5.5
CG 120 1.0 + 0,0 -0.707 -0.707 0.000 0.002 0.002 0,002 -9.0
CF 169.7 1.0 + 0,0 +1.0 0.000 0.00566 +6.3
CH 169.7 1.0 + 0.0 +1.0 0.000 - - - - - - - - 0.00566 +6.3
~= -0.1224 -0.1224 0.01732 0.002 0.01732
=Ii1 = Ii2
"Force units are kips; length units are inches; E = 30,000 ksi.
1-26 Structural Analysis

culate the horizontal displacements 15 1 andf11 at the support Din the primary structure
consists of a unit horizontal force at D. If only flexural deformations of the frame are taken
into account, the horizontal displacement at the support D in the primary structure can
be calculated as indicated in Fig. 14. Table 3 gives formulas for integrals JM1 M2 dx. Using
these results, 15 1 wL3 H/12Ebh, andfi 1 = 2H3/3E,1c + H2L/Ebh- Thus the redundant
force R 1 is
1 wL2 1
----
H 12 1 + 2a/3
in which a = (Ebh/L)/(E,lc!H) is a measure of beam-to-column relative stiffness. The max-
imum positive and maximum negative bending moments in the frame are shown in Fig.
15 as a function of the beam-to-column relative stiffness. Shear and axial deformations
can be taken into account by adding the appropriate terms shown in Eq. (25).
17. Three-Moment Equation for Continuous Beams The three-moment equation is a special
force-method solution for continuous beams in which the bending moments over the inte-
rior supports are chosen as redundants. The general form of the equation relates bending
moments in the continuous beam at three adjacent interior supports. This results in a
standard approach for continuous beams which can be easily applied and which gives
well-conditioned simultaneous equations with a narrow bandwidth (the bandwidth is 3).

r
H le

L A D

Curvatures

Moments

Fig. 14 Statically indeterminate portal frame by force method.

 Three-Moment Equation for Continuous Beams 1-27

0.8
Positive moment

0.6

Negative moment

0.2

0.0
0.0 0.2 0.4 0.6 0.8 1.0
a
1+a
Beam-to-column stiffness ratio
Fig. 15 Effects of proportions on maximum bending moments in hinged-base
portal frame.

The primary structure formed by releasing internal bending moments over all interior
supports consists of a number of adjacent simply supported spans (Fig. 16h). Therefore,
the basic building block in this approach is a simply supported beam, and the generalized
displacements D; and/;1 of the force method are easily constructed from calculated rota-
tions at the end supports. A small core of standard results suffices for many problems of
practical interest (Fig. 17). In this regard the three-moment equation possesses the
advantageous characteristics usually associated with displacement methods.
The example shown in Fig. 16 is a four-span continuous beam with equal span lengths
L and constant stiffness EI in all spans. The external load Pis applied at the center of the
end span AB. The flexibility coefficientsf11 ,f21 ,f31 are the relative rotations at the releases
for the unit-load system R 1 = 1, R2 = 0, R3 = 0 (Fig. 16c). These are found from Fig.
17 as f 11 = L/3EI + L/3EI,f21 = L/6EI,f31 = 0. The two contributions to f 11 are the end
rotations at B of spans AB and BC, respectively, under the pair of unit moments R1 = 1.
If the spans AB and BC have different lengths and elastic properties, thenf11 = (L/3El)AB
+ (L/3El) 8 c, f 21 = (L/6El) 8 c, andf31 = 0. The generalized displacement D 1 of the pri-
mary structure is the rotation at B of span AB, due to the external load P. This is given in
Fig. 17 as PL2/16EI. Additional loads are treated in the same way.
Once the redundants R; are calculated, the bending moment diagram for the continuous
beam is constructed using superposition, as shown in Fig. 16.
Displacement Method
The steps involved in the displacement method are (1) identification of the degrees of
freedom; (2) calculation of generalized forces P; in the fully restrained structure due to
element loads (the fixed-end forces); (3) calculation of generalized forces (the stiffness
coefficients k,1) due to the n unit degree of freedom displacement cases; (4) imposition of
equilibrium conditions work-conjugate to the n degrees of freedom (assembly of stiffness
equations); (5) solution of the simultaneous equations for the values of the n degrees of
freedom U;; (6) back substitution for the member forces and deformations.
1-28 Structural Analysis

The displacement method is generally preferred over the force method as the basis for
general-purpose structural analysis computer codes. This is primarily due to the fact that
the operations associated with the displacement method can be systematically compart-
mentalized so that all structural elements, whether they represent trusses, frames, plates,
shells, or solids, can be handled in the same way and the same procedures and algorithms
can be utilized. Details which are specific to a particular type of structural member can
be confined to a small separate part of the computer code.
18. Structure Stiffness Coefficients The stiffness coefficients k,1 of a structure are deter-
mined by direct summation of stiffness coefficients of individual members. This is illus-
trated by the planar truss shown in Fig. 18. The two degrees of freedom U1, U2 are
defined as the horizontal and vertical displacements, respectively, of joint 4 of the truss.
The stiffness coefficients are determined from the basic relations for the truss shown in
Eqs. (1) to (3). Compatibility relations shown in Eq. (2) determine member elongations
given the imposed joint displacements; constitutive relations (elastic behavior) shown in
Eq. (3) determine member forces from elongations; and equilibrium relations shown in
Eq. (1) determine the stiffness coefficients from the member forces.
The stiffness coefficients are displayed in Fig. 18 in matrix form, in which the separate

,& C ~ A B C D E
(a) (b)

2
2LJ3EI
lJ 6EI
LJ6EI O
2LJ3EI lJ 6EI
J [RI.
R2 ;
[-PL /16Elj
O
[
0 lJ 6EI 2lJ3EI R O
3

(c)
I R
R
1]
R2 ; _fb._ ·

3
448
[- 45~
12 ; [-·.0268PL
- 3

(d)
1004Plj
-. 0067PL

+
0.0268PL
LS~
-0.1004PL
--r===- --==--=lf
-0.0067PL

~
0· 1998PL II
0. 0268PL
!::. ~ «i ~
-0.1004PL -0.0067PL
(e)

Fig. 16 Three-moment equation for continuous beam: (a) 4-span continuous beam, (h) pri-
mary structure, (c) unit-load systems, (d) compatibility equations of force method, (e) construc-
tion of bending-moment diagram.
 Member Stiffness Matrices 1-29

contributions of the three members appear as member stiffness matrices. Member 2 is

oriented at an angle a with respect to the cartesian coordinate system (x, y). Member
stiffness matrices are usually expressed in a local or element coordinate system, which in
this case would have a local x-axis pointing along the member axis. A general coordinate
transformation is then used to describe member stiffness information in the global coor-
dinate system.
19. Member Stiffness Matrices The member stiffness matrix is the basic building block
of the displacement method. Member stiffness information 5 is required for every type of
structural member (model) from which a given structural model is to be synthesized.
Planar Truss. The stiffness matrix shown in Fig. 19 for a general planar truss member
is referenced to four degrees of freedom in order to cover the most general situation. The
four degrees of freedom u 1, ut u~, u 4 are defined with respect to the member axes (the
element coordinate system). The truss member has no stiffness associated with displace-
ments normal to its centerline as long as the displacements are small. The transformation
matrix T expresses the relationship between the displacements in the two coordinate sys-
tems as u' = T · u.
Since u'T · p' = uT · p, the relation between forces in the two coordinate systems fol-
lows. The transformation
u' =T· u (34)
is known as a contragradient transformation. The stiffness matrix of the member in the
global coordinate system is then expressed as
k = TT · k' ·T (35)
For a planar frame the degrees of freedom at a joint include a rotation as well as two
displacements. The transformation matrix T is then a 6 X 6 rather than a 4 X 4 matrix,
which is obtained from the matrix shown in Fig. 19 by inserting rows and columns cor-
responding to degrees of freedom 3 and 6, with 1 on the diagonal and O elsewhere.
Planar Frame. Member response in the displacement method is separated into two
component parts: (1) the response of the fixed-end member (i.e., with member degrees
of freedom constrained) subjected to member loads, and (2) the response of the unloaded
member to the degrees of freedom. The stiffness coefficients are related to this latter part
of the member deformation. Either the direct approach or the work-energy approach can
be used to develop the stiffness information for one-dimensional members (line ele-
ments). For plates, shells, and solids, however, the work-energy approach is
indispensable.

w p
L/2 t L/2

p
a
LS,....,
+ b

------------ , , . . . . ~

Pab (a+2b) L, EI Pab(b+2a)

6EIL 6EIL
Fig. 17 Support rotations in a simply supported beam.
1-30 Structural Analysis

[I]
Member l U = l
l>--------1- -
@J

,,/'"
k21
t- k
11
I
I
3 /
Joint OJ rn I
I

(a)

(b)

cos a sin a]+ AJ EJ . [O ol

sirfa L3 0 1J

(c)
Fig. 18 Structure stiffness coefficients: (a) Unit degrees of freedom and definition of struc-
ture stiffness coefficients, (b) Joint equilibrium. (c) Structure stiffness assembled from mem-
ber contributions,

The stiffness coefficients for a plane beam can be developed using the approach illus-
trated in Fig, 20. The degrees of freedom corresponding to axial displacements are not
shown since the axial and bending responses uncouple from each other. The determina-
tion of the stiffness coefficients involves the solution of a statically indeterminate beam,
which can be carried out by the force method. In order to determine the stiffness coef-
ficients k33 and k43 , the force method gives

[f33 f43] [ k33] [O1]

f43 f44 k43

and for the determination of the stiffness coefficients k34 and k44 ,
 k'll k'12 k'1a k'14 a -1 a
k'21 k'22 k'2a k' 24 AE a a a a
T
Y\\Y _x' k'a1 k'a2 k'
33
k'
34
-1 a
a a a a
a
k'41 k'42 k' k'
43 44
~x

Element coordinate system (x', y')

u'l c a a
Global coordinate system (x, y) u'2 -s c a a
c = cos a, s = sin a u'a a a c
u'4 a a -s c

Fig. 19 Stiffness matrix for planar truss member.

k33

'•t~-------_-_,_~-~-~_i_,

(a) (b)

<I>~ ___!,!!_
2
A 5 GL

Fig, 20 Coefficients for plane beam: (a) stiffness, (b) flexibility.

1-31
1-32 Structural Analysis

In these two sets of simultaneous equations, the flexibility coefficients are assumed to be
known and the stiffness coefficients are sought. The coefficient matrix (the flexibility
matrix) is the same in both cases, and so they can be combined into

or

from which it is found that kbb = fhb1• The flexibility coefficients can be found by using
the dummy unit-load method in which, for example, shear deformations can be included
as well as flexural deformations.
The full 4 X 4 stiffness matrix is found by using equilibrium and the fact that the stiff-
ness matrix must be symmetric. Equilibrium requires that kab = Tab · kbb and kaa = Tab ·
kb,,, where
-1
Tab= [ -L
and symmetry requires that kab = kba· Therefore the 4 X 4 stiffness matrix for the planar
beam is

(36)

Figure 21 shows the explicit form of this stiffness matrix expanded to a 6 X 6 matrix
to include the axial degrees of freedom.
Direct Method for Determining Member Response. The response of one-dimer;1sional
(line) elements can also be determined directly from the differential relations based on
equilibrium, compatibility, and the constitutive law. For a planar beam in which only
flexural deformations are considered, Eqs. (20) and (22) are the starting point. If shear
deformation effects are to be considered, the second differential relation in Eq. (22) is
modified to include the transverse shear strain 'Y, which is assumed to be constant through
the depth of the beam. In addition, the constitutive relation between shear force and
shear strain is needed. The result is
dfJ = M(x)
dx EI
~~ = O(x) - -y(x) (37)

-y(x) = V(x)
AsG
In this theory, sometimes called Timoshenko beam theory, the rotation() of the line which
is originally normal to the centerline differs from the rotation dy/dx of the centerline by
the shear angle -y.
The solutions of Eqs. (20) and (37) give the complete response of the beam. The por-
tion of the response involving the applied load p(x) determines the fixed-end forces. The
portion of the response for p(x) = 0 determines the stiffness coefficients.
Figure 22 shows the solutions ofEqs. (20) and (37) for p(x) = 0 and with nodal degrees
of freedom u 1 = y(O), u 2 = 0(0), u 3 = y(L), u 4 = fJ(L). The internal generalized beam
forces are determined from Eq. (37) as V(x) = AsG-y(x), M(x) = EI dfJ(x)/dx, and the nodal
generalized forces are determined from p 1 = V(O), p 2 = -M(0), p 3 = -V(L), p 4 = M(L).
The coefficients of the generalized coordinates u1, u 2 , u 3 , u 4 in these expressions for p1,
p 2 , p 3 , p 4 are the stiffness coefficients shown in Fig. 21. Axial displacements are included
by expanding the stiffness matrix and inserting the axial stiffness coefficients as in Fig. 21.
The analytical expressions for the beam deformations y(x), fJ(x), -y(x) in terms of the
degrees of freedom involve shape fi:nctions, shown in Eq. (38) for the beam deflection
y(x).

(38)
 Member Fixed-End Forces 1-33

u's· p's
u'4• p'4
u'6, p'6

2 3 4 5 6
, - - - - - - ------ - - - - - - -----,
AE -AE
I L L I
I
'
12EI 6EI - 12EI 6EI
2 I 2 2 I
I L\l+ ti>) L (1 + <!>) L\l+ ti>) L (1 + ti>)
I 6EI -6EI
(4-<!>)EI (2 - ti>) El
3 2
I 2
L (1 + ti>) ~ L (1 + ti>) ~
I
-AE AE
4
I L L I
I
I
- 12EI -6EI 12EI -6EI
5 I 2 3
L (1+ ti>)
2
L (1 + ti>)
I
I L\l+ <!>) L (1 + ti>)
'
I
6EI -6EI
6 I 2
(2-<1>)EI
2
(4- ti>) El J

L ___ L (1 + ti>) ~ L (1 + <!>) ~+.!2_ L_J

- - - L__ - - _L_ - - - ----L...-

Note <1>~~
2
L A,G

Fig. 21 Stiffness matrix for plane-frame element.

Shape functions M for O(x) and N7 for -y(x) are defined in a manner similar to that shown
in Eq. (38), using the information given in Fig. 22.
The shape functions Ny each contain two distinct components, the first proportional to
1/(1 + <J,), which is the flexural deflection, and the second proportional to <J,/(1 + <J,),
which is the shear deflection. The parameter <J, = 12EI/A,GL2 measures the importance
of shear deformation; <J, - 0 recovers the flexural beam theory, and <J, - oo is the case of
a purely shear-deformable beam.
The flexural shape functions for beam deflection, denoted simply as N;, are shown in
Fig. 23. These shape functions are part of a general family of interpolation polynomials
called Hermitian shape Junctions. 6 •7 Hermitian interpolation uses nodal values of displace-
ment derivatives (u 2 , u 4) as well as nodal values of the displacements (ui, u 3) themselves.
These shape functions, together with the Lagrangian shape functions, 6 •1 play an important
role in finite-element analysis. (Lagrangian interpolation uses only nodal function values.)
20. Member Fixed-End Forces The fixed-end member response to element loads, i.e.,
p(x) "i' 0 with u; = 0, is superposed on the unloaded member response to the degrees of
freedom, i.e., p(x) = 0 with u; "i' 0. Therefore, the member generalized forces are
expressed as shown in Eq. (39) in terms of the stiffness coefficients and the fixed-end
forces.
1-34 Structural Analysis

2
U(x) = u 1[ ; /-6~ +6~ )]
1
1 2
+ Lu 2 [ -- ( 1 - 41; + 3~ ) + _<I>_ (1 -
l+cj, l+cj,
1;)]
2
+ U3 [
1
; / o; - o; )]
+ Lu4 [ 1; / - 2~ + ~2) + 1: <I> W]
Ly(x) = u 1[ l : <I> (1)]
+ Lu 2 [-JJ½)]
+ U3 [
1
0]
; cj, (-

+ Lu 4 [~½)]
Note: <I>=~
A 5 Gl.2
Fig. 22 Beam shape functions.

(39)

The member fixed-end forces are found from the solution of a statically indeterminate
structure, the fully restrained beam, subjected to the specified loading. The specified
element loading may consist of mechanical loads (distributed or concentrated forces, and
distributed or concentrated moments); distortions (relative displacements or rotations,
useful for generating influence lines); or inelastic deformations, such as those due to tem-
perature changes.
The solution for member fixed-end forces can be carried out by any of the methods
used to determine the stiffness coefficients. A convenient formulation is obtained using
the reciprocal theorem (Art. 10) considering, in turn, each of the unit displacement cases
shown in Fig. 23, paired with the fixed-end loaded member. This leads to an expression
for member fixed-end forces due to a distributed force p(x).
 Portal Frame by Displacement Method 1-35

[~:]=-LL[~~]
p4 N~
p(x)ax (4o)

Figure 24 summarizes member fixed-end forces in a planar flexural beam (i.e., one in
which there are no shear deformations) for several loading conditions, including two cases
in which unit distortions are applied. In the first of these cases, a unit relative rotation is
imposed, with zero relative displacement. In the second, a unit relative vertical displace-
ment is imposed, with zero relative rotation, i.e., the slope of the beam is the same on
both sides of the section at which the relative displacement is applied. Figure 25 sum-
marizes member fixed-end forces in a planar shear beam (i.e., one in which there are no
flexural deformations) for the same loading conditions. In the general case in which there
are both flexural and shear deformations the fixed-end forces are
- 1 - <I> -
P = 1 + </> • Pflex + l + </> • Pshear (41)

in which</> = A,G/12EIL2 is zero when there is no shear deformation and infinite when
there is no flexural deformation.
21. Portal Frame by Displacement Method The portal frame shown in Fig. 26 is fixed at
the base and carries a horizontal force P at girder level. The unsupported frame (i.e.,
before applying boundary conditions at A and D) has 12 degrees of freedom (Fig. 27b).
The boundary conditions constrain six of these to zero, and there remain six degrees of
freedom (two displacements and a rotation at Band at C) as shown in Fig. 27 c. Often in
analyses of structures of this type, it is assumed that axial length changes in the members
are negligible; this assumption introduces an additional three constraints, one of which is
a relative constraint; i.e., the horizontal displacements at B and at C are equal. For an
analysis in which axial length changes are assumed to be zero, there are three degrees of
freedom as shown in Fig. 27 d.
The structure stiffness coefficients shown in Fig. 28 are determined as the sums of the
appropriate member contributions for each of the imposed unit degrees of freedom. Axial

y(xl

U4
'-l~------•-,1----x
EI,L
Fig. 23 Beam (Hermitian) shape functions.
1-36 Structural Analysis

p P1 P2 P3 P4

ca

P2
t !
p
L
b
~) P4
Pbba + b)
L3 7
Pab
2
Pa2<a+3b)
L3
Pa2b
-7

cf ! f)
P2
L/2 L/2

i5 4
.E.
2
fl,_
8
.E.
2
PL
-8

(/fl 11 ! f_) l wl l l l
2
wl wL
12
2
2
wl wL
-72
2

P2 P4

d~:eqi)
P2 P4
3wl
20
wL
30
2
7wl
20
wL
-20
2

Ct~' f) -~~-b
P4
-6EI (b-a) -2El(a-2b) .2¼b-a) ~2a-b)
L3 L2 L3 L2
P2

c/P2
,
...... .......
12EI
7 ?
6EI -12EI
7
6EI
7
a b

- l - cp -
P = l + cp • P flex + l + cp • P shear
Fig. 24 Fixed-end member forces for planar flexural beam.

deformations of the members are taken into account, but shear deformations are
disregarded.
The joint equilibrium equations for the portal frame are shown in Fig. 29. If axial defor-
mations are assumed to be zero, then the joint equilibrium equations are as shown in
Fig. 30.
22. Slope-Deflection Equations The slope-deflection equations are a set of stiffness rela-
tions for a planar flexural beam. The expression for the end moment p 2 , Eq. (39), is p 2 =
(4EI/L)u 2 + (2EI/L)u4 + (6EI/L2)(u3 - u 1) + p2 • In the usual notation associated with
the slope-deflection method, p 2 = Mab, u 2 = IJ., u 4 = /Jb, and the chord rotation ¥lab is
defined as Vlab = (u3 - u 1)/L. The fixed-end moment p2 is usually written as M!b- With
these changes in notation, the generic slope-deflection equation becomes
2EI ( )
Mab =L 2/Ja + /Jb - 3,/;ab + MabF
Expressions for the end shears are usually not written explicitly in terms of the degrees
of freedom; they are subsequently found from equilibrium. Axial deformations are
 Slope-Deflection Equations 1-37

p P1 P2 P3 P4

f
(1/~
j5 2
a ! L
b
f/,) P4
Pb
L
J_Pab
2 L
Pa
L
1 Pab
-2L
p

ct
P2
L/2 ! L/2
~) P4
p
2
PL
8
p
2
_ £b_
8

i5 1 w i5 3
2 2
(~ l l l l l l l l l ~) wl wL wl wL
T 12 T -72
P2 P4

2 2
(~ 6
wl wL
24 3
wL wL
-24
P2 P4

c,,f~ ___.~'
P2
, ,_ ---
a-b
r) j5 4
0 0 0 0

('

P2
t~-
~
f) w_l . . _ ',, __

b j5 4
0 0 0 0

- 1 - <I> -
P = 1+ <I> • P flex + 1 + <I> • P shear
Fig. 25 Fixed-end member forces for planar shear beam.

Fig. 26 Fixed-base portal frame.

1-38 Structural Analysis

(bl

,..:;...;..._ _ _Us
_.T._U4

( C) ( d)

Fig. 27 Degrees of freedom in portal frame: (a) member, (b) unsupported frame, (c)
supported frame, (cl) supported frame with axial length changes constrained.

6Eic 0
0 0~
H2 12Eic AE
- - + -0-
I H3 L
I I
I I
I

1_2_EI_g + _Ac_E 12Eiq + AcE

~L
3
_sy_114t _L-30
Tr-: I
0

',
H
6Elg
,2
_ _ _ _ i.;;
-12Eio
~
0 6Elg
~
~
-12Eio

0
rJ:=A-=-~-~----,1
~i.;; -r H

-6Elg
2Eig
o L
l
I

Fig. 28 Structural stiffness coefficients for portal frame.

 Moment Distribution 1-39

4 6
,---
ll2Elc + ~
--- ----
6Elc
-------
-AgE
---, p
I u,
1 HJ L 7 L

l2Elg ACE 6Elg - l2Elg 6Elg

I --+-
LJ H y -LJ- y I u2 0
I
I 6Elg 4Elg 4Elc -6Elg 2Elg
-+-
I y L H
T -L- 0
I
-AgE l2Elc AgE 6Elc
I -L- --+-
I u, 0
I
HJ L 7
- l2Elg -6Elg l2Elg AcE -6Elg
I --+- I 0
I
-LJ- -L2- LJ H T
I 6Elg 2Elg -6Elg 4Elg + 4Elc I
6
L ___ - r_ - ----
-L-
._ ___ _t _ _ :_ _H_J
0

Fig. 29 Joint-equilibrium equations for portal frame.

2 3
--- --7
r 24Elc 6Elc 6Elc p
I - H3 -:-r I Ul
7 H

6Elc 4Elg 4Elc 2Elg

--+-- -L- u2 0
7 L H
2Elg 4Elg 4Elc
I~ -L- --L- + H 1 U3 0
L.i _ _ _ _ _ _ _ _J

3
U ; ~ - l+l.5a
1 6Elc l+ 6a

M ; PH. l + 3a M ; PH.~
base 2 1+6a top 2 1+6a

Note a ; Elg/L
Elc/H
Fig. 30 Portal frame with zero axial deformations.

ignored, and joint moment equilibrium equations are written. When joint translations
occur (sway), the corresponding equilibrium equations are found by first expressing joint
forces in terms of member end moments.
23. Moment Distribution The moment-distribution method8 is a numerical solution tech-
nique which can be employed to quickly obtain solutions for certain classes of framed
structures. It is most useful as a hand-calculation method for small structures, particularly
continuous beams, in which joint translations (sway) do not occur. The method proceeds
by first calculating the fixed-end moments in the fully restrained structure. Then one joint
at a time is "unlocked," i.e., allowed to rotate to an equilibrium position, with all other
joints restrained. This unlocking of a single joint produces a redistribution of bending
moments and shears in members which connect to that joint. This is termed balancing
the joint. The balanced joint is then relocked in its new equilibrium position, and the
process is repeated for a new joint. Each cycle of joint balancing improves the satisfaction
of equilibrium in the structure, until the process is terminated. The procedure is approx-
imate only to the extent that it is usually terminated when there is still some small equi-
librium imbalance. This is one of the attractive features of the method; rough results can
1-40 Structural Analysls

be obtained very quickly, and if more accurate results are desired, additional cycles of
joint balancing can be carried out.
The effect of balancing a single joint is calculated using distribution factors and carry-
over factors which are determined as shown in Fig. 31. The equilibrium imbalance, say
- M., at a generic joint i is removed by applying a concentrated joint moment M, to the
joint. This produces end moments in each member which is connected to joint i. The sum
of these member end moments is M,, and the fraction of the imbalance which is trans-
ferred to each member is given by its distribution factor; the distribution factor for a
member at a joint is the ratio of its rotational stiffness to the total rotational stiffness of
the joint. For prismatic members, this member rotational stiffness is 4EI/L. Since the far
end of each member connected to joint i is fixed against rotation while the balancing is
taking place, a moment is induced there as well. This moment is calculated as a carryover
factor times the member end moment at the balanced joint. For prismatic members, the

81
......::-:.=-:..:-=--=----v/) M cl
'A---===---ir--7k-+-..._'_
C

Balancing joint

Distribution factors

Carryover factors
2El 1a
M al =---
L
0·1
la

2El 1a
-L-
Mal = ~~ . Mia or Mal= COFla, Mia
__
al

Lal
Fig. 31 Distribution and carryover factors for moment distribution.
 Matrix Formulation of the Displacement Method 1-41

3kips/ft i20 kips i 30 kips

,.>-:f l l l l l l lJs 2S 6
·'1
I--- 20' •
I:= -I- =t=
10' 10' 10' -+5'-J
20' ---i------ 15' --J

[Q[]
!FEMI ft-kip 100 -100 50 -50 33.3 - 66.7

-3.6
---- 33.3- 66.7
-7.1 -9.5 ----4.8
13.4 26.8 26.8
-2.9
- 13.4
-5.7 -7.7 ----3.9

- - 1.0
---- 4.4-
- 1.9 -2.5 ---8.7
- 1.3
1.0 1.9 2.0
-0.2
- 1.0
-0.4 -0.6 ----0.3

0.1
- 0.2
-0.2
0.2
- 0.1
0.8-

0.0 -0.1
1.6
-0.3 -0.5 ----o.3

0.0
0.2- 0.3
-0.1 -0.1

114.5 -71.1 71.1 -51.0 51.0 0.0

83.0

Bending
moment
ft-kips

-114.5 -51.0
Fig. 32 Moment distribution for continuous beam.

carryover factor is 0.5. An example of the use of moment distribution for a continuous
beam is shown in Fig. 32.
When joint translation (sway) occurs, it is generally necessary to carry out a separate
distribution for each sway degree of freedom and then to combine these cases appropri-
ately to satisfy equilibrium. This involves the solution of a set of simultaneous equations,
equal in number to the number of sway degrees of freedom. The method loses much of
its appeal in these cases. However, special procedures have been developed for multi-
story frames, in which the solution of simultaneous equations is not necessary; the sway
is instead handled by a numerical shear balancing procedure. 8
Various refinements of the basic moment-distribution procedure have been developed:
A form of stiffness condensation can be used when the bending moment in a member at
a joint is known; member rotational stiffnesses can be modified to recognize symmetry or
antisymmetry. 8
24. Matrix Formulation of the Displacement Method The displacement method can be for-
mulated so that a general set of procedures applies in common to all types of struc-
tures. 9·10 The details of element formulation are confined to the lowest level of the pro-
1-42 Structural Analysis

cedure in which element stiffnesses and fixed-end forces (or element equivalent nodal
loads) are computed, and element stresses are back-calculated. Transformations of ele-
ment stiffnesses and loads from the element coordinate system to the global coordinate
system; assembly of the structure stiffness matrix and structure load vector; and solution
of the system of joint equilibrium equations are core algorithms used for all structures.
The element nodal displacements u;,) in the local element coordinate system are trans-
formed to the global structure coordinate system by the transformation matrix T(,) accord-
ing to
(42a)
The transformation matrix for a plane truss member is shown in Fig. 19.
The relation between the n X 1 nodal displacement vector of an element and the N X
1 structure nodal displacement vector is
U(e) = L(,) • U (42b)
in which then X N localizing matrix L(,) contains only ls and Os and is very sparse. This
set of matrices, one for each element, contains the element connectivity information
needed to assemble the structure stiffness matrix and load vector. In practice, this infor-
mation is stored for each element in a compact form, the n X 1 destination vector, rather
than in the sparse localizing matrix. The destination vector is a list containing the struc-
ture degree of freedom corresponding to each element degree of freedom.
A virtual-work formulation is preferred since it can be carried through for all types of
structural elements. The virtual work of an assembly of elements (a structure or substruc-
ture) is the sum of the virtual-work contributions of the individual elements. Thus

liW= I:ow(,)
/iWint = t e
/iW(,)int (43)

/iW.,, =L /iW(e)e,t

The internal virtual work of an element is expressed as

liW(,)int =- J oi[.) ·
V(e)
u(e) dV(,J (44)

in which E(,) is the generalized strain vector for the element and "(,) is the generalized
stress vector, Eq. (14).
The element displacements are interpolated from the element nodal displacements
using shape functions (Art. 19), which are usually polynomials, in the form
u(x) = N · u;,) (45)
in which u(x) stands for any relevant displacement component in the interior of the ele-
ment. In some cases there may be more than a single displacement quantity which is
interpolated; for example in a two-dimensional continuum, displacements u(x) and v(x)
are needed to describe deformation. The interpolation functions for a planar beam are
shown in Fig. 23, in which the beam displacement y(x) normal to its centerline is inter-
polated in terms of the four generalized nodal displacements ui, u 2 , u3 , u 4 • The interpo-
lation functions for one-dimensional line elements are usually chosen as the exact solu-
tions for the unloaded member subjected to imposed nodal degrees of freedom. The
cubic polynomials shown in Fig. 23 are of this type; they are exact solutions for an
unloaded beam. Similarly, for a truss element the shape functions are taken as linear poly-
nomials N1 = 1 - x/L, N2 = x/L. For this class of elements the direct approach yields
the same result as virtual work. However, for plates, shells, and two- or three-dimensional
solids the direct approach is not feasible; exact solutions are not available for element
deformed shapes under imposed unit nodal degrees of freedom. It is for this latter class
of structural elements that the virtual-work approach is indispensable. Now, Eq. (45) rep-
resents an assumed approximate element displacement field. The strain-displacement rela-
 Matrix Formulation of the Displacement Method 1-43

tions for the particular structural element are then used to obtain the element generalized
strains as
(46)
Equations (45) and (46) represent the key step in the finite-element method. The ele-
ment shape functions in Eq. (45) cannot be chosen arbitrarily; they must satisfy certain
requirements in order to provide reasonable results (Art. 29).
The internal virtual work for the element can now be expressed as

oW(,);n, = -ou,~ · f er,, · u<,J dV<,)

Jv(e) (47)
or oW(,);n, = -ou,~ · It,) = -our,, · I(,)

in which the element internal resisting force vector I(,) is the nodal force vector which is
equivalent, in a work-energy sense, to the element stresses u(,)· The element internal
resisting force vector is defined as

I<•l = T(e) · f
Jv(e)
er,, · u(,) dV<•l (48)

The element external virtual work can be expressed as

oW(e)ext = ou(~ · ( -ii(e)) = OU(e) · ( -ii(e)) (49)
in which iit,) are the element fixed-end forces, and -ii;,) are the element equivalent nodal
loads. The equivalent nodal loads are the fixed-end forces with reversed sign, since the
equivalent nodal loads are simply the work equivalents of the applied element loads, and
the fixed-end forces are the reactions to these applied element loads.
The virtual work for the entire structure can now be assembled from element contri-
butions as

oW;nt = L oW(,);nt = - oUT. L L(e). I(,) = -oUT. I

oWext = t oW(,)exl = oUT . [P0 ~ L L(e) . (-ii,.,)] = oUT . p
(50)

in which

I =L L(e) ' l(e)

(e)

=L L(e) · T(e) ' f er,) · O'(e) dV(,) (51)

(e) Jv(e)
P = P0 + L L(e) · T(e) · (-ii;,,)
and the applied joint load vector is P0 •
The principle of virtual work now yields
oW = - oUT · (I - P) = 0 (52)
from which the basic set of equilibrium equations of the displacement method is obtained
as
I= p (53)
Equation (53), in which the internal resisting force vector I and the load vector P are
defined in terms of element contributions by Eq. (51), depends only on equilibrium and
compatibility relations. It is valid for any type of material behavior law.
If the element material behavior is elastic, element stresses and strains can be related
by a constitutive relation
I\
(54)
1-44 Structural Analysis

and the element internal virtual work can be written as

/JW(,)int = (55)

where

(56)

is the element stiffness matrix in element local coordinates. The internal resisting force
vector becomes
I= K · U (57)
where the structure stiffness matrix is

K =L LT,) • k(e) • L(e) (58)

(e)

and the element stiffness matrix in global coordinates is

k<,) = T[.) · kc,) · T<,) (59)
Equation (58) displays the structural stiffness assembly process. It is carried out very
efficiently in practice using destination vectors, instead of the sparse matrix multiplica-
tions shown in Eq. (58). A similar technique is used for load vector assembly, or internal
resisting force vector assembly in nonlinear analysis, instead of the matrix multiplications
shown in Eq. (50).
25. Frame Element Stiffness Properties and Equivalent Loads Using Virtual Work Derivation
of element properties by the virtual-work approach, even for line elements where alter-
native procedures are available, provides a unified basis for the displacement method.
The stiffness formulation using virtual work described in Art. 24 is applied to a plane
flexural beam element in this article.
The beam displacement y(x) is interpolated using the cubic Hermitian polynomials
shown in Fig. 23.

(60)

Equation (24) gives the expression for the internal virtual work of the planar flexural
beam as

IJW(e)int =- (L MOK dx (61)

Jx-0
The beam curvature is the generalized strain and is determined from

' - -,w - -[Nl, Nl, Nl, Nll [ ::

U4
l (62)

in which ( )H denotes d 2 ( )/dx2 • The nodal displacement-to-element strain transformation

B(,) is

1
B(,) =- L2 [(-6 + 12~) L(-4 + 6~) (6 - 12~) L(-2 + 6~)] (63)

The elastic constitutive matrix E(,) = EI, a scalar. Therefore, the element stiffness
matrix is given by

(64)
 Frame Element Stiffness Properties and Equivalent Loads Using Virtual Work 1-45

The stiffness matrix given by Eq. (64) is the same as that shown in Fig. 21 for the four
flexural degrees of freedom, with the shear parameter </> set equal to zero.
The element equivalent load vector corresponding to a distributed load p(x) is found
from

-p(,) r1 [ ~N:3
= L J,=o
N4
l p(~) d~ (65)

The load vector P(,) given by Eq. (65) is shown for several cases in Fig. 24.
 Part 2. Continua: The Finite-Element
Method
WILLIAM C. SCHNOBRICH
Professor of Civil Engineering, University of Illinois at Urbana-
Champaign

FINITE-ELEMENT METHOD
In the finite-element method the actual structure or structural component is conceived
of as being subdivided into a number of subregions. These subregions can be in the form
of line elements used to model frame members, through planar two-dimensional trian-
gular and/or quadrilateral shapes to three-dimensional solid subdivisions. The method
gets its name from the fact that subregions are called finite elements. The equations that
apply to these elements are the equations appropriate to the particular theory applicable
to the problem at hand, i.e., plane stress, a plate theory, a shell theory. Next, the dis-
placement response of each of these subregions is assumed to be prescribed within a
selected shape defined by the magnitudes of the displacements at selected points
throughout the element. These displacement shapes are described through the summa-
tion of interpolation functions multiplied by the magnitudes of the displacements of the
selected points located mostly around the perimeter of the element. The points where
the displacements are defined are called node points. The interfaces between adjacent
elements are called nodal lines in two-dimensional problems (Fig. 33) and nodal planes
or nodal surfaces in three-dimensional problems. At the nodes, degrees of freedom
(which are usually in the form of the nodal displacements and/or their derivatives,
stresses, or combinations of these) are assigned. Displacements are the most commonly
used nodal variables, with most general-purpose programs limiting their nodal degrees of
freedom to just displacements. The magnitudes of the nodal variables are then established
based on the satisfaction of virtual-work principles applied to the assemblage of elements
which are considered to make up the structure.
Models which use displacements are logically called "displacement models." Some
models use stresses defined at the nodal points as unknowns. Models based on stresses
are called force or equilibrium models, while those based on combinations of both dis-
placements and stresses are termed mixed models or hybrid models depending upon the
variational procedures used and the quantities defined at the perimeter nodes. Some
advantages of hybrid models for plate and shell problems have been cited, but few gen-
eral-purpose programs have opted to provide this capability, PAFEC being one of the
exceptions. Consequently the displacement model will be the only model discussed in
this section.
With displacement models the form of the displacements within the elements is
selected based on the number of nodes in the element. These displacements prescribe
the internal work which through virtual work defines a limited set of equations for the
structure. Their solution satisfies the displacement compatibility requirements set down
for the structure but only approximates the satisfaction of equilibrium conditions for that
structure; that is, the stresses calculated to exist within the element as a consequence of
1-46
 Requirements for Finite-Element Analysis 1-47

(a) z {b)

Fig. 33 (a) Angle weld to channel used in association with lift mechanism. (b) Finite-element
model of angle.

its deformations may not satisfy either local or global equilibrium. The strains and con-
sequently the stresses and their variations are denned on the basis of the selected assumed
displacement shapes. Stresses cannot take on a variation outside that associated with the
assumed displacements. Some gage of the quality of the solution can be established by
checking the order of violation of equilibrium of freebodies cut from the structure.
Each element within a finite-element model can have its own shape, thickness, material
properties, and even external load effects, so that it is possible to handle irregular geom-
etries and complex or isolated support conditions, as well as a variety of external load
effects. This provides the system with the versatility to treat very complex problems.
26. Requirements for Finite-Element Analysis There are four aspects to the establishment
of a sound finite-element analysis.
1. The real structure must be converted into a mathematical model. The analyst
should be familiar with the errors associated with this aspect. The model should entail
the proper structural mechanics theory. For example, when using plane-stress assump-
tions, the real problem must be suitable to that choice, i.e., suitable to a planar stress field
with the stresses through the thickness directions being negligible. When investigating a
slab with eccentric support members, a shell theory or a theory which contains combined
direct stress and bending must be used to model the slab, etc.
2. The actual input must correctly describe the model as conceived by the analyst.
This phase involves significant amounts of data, particularly node point locations and con-
nectivity. Input errors in these data can easily occur. Graphical display of the structure
and its components should be viewed for correctness before any analysis is performed.
With the availability of commercial graphical preprocessors to produce the element
model, the probability of input error is markedly reduced. Their use is recommended.
3. The appropriate solution techniques and procedures must be utilized. If the ana-
lyst uses a commercial program for a linear problem, solution techniques are usually easy
to implement. Some examples where difficulties might exist are, for example, seeking a
solution for a problem involving a nearly incompressible material without using special
procedures such as penalty functions, or having degrees of freedom which have no stiff-
ness (or nearly a zero stiffness) and for which no constraints are specified. This can happen
in assembling dissimilar types of elements such as a beam element connected with a
plane-stress element (Fig. 34). In such cases it is necessary to write constraint equations
1-48 Structural Analysis

Coupling 35
plane Beam

f
beams
stress element
14--e lements
Shear 30
wall
Lse earn 25

- ::7 830=
U35 - u25
2h

,
(al ( bl (cl
Fig. 34 Assembly of dissimilar elements: (a) problem, (b) finite-element model, (c) constraint
equations.

which relate the rotation at the node to the planar displacements of adjacent nodes or to
incorporate rotation around the normal as a nodal degree of freedom. However, com-
mercial programs do not have such a special element.
4. The final phase is the interpretation phase. The output may be displacements and
stresses at node points. Displacements are normally given in the global reference frame,
while stresses are frequently given in a local reference system. The stresses for the various
elements meeting at a particular node are in general not equal. For low-order elements,
such as the constant-strain triangle and the plane-stress rectangle (Table 5), this stress
disparity can be quite large, in some cases even involving a difference in sign. The analyst
must therefore use some sort of averaging procedure to decide on what stress exists at
that particular location. Postprocessors which include the ability to plot stress contours,
such as PATRAN, make the job a lot easier. The user should realize that all postprocessors
use some sort of averaging procedure or other interpretation to get to a common value
of stress.
27. Discretization of the Structure This step, wherein the physical space which makes up
the structure is divided into the chosen finite elements that are to approximate that struc-
ture's behavior, is critical in establishing the accuracy that will be achieved in the anal-
ysis. The choice of the type, number, shape, and size of the elements is a matter of judg-
ment and past experience with particular element types. Some guidelines regarding these
selections can be found in Art. 28. Needless to say, some understanding of the problem
under investigation is crucial to the ability to establish the best element to use to achieve
a correct and efficient solution.
Normally the basic nodes of an element are at the corners of that element. However,
elements defined using only corner nodes are often far too stiff, so additional degrees of
freedom are added by using side nodes, usually at midside, and/or in some cases interior
(internal) nodes (Fig. 35). For plane-stress and plane-strain problems the 8-node quad-
rilateral element is often the best choice. Additional degrees of freedom can also be
added by specifying higher-order displacement derivatives at the corner nodes. This has
a two-fold benefit because displacement derivatives are the basic ingredients for stress
evaluation, their use as nodal unknowns means a more accurate stress determination, and
also the bandwidth of the equations is normally narrower than for the case of elements
which use just the standard displacements. On the negative side is the need to specify
what these derivatives might be for the boundary nodes. A more severe shortcoming is
the need for the program to be able to handle variable numbers of degrees of freedom
per node depending on the element being used. Most commercial programs will use only
engineering displacements, i.e., translational displacements and rotations as nodal
degrees of freedom.
28. Guidelines for the Selection of Grid The first step in producing a finite-element solu-
tion is the selection of the element type and shape that are to be used; then the network
or layout of these elements that covers the plan or volume of structure is chosen. Prepro-
 Guidelines for the Selection of Grid 1-49

TABLES Plane-Stress Elements

Nodal DOFper Polynomial Name and

Element parameters element degree reference Remarks
u,v 6 Constant- Not a good element to

~
strain use when any significant
triangle strain gradient exists.
(CST) Directional, consider
13 stresses at centroid
u,v 12 2 Linear-strain Significantly better than

~
triangle CST. Stresses calculated
(LST) at node points are much
14 better estimates

18 plus 2 15 Uses nodal displacement

~
U, U1x, U1y 3
V, V}x, V}y internal derivatives, not
standard engineering
displacment. Not
present in most general-
purpose programs
u,v 8 Bilinear Plane-stress Much better than CST.

□
rectangle Accommodates flexure
(PSR) through shear. Standard
13 element of most
programs
u,v 8 Bilinear in 4-node Lowest-order

D
natural quadrilateral isoparametric element.
coordinates isoparametric Yields PSR element
7 when rectangular.
Decline in accuracy
with shape distortion
u,v 16 Quadratic 8-node Good standard element

]j.
in natural quadrilateral to apply to most
coordinates isoparametric problems. Numerically
7 integrated. Reduced
order of integration
should be considered
u,v 8 plus 2 Quadrilateral Not as good as PSR or

~
internal from 4 CST 4-node quadrilateral.
elements Interior node is
15 condensed. Same results
as 4 CSTs

cessors such as PATRAN provide a graphical generating capability that allows the analyst
to produce a finite-element model much more rapidly and at the same time relatively free
of the input errors that proved to be so costly when the operator had to input all node
and element data by hand. The necessary input data is quite substantial, so errors were
not uncommon in analyst-generated data. Figure 36 is indicative of the sort of complexity
that can be readily dealt with using commercially available software. The following guide-
lines on the establishment of a grid have evolved from experience with applying the
method. However, these guidelines are not subscribed to by all analysts. Some prefer to
use a very high density of simple triangular or 4-node quadrilateral elements for all
problems.
Element Shape. Quadrilateral elements should be used where at all possible. Trian-
gular elements should only be used where necessary to refine the grid or as necessitated
by the geometry of the external boundary of the structure. Rectangular elements accom-
modate sharp stress gradients better than corresponding triangular elements. In regions
of primary concern, elements should be as uniform as possible, and square if possible.
The quality of the solution is best for elements which have dimensions of the same order,
1-50 Structural Analysis

3 r

Corner nodes 1, 2, 3 ,4
y,v

Lx,u
D -x
Corner nodes 1,2,3,4
2

Displacements at each node

Side nodes 5,6,7,8
u -Su -Su
Interior node 9 ' 6x ' 6y
Displacement ot each node u,v 6v 6v
v- -
' 6x ' 6y
(al (bl

Fig. 35 Higher-order elements with additional degrees of freedom per node

or from added nodes: (a) 18 degrees of freedom, (b) 24 degrees of freedom.

i.e., rectangles which are nearly square. The aspect-ratio limit (ratio of the length of one
side to that of the other) varies somewhat with the type of element being used; a good
general guideline is to keep it below 10 for deformation analysis. This limit may have to
be reduced to 5 when stress evaluation is of prime concern.
Any departure from that of a rectangular shape will entail some degradation in the
quality of results. The more distorted the shape of the quadrilateral, the worse the results.
If any corner has an angle in excess of 135 deg., major errors can be expected. A good
basic element for planar problems is the 8-node serendipity or the 9-node Lagrangian
isoparametric element. If a three-dimensional analysis is necessary, the comparable ele-
ment is a 20-node one. However, because of the bandwidth associated with such ele-

Fig_ 36 Tubular joint modeled by PATRAN.

 Guidelines for the Selection of Grid 1-51

ments, the standard 8-node element is more often selected. These quadratic elements
exist in any first-rate general-purpose program. Some programs maintain higher-order
elements in their library, including cubic elements in the form of 12- or 16-node planar
elements. The corresponding three-dimensional element would be a 32-node element,
but since the bandwidth is excessive, the element is rarely used. With the quadratic (S-
node) element, the side node should be located within the middle third of any edge.
These elements are capable of handling even rather severe or sharp gradients in the stress
field. On the other hand, if the gradient in stress can be expected to be quite shallow (low
rate of change), even the use of constant-strain triangles will prove to be adequate. Most
elements respond quite well to uniform stress, and simple elements are as effective for
these cases.
Geometric Approximations. A curved boundary can be approximated as piecewise lin-
ear by the edges of straight-sided elements. However, for some cases this can result in a
singular condition, as for example at a simply supported node of a circular plate in a flex-
ural case. Isoparametric elements or similarly defined elements whose edges are
described by three or more nodes are capable of describing parabolic or higher-order
curves along their edges. Thus these elements are better able to describe curved bound-
aries. However, the edges of these isoparametric elements should not be curved except
where necessary in the vicinity of the curved edge, as this configuration with its curved
sides also causes some degradation in results.
Mesh Layout. Unless the analyst is familiar with the problem and knows where all the
regions of high stress will occur, it is desirable to first solve it with a coarse grid to delin-
eate those areas of significant stress. To obtain more accurate values in the critical regions
of stress concentration and to examine convergence, the solution process must be
repeated. For convergence studies the refined mesh must include the original mesh; oth-
erwise a new approximating sequence is started and with it a new convergence pattern.
The network should be kept relatively uniform in the various regions of the structure.
Of course this is not possible if a grid refinement is needed to accommodate sharp stress
gradients within a problem because of geometric or constraint reasons or because ofload-
ing. Refined grids are necessary where any sharp stress gradients are anticipated and are
to be used in all areas of significant stress.
To make the transition from one grid density to another, a band arrangement such as
shown in Fig. 37 can be employed. This transition band should be placed in an area where
stresses are not important since there will be some distortion of the stress field. A tran-
sition can also be established by writing constraint equations for the nodes along the
boundary between the two grid densities, restraining the higher number of nodes to have
displacements consistent with the fewer nodes (Fig. 38). However, this is not the rec-
ommended procedure because relative constraint equations seriously increase equation
solving time. To allow displacement incompatibility along the transition line may be the
lesser of the two evils.
Subdivisions at Discontinuities. In establishing the layout of elements and nodes, sub-
division lines or planes should be located at points where there are abrupt changes in
geometry, loading, and material properties.

I I I
Coarse
grid area I I I Coo rse
grid area
U16 = 1/2 ( U15 + U17) +C--

5 16 17 18 19

Fine Fine
grid area grid area
I+"--

Fig. 37 Use of topological zoom to Fig. 38 Use of constraint equations

produce grid refinement. to produce grid refinement.
1-52 Structural Analysis

The grid may introduce artificial discontinuities. Connecting higher-order elements to

lower-order elements can cause irregularities in the stresses, which may rey_uire smooth-
ing. If overconforming elements (Art. 29) are used, conflicts may occur between the
requirements for conformity and the presence of stress discontinuities.
29. Element Models These can consist of line elements (truss, beam, column, etc.), two-
dimensional plane-stress or plane-strain elements, flexural, plate, and/or shell elements,
axisymmetric elements, general three-dimensional solid elements, etc. Figure 39 shows
some frequently used elements. The line elements are used for truss, beam or spaceframe,
column, grid, and stiffener members. Most are 2-noded elements whose nodal degrees of
freedom are displacements and, where necessary, rotations. Some 3-node beam or
spaceframe members can be found in some programs. These are used mostly for a beam
or stiffener which is used in conjunction with plate or shell flexural members. They are
necessary to the maintenance of compatibility when quadratic or higher-order elements
are used to model the surface structure.
For two-dimensional elements the constant-strain triangle and plane-stress rectangle
(Table 5) can, in sufficient numbers, reproduce in acceptable detail the stress field for
most problems. However, their tendency to overestimate shear response in association
with in-plane bending has led to more frequent use of the 8- and 9-node isoparametric
elements. The 8-node quadrilateral element (Table 5) is one of the more efficient and is
recommended for general application. It is a numerically integrated element. This inte-
gration is normally accomplished using gaussian quadrature. A 3 X 3 grid of integration
points evaluates the stiffness matrix of an 8-node element to the accuracy of the assumed
displacement shape. However, the use of a reduced integration 2 X 2 grid normally pro-
duces a more efficient element which generates a better stress evaluation. Consequently
this is recommended. Table 6 lists the order of integration to be used for a number of
plane-stress elements. Stress output for these numerically integrated elements are nor-
mally given at their integration points. Stresses computed at integration points are sig-
nificantly more accurate than those computed at nodal points. These 8-node elements
degrade as their shape changes from that of a rectangle. The 9-node element behaves

(a) Line element

A
,~, ,D, ,~,
4 3 4~

Triangle Rectangle Quadrilateral

( b) Two-dimensional elements

4),
4 8 7

4
6
1

2
Tetrahedron Rectangular prism

(c) Three-dimensional elements

Fig. 39 Element models.
 Element Models 1-53

much better when distorted shapes are involved (significant departures from a rectangle).
It is also a numerically integrated element. A 2 X 2 grid is the most efficient, but both
the 8- and the 9-node elements are rank deficient when underintegrated (2 X 2). That
is, their displacement shapes contain some zero-energy modes in excess of those associ-
ated with rigid-body movement. For the 8-node element this condition disappears as soon
as the structure involves two or more elements. The 9-node element does not lose its
zero-energy modes as easily, so some form of suppression procedure is necessary when
using 9-node elements in a reduced integration format. This can be accomplished by
using one or more elements which are fully integrated (3 X 3), the whole grid being
overlaid with a second grid of very soft fully integrated elements, or each node can have
an additional spring stiffness added to restrain such modes.
Three-dimensional tetrahedron- or prism-shaped elements are to be used for those
problems which cannot be simplified into a two-dimensional case. The 8- and 20-node
prism elements are among the more efficient. They are both numerically integrated ele-
ments, for which a 2 X 2 X 2 grid is recommended. However, a word of caution must
be given. If the 20-node element is distorted (changed from a prism shape), this order of
integration produces an element which violates the "patch test," which is discussed later
in this article. This means it is possible that with refinement the solution may not con-
verge in the limit to the exact solution. Therefore, distorted 20-node underintegrated
elements must be used with caution. Their suitability should be verified by application to
some simple similar problems before the major problem is undertaken.
For all elements the node numbering sequence specified by the program must be
adhered to precisely. This establishes the incidence array that the program uses to deter-
mine stiffness contributions each element makes to the overall structural stiffness array.
If a preprocessor is utilized, satisfaction of the correct nodal sequencing is automatic. On
the other hand, the element numbering scheme is arbitrary, chosen by the analyst as
whatever is convenient. Figure 39 shows the incidence array an 8-node element would
have for the indicated global node numbering. The numbering scheme used for the global
nodes can be very important if the program equation solver is bandwidth-dependent,
which is most likely the case. This means the nodes should be numbered in the shortest
directions. Most commercial preprocessors have bandwidth optimizers so that the analyst
need not be preoccupied with bandwidth.
The displacement of any point within the element is approximated by expressing it in
terms of the nodal displacements by using polynomials whose coefficients are the gener-
alized coordinates (equations of the form listed in Arts. 31 and 32) or by interpolation
(shape) functions (Tables 6 and 7). The order of these functions defines the deformations
that the element can sustain. By knowing what displacement shapes are used in the for-
mation of the element stiffness matrix, the analyst can use these in a more intelligent
manner. The following restrictions must be placed upon the form of the polynomials or
interpolation functions used:
1. Number of terms. The number of terms in the polynomial must equal the number
of nodal degrees of freedom associated with the element.
2. Constant-strain and rigid-body modes (completeness). To ensure convergence to
the true solution as the number of elements is increased, the displacement function must
be capable of representing (1) constant-strain states and (2) rigid-body displacements. To
satisfy these conditions, the displacement function must be a complete polynomial to a
degree at least equal to the order of the highest derivative appearing in the equations
relating strain to displacement. For a polynomial to be complete, it must contain all pos-
sible terms through the order specified. Thus, a general complete polynomial of degree
n requires n + 1 terms in one dimension, ½(n + l)(n + 2) terms in two dimensions, and
¼(n + l)(n + 2)(n + 3) terms in three dimensions. For example a complete quadratic in
x and y requires ½ X 3 X 4 = 6 terms, which means the polynomial
ao + a1x + a2y + a3x2 + a4xy + a5y 2
3. Interelement compatibility. Compatibility of displacements at the nodes is guar-
anteed by the assembly process, which enforces common displacements at common
nodes. If monotonic convergence is to be achieved, the displacement function must main-
tain compatibility of displacements and their derivatives along interelement boundaries
up to one order less than that of the highest derivative in the strain-displacement rela-
tions. Elements which satisfy this condition are called conforming or compatible ele-
ci,
• TABLE& Two-Dimensional (Plane) lsoparametric Elements (fo = ~~h 110 = 1111, for node I)
Integration
Element
name Configuration DOF Reduced Full Shape functions Advantages Disadvantages
Four-node
plane
CJ}ladrilateral (-1,1194
r 39 <1, 11
u,v 1 X 1 2X2 N;
i
= ¼(l + ~o)(l + 110)
= 1,2,3,4
Simple
geometric
shape
Poor
approximating
power
Decline of
accuracy
with shape
distortion
t
1 2
(-1,-1)1 L1,-1l

Eight-node 'T} u,v 2 X 2 3X3 N, = ¼(l + fo)(l + 110)(~0 + 110 - 1) Curved sides Decline of
plane (0,1)
i = 1,3,5,7 possible accuracy
quadrilateral (-1,1) 5 (1,1l N, = ½(l - ~2)(1 + 110) Easily with
7 6 i = 2,6 adaptable to excessive
N; = ½(l - 11 2)(1 + fo) crack tip by shape
i = 4,8 moving distortion
(-1,0)¢8
~t midside nodes
Ideal for
nonlinear
(-1 -1)
'
~
(0,-1)
(1,-ll formulations

Twelve-node 'T} u,v 3X3 4X4 N, = ¼.(l + fo)(l + 110)(-10 + 9(~ 2 + 112 )] Curved sides Decline of
plane (-,/3,1) (1/3, 1) i = 1,4,7,10 possible accuracy
quadrilateral (-1,1) 10 7 (1,1) N, = %.(1 + fo)(l - 11 2)(1 + 9110) Easily with
9 8 i = 5,6,11,12 adaptable to excessive
(-1,,/3) 11 6 (1, 1/3) N, = %,(1 + 110)(1 - ~2 )(1 + 9fo) fracture- shape
i = 2,3,8,9 mechanics distortion
applications
(-1,-1/3) 12 5 t Ideal for
(1,- 1/3) nonlinear
2 3 4
(-1,-1) (1,-1) formulations
(-,/3,-1 l (1/3 ,-1)
 Nine-node "7 u,v 2X2 3X3 N, = ¾(~)(l + fo)(11)(l + 110) Curved sides Zero energy
plane
(1,0)
i = 1,3,7,9 possible nodes with
quadrilateral (-1,1) (1,1) N, = ½(l -e)('l)(l + 110) Better suited reduced
9
Lagrangian F 8 i
N,
= 2,8
= ½(~)(l + fo)(l - 11 2)
to shape
distortion
integration

(-1,0),;>4
15 6 (1,0) i = 4,6
(O,Q) N, = (1 - ~2)(1 - 11 2)
€ i =5
2 3
H,-11 1 ' (1,-1)
(0,-1)

Seven-node "7 u,v 2X2 3X3 N1 = -¾(l - m1 - 77)(1 + ~ + 77) Curved sides Decline of
plane
(0,1)
N2 = ½(l - 77)(1 - e) possible accuracy
quadrilateral (-1,1) 6 (1,1) ~ Ideal for with
5 4 N3 =
4 (1 + m1 - 11) nonlinear
formulations
excessive
shape
.€ N4 = 4~ (1 + m1 + 11) Useful for distortion
H,0191 I transitioning
Ns = ½(l + 11)(1 - ~2) from
N5 = -¾(l - m1 + 11)(1 + ~ - 11) parabolic to
H,-1) ~ (1,-1l N1 = ½(l - m1 - 11 2) linear
(0,-1) elements or
vice versa
Three-node u,v --- One point or N1 = L1 Simple Only constant
plane explicit N2 = L2 geometric strain

ti
triangle N3 =¼ shape possible
Good modeling
aid for
complex
shapes with
2
sharp corners
(1,0,0) (0,1,0)

...
g:
...u,
OI

TABLE 6 Two-Dimensional (Plane) lsoparametric Elements (E = EEh 'lo= 'l'l1for node i) (Continued)
Integration
Element
name Configuration DOF Reduced Full Shape functions Advantages Disadvantages

Six-node u,v Four Six points or N1 = L1(2L1 - 1) Curved sides Decline of

plane x..·0.1) points exp1icit with N2 = 4L1L2 possible accuracy
triangle straight sides N3 = L2(2L2 - 1) Ideal for with
N4 = 4L2L3 nonlinear excessive
Ns = L3(2L3 - 1) formulations shape
N6 = 4L3L1 Good modeling distortion
aid for
(1,0,0)~{0,1,0) complex
shapes with
( 1/2, 112 ,0)
sharp corners
Ten-node u,v --- Twelve points N1 = ½(3L1 - 1)(3L1 - 2)L1 Curved sides Decline of
plane h0.1) or explicit N2 = %L1L2(3L1 - 1) possible accuracy
triangle (,/3,0,%) (0, 1/3,2/3) with straight N3 = %L1L2(3L2 - 1) Ideal for with
sides N 4 = ½(3L2 - 1)(3L2 - 2)L2 nonlinear excessive
,213,,/3)
Ns = %L2L3(3L2 - 1) formulations shape
N6 = %L3L2(3L3 - 1) Good modeling distortion
N1 = ½(3L3 - 1)(3L3 - 2)L3 aid for
(1,0,0)~(0,1,0) N8 = %L1L3(3L3 - 1) complex
( 2/3, 1/3,0) ( 1/3, 213,0) N9 = %L3L1(3L1 - 1) shapes with
N10 = 27L1L2L3 sharp corners
Source: Adapted from H. Kardestuncer and D. H. Norrie (eds.), Finite Element Handbook, McGraw-Hill, New York, 1987.
 Plane-Stress and Plane-Strain Elements 1-57

ments. Most two- and three-dimensional stress-analysis elements retain maintenance of

compatibility along interelement boundaries. Thus, being conforming and complete, they
converge monotonically while they provide an upper bound to the stiffness of the struc-
ture (more properly, to the energy stored in the structure). To establish the rate of con-
vergence, each succeeding discretization must contain the grid of the preceding analysis
(this is termed a reducible net). Furthermore, the form of the element displacement
expansions must be independent of the orientation of the element (this property is called
geometric isotropy or geometric invariance). Monotonic convergence is desirable if the
analyst wishes to use extrapolation to achieve a better estimate of the final stresses.
Almost any plane-stress or plane-strain, axisymmetric solid, three-dimensional solid,
etc., element will be conforming, as conformity only requires continuity of the displace-
ment itself across an interelement boundary (this is termed C° compatibility). To have
nonconformity in these stress-type problems requires a premeditated violation of com-
patibility by the analyst. This is achieved by intentionally incorporating some additional
shapes which are nonconforming, or, for numerically integrated elements, integrating to
an order less than that necessary to integrate to the order of the assumed displacement
shape (reduced integration). Displacement shapes are said to be nonconforming if, when
evaluated along the interelement boundary, the displacement description involves nodes
other than those on the boundary. Intentional nonconformity is used because the error
associated with relative movement along interelement boundaries is counter to the error
associated with using a finite constrained displacement shape. Although monotonic con-
vergence is lost, the presence of counterbalancing errors often results in lower gross
errors in the coarse-grid solutions. However, with nonconforming elements the guarantee
of convergence is lost. To establish that the system will converge, a test known as the
patch test should be performed. This is a necessary but not sufficient test for convergence
of such nonconforming elements. In this test a "patch" of several elements of the struc-
ture is isolated and then subjected to all possible uniform strain states. For each of these
strain states there must be no residual forces present at the interior nodes of the patch.
When dealing with flexural-type problems, such as plate and shell elements, compatibility
also involves rotational degrees of freedom (continuity of displacements and first deriva-
tives thereof is termed C1 compatibility). This is much more difficult to achieve. Because
conforming elements meeting this C1 requirement are often far too stiff, nonconforming
modes are sometimes inserted into conforming elements in an effort to accelerate con-
vergence. The use of such eleme1;1ts must be carefully monitored, however, because in
irregular grids this may result in convergence problems. Nonconforming elements that
converge may not do so monotonically. This is the reason for their success-they oscillate
around the true solution with the error due to nonconformity partly compensating for the
overstiffness that results from using a finite discretized system.
Nonconforming elements have been used most successfully in plate and shell struc-
tures, for which the complex displacement expansions required to satisfy the conforma-
bility requirement can be replaced by relatively simple expressions if this requirement is
relaxed. Plate or shell elements based on a shear theory, such as the Mindlin plate theory,
use rotations as independent unknowns, thereby reducing the continuity requirement
back to that of just the displacements themselves (C° compatibility) with rotation now
considered as an independent displacement. These nonconforming elements are some of
the more commonly used plate elements.
Overconforming elements may come about as a result of using higher-order derivatives
as nodal unknowns to improve the performance of the element. This means that these
derivatives become nodal degrees of freedom and so must be specified at boundaries.
What this specification should be for a particular support condition is not always obvious.
30. Plane-Stress and Plane-Strain Elements In evaluating the local stress state in detail, a
number of structural members can be investigated as plane-stress or plane-strain prob-
lems. Because of this general applicability many such elements have been developed. A
few are listed in Table 5. The constant-strain triangle (CST) and the plane-stress rectangle
(PSR) were developed first. These elements utilize only corner node displacements, so
they include a very limited number of deformation modes. They tend to be excessively
stiff and to emphasize the shear mode. They also suffer from aspect ratio problems. When
the aspect ratio of the element becomes excessive, the quality of the results degrades.
Aspect ratios of 10 or more should be avoided. Stress solutions degrade more rapidly than
displacement solutions. When used to solve planar problems, most general-purpose pro-
.....
ci,

TABLE 7 Location and Weighting Factors for Gaussian Integration

0.577
T/1~77:1
=A=1· --- ' 71 1-
p
I
3

Li
•3 6. 9• •4 •a •12:,
e2

.,
-~
4e

t
577

•2
.,
5

4•
8

7e
-
=-:r,774
t
•••• ~

• • • • t
• , .5 .9 t1 3
;

0.34
!0,861

1
• 2
»x = 2, ny = 2 Rr = 3, "y = 3 »:r = 4. »y = 4

II
11 J_ n m
Rectangular J_ 1
f(xy) dxy = i~ ~ HiHJ'(a1, aj)
1

±a H
n=2
0.57735 02691 89626 1.00000 00000 00000
n=3
0.77459 66692 41483 0.55555 55555 55556
0.00000 00000 00000 0.88888 88888 88889
n=4
0.86113 63115 94053 0.34785 48451 37454
0.33998 10435 84856 0.65214 51548 62546
n=5
0.90617 98459 38664 0.23692 68850 56189
0.53846 93101 05683 0.47862 86704 99366
0.00000 00000 00000 0.56888 88888 88889
 Triangular I
JA
f(x, y) dx dy = t
i=l
H,f(L1, L2, L3)

L1 L2 L3 H

1-point formula, degree of precision 1

0.33333333 0.333333 0.333333 1.0000000

4-point formula, degree of precision 3

0.33333 33333 33333 0.33333 33333 33333 0.33333 33333 33333 -0.56250 00000 00000
0. 60000 00000 00000 0.20000 00000 00000 0.20000 00000 00000 0.52083 33333 33333

6-point formula, degree of precision 4

0.81684 75729 80459 0.09157 62135 09771 0.09157 62135 09771 0.10995 17436 55322
0.10810 30181 68070 0.44594 84909 15965 0.44594 84909 15965 0.22338 15896 78011

12-point formula, degree of precision 6

0.87382 19710 16996 0.06308 90144 91502 0.06308 90144 91502 0.05084 49063 70207
0.50142 65096 58179 0.24928 67451 70910 0.24928 67 451 70911 0.11678 62757 26379
0.63650 24991 21399 0.31035 24510 33785 0.05314 50498 44816 0.08285 10756 18374

...
:s
1-60 Structural Analysis

grams have several elements to choose from. However the 8-node isoparametric element
has replaced these low-order elements as the standard for plane-stress analyses. The lin-
ear-strain triangle (LST, Table 5) and the 9-node Lagrangian (Table 6) are special ele-
ments that can be useful. The 9-node element is particularly desirable if highly distorted
elements are involved in the grid.
31. Plane-StreBS Analysis Because strain-displacement equations involve only first
derivatives, it is only necessary that the displacements u and v be finite and continuous
across interelement boundaries in order to satisfy interelement compatibility. Also, to
satisfy the constant-strain and rigid-body conditions, the series must contain a complete
linear polynomial. These conditions are satisfied by the functions:
u = a0 + a 1x + a2y + additional terms
v = a3 + a 4x + a5 y + additional terms
The additional terms must be sufficient in number to make the number of terms in the
polynomial equal to the number of degrees of freedom for the element. Furthermore, if
the interpolation functions are used, they must contain these polynomials. Table 6 con-
tains the interpolation functions for several commonly used elements. The CST and PSR
satisfy these requirements but are much too stiff to accommodate sharp stress gradients
since they respond with too much shear-strain energy. If a unit length of beam which is
modeled by CSTs is subjected to a constant moment, it has about 4.5 times the strain
energy it should have (responding primarily with shear) but only about 1.5 times the
correct strain energy if modeled by PSRs (but still heavily in shear). This means the rec-
tangular element is preferred, but if a triangular shape is needed, a higher-order displace-
ment assumption is mandatory. This led to the development of the linear-strain triangle
(LST, Table 5) and the isoparametric elements.
Isoparametric elements can be expanded from the simple 4-node quadrilateral element
to higher-order elements by the addition of side nodes. Most programs include the 8-
node element which uses midside nodes. These higher-order elements have the advan-
tage of being able to adapt to the geometry of the problem. However, curving the edges
of the element should be done only as needed for the geometry. There is some loss of
accuracy with curved edges. Computation of the element stiffness requires integration
over the area of the element, but this can be done only approximately using numerical
integration techniques. The integration is normally performed by gaussian quadrature
using sampling points at the locations as listed in Table 7. Also listed are the weighting
factors by which the values of the function at the integration points must be multiplied
before summing. There are limits on the order of integration. The number of integration
points in any direction cannot be less than that necessary for the integration of the area
to be done correctly. There is an upper limit on the number of integration points, above
which the stiffness matrix does not change; that number is the one necessary to integrate
the stiffness matrix to the same order as that of the order of the assumed polynomials used
for the displacement definition. These limiting orders of integration are listed in Table 6.
Most programs select the order of integration for the user in order to avoid the problems
that may be present if too low an order is chosen. This means that the user cannot take
advantage of the increased efficiency of the reduced integration order. The general quad-
rilateral element with only corner nodes has the same overstiffness as the PSR element.
For this reason some programs allow the addition of nonconforming modes if reduced
integration is not allowed. When either of these options is used, the patch test should be
applied because convergence is no longer guaranteed. Because nonconforming modes
violate some of the boundary conditions along lines of symmetry and support, the user
may not want to use them in elements adjacent to these lines.
Most plane-stress elements were developed using a theory which neglects rotation
about the normal axis. Thus when they are used in conjunction with flexural-type mem-
bers, the analyst must make special provisions where the flexural member frames into the
plane-stress grid (Fig. 34). Unless something is done, there will be no rotational restraint
at the joint, and the flexural member will see it as a simply supported joint. Constraint
conditions must be used to define the rotation at the node in terms of the planar displace-
ments of adjacent nodes. Another edge condition of plane-stress problems needing con-
sideration is the fixed case (both displacements set to zero along that edge). Such an edge
no longer breathes in the perpendicular direction, so there will be a significant disruption
of the stress state in the vicinity of the edge.
 Beam and Plate Bending 1-61

UNIAXIAL

Fig. 40 Longitudinal stress contours for plate with circular hole in center;
plate loaded by axial force.

When higher-order elements (8-node and above) are used, work-equivalent or consis-
tent loads must be used. If the program does not provide such element loads, it will be
necessary for the analyst to insert them as nodal loads. Failure to use consistent loads for
high-order elements severely distorts the results. Midside nodes represent a dominant
behavior.
With preprocessors such as PATRAN the development of an error-free model is a rel-
atively routine matter. However, the quality of the model still depends upon the selection
of a proper model based upon the analyst's knowledge of the problem. Many systems
have the capability of graphical presentation of the results. For example, the stress con-
tours for a plate with a central hole, loaded with a uniformly distributed axial force, can
be quickly plotted. The longitudinal stress contours of such a plate are shown in Fig. 40
and the in-plane shear contours in Fig. 41. Similarly, the deflected shapes can be assem-
bled. The analyst often finds the deflected shape very informative. These plots were made
using the 8-node isoparametric element. Similar plots developed with the simpler ele-
ments such as the constant-strain triangle would show more erratic contour plots.
32. Beam and Plate Bending In these problems the strain-displacement equations
involve second-order derivatives. Therefore, to achieve element compatibility, the dis-
placement w together with the derivatives aw/ax and aw/ay must be finite and continuous

Fig. 41 Shear-stress contours for plate with circular hole in center; plate
loaded by axial force.
1-62 Structural Analysis

on the element interfaces. Also, to satisfy the constant-strain and rigid-body condition the
series must contain a complete quadratic polynomial. These conditions are satisfied by
the functions:
One dimension: w = a0 + a x + a x + additional terms
1 2
2

Two dimensions: w = a0 + a 1x + a2 y + a3 x2 + a4 xy + a5 y2 + additional terms

The requirements for additional terms and interpolation functions are the same as for the
plane-stress polynomials (Art. 31).
In plate-bending elements, standard plate theories, some with shear deformations, are
used to compute middle-surface displacements and bending-moment stress resultants.
This is an area where there is little agreement on the best element. Numerous elements
have been developed; Table 8 lists a few of these. In general the rectangular elements
are far superior to triangular shapes. The excess stiffness the triangular shapes display in
plane-stress applications is still evident in plate-bending problems. Thus rectangular ele-
ments should be used in the major regions of the structure if at all possible.
The nonconforming rectangulru- plate model (12 degrees of freedom (DOF)] is very
efficient. Its major limitation is that of its shape constraint. A number of 4-node quadri-
lateral plate elements exist in several commercial programs, which allows more gridding
flexibility. However, they retain the locking problem and can produce significant under-
estimates of the response. When shaped in the general quadrilateral form, these elements
are normally established by numerical integration. They are therefore much more expen-
sive to use than are the pure rectangular elements which are explicitly integrated. The
conforming rectangular element (16 DOF listed in Table 8) is also very efficient, but it is
only available in a rectangular form. The analyst must specify not only the deflection and
slope on boundaries that are constrained but also any constraints on the twist. The cor-
responding triangular elements are more poorly behaved. The nonconforming triangular
element with 9 DOF has worked very well in many cases, but it experiences convergence
problems with some meshes when the element fails to pass the patch test. Some arrange-
ments of the 9-DOF plate element do converge to a result which is slightly in error. For-
tunately, the error is normally not of engineering significance. The corresponding con-
forming triangular element is far too stiff and requires rather dense grids to achieve any
reliability.
Higher orders are normally used to produce efficient conforming triangular plate-bend-
ing elements. The complete fifth-order polynomial has performed well, but the band-
width of the equations is too high, so the element has been relegated to an inactive role.
Some intermediate DOF elements have performed satisfactorily and still exist in pro-
grams with a SAP parentage.
The most popular elements are the isoparametrically based degenerated elements.
Constraints are applied to a general three-dimensional formulation to develop these plate
elements (last four elements in Table 8). These elements have the shape generality of
numerically based elements, but the medium-thick to thin plate versions have locking
problems. Reduced integration procedures are used to remedy the locking problem, but
this creates another equally serious problem of possible zero-energy modes. Because of
this latter problem, some general-purpose programs do not allow any reduction in the
order of integration less than that necessary to integrate the stiffness matrix exactly. Plate
elements have been developed for both the Kirchhoff and the Mindlin (shear deforma-
tions included) theories. One of the ways to try to soften a thin element is to enforce the
Kirchhoff conditions at only discrete points. Several 4-node and some higher-order ele-
ments have discrete Kirchhoff versions implemented in their programs.
33. Shells and Combined Direct Stress and Bending Many problems involve a combination
of plane-stress or membrane forces with flexure. Stiffened plates and shells in the form of
either curved or prismatic surfaces are examples. Three types of elements can be used
for shells:
1. Flat-plate elements that model the shell as a faceted surface
2. Curved shallow or deep shell elements derived from a general shell theory
3. Three-dimensional elements, either directly or after they have been adapted by
constraint equations (degenerated elements)
Several of these elements are listed in Table 9. The flat elements are obtained by super-
imposing plane-stress and plate-bending elements. For low-order elements this process
is adequate. CST plane-stress with the 9-DOF plate-bending element is a common tri-
 Three-Dimensional Problems 1-63

angular shell element. A PSR plane-stress with the 12-DOF plate-bending element is a
good element for developable shells (cylinders, etc.). Curved shell elements, which use
an assumed displacement field directly in whatever shell theory is desired, produce good
elements. A coordinate transformation is normally needed to convert from element to
global coordinates. However, such elements frequently violate rigid-body motion con-
ditions. If the enclosed angle is less than about 10 deg, the rigid-body violation is not of
engineering significance. Even so, there may be some geometric differences between the
shell and its element array, even with grids of small elements, so some analysts use only
flat elements. This restricts the shape for general shell elements to triangles or
quadrilaterals.
The degenerated element formed from an isoparametric formulation is one way around
the geometry problem. This class of element has become the standard plate and shell
element in most programs. The 8-node serendipity element (Table 8) is a good basic ele-
ment to use to model most shell problems. The order of integration for this element can
be most successfully performed using a 2 X 2 reduced order of integration. When fully
integrated, 3 X 3, the element becomes very stiff when used for a thin shell. This excess
stiffness is termed locking. Use of reduced integration is one remedy for locking problems,
but this can induce zero-energy mode problems. The zero-energy mode problem does
not manifest itself for this (8-node) element when the structure involves more than one
element. The element does degrade in quality, however, when the shape departs signif-
icantly from that of a rectangle. For severely distorted element shapes a 9-noded Lagran-
gian element is recommended, but this element does suffer significant zero-energy mode
problems. Techniques to suppress the zero-energy modes have not found their way into
most general-purpose programs, so the user must implement his or her own suppression
scheme. One method is to fully integrate selected elements in the grid. The location of
the elements to be fully integrated should be chosen in an area where there are relatively
slow stress changes.
One area where finite-element models need to be employed is in the investigation of
plates and or shells with eccentric stiffeners or support beams. Most programs allow the
inclusion of the eccentricity through a rigid link between the elevation of the beam axis
and the level of the middle surface of the shell. The element used to model the surface
structure must include both in-plane or membrane forces as well as flexure, so simple
plate elements are not enough. The composite member should produce T-beam action
for the shell-stiffener combination. This manner of handling the eccentricity in effect
transfers the stiffness characteristics of the beam or stiffener up to the elevation of the
shell middle surface. If support conditions are now prescribed, those conditions apply at
the level of the shell nodes. If it is desired to provide the support say at the base of the
beams, a constraint relationship must be written relating the shell node displacement to
that of the bottom of the beam. This manner of treating eccentricity results in an incom-
patibility condition along the junction line between the beam and the shell.
The fact that rotation about the normal is neglected in most plate and shell elements
causes some problems. When modeling two plates which intersect at right angles, what
is flexure in one plate is rotation about the normal in the perpendicular plate (Fig. 42).
If no constraint is placed on these degrees of freedom, the flexural response of the plate
will see no restraint from the perpendicular plate. To overcome this, some attempts to
introduce a drilling mode in the stiffness matrix for shell elements have been undertaken
but have not been overly successful. As the flexural response of the orthogonal element
interacts with an in-plane stiffness of the plate in the perpendicular plane, that stiffness
severely restrains the flexural rotation. One alternative is to constrain the flexural rota-
tion; another is to equate the rotation of the node to the average in-plane displacements
of the adjacent nodes.
34. Three-Dimensional Problems Two three-dimensional elements are in common use:
the tetrahedron and the isoparametric solid element in an 8- or 20-node version. The
latter is usually the much more economical choice. The tetrahedron, like its degenerate
case the constant-strain triangle, exhibits far too stiff a response to any stress field with a
gradient. The isoparametric element exhibits a much smaller trace. Three-dimensional
problems are considerably more expensive to solve than are their two-dimensional coun-
terparts. This is a consequence of a dramatic increase in bandwidth along with a signifi-
cant increase in the computational time necessary to determine the element stiffness
matrices due to the need to integrate in the third direction. A few of the available ele-
...
t

TABLES Plate-Bending Elements

Nodal DOFper Polynomial

Element parameters element degree Compatible Reference Remarks

WJ,W!x,WJy 9 Incomplete No 16 Good nonconforming element.

~
3 Fails patch test in some
configurations.
Generally available

9 Yes 16 Too stiff. Significantly

~
Wj, WJx, WJy
underestimates results

9 Incomplete . Yes 17

~
WI, WJx, WJy
3

1. WJ, WJx, W]y, 12 3 Yes 7 Use of variable number of

4 2. WJn

21 5 Yes 15,18
DOF per node means it is not
suitable for most general-
purpose programs

Good results but too many

~
1. WJ, W1x, WJy,
Wtxx, WJxy, DOF per element. Nodes
WJyy have different DOF, so there
2. W]n is very limited availability of
curvature continuity
 WJ, Wix, WJy 12 Incomplete No 19 Simple, yet good, basic plate

□
4 element.
(bicubic) Shape is a limiting constraint

Wt, WJx, W\y, Wtxy 16 Incomplete Yes 20 Need to specify Wtxy on

□
6 boundary.
(Hermitians) Excellent results but shape
constraint is limiting

Wi, Wix, Wty 12 4 No 7 Generalization of rectangle.

Ll Some deterioration of results

when used in distorted shapes

Wi, Wix, Wty 12 plus 7 Yes 21

CJ
internal

24 Yes 7 Good basic plate element. In

Ll
W1, W1x, Wly
both Mindlin and Kirchhoff
versions. There are some
shape distortion effects.
W, Wix, W1y 27 - - Available in both Mindlin and
Kirchhoff versions. Element is

□
better able to take shape
distortions. Reduced
integration will have zero-
energy modes .

...
a,
en
-
0,
a,
TABLE9 Shell Elements

Nodal DOFper
Element parameters element Flat or curved Compatible References Remarks

Li
u,v, w 15 Flat triangular u, v, w, yes 22 Acceptable element for
W1x, Wty w 1, no problems without sharp
stress gradients. CST
limits membrane
response
u,v, w 15 Curved --- 23 Not a good element for

A a,{J

u,v, w 20
triangular

Flat u, v, w, yes
general use. Limited
availability

Good element for cylinder

0
Wx, Wy rectangular w, no problems. Sometimes in-
plane shear distribution
is distorted. Must use
nodal averaging for
stresses
U, V, W 20 Curved u, v, w, yes 24,25

Q a, fJ parallelogram a,{J,no

20 Faceted --- --- Acceptable low-order

l5;J
u,v, w
a, fJ plus quadrilateral element
5

40 Curved --- 7 Good efficiency. Excellent

0
U, V, W
a, fJ arbitrary element for general use.
Both Mindlin and
Kirchhoff versions.
Reduced integration is
desirable. Can have
zero-energy mode
 Element Stiffness Matrix and Nodal Force Vector 1-67

-----Deflection
xz plane

By is flexural for plate a.

By is rotation around normal for
plate b (activates in-plane forces)

Fig. 42 Problem associated with plates intersecting at an angle.

ments applicable to three-dimensional problems are shown in Tables 10 and 11. The 8-
node hexahedron is a good choice as a standard element. When sharp gradients are pres-
ent, the 20-node element is more suitable because it is better able to adapt to the stress
gradient. With the higher order, the use of consistent loads is more imperative.
35. Element Stiffness Matrix and Nodal Force Vector The element stiffness matrix k and the
nodal force vector p are evaluated by the following procedure, described here for a two-
dimensional element with e nodes. The extension to three dimensions and/or plate and
shell flexural problems is obvious. The steps involved are outlined below to provide the
user with some idea of what is involved in the development of an element solution of a
particular problem.
1. Each component of the displacement vector u(x, y) or a point P(x, y) within the
element is described by a polynomial chosen to meet the conditions of Art. 29. Normally,
identical functions apply for each displacement component. Thus

u(x, y) = [ ~] = [ t' ~•] {a} = Ma (66)

where ti>' = (1, x, y, x2 , xy, ... , toe terms]

a, = {a0, ai, a2 , ••• , 2e terms}
Set a, contains the unknown coefficients of the functions ti>'.
2. Nodal displacements u = {u., v,) are expressed in terms of the generalized coor-
dinates a by substituting the spatial coordinates of the nodes into the matrix M ofEq. (66)
to obtain
(67)
where u! = {ui, u 2, ••• , u.} and v! = {vi, v 2 , ••• , v.)
3. Displacements u within the element are expressed in terms of the nodal displace-
ments q. This involves substituting a from Eq. (67) into Eq. (66) yielding
(68)
The matrix N is called an interpolation matrix. In many cases the constituents of this
matrix can be selected directly without the need for going through steps 1 and 2. It can
be written directly because, by definition, an interpolation function is a function which
has a unit value at a given node and zero values at all other nodes. Shape functions for a
number of isoparametric elements are listed in Tables 6 and 11. Thus, a particular inter-
polation function can be derived by writing an equation which is the product of the equa-
...
a,
00

TALBE 10 Three-Dimensional Elements

Nodal DOFper Polynomial Name and Remarks

Element parameters element degree reference
u,v. w 12 Linear Tetrahedron Poor element for general

© u, v, w 30 Quadratic in
26

26
use. Low efficiency.
Simple geometric
shape

Linear strain, marked

©
natural improvement over 4-
coordinates node tetrahedron.
Curved edges possible

24 Bilinear Hexahedron Limited strain variation.

~
U, V, W
or 8-node Decline in accuracy
isoparametric with shape distortion
7

U, V, W 60 Quadratic in 20-node Good general element.

®
natural isoparametric Some decline in
coordinates 7 accuracy with
excessive shape
distortion. Curved
edges possible
 TABLE 11 Three-Dimensional (Solid) lsoparametrtc Elements (~o = t~. 110 = 11;11, to = t;t for node ,1
Element Integration
name Configuration DOF Reduced Full Shape functions Advantages Disadvantages

Eight-node
~ u, v, w 2X2X2 2X2X2 N; = ¼(I + fo)(I + ~o)(I + !O) Simple Poor
3D solid One i = I, 2,, .. , 8 geometric approximating
point if shape power
cube Decline of
accuracy
with shape
(-1,-1,1)5Q" IQ"" distortion
I
I
I -t
I
4~---+--- 3(1,1,-1)
,,,.,,,.(-1,1,-1)
,,,.
1
(-1,-1,-1) 2(1,-1,-1)

Twenty-
~
U, V, W 3X3X3 3X3X3 N; = ¼(l + fo)(l + ~o)(l + 10)(~0 + ~o + !O - 2) Curved sides Decline of
node 3D 2X2X i = 1, 3, 5, 7, 13, 15, 17, 19 and faces accuracy
solid 2 if cube N; = ¼(l - ~2)(1 + ~o)(l + 10) possible with
i = 2, 6, 14, 18 Easily adaptable excessive
7) N; = ¼(I - ~2)(1 + fo)(l + 10) to fracture- shape
i = 4, 8, I 6, 20 mechanics distortion
N; = ¼(I - 12)(1 + fo)(I + ~o) applications
i = 9, 10, 11, 12 Ideal for
nonlinear
I t formulation
9
(-1,1,-1)
,,,. k
7 --
,,,..Pa
1 2 3(1,-1,-1)
(-1,-1,-1)

...a,
<O
...
~

TABLE 11 Three-Dimensional (Solid) lsoparametric Elements (~0 = U, 110 = "'"' io = i,i for node i) (continued)
Inte_g_ration
Element
name Configuration DOF Reduced Full Shape functions Advantages Disadvantages

Four-node 4(00,0,1) u, v, w One point N1 = L1 Simple Only constant

linear or N2 = L1 geometric strain
tetrahedron explicit N3 = L3 shape possible
N4 = L4 Good modeling
aid for
complex solids
<1,0,0,0)1 3(0,0,1,0) with sharp
corners
~
2 <0,1,0,0)

Ten-node U, V, W N1 = L1(2L1 - !); N2 = 4L1L2 Curved sides Decline of

parabolid N3 = L2(2L2 - !); N4 = 4~L3 and faces accuracy
tetrahedron Ns = L3(2L3 - !); Ns = 4L3L1 possible with
N7 = 4L1 L4; Ns = 4L2L4 Ideal for excessive
Ng = 4L3L,; N10 = L4(2L4 - I) non1inear shape
formulations distortion
Good modeling
5(0,0,1,0) aid for
complex solids
with sharp
3 (0,1,0,0) corners

Source: Adapted from H. Kardestuncer and D. H. Norrie (eds.), Finite Element Handbook, McGraw-Hill, New York, 1987.
 Element Stiffness Matrix and Nodal Force Vactor 1-71

tions of the lines or surfaces through all nodes other than the one for which the function
is sought. For example, to determine the interpolation function for node 1 of the rectan-
gular plane-stress element shown in Fig. 39b, with the x-axis alongside 1-2, the y-axis
alongside 1-4, with side 1-2 of length a and side 1-4 of length b, the equations x - a
= 0 of line 2-3 and y - b = 0 of line 3-4 are multiplied. This gives n 1 = c(x - a)(y -
b), which is O at nodes 2, 3, and 4. The constant c is set equal to I/ab to give n = I at
node 1. The resulting equations are

u(x, y)} _ [Nl(x, y) 0 ] {u•} (69)

{ v(x, y) - 0 Nl(x, y) v,

where Nl(x, y) = [ni, n2 , ••• nr n,] is a row vector containing the interpolation
or shape functions ni(x, y), j = 1, 2, ... , e. The polynomial description in these shape
functions sets the possible displacement variations that may be present in an element and,
therefore, the strain variations sustainable within an element. Interpolation or shape func-
tions for nonrectangular elements can be expressed most conveniently in terms of natural
coordinates. Figure 43 shows some systems in common use.
4. The strains within an element E(x, y) = {E,, Ey, "Yr y} at a point P(x, y) are set by the
kinematic conditions described by the strain-displacement equations

which is written in symbolic form as

E = Bu (70)
where the strain-displacement matrix B is obtained by appropriate differentiation of the
interpolation functions N. If the interpolation functions for the particular element being
used have been written in a natural coordinate system, a transformation to cartesian coor-
dinates must be made in establishing the strains, because those strains are defined in an
engineering sense in cartesian coordinates.

(-1,1)4 t"I 3 (1,1 l 3(0,0,1)

I
I

2b +-- t
20
(-1,-1 )1 2(1,-1)
X y
t= ci' , 77= b 2(0,1,0)
(al Rectangle
t;= ~I ,i=1,2,3
A = A1 + A2 + A3

(bl Triangle

E=¾,77=f, t=f
(cl He~ohedron
Fig. 43 Natural coordinates.
1-72 Structural Analysis

5. With the strains defined, the element stresses can be determined, s(x, y) = (u,,
uy, 'T xy) corresponding to the strains e. To evaluate the stresses the strain matrix is pre-
multiplied by D, the stress-strain matrix, to generate the stresses.
s = DBu (71)
The ingredients of the D matrix are whatever the constitutive equations are for the par-
ticular material being used. Whatever assumptions pertain to the theory being used apply
throughout the analysis. .
6. The element stiffness matrix k and the nodal force vector p are calculated using
the virtual-work principle requiring that no net work be done when each of the nodal
degrees of freedom of the system is given a virtual displacement. The element stiffness
matrix is established based on the internal virtual work done (product of the internal
stresses times the virtual strains de = {oEx, iiEy, ii-y xy) corresponding to a set of virtual dis-
placements du).

dW, = JI l [iiE)'u dV

which becomes, on substituting for the expressions for stress and strain,

The corresponding external virtual work iiW, due to the work of the applied body force
X and surface tractions T going through the virtual displacements of the nodes is

iiW, = (ou)' JJJ. N'X dV + fl (iiu)'T dS

Equating the virtual internal work to the virtual external work yields
kq = P (72)

where (73)

(74)

The expression for the stiffness matrix ofEq. (73) is the same for all cases: line elements,
plane stress, plane strain, plates, and shells, to three-dimensional cases. The actual terms
in the B and D matrices change depending on the theory being used and the material
making up the structure. Explicit forms of the stiffness matrices for a number oflow-order
elements can be found in Ref. 29. Stiffness matrices for the higher-order elements are
usually not available since they are usually established numerically in the computer with
no explicit form existing.
36. Equilibrium Equations for the Assemblage With the element stiffness properties estab-
lished, the total work is simply the summation of the stiffness matrices of all the elements.
This set of equations can be thought of as equilibrium equations of the nodes. To perform
the addition it is necessary to
1. Transform the element-stiffness matrix k and nodal load vector p for each element
to a common coordinate reference system, that is, into the global axes of the system. Most
large, general-purpose programs do this without the user being required to initiate it.
The user inputs the coordinates of the node points in the global reference frame, or bet-
ter, uses a preprocessor to do this, indicates what the element type is and its material
properties, and then specifies the incidence matrix which relates the nodes of the element
to the nodes of the structure. The program will use this information to relate the local
reference frame to the global axes. The required coordinate transformation matrix to
accomplish this is the rotation matrix T made up of the direction cosines that describe
the orientation of the two axes systems. The stiffness and load transformations are
k0 = T11<1T (7,5)
(76)
 Equilibrium Equations for the Assemblage 1-73

where the subscripts I and O refer to the local and the global axes, respectively. Diagonal
matrix T is made up of submatrices t which contain for each node the direction cosines
for the local directions with respect to the global directions. Matrix T is of the order of
the total number of displacements q for the element, while t is of the order of the number
of displacements at a node.
2. Write Eqs. (75) and (76) in the global directions, this being done within the program.
The result is a system of equations for each node which is the sum of the contributions of
all elements meeting at that node:
KU= P (77)

where K = structure stiffness matrix: sum of all element stiffness matrices

U = vector of all nodal displacements
P = vector of m + r nodal forces
m = number of active displacements of structure
r = number of specified support displacements
The matrix K, which is symmetrical and of order m + r, is given by

(78)

TABLE 12 Consistent Loads for Quadrilaterals

Element Load Nodal loads

Four-node quadrilateral Body force, q
qobh-□
-qobh

2b

2
qobh- 0. -qobh

Uniform edge, p
□ -pbh
b

-pbh

Eight-node quadrilateral Body force, q

Uniform edge, p
1-74 Structural Analysis

where g = subscript denoting a typical element and s = number of elements. Likewise

the load vector involves the sum of P./l, the corresponding consistent (i.e., work equiva-
lent) load, as well as any concentrated nodal loads. Table 12 contains a list of the consis-
tent loads that apply to the 4- and 8-node quadrilaterals.
s
p = LPg (79)
g=l

Equations (78) and (79) constitute the assembly rules for Kand P. The rules are applied
by the direct stiffness method. In this method the assembly is controlled by the incidence
table which relates the element nodes to the structure's global nodes thereby indicating
which terms of the stiffness matrix to add and where to add them, in other words, insert-
ing the appropriate submatrices of the element stiffness into the proper locations of the
global matrix.
37. Solution for the Displacements The structure stiffness matrix K is singular; hence the
system equations [Eq. (77)] cannot be solved until boundary conditions which eliminate
all possible rigid-body motions are enforced. In the displacement method, it is not pos-
sible to explicitly enforce boundary conditions involving prescribed forces, but their
effect is included in the appropriate equilibrium equations. The geometric (displacement)
boundary conditions must be explicitly satisfied. These conditions must be inserted into
the equilibrium equations in order to form the set of nonsingular equations for the assem-
blage of elements. These essential boundary conditions can represent absolute conditions
where the displacement component is assigned a specific value, most commonly zero, or
be of the relative case where a relationship exists between various displacements and/or
their components.
38. Solution for Element Strains and Stresses The solution of the equilibrium equations
for the structure (Eq. (77)] evaluates the displacements at all the interelement nodes.
With the displacements established, the strain conditions are prescribed at all locations
of the structure. Thus the components of strain and of stress at a point within an element
can then be computed using Eqs. (70) and (71). In general, owing to the approximations
involved, the predicted stresses evaluated for each of the elements meeting at a common
node are not equal, so some sort of averaging must be employed. It follows that the local
equilibrium conditions are not satisfied at the points within an element. The differences
in nodal stresses and in local equilibrium can be quite significant for low-order elements.
The disparity in stresses generally decreases with increases in the order of the approxi-
mations used for the element so that less manipulation and interpretation of stresses is
needed in the higher-order elements. This is one of the advantages of using higher-order
elements. For such elements the stresses calculated at the midside nodes are normally
significantly better behaved and more reliable than are the similar quantities evaluated
at the corner nodes. The corner nodes are extremities of the region, so the quality of the
approximating functions should be expected to be poorer.
It is common practice to average the stresses for all elements meeting at a node to
determine the stresses at that node. The average may be achieved by simple direct aver-
aging, or its calculation may involve some weighting factor based upon element areas,
distance to the centroid, etc. For low-order elements one common practice is to consider
the stresses only at the centroid of the element. For isoparametrically based elements,
the gaussian integration points where the stresses are evaluated are the best points at
which to view the stress state. However, when contouring plots are produced by a post-
processor, nodal point stresses are used. This means that some extrapolation process is
needed to shift the stresses from integration points where the stresses are calculated to
nodal points where they are averaged.
If any of the stresses calculated by any of the procedures just described are to be used
as the basis for design, the stresses should be checked by evaluating the equilibrium of
selected freebodies cut from the structure and loaded with the calculated stresses applied
to the cut edges. Some form of numerical integration, such as Simpson's rule, will be
necessary to carry out the evaluation of the stress resultants. Since the solution is approx"
imate, the equilibrium equations will not be exactly satisfied. The displacement model in
a finite-element procedure does not guarantee satisfaction of equilibrium even though
the global equations are exactly satisfied. Errors of several percent are possible, but dou-
ble-digit errors should be questioned.
In problems involving both in-plane and out-of-plane displacements, such as occur in
 Solution for Element Strains and Stresses 1-75

combinations of beams, plates, and shells, the in-plane displacements are normally of a
much lower order than the out-of-plane displacements. Thus, if elements join at an angle,
if their nodal points are eccentric to one another, etc., not only will the displacement
compatibility that exists in the flat be lost, but stress interpretation along the edges
between elements becomes more difficult. There will be differences in the variation of
the in-plane forces and that of the bending moments along these edges. Because of the
lack of a degree of freedom associated with the rotation about the normal, the in-plane
restraint on the bending along the edge is not activated, so more flexure is calculated than
is actually present. If the problem involves beams and they are eccentric to the plate, the
fiber stress in the beam along its connection to the plate will not vary in the same manner
as the membrane (in-plane) force in the plate. Averaged nodal stress values may be better
in this case. One must know which plane (plate or beam) is the reference for the displace-
ments and stresses; normally the plate reference frame is used. This means that a coor-
dinate transformation may be necessary in order for the boundary conditions to apply,
say, to the bottom of the beam rather than to a point in the plate reference frame. Such
special treatments are required in most general-purpose programs.
Few elements include a degree of freedom associated with rotation about a normal to
the surface of the element. Therefore, if two elements in combined direct stress and
bending intersect orthogonally, or at a significant angle, additional constraints should be
placed on the associated rotational degrees of freedom on the line of intersection. This is
necessary to take better account of the constraint of the orthogonal plate on the normal
displacements of the intersecting plate. There are also difficulties if a beam element inter-
sects a network of plane-stress elements in the same plane. A solution is to write special
constraint equations relating the rotational degree of freedom to the difference in in-
plane displacements of adjacent nodes divided by the distance between those nodes. An
alternative partial solution is to extend the beam a couple of elements into the plane-
stress grid. This allows a nonzero bending moment to exist at the intersection face.
Comments on applications, and in particular on stress interpretation, are found in Refs.
6 and 7.
Examples PLATE-GIRDER HAUNCH. Finite-element analysis is useful for investigating stresses in struc-
tures, or parts of structures, of irregular configuration. Figure 44 shows some of the results of an
analysis of the haunch of a continuous plate girder. The haunch grid is shown in Fig. 44c. The web
was modeled with 8-node quadratic isoparametric elements (Table 5), using line elements (truss mem-
bers) for the flanges and stiffeners. The 8-node element is a good choice for this problem for reasons
discussed in Art. 31. The remainder of the girder entered the solution in the form of the stress resul-
tants shown at each end of the haunch section: these were determined by conventional frame analysis.
The springs shown at the juncture of the haunch and inclined leg were used to determine the stress
resultants at this section.
The floor-beam concentrations (denoted by FB 1 , FB 2 , and FB 3 in Fig. 44c) were not input directly.
FB 1 was accounted for approximately by adjusting the shear and moment at the left end of the grid
so as to give the correct shears and moments at FB 1 and FB 2 . FB 2 was omitted, which did not affect
the shears and moments at the section, but the local effects of the concentration were lost. FB3 was
accounted for correctly by inputting the shear immediately to the left.
FINITE 29 was used for the solution. Bending stresses at two sections and the axial stress along one
flange, all for a negative-moment condition, are shown in Fig. 44d, e, andf
HYPERBOLIC PARABOLOID ROOF. Finite-element analysis is used extensively in shell structures, partic-
ularly when the answers sought pertain to shell-beam or shell-stiffener interaction. Problems of this
type occur in roof structures, reactors, offshore structures, etc.
Figure 45a shows part of the roof of a water-pollution control structure. The roof consists of 44
hyperbolic paraboloids of the configuration shown in Sec. 24, Fig. 40, and is 4 units wide by 11 units
long (250 X 660 ft). Each shell has a rise of 9 ft and is 4 in thick. Ridge beams are 10 in deep by 18
in wide, and edge beams are tapered from 10 in deep at the ridge to 33 in at the corners. The roof is
supported on columns spaced 60 ft in each direction, with tie rods at the column tops.
For economy in construction each shell was cast as a unit. Therefore, since there was little or no
continuity of adjacent units, a single unit could be analyzed independently, and because of symmetry
one 30 X 30 ft quadrant could be considered. Figure 45c shows a projected view of one quadrant,
with an 8 X 8 grid. Because there is no region of sharp stress gradients this structure can be modeled
with a uniform grid of square elements. Subdivision of the mesh to investigate a boundary layer effect
usually results in a distorted grid with complicated elements shapes, and might also make it impossible
to use any data-generator capabilities of the program. Furthermore, data preparation can be time-
consuming, and mistakes can necessitate a rerun, which could be a serious loss.
The curved parallelogram element of Table 9 was used in this analysis. The flat triangular element
in the table could have been used with a shell as shallow as this one, but the resulting accuracy was
not expected to be competitive with the curved element.
 457.5m
46.25m 73m 73m 73m 73m 73m 46.25m
I• •I• •I•

(al

650 x 30 flange
24 x 3000 web
650 x 30 flange

E 24 x 1100 web
IC)
l'-

N
NOTE: Dimensions in
millimeters unless
noted.

(bl
r+A
FB1 FB2

NOTE: Heavy lines show

flanges and stiffeners.

-10 -5 0 5 10 15 -10 -5 0 5 10 15 -6 -4 -2 0
Stress, ksi Stress, ksi Stress, ksi

(dl (el (fl

Fig. 44 (a) Continuous plate-girder bridge over Wadi Qaddiyah for Ministry of Communications,
Kingdom of Saudi Arabia. Designed jointly by Hanson Engineers, Inc. and Wilson & Company. (b)
Typical segment. (c) Finite-element grid for haunch. (d) Stress normal to section A-A. (e) Stress normal
to section B-B. (f) Flange stress along section C-D.

1-76
 References 1-77

Distance along ridge, ft

(cl (d)
Fig. 45 (a) Portion of hyperbolic paraboloid roof for water pollution control structure, Oakwood
Beach, Staten Island, New York City. Greeley and Hansen, Engineers; Klein & Hoffman, Structural
Engineers. (b) View of one quadrant. (c) Finite-element grid for one quadrant. (d) Axial force and
bending moment in ridge beam.

A 4 X 4 grid will provide adequate preliminary design information for a structure of this type. This
results in a problem with 16 shell elements, 16 beam elements, 125 unknown displacements, and a
half-bandwidth of 65. Because solution time increases with the square of the bandwidth, it is important
to keep it as small as possible. For example, a 6 X 6 grid gives 36 shell elements, 24 beam elements,
245 unknown displacements, and a half-bandwidth of 85, and computation cost will be almost double
that of the 4 X 4 grid. The 8 X 8 grid used for the final analysis of this shell has 64 shell elements,
32 beam elements, 405 unknown displacements, and a half-bandwidth of 105. This translates into a
computation cost of about 3 times that of the 4 X 4 grid.
The solution gives the deflected shape, the shell forces, the beam forces, and, if the supports are
modeled by stiff springs, the reactions. Shell stresses are usually low, and the primary reason for the
analysis is to establish the stresses in the various supporting beams. The axial force and bending
moment in the ridge beam are of particular concern, and cannot be determined by an elementary
analysis. Because of its interaction with the shell, the ridge beam behaves as a beam on an elastic
foundation, which accounts for the wave form of the moment curve in Fig. 45d. There have been
several failures of hip-roof ridge beams because of inadequate reinforcement due to gross underesti-
mates of the moment, and in at least one case the result was a complete collapse of the roof.

REFERENCES
1. Timoshenko, S. P., and D. H. Young: "Theory of Structures," 2d ed., McGraw-Hill Book Com-
pany, New York, 1965.
2. Langhaar, H. L.: "Energy Methods in Applied Mechanics," John Wiley & Sons, Inc., New York,
1962.
3. Norris, C. H., J. B. Wilbur, and S. Utku: "Elementary Structural Analysis," 3d ed., McGraw-Hill
Book Company, New York, 1976.
1-78 Structural Analysis

4. West, H. H.: "Analysis of Structures," 2d ed., John Wiley & Sons, Inc., New York, 1989.
5. Przemieniecki, J. S.: "Theory of Matrix Structural Analysis," McGraw-Hill Book Company, New
York, 1968.
6, Cook, R. D.: "Concepts and Applications of Finite Element Analysis," 2d ed., John Wiley & Sons,
Inc., New York, 1981.
7. Zienkiewicz, 0. C.: "The Finite Element Method," 3d ed., McGraw-Hill Book Company, New
York, 1981.
8. Lightfoot, E.: "Moment Distribution," E. & F.M. Spon Ltd., London, 1961.
9. McGuire, W., and R. H. Gallagher: "Matrix Structural Analysis," John Wiley & Sons, Inc., New
York, 1979.
10. Holzer, S. M.: "Computer Analysis of Structures," Elsevier, New York, 1985.
11. Pian, T. H. H., and P. Tong: Basis of Finite Element Methods for Solid Continua, Int. ]. Numer.
Methods Eng., vol. I, pp. 3-28, 1969.
12. Irons, B. M., and A. Razzaque: Experience with the Patch Test for Convergence of Finite Ele-
ments, in A. K. Aziz (ed.), "The Mathematical Foundations of the Finite Element Method with
Applications for Partial Differential Equations," Academic Press, Inc., New York, 1972.
13. Turner, M. J., et al.: Stiffness and Deflection Analysis of Complex Structures. J. Aeronaut. Sci.
September 1956.
14. Fraeijs de Veubeke, B.: Displacement and Equilibrium Methods in the Finite Element Method,
in 0. C. Zienkiewicz and G. S. Holister (eds.), "Stress Analysis," John Wiley & Sons, Inc., New
York, 1965.
15. Felippa, C. A.: "Refined Finite Element Analysis of Linear and Nonlinear Two-Dimensional
Structures," Structural Engineering Laboratory, University of California, Berkeley, 1966.
16. Bazeley, G. P., et al.: Triangular Elements in Plate Bending-Conforming and Nonconforming
Solutions, Proc. 1st Conf Matrix Methods in Structural Mechanics, Wright Patterson Air Force
Base, 1965.
17. Clough, R. W., and J. L. Tocher: Finite Element Stiffness Matrices for Analysis of Plate Bending,
Proc. 1st Conf Matrix Methods in Structural Mechanics, Wright Patterson Air Force Base, 1965.
18. Bell, K.: Triangular Plate Bending Elements, in I. Holand and K. ,Bell (eds.), "Finite Element
Methods in Stress Analysis," Tapir, Trondheim, Norway, 1969.
19. Melosh, R. J.: Basis of Derivation of Matrices for the Direct Stiffness Method,]. AIAA, July 1963.
20. Bogner, F. K., R. L. Fox, and L. A. Schmit: The Generation of Interelement Compatible Stiffess
and Mass Matrices by Use of Interpolation Formulae, Proc. 1st Conf Matrix Methods in Structural
Mechanics, Wright Patterson Air Force Base, 1965.
21. Clough, R. W., and C. A. Felippa: A Refined Quadrilateral Element for the Analysis of Plate Bend-
ing, Proc. 2d Conf Matrix Methods in Structural Mechanics, Wright Patterson Air Force Base,
1968.
22. Clough, R. W., and C. P. Johnson: A Finite Element Approximation for the Analysis of Thin Shells,
]. Solids Structures, vol. 4, 1968.
23. Strickland, G. E., and W. A. Joden: A Doubly Curved Triangular Shell Element, Proc. 2d Conf
Matrix Methods in Structural Mechanics, Wright Patterson Air Force Base, 1968.
24. Connor, J., and C. A. Brebbia: Stiffness Matrix for Shallow Rectangular Shell Element, J. Eng.
Mech. Div. ASCE, October 1967.
25. Pecknold, D. A., and W. C. Schnobrich: Finite-Element Analysis of Skewed Shallow Shells, J.
Struct. Div. ASCE, April 1969.
26. Argyris, J. H.: Continua and Discontinua, Proc. 1st Conf Matrix Methods in Structural Mechanics,
Wright Patterson Air Force Base, 1965.
27. Pawsey, S. F., and R. W. Clough: Improved Numerical Integration of Thick Shell Finite Elements,
Int.]. Numer. Methods Eng., vol. 3, pp. 575-586, 1971.
28. Zienkiewicz, 0. C., R. L. Taylor, and J.M. Too: Reduced Integration Technique in General Anal-
ysis of Plates and Shells, Int.]. Numer. Methods Eng., vol. 3, pp. 275-290, 1971.
29. Lopez, L. A., J. Urzua, R. H. Dodds, and D.R. Rehak: Finite: A Polo II Subsystem for Structural
Mechanics, Civil Engineering Systems Laboratory, University of Illinois, Urbana, October 1976.
 Section 2
Computer Applications in
Structural Engineering
STEVEN J. FENVES
Sun Company University Professor of Civil Engineering, Carnegie-
Mellon University

INTRODUCTION
1. The Structural Design Process In order to place computer applications into proper
context, a brief review of structural design is presented first. A highly simplified repre-
sentation of the structural design process is shown in Fig. 1. It begins with information
concerning the client's needs and resources, limitations to be imposed on the project
(e.g., technical, social, legal), and criteria to be used to evaluate designs. It proceeds
through a series of activities of project planning, preliminary design, analysis, and pro-
portioning. Each process generates additional information to be used by the succeeding
process. Typically, the information produced by each process is evaluated for consis-
tency, economic and technical feasibility, and the like; if the results turn out to be unsat-
isfactory, one or more of the preceding processes must be repeated. Certain decision
points are explicitly shown in the figure: an iterative analysis-proportion process must
converge to some acceptable tolerance before the design can proceed, and the final
design must satisfy the original technical and economic criteria (or, ideally, be optimal
under the limitations and constraints imposed) before the final information, that is, the
design documents, can be produced. This review serves two purposes: (1) to illustrate
that computer programs are used for essentially every phase of each of the design proc-
esses shown in this figure; and (2) to set the framework for discussing the development
of a computer program.
2. The Program-Development Process The use of the computer requires that problem
solving be separated into two phases: development, during which the program is gener-
ated, and production, involving repeated use of the program. Program development is a
major design activity which follows a process similar to the structural design process, as
illustrated in Fig. 2. The degree to which the steps given below are followed will naturally
depend on the scope and importance of the program, the resources available, and the
intended use and distribution of the program. These steps must be followed even if the
program is obtained elsewhere and it is only intended to modify or adapt it to the orga-
nization's needs, resources, and practices.
2-1