FULLTEXT01 Hydraulique

Load Unit Geometry Optimization
for Heavy Duty Machinery
Master Thesis
Ted Samuelsson
Department of Physics
Umeå Universitet
June 12, 2015
Supervisor Examiner
Bobbie Frank Claude Dion
[email protected] [email protected]
Abstract
The construction equipment industry is developing at a fast pace, increasing the
expectation on the next-generation machines. Wheel loaders and backhoe loaders are
part of this evolution and all subsystems in those machines need to be developed to
meet the high demands in energy efficiency and productivity.
One of the most important parts of the wheel loader is the loading unit. This
is traditionally designed using highly experienced engineers and CAD software. To
simplify the early stages of this process was an optimization tool developed to gen-
erate a design outlay. The optimization will minimize the mass of the linkage since
unnecessary weight will lower the efficiency. The minimum can be found by moving
the joints and adjusting the shape of the device. The optimization will also include
constraints to assure the correct performance of the final linkage.
Since there are a high number of design variables, a gradient-based optimization
method was used. A finite element solver was also implemented to calculate the forces
and stresses in the linkage.
The linkages studied in this report are one from a typical wheel loader and one
from a backhoe loader. Since these machines are extremely versatile, and used for
many different tasks, two sets of constraints are compiled. One of the constraint
sets yields a linkage suitable for machines only equipped with bucket, while the other
results in an all-round linkage suitable for most tools and applications.
The optimized linkages are compared to existing devices. The results show that
there are some improvements possible and that the software could be used to help
designers. However, the optimization problem is hard to solve due to non-smooth
constraints functions and numerical instabilities. This issue could be overcome by
different means, like using automatic differentiation, a non-gradient based optimiza-
tion method, decreasing the number of constraints or decreasing the number of design
variables.
i
Sammanfattning
Utvecklingen av anläggningsmaskiner sker i snabb takt och detta ökar förväntningarna
på framtidens maskiner. En stor andel av alla anläggningsmaskiner är hjullastare och
traktorgrävare och alla delsystem på dessa maskiner måste följa med i utvecklingen.
En av de viktigaste delarna på en hjullastare är lastaggregatet. Det designas
traditionellt med hjälp av CAD mjukvara och mycket erfarna konstruktörer. För att
underlätta denna process har en optimeringsrutin utvecklats, som generarar ett design
förslag. Optimeringen minskar länkagets massa genom att flytta lagringspositioner
och ändra delarnas dimensioner. Detta ökar effektiviteten hos maskinen eftersom
den slipper köra runt på onödig vikt. Optimeringen innehåller även villkor för att
säkerställa god prestanda hos det optimerade aggregatet.
Eftersom det ingår väldigt många designvariabler i optimeringen används en gra-
dientbaserad metod. En finita element approximation används för att beräkna krafter
och spänningar i länkaget.
De länkage som undersöks i detta projekt är ett typsikt hjullastaraggregat och
ett typiskt traktorgrävaraggregat. Eftersom dessa maskiner är väldigt mångsidiga
sammanställdes två olika uppsättningar av villkor. Den ena uppsättningen används
för att optimera ett aggregat som endast ska användas med skopa, medan den andra
uppsättningen används för att ta fram ett mer mångsidigt aggregat avsätt för att
kunna klara av de flesta situationer och verktyg.
De optimerade lastaggregaten är jämförda med produktionsaggregat och det visar
sig att vissa förbättringar är möjliga. Slutsattsen är att optimeringsrutinen kan bli
ett bra hjälpmedel för konstruktörer men att den behöver lite mer verifiering. Villko-
rsfunktionen som optimeringen måste lösa är inte helt slät vilket är ett problem för
en gradientbaserade metod och dessutom finns vissa numeriska instabiliteter. Dessa
svårigheter kan undkommas på olika sätt, t.ex. genom att använda automatisk de-
rivering, byta optimeringsalgoritm, minska antalet villkor eller minska antalet vari-
abler.
ii
Contents
1 Introduction 1
1.1 Problem Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Machine Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Linkage Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2 Optimization problem 7
2.1 Mathematical formulation . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Design Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3.1 Geometrical . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3.2 Force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3.3 Stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 Objective Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4.1 Preliminary Functions . . . . . . . . . . . . . . . . . . . . . . 15
2.4.2 Final Function . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3 Method Overview 19
3.1 Optimization Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Stress Calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4 Result and Analysis 25

4.1 Z-bar linkage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
iii
4.2 TPC linkage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.3 Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.4 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.4.1 Objective and Constraint Function . . . . . . . . . . . . . . . 36
4.4.2 Optimization Routine . . . . . . . . . . . . . . . . . . . . . . 37
4.4.3 FEM Mesh Resolution . . . . . . . . . . . . . . . . . . . . . . 39
5 Discussion 41
5.1 Result Reliability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.2 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
A Design Variables 45
B Finite Differences 49
C Additional Results 51
D Function Analysis 53
iv
List of Figures
1-1 Schematic figure of a wheel loader. . . . . . . . . . . . . . . . . . . . 3

1-2 The common Z-bar linkage set-up. . . . . . . . . . . . . . . . . . . . . 4
1-3 Volvo invented TP linkage. . . . . . . . . . . . . . . . . . . . . . . . . 5
1-4 Common backhoe loader linkage. . . . . . . . . . . . . . . . . . . . . 6
2-1 Definitions of attachment angles (AA) and hinge pin height (HPH). . 10
2-2 Possible collision for a both a Z-bar and TPC linkage. . . . . . . . . . 12
2-3 Geometric performance constraints. . . . . . . . . . . . . . . . . . . . 13
2-4 Break-out torque definition. . . . . . . . . . . . . . . . . . . . . . . . 13
2-5 Optimization of Z-bar (above) and TPC (below) linkage with mass as
objective function. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2-6 Optimization of Z-bar with respect to tipping force. . . . . . . . . . . 16
2-7 Forces acting on a stationary machine. . . . . . . . . . . . . . . . . . 17
3-1 Simplified schematic of the optimization routine. . . . . . . . . . . . . 20

3-2 Showing elements of initial and optimized Z-bar linkage. . . . . . . . 21
3-3 The three load cases used in the optimization. . . . . . . . . . . . . . 23
4-1 Initial state of Z-bar linkage. . . . . . . . . . . . . . . . . . . . . . . . 27

4-2 Optimized Z-bar linkage with all-round conditions. . . . . . . . . . . 28
4-3 Optimized Z-bar linkage with bucket constraints. . . . . . . . . . . . 29
4-4 Initial state of TPC linkage. . . . . . . . . . . . . . . . . . . . . . . . 32
4-5 Optimized TPC linkage with all-round conditions. . . . . . . . . . . . 33
4-6 Optimized TPC linkage with bucket constraints. . . . . . . . . . . . . 34
v
4-7 Analysis of the objective function for a Z-bar linkage. . . . . . . . . . 37
4-8 Analysis of the constraint function for Z-bar linkage. . . . . . . . . . 38
4-9 Influence of FEM mesh resolution on execution time and calculated
mass. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
B-1 Optimized TPC linkage with all-round conditions. . . . . . . . . . . . 49
C-1 All-round optimized Z-bar linkage. . . . . . . . . . . . . . . . . . . . 51

C-2 Bucket optimized Z-bar linkage. . . . . . . . . . . . . . . . . . . . . . 52
C-3 All-round optimized TPC linkage. . . . . . . . . . . . . . . . . . . . . 52
C-4 Bucket optimized TPC linkage. . . . . . . . . . . . . . . . . . . . . . 52
D-1 Z-bar constraint functions for problematic design variables. . . . . . . 53

D-3 Evaluation of TPC objective function. . . . . . . . . . . . . . . . . . 55
vi
List of Tables
2.1 Minimal dump and rollback angles at different hinge pin heights. . . . 10
2.2 Critical geometrical collisions which need to be considered during the
optimization of a Z-bar linkage. . . . . . . . . . . . . . . . . . . . . . 11
2.3 Critical geometrical collisions which needs to be considered during the
optimization of a TPC linkage. . . . . . . . . . . . . . . . . . . . . . 11
2.4 Performance metrics used to validate the correct function of the linkage. 12
2.5 Breakout torque constraints for an all-round and a bucket machine at
different linkage positions. . . . . . . . . . . . . . . . . . . . . . . . . 14
2.6 Lift force constraint for both bucket and all-round machine. . . . . . 14
4.1 Resulting masses for optimized Z-bar. . . . . . . . . . . . . . . . . . 26

4.2 Force, torque and attachment angle constraint fulfilment for Z-bar link-
age. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.3 Performance constraint fulfilment for Z-bar linkage. . . . . . . . . . . 26
4.4 Resulting masses for optimized TPC. . . . . . . . . . . . . . . . . . . 30
4.5 Force, torque and attachment angle constraint fulfilment for TPC link-
age. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.6 Performance constraint fulfilment for TPC linkage. . . . . . . . . . . 31
4.7 Linkage masses inclusive cylinders. . . . . . . . . . . . . . . . . . . . 35
4.8 Adjusted linkage masses. . . . . . . . . . . . . . . . . . . . . . . . . 35
4.9 Sensitivity analysis for Z-bar linkage. . . . . . . . . . . . . . . . . . . 38
4.10 Sensitivity analysis for TPC linkage. . . . . . . . . . . . . . . . . . . 39
A.1 Design variables for Z-bar linkage. . . . . . . . . . . . . . . . . . . . . 46
vii
A.2 Design variables for TPC linkage. . . . . . . . . . . . . . . . . . . . . 47
viii
CHAPTER 1. INTRODUCTION
Chapter 1
Introduction
This project is carried out at Volvo Construction Equipment who are developing and
manufacturing wheel loaders, excavators, articulated haulers and other types of heavy
duty machinery. The Eskilstuna site, were this project is based, is primary focusing
on wheel loaders, as is this report.
High energy efficiency and productivity is one of the most important attributes
of a modern wheel loader. To achieve this in next-generation machines must all
subsystems be optimized to fulfil these expectations. However, the loading unit (i.e.
the device holding the bucket, also called linkage or working device) of the current
machines are traditionally made using CAD tools and experienced designers [1, 2].
Shorter project times, higher material costs and higher demands on performance call
for a more effective design process. Therefore aims this report at improving the early
stages of this process by implementing an design optimization tool. The tool will
optimize the bearing positions and linkage dimensions with respect to a specified
objective. This will streamline the process and at the same time strive for a better
result.
1.1 Problem Definition
The optimization problem consists of adjusting all bearing positions of the loading
unit to minimize the linkage mass. Furthermore, the width and height of all arms
1
1.2. RELATED WORK
should be adjusted to hold any likely external load. An attempt of this has been done
by Volvo CE in 2009 [1] and is used as a starting point for this work.
The linkages considered in this report will be one for a typical wheel loader and
one for a backhoe loader. These two linkages have different design, which will be
further described in section 1.4.
1.2 Related Work
There is a great interest in linkage optimization for heavy duty machines and some
of the research will be presented here. An optimization of lift cylinders placement
and size has been done by Volvo CE in 2003 [3]. A complex optimization method
and an exhaustive search method were used. Both worked well for the problem which
only consists of four design variables (two positions in a two dimensional domain).
Another attempt using an exhaustive search was done in [4], optimizing the power
consumption. Others have also tried to optimize the design of the linkage using
a genetic algorithm on a backhoe loader [5] and a wheel loader [6]. A sequential
quadratic programming method was used in [7].
However, the above-mentioned authors do not consider the internal stresses in
the linkage and treat the problem as two dimensional. This approach is not good
enough to create a trustworthy and usable result but could be used for evaluating
new concepts. A 3D parametric finite element approach has been performed, but
only by minimizing the material on an existing excavator working device (no coupling
between joints) [8].
The method described by Ekevid in [1] is three-dimensional and has both joint
positions and arm dimensions as parameters resulting in more than hundred design
variables, more than 5 times the amount of the other papers. The stresses in all
linkage parts are calculated using a finite element method (FEM) and the result
looks very promising. However, the project did not really get all the way to achieve
a usable software. One of the largest issues was the objective function, which did
not include all required physics. This work adresses this issue and the changes to the
2
function are presented in section 2.4.
1.3 Machine Design
Wheel loaders are multi-purpose vehicles which are used for many different tasks, e.g.
transportation of gravel, pallet handling or logging applications. This is especially
true for small and mid-sized machines. They are often equipped with an attachment
bracket for easy exchange of buckets, forks, handling arms, logging grapples, etc.
The largest vehicles are, to the contrary, often used for a specific task, like bucket
handling.
The backhoe loader is also a versatile vehicle, but in general somewhat smaller.
The backhoe loader has, in addition to the front working device, also a rear-mounted
excavator linkage. To simplify the comparison between the linkages, only one kind of
machine will be used in the optimizations. This is possible since the machines are very
similar, not least since this report only considers the front devices. Therefore, the 20
ton Volvo L120H wheel loader is chosen as the reference machine for the optimization.
Loader Linkage
Operator Cabin
Counter
Weight
Front Frame Hinge Rear Frame
Figure 1-1: Schematic figure of a wheel loader.
The complete machine can be divided into a small number of basic parts, see
Figure 1-1. There is a front and a rear frame joined by a hinge used for the articulated
steering. On the rear frame a cabin and a counter weight are attached. The counter
weight prevents the machine from tipping forward when a large force is exerted on
3
1.4. LINKAGE DESIGNS
the bucket, e.g. when a heavy load is lifted. The rear frame includes most of the
parts related to the engine and hydraulic system. On each frame there is also one
pair of wheels and their related axles. The loader linkage and bucket are positioned
on the front frame.
1.4 Linkage Designs

There are many different linkage designs available for these kind of machines. They
differ in complexity, weight and performance. They can, with the same hydraulic
system, have different lift force, breakout torque and parallel alignment. Good parallel
alignment means that the tool keeps its angle to the ground while the linkage is raised
or lowered, important when using forks and logging grapples.
The most common linkage on a wheel loader is called the Z-bar, see Figure 1-2.
It is used by most manufacturers and is standard on Volvo’s largest models.
E
O F
C
D
J
P
G A
Figure 1-2: The common Z-bar linkage set-up.
The design owns its popularity to a simple and light construction with few joints,
but also because of its strong lift force. The downsides are bad parallel alignment and
low breakout torque at some positions. This type of linkage works great for bucket
applications, but less for fork handling.
A linkage used on most of Volvo’s loaders is called TP (Torque Parallel), see
Figure 1-3. It is a heavier and more complicated setup but offers excellent parallel
4
alignment and at the same time high breakout torque throughout the lifting range.
This linkage is suitable when good parallel alignment is of importance, like fork and
logging grapple applications. However, the higher weight decreases the maximum
possible bucket load. This linkage will not be considered in this report more than as
a reference.
O
C H
E D I
J
F
B
P
A
Figure 1-3: Volvo invented TP linkage.
The third linkage, common on backhoe loaders, is called TPC and utilizes a simple
uncoupled design. This linkage has no mechanical coupling adjusting the parallel
alignment of the bucket when the unit is raised or lowered. Instead it often uses
hydraulic coupling to address the problem.
This linkage can, contrary to the others, roll-back the bucket enough to dump
material over the cabin. This is a safety issue that needs to be dealt with.
The version considered in this report will be a single boom unit, i.e. it only has
one lift arm instead of the more common set-up with two arms. The boom will also
feature a hollow-cross section and one centred lift cylinder. This yields a compact
and light-weight linkage without decreasing the performance.
5
1.4. LINKAGE DESIGNS
O E
C
P
F
A
Figure 1-4: Common backhoe loader linkage.
6
CHAPTER 2. OPTIMIZATION PROBLEM
Chapter 2
Optimization problem
The optimization problem consists in minimizing the total weight of the machine and
at the same time improve the operating performance, like lift force and maximum
dump angle. To achieve this, the placement of the joints in the loading unit (defined
in Figures 1-2 and 1-4) are moved towards their optimum, with respect to a specified
goal. All joints are bounded to boxes (often, but not restricted to, 2 dimensional).
At the same time the dimensions of the different parts in the linkage are adjusted to
confine the von Mises stress [9]
r
1
σevM = (σx + σy )2 + (σy + σz )2 + (σz + σx )2 . (2.1)
2
The linkage weight is then calculated and the counter weight is sized to satisfy the
tipping force.
7
2.1. MATHEMATICAL FORMULATION
2.1 Mathematical formulation
Let f (x) be the objective function that should be minimized, where x exists in the
domain Ω. Then the optimization problem can be stated as
minimize f (x)
subject to
c(x) ≤ 0
ceq (x) = 0 (2.2)
xlb ≤ x ≤ xub .
where c(x) ≤ 0 and ceq (x) = 0 are the inequality and equality constraints to the prob-
lem, respectively. The upper and lower bound are defined by xlb and xub . Inequality
constraints can be rewritten as equality constraints on the form c(x) + s = 0, where
s ≥ 0 are called slack variables.
Most gradient based optimization algorithms for constrained non-linear problems

use the Lagrangian
L(x, λ, λlb , λub , λeq , s, slb , sub ) = f (x) + λT (c(x) + s) + λTeq (ceq (x) + s)
(2.3)
+ λTlb (xlb x + slb ) + λTub (xub x + sub )
where λ ≥ 0, λlb ≥ 0,λub ≥ 0 and λeq ≥ 0 are called Lagrange multipliers. The s, slb
and sub are the slack variables.
Minimizing the Lagrangian is considerably easier since the constraints are in-
corporated in the Lagrangian function. At an optimum, the Karush-Kuhn-Tucker
conditions [10]
∂L ∂L ∂L
= 0, = 0, =0 (2.4)
∂x ∂λ ∂s
λ ≥ 0, s ≥0
c(x) ≤ 0, ceq (x) = 0 (2.5)
xlb − x ≤ 0, x − xlb ≤0
8
λT (c(x) + s) = 0
λTlb (xlb − x + slb ) = 0 (2.6)
λTub (x − xub + sub ) = 0
must be satisfied [11].
2.2 Design Variables

The model parameters, which should be optimized, are called design variables. These
should describe the linkage without any additional information. They will, under the
optimization, be adjusted to satisfy equation 2.2 and will then be returned as the
solution.
For this linkage optimization there are four different groups of variables: Joint
positions, cylinder dimensions, element sizes and counter weight size. The joint po-
sitions describe the simple shape and lengths of all linkage parts. They are defined
in each linkage parts local coordinate system. Parameters for the cylinders are for
instance their extended and contracted lengths as well as diameters. The element
sizes, which are the largest part of the design variables, describe the width and height
of all linkage parts. They will also describe the thickness were it applies, e.g. the
hollow boom of the TPC. The number of variables in this group is dependent on the
FEM mesh resolution. The last group only consists of the counter weight mass.
The variables for the joints, cylinder lengths and counter weight are stated in
Appendix A.
2.3 Constraints
The linkage needs to fulfil a certain number of constraints related to geometrical
conditions, lift forces and stresses. To deal with the multi-purpose behavior of the
machine, two sets of constraints have been compiled. One for an all-round machine,
which need high overall lift performance and breakout torque, and one for a bucket
handling machine, which only needs high performance at ground level when operating
9
2.3. CONSTRAINTS
in a pile of material. The two different sets of constraints will henceforth be referred
to as all-round and bucket.
2.3.1 Geometrical
The geometrical constraints consist of three subgroups: Attachment angles, collisions

and linkage performances. None of these will differ between the two versions of the
constraints. However, there will be some difference between the Z-bar and the TPC
linkage due to the variation in design.
Attachment Angles
The machine needs to be able to tilt the bucket sufficiently so that the operator can
easily empty the material. It must also be able to tilt the bucket backwards to keep
the material in the bucket while travelling on rough roads. To assure these properties
the linkage must manage the dump and roll-back angles (see Figure 2-1) stated in
Table 2.1.
Table 2.1: Minimal dump and rollback angles at different hinge pin heights.
HPH [m] Min rollback angle [°] Min dump angle [°]
min -55 35
0.58 -55 49
2.00 -55 60
max -45 58
Roll-back
Angle
Dump
Angle
Hinge Pin
Height
Figure 2-1: Definitions of attachment angles (AA) and hinge pin height (HPH).
10
Collisions
To assure that the optimized design works satisfactorily, no linkage parts are allowed
to collide with each other when the machine is operated. Therefore, it is important
to localize the critical linkage positions were this could happen. These different cases
are visualized in Figure 2-2 and Tables 2.2 - 2.3.
Table 2.2: Critical geometrical collisions which need to be considered during the
optimization of a Z-bar linkage.
No Description Condition [m]

A1 Bucket A ↔ Bucket ridge > 0.10
A2 Bucket J ↔ Bucket ridge > 0.10
A3 GDF (D) ↔ Bucket ridge > 0.20
A4 GDF (DF) ↔ Bucket top > 0.20
A5 GDF (G) ↔ Front axle > 0.34
A6 Lift cylinder ↔ Front axle > 0.25
Table 2.3: Critical geometrical collisions which needs to be considered during the
optimization of a TPC linkage.
No Description Condition [m]

B1 Bucket A ↔ Bucket ridge > 0.05
B2 Bucket J ↔ Bucket ridge > 0.05
B3 Tilt cylinder (EF) ↔ DEG (G) > 0.20
B4 Tilt cylinder (E) ↔ Boom (OC) > 0.20
B5 Lift cylinder (C) ↔ DEG (D) > 0.34
B6 Lift cylinder ↔ Front axle > 0.25
In addition to the constraints stated above, the cylinders inner diameter must not
become larger than the outer diameter since this would yield a negative mass.
Performance
The linkage needs to handle some basic tasks. For instance, it needs to be able to lift
the bucket enough to dump in a hauler and also lower the bucket enough for some
digging capabilities. These situations are defined in Table 2.4 and Figure 2-3.
11
2.3. CONSTRAINTS
B3
A4
B4
A3
B5
A2 B2
A6 A5 B6
A1 B1
Figure 2-2: Possible collision for a both a Z-bar and TPC linkage.
Table 2.4: Performance metrics used to validate the correct function of the linkage.
Measure Value [m]

Digging depth < -0.10
Reach > 1.20
Dump Clearance > 2.88
2.3.2 Force
The force constraints treat the different forces and torques the linkage need to be
able to handle. Those are divided into two sub categories: Lift forces and breakout
torques.
Breakout Torque
The breakout torque is the maximum torque available at the bucket tip and is defined
as
M = |r1 × F | = |r2 × R| (2.7)
where F is the reaction force when a tilt cylinder force is present, see Figure 2-4. R
is the corresponding force in the GJ link.
The breakout torque is mostly used at ground level when filling the bucket. How-
ever, when the machine is used with other equipment, like logging grapples, other
positions are also of interest. The constrains are collected in Table 2.5.
12
Reach
Clearance
Dump
Digging Depth
Figure 2-3: Geometric performance constraints.
R
F
r2
r1
Figure 2-4: Break-out torque definition.
Lift Force
Lift force is defined as the force exerted by the linkage on the load, measured at the
loads center of gravity. To achieve the wanted lift behavior three constraints for the
minimum lift force are sufficient. These are collected in Table 2.6.
The tipping force is also of interest, in addition to of the constraint above. A

large tipping force makes it possible to lift a heavy load without risking that the rear
wheels lift of the ground. The tipping force is calculated with the linkage positioned
so that the bucket load is as far forward as possible. For this size of machine, the
tipping force should be at least 13380 N including the bucket.
13
2.4. OBJECTIVE FUNCTION
Table 2.5: Breakout torque constraints for an all-round and a bucket machine at
different linkage positions.
HPH [m] Angle [°] Torque

All-round Bucket
0 min 1.00 0.45
0 0 1.00 1.12
0 max 1.00 0.49
max min 1.00 0.70
max 0 1.00 0.50
The values have been normalized for each
position to protect proprietary information.
Table 2.6: Lift force constraint for both bucket and all-round machine.
HPH [m] Lift force

All-round Bucket
min 1.00 0.88
2 1.00 0.97
max 1.00 0.88
The values have been normalized for each
position to protect proprietary information.
2.3.3 Stress
To assure that no parts of the linkage break must the von Mises stress, due to twisting
and bending, be under a predefined value. This value is chosen according to the
materials properties. During the optimization the stresses in the linkage are calculated
by a finite element method described in section 3.2.
2.4 Objective Function
The function defining the goal of the optimization is called objective function. Given
a set of design variables the function should return a scalar value, representing the
fitness of the input. Often an optimization has more than one goal, e.g. both min-
imizing the mass and maximizing the rigidity of a mechanical structure. This can
be handled by combining the goals into one objective function. The different goals
14
can then be weighted individually dependent on their importance. For n goals the
function becomes
n
X
f (x) = wi gi (x) (2.8)
i=1
where g(x) is the individual objective functions and wi are each goals weight.
The weighting process can however be an issue since the exact relative importance
between the goals often are hard to estimate or even unknown. To avoid this issue,
some goals can be converted to constraints. Then the issue instead becomes choosing
the constraint limitation value, which sometimes is easier [12].
2.4.1 Preliminary Functions
The first model minimized the total mass of the linkage, including cylinders. The
result from that kind of formulation is not satisfactory even though it becomes light
weight. The problem is that there are no tendencies to move the center of gravity
towards the front axle and thus increasing the tipping force. Instead the linkage
becomes rather long and slim, see Figure 2-5. This is not the appearance, nor the
performance, a linkage should have and therefore another objective function was
developed.
2
z
0
−3 −2 −1 0 1 2 3 4 5
3
x
2
z
Figure 2-5: 0Optimization

−3 −2
of −1
Z-bar (above)
0 1
and TPC
2
(below)
3
linkage5 with mass as
4
objective function. x
15
The second function maximized the vehicles tipping force, i.e maximized the pos-
sible load in the bucket before the vehicles rear wheels ease from the ground. This
is done by calculating each linkage parts contribution to a moment around the front
wheel. This causes the optimization to push the linkage towards the rear. The place-
ments of the joints looks better with this model, however all elements behind the
tipping point will be maximized in size resulting in a very odd looking and heavy
linkage, see Figure 2-6.
0
−4 −3 −2 −1 0 1 2 3 4 5
Figure 2-6: Optimization of Z-bar with respect to tipping force.
2.4.2 Final Function
The final function uses the above concepts and adds a new part to the optimization, a
counter weight. This yields the correct behavior; with a beefy, short and back-pushed
linkage. Since the size of the counter weight decides the tipping force, it must be
considered during the optimization process. In a normal design process the counter
weight is actually designed in parallel with the linkage to yield the expected tipping
force.
The tipping load is now set as a constraint. Consequently, the linkage must be
able to handle a specific load and the center of gravity will be pushed backwards. The
optimization function then consists of minimizing the total weight of the linkage and
the counter weight. Since the counter weight is farther from the tipping point (triangle
in Figure 2-7) it will be cheaper to add weight there than on the linkage. Since the
16
Fload
Flink
Frear Fcw
Figure 2-7: Forces acting on a stationary machine.
function is a summation of two masses these can be individually weighted, which will
be considered in the result section. However, to minimize the actual physical weight
of the machine the ratio should be set to one. To minimize the product cost another
ratio could be used to move weight to the cheaper manufactured counter weight. The
final function is
X
f (x) = w1 mi + w2 mcw (2.9)
i
where mi is the mass of linkage element i, mcw is the counter weight mass and
w = (w1 , w2 ) are the goal weights.
17
THIS PAGE INTENTIONALLY LEFT BLANK
18
CHAPTER 3. METHOD OVERVIEW
Chapter 3
Method Overview
The optimization routine will under the optimization evaluate the objective function
and check the constraints. The first task is accomplished by passing the design
variables to the objective function which then evaluates and returns the fitness value.
The constraints are checked by passing the design variables to a constraint evaluation
function. This function checks all constraints by evaluating the different constraint
types individually. The stress constraint evaluation part invokes a FEM routine
for the stress calculations. The FEM routine also generates important information
used by the force constraint evaluation function. The constraints are then combined
together and returned to the main optimization routine, which uses the information to
decide if the optimization should terminate. Termination is done if the current design
variables satisfy the constraints and represent a minimum for the objective function,
otherwise the routine is iterated with modified design variables. An overview of the
system is represented in Figure 3-1.
3.1 Optimization Algorithm
How to move the design variables inside the calculation domain, at each iteration, de-
pends on the chosen optimization algorithm. There exist several algorithms to choose
from for constrained non-linear optimization, with different advantages. For the task
in this report a gradient based algorithm has been chosen, called interior-point. It
19
3.1. OPTIMIZATION ALGORITHM
Material
Properties FEM R
model
Load Cases
d
DV
Geometrical Stress Force
Constraints Constraints Constraints
Evaluation Evaluation Evaluation
CG CS CF
DV DV DV
Constraint Evaluation
DV
Design Terminate? Optimized

Variables Optimization Routine Yes
Design
No Variables
DV
Objective Function
Figure 3-1: Simplified schematic of the optimization routine.
moves the design variables in the negative objective gradient of the current point.
If an expression for the gradient is unknown can it often be approximated by some
numerical method. In this report, the gradient is approximated with central finite
differences. Forward finite differences were investigated since they are less compu-
tationally heavy but they introduced numerical disturbances in the calculation, see
Appendix B. The gradient-based method is a good choice due to its robustness and
possibility to handle the large number of variables. However, this method only works
on problems with an objective function and constraint functions that have continuous
derivatives, see Section 4.4.1 for an analysis of the functions used. Other alternatives
considered were the Complex method, genetic algorithms and exhaustive search. Both
the exhaustive search and the complex method are not well suited for the size of this
system, and were therefore discarded. The genetic algorithm is a suitable choice but
20
was unfortunately not tested.

Feeding the algorithm with analytic Jacobian and Hessian will decrease the compu-
tational time considerably. However, the expressions are often complex and sometimes
they become too large to be used. A work-around is to use automatic differentiation
software, e.g. the ADMAT and ADiMat toolboxes for MATLAB [13]. This technique
automatically differentiates a function by using the chain rule on its elementary arith-
metic operations. The calculations have an accuracy of working precision and uses
only a constant factor operations more than the original function, often resulting in
shorter computational time than using analytic expressions [14]. Such software would
also add robustness and accuracy to the calculations of the derivatives compared with
usual finite differences, but were not used due to the limited project time.
3.2 Stress Calculations

The stresses in the linkage are calculated using a finite element method. The linkage
parts are divided into small beam elements, as seen in Figure 3-2. The calculation
is provided with information about the linkage: details about all part positions and
material properties. It also includes all external forces present and boundary condi-
tions. The displacement of each element is returned, which can be recalculated to
stresses and forces in the linkage.
Figure 3-2: Showing elements of initial and optimized Z-bar linkage.
21
3.2. STRESS CALCULATIONS
The elements are approximated by Euler-Bernoulli beam elements [15]. It is a

classical beam theory for elastic beams subject to forces yielding small deflections.
The Euler-Bernoulli equation is defined as
d2 d2 w
(EI )=q (3.1)
dx2 dx2
where w is the deflection, q is the force per unit length, I is the second moment of
inertia and E is Young’s modulus [9]. For a complete calculation the displacement
equation
d du
(EA ) = U (3.2)
dx dx
and the equation for Saint-Venant’s torsion
d2 dτ
(GKv ) = T (3.3)
dx 2 dx
are also needed. A is the cross-section area, U is the displacement along the beam, Kv
is the Saint-Venants torsion constant, G is shear modulus and T is the torque along
the beam. For a deeper insight on how to formulate the finite element approximation,
see [16].
With the use of the bending moment
d2 w
M = −EI (3.4)
dx2
the normal stress can be calculated as
N
σx =
A
Mb
σy = (3.5)
2Iz
Mh
σz =
2Iy
where h is the element height, b is the element width and N is the normal force.
22
The shear stress is calculated with
1 dM
τ= (3.6)
Wx dx
where Wx is the section modulus [9].

Since the stress in the linkage is dependent on the cross-section area will the
optimization add material where the stress is high and cut off where the stress is low.
This will make the optimization strive towards minimizing the stress and by that
satisfy the goal of decreasing the mass.
To calculate the maximum stress that the linkage may be subject to under normal
operation are a number of so called load cases used. Load cases are extreme situations
that could happen when a wheel loader is operated. They state external forces,
hydraulic cylinder pressures and linkage positions [17]. In addition to the load cases,
the stress is also measured during the evaluation of the force and torque constraints.
The load cases with the most significant influence, i.e. yielding the largest stresses,
on the linkage geometry were used during the optimization and are defined in Figure
3-3.
Fh Fv
Fv Fh
Fh Fd
Fd Fd Fd
Fh Fh Fh
Fs
(c) The machine is crowded

(a) The machine is crowded (b) The machine is crowded with level bucket at ground.
with the bucket in dump with level bucket at ground. Bucket is lifted with max-
position. External force Bucket is lifted with max- imal pressure. Assymet-
on bucket tip center. Not imal pressure. External ric external force on bucket
against mechanical stop. force on bucket tip. tip.
Figure 3-3: The three load cases used in the optimization.
23
3.2. STRESS CALCULATIONS
24
CHAPTER 4. RESULT AND ANALYSIS
Chapter 4
Result and Analysis
In this chapter, the results from the optimizations are presented. The mass of each
linkage part are stated as well as a constraint fulfilment review.
There will also be a comparison between the different types of linkages and con-
straints. However, the quality of the result will be further discussed in chapter 5. An
analysis of the robustness and performance are included in section 4.4.
4.1 Z-bar linkage

The final Z-bar linkage masses are stated in Table 4.1. The differences between the
constraints can easily be observed. In Table 4.2 and 4.3 can the constraint fulfilment
be studied. Additional result, containing data about variable positions and constraint
convergence, is presented in appendix C.
The initial shape of the linkage is showed in Figure 4-1 and the final design of the
linkages is visualized in Figure 4-2 and 4-3.
There is some unexpected behavior in the results. Even though the all-round
linkage has a higher breakout torque for all attachment angles and a higher lift force
for all hinge pin heights it also has a lower weight. This indicates that the optimization
is sensitive to the specified constraints. However, the hydraulics are much lighter for
the bucket version and the maximum dump and roll-back angles are a bit better.
25
4.1. Z-BAR LINKAGE
Table 4.1: Resulting masses for optimized Z-bar.
All-round Bucket
Part Original Mass
Optimized Mass Optimized Mass
Boom 100% 78% 80%
GDF 100% 58% 65%
GJ 100% 127% 156%
Lift Cylinder 100% 91% 80%
Tilt Cylinder 100% 170% 72%
Total, w/o cylinders 100% 77% 81%
Total, w/ cylinders 100% 83% 80%
The masses are presented in percent of original
mass to protect proprietary information.
Table 4.2: Force, torque and attachment angle constraint fulfilment for Z-bar linkage.
Value
Measure AA HPH Unit
All-round Bucket
Tilt Torque 0 0 1.88 1.00 kNm
max 0 1.09 1.40 kNm
min 0 1.00 1.00 kNm
0 max 1.39 1.28 kNm
Lift Force max min 1.00 1.11 kN
max 2 1.01 1.00 kN
max min 1.00 1.02 kN
Tipping Force - - 1.00 1.00 kN
Attachment Angle roll-back min 35.30 35.30 deg
dump min 56.60 56.60 deg
roll-back 0.7 53.30 54.60 deg
dump 0.7 55.00 55.10 deg
roll-back 2 60.00 60.00 deg
dump 2 55.00 55.10 deg
roll-back max 58.00 58.00 deg
dump max 40.90 42.00 deg
Table 4.3: Performance constraint fulfilment for Z-bar linkage.
Value [m]
Measure
All-round Bucket
Digging depth -0.10 -0.10
Reach 1.48 1.48
Dump Clearance 2.88 2.88
26
0
−3 −2 −1 0 1 2 3 4 5
0
−3 1
−2
−1
0 0
1
2 −1
3
4
5
Figure 4-1: Initial state of Z-bar linkage.
27
4.1. Z-BAR LINKAGE
0
−3 −2 −1 0 1 2 3 4 5
0
−3 1
−2
−1
0 0
1
2 −1
3
4
5
Figure 4-2: Optimized Z-bar linkage with all-round conditions.
28
0
−3 −2 −1 0 1 2 3 4 5
0
−3 1
−2
−1
0 0
1
2 −1
3
4
5
Figure 4-3: Optimized Z-bar linkage with bucket constraints.
29
4.2. TPC LINKAGE
4.2 TPC linkage

The TPC linkage results are stated in Table 4.1. The constraint fulfilment can be
observed in Tables 4.5 and 4.6. Constraint convergence and design variable domain
position are availble in appendix C.
Table 4.4: Resulting masses for optimized TPC.
All-round Bucket
Part Original Mass
Optimized Mass Optimized Mass
Boom 100% 53% 55%
GDF 100% 56% 67%
GJ 100% 20% 21%
Lift Cylinder 100% 109% 115%
Tilt Cylinder 100% 72% 97%
Total, w/o cylinders 100% 51% 53%
Total, w/ cylinders 100% 65% 72%
The masses are presented in percent of original
mass to protect proprietary information.
The final design of the linkages is visualized in Figure 4-5 and 4-6. Also on this
linkage there are some disturbance in the stress calculations.
Also for this linkage are the bucket version heavier, but it also performs better for
force and torque constraints at some positions than the all-round version. Another
interesting result is that the boom masses for the optimized linkages are about half,
compared with the original boom. The optimization algorithm does not take into
account for needed padding around the bearings and link ends. That would approxi-
mately add about 10% to the mass. Another reason for the original linkage boom to
be so much heavier is that it was designed to be equipped with a hydraulic parallel
alignment system. Last the optimizations are considering a molded boom in contrast
to the original welded boom.
30
Table 4.5: Force, torque and attachment angle constraint fulfilment for TPC linkage.
Value
Measure AA HPH Unit
All-round Bucket
Tilt Torque 0 0 1.20 1.00 kNm
max 0 1.31 2.19 kNm
min 0 1.00 2.45 kNm
0 max 1.87 2.75 kNm
Lift Force max min 1.15 1.32 kN
max 2 1.11 1.09 kN
max min 1.00 1.00 kN
Tipping Force - - 1.15 1.01 kN
Attachment Angle roll-back min 35.30 35.30 deg
dump min 56.50 56.50 deg
roll-back 0.7 50.00 50.00 deg
dump 0.7 56.50 56.50 deg
roll-back 2 63.30 63.30 deg
dump 2 56.50 56.50 deg
roll-back max 58.30 58.30 deg
dump max 40.50 40.50 deg
Table 4.6: Performance constraint fulfilment for TPC linkage.
Value [m]
Measure
All-round Bucket
Digging depth -0.21 -0.21
Reach 1.59 1.48
Dump Clearance 2.89 2.88
31
4.2. TPC LINKAGE
0
−3 −2 −1 0 1 2 3 4 5
0
−3 1
−2
−1
0 0
1
2 −1
3
4
5
Figure 4-4: Initial state of TPC linkage.
32
0
−3 −2 −1 0 1 2 3 4 5
0
−3 1
−2
−1
0 0
1
2 −1
3
4
5
Figure 4-5: Optimized TPC linkage with all-round conditions.
33
4.2. TPC LINKAGE
0
−3 −2 −1 0 1 2 3 4 5
0
−3 1
−2
−1
0 0
1
2 −1
3
4
5
Figure 4-6: Optimized TPC linkage with bucket constraints.
34
4.3 Comparison
The assembled results are compared and analyzed here. From Table 4.7 it can be
seen that the TPC linkage is lighter for all categories. But bear in mind that the
TPC linkage needs some kind of help for the parallel adjustment. A hydraulic help
system would add about 15% in weight. An electronic help system would not add as
much weight, but has poor energy efficiency.
Table 4.7: Linkage masses inclusive cylinders.
Type TP Z-bar TPC

Original 100% 93% 90%
Optimized All-round 77% 58%
Optimized Bucket 74% 65%
For a more accurate comparison the bearing padding, mentioned in section 4.2,
must also be considered in the result. The maximum added weight due to this is
calculated to 10%. The final adjusted masses for all linkages, also including the
hydraulic help system for the TPC, are stated in Table 4.8.
Table 4.8: Adjusted linkage masses.
Type TP Z-bar TPC

Original 100% 93% 104%
Optimized All-round 85% 74%
Optimized Bucket 81% 82%
Added bearing padding(+10%) and parallel
alignment hydraulics (+15%, TPC only).
It is easy to see that a mass reduction is possible, for both Z-bar and TPC, without
decreasing the performance of the linkages. The best choice for bucket applications
is the Z-bar were the mass reduction is 10%. For a all-round machine the TPC seem
to be lightest (reduction of about 20%) and therefore the best choice. However, the
linkages are modeled as fully molded, meaning that no welded seams are present. The
yielding strength for a welded seam is about half the strength of a molded. Since real
linkages needs to be welded in some sections must material be added to strengthen
35
4.4. ROBUSTNESS
these parts. For the TPC linkage considered in this report the boom is welded at
several large seams.
But trying to get an exact mass from this optimization is not the point of the
software. The software should not substitute a design engineer, instead it should
generate an optimal design layout for the engineer to use in the linkage development.
This is an important distinction that needs to be understood.
4.4 Robustness
To verify the reliability and correctness of the used approach the robustness of the
optimization was analyzed. The objective and constraint functions behaviour were
inspected since these need to be smooth and have continuous derivatives. Later the
complete optimization routine is analyzed. Different starting points, domain sizes
and goal weights are tested to investigate the impact on the result. In section 4.4.3
different mesh resolution are compared to see when the FEM calculations converges.
4.4.1 Objective and Constraint Function
A parametric analysis of the constraint function was performed by studying one design
variable at a time. The function was evaluated with the variable at different points
between the lower and upper boundaries. This was done for all design variables with
20 evenly spaced points on the interval [xlb , xub ]. The other variables were fixed at
their center position. The top row in Figure 4-7 shows the result for the variables
handling the joint positions, cylinders and counter weight. For some variable sets
the linkage can not be geometrically assembled by the software, this can be viewed
in Figure 4-7 as missing data. In graph II the last variable, the counter weight,
is removed and the z-axis is zoomed to make it easier to see the curvature of the
other variables (since they have much smaller amplitude). In III, the 15th variable
(stabilizers z position) is also removed. The second row shows the variables connected
to the beam elements. In graph V the 60th variable is removed.
All the curves analyzed seem to be continuous and smooth with the step size
36
4
x 10 4
x 10
4
x 10
1.6
1.4 I 1.12
1.11 II
1.09
1.085 III
1.2 1.1
1.08
f(DV)
f(DV)
f(DV)
1.09
0.8 1.075
1.08
0.6 1.07 1.07
0.4 1.06 1.065

20 20 20
15 50 15 40 15 40
40
10 30 10 30 10 30
5 20 20 20
10 5 10 5 10
DV − position 0 0 0 0
DV # DV − position 0 DV − position 0
DV # DV #
4
x 10 4
x 10
1.35
1.3 IV 1.13
1.12 V
1.25 1.11
f(DV)
1.2
f(DV)
1.1
1.15 1.09
1.1 1.08
1.05 1.07
20 20
15 120 15 120
10 100 100
80 10
5 80
60 5 60
DV − position 0 40 0
DV # DV − position 40
DV #
Figure 4-7: Analysis of the objective function for a Z-bar linkage.
used. That means that the derivatives should be easy to compute and generate a
fairly accurate result. However, since the analysis was only considering one variable
at a time this result should be considered with caution.
The constraint function was analyzed in a similar way, but there were some non-
smooth behaviour. A few of these problematic variables are plotted in Figure 4-8, the
rest are collected in appendix D. Some of the variables seem to have a singular point
in the interval [xlb , xub ], they include the booms Ax and Ay variables (see left graph
in Figure 4-8). The singularities do presumably depend on physical conditions of the
linkage and may be avoided by adjusting the boundaries (if they cause problems to
the solver). Other variables, e.g. bucket x-coordinates, were subject to numerical
perturbations (see center and right graph in Figure 4-8). The FEM calculations
are probably the reason for these instabilities, however, it is hard to establish the
exact cause without a more thorough investigation (which is beyond the scope of this
report).
4.4.2 Optimization Routine
The sensitivity of the optimization has been analyzed by changing the domain size,
goal weights and initial starting variables. As reference are the all-round optimized
37
4.4. ROBUSTNESS
90 3 3
80 2 2
70 1
Constraint Value 1
Constraint Value
60 0
0
Constraint Value
50 −1
−1
40 −2
−2
30 −3
−3
20 −4
−4
10 −5
0 −6 −5
−10 −7 −6
xlb xub xlb xub xlb xub
Figure 4-8: Analysis of the constraint function for Z-bar linkage.
versions for each linkage used.
The analysis for the Z-bar linkage is collected in Table 4.9. With expanded domain size
the optimization does converge but to another minimum. The run with altered goal
weight to 1:1.2 is the one nearest to the normal optimized linkage, with a maximum
joint position offset of 12%. The optimization started from the upper boundary
did not, unfortunately, converge during the first 5000 function evaluations but the
constraint violation become very low.
Table 4.9: Sensitivity analysis for Z-bar linkage.
Goal Constraint Max variable No of variables

Domain Size Initial Variables
Weight Violation Deviation > 1% dev.*
Large (+100%) 1:1 Original Z-bar 0% 66.0% 26
Normal 1:1.2 Original Z-bar 0% 12.3% 9
Normal 1:5 Original Z-bar 0% 25.0% 16
Normal 1:1 Lower Bound 0% 58.0% 14
Normal 1:1 Upper Bound 1% 44.0% 19
*Number of variables with constraint deviation larger than 1%
TPC
The analysis for the TPC linkage is presented in Table 4.10. Increasing the domain
size causes the optimization to fail in founding a feasible solution (with a constraint
violation of 6%). Changing the goal weight does not influence the result much, just
38
moving a few joints some millimeters. The same happens when alternating the start-
ing point to any of the boundaries. This is an important feature for a stable solution
and implies that the minimum may be global on the domain.
Table 4.10: Sensitivity analysis for TPC linkage.
Goal Constraint Max variable No of variables

Domain Size Initial Variables
Weight Violation Deviation > 1% dev.*
Large (+100%) 1:1 Original TPC 7% 25.0% 11
Normal 1:1.2 Original TPC 0% 7.7% 4
Normal 1:5 Original TPC 0% 3.1% 3
Normal 1:1 Lower Bound 0% 8.3% 4
Normal 1:1 Upper Bound 0% 7.0% 4
*Number of variables with constraint deviation larger than 1%
4.4.3 FEM Mesh Resolution
The FEM mesh resolution for the stress calculations needs to be fine enough to give
a reliable result. Therefore are 5 different resolutions tested. The finest resolution is
8 times denser than the coarsest. The result from the measurements are plotted in
Figure 4-9.
Mesh Resolution Analysis

4
Execution time
3.5 Optimized Mass
2.5
1.5
0.5
0
Coarse Normal Fine Extra Fine Extremly Fine
Mesh Resolution
Figure 4-9: Influence of FEM mesh resolution on execution time and calculated mass.
39
4.4. ROBUSTNESS
The best mesh to use, with respect to both execution time and result reliability,
is the ”Fine” mesh which is 4 times denser than the coarsest one. It is the mesh with
the lowest execution time where the computational results had converged satisfactory.
That mesh was used for all the results in this chapter.
40
CHAPTER 5. DISCUSSION
Chapter 5
Discussion
This chapter will cover a discussion about the overall performance of the optimization
software developed and the result it produces. Last are some suggestions in improving
the software.
5.1 Result Reliability
The sensitivity to the initial conditions (see Section 4.4.2) is quite low and suggests
that the optimization works as intended to. That is an important feature to mini-
mize the time spent on tuning the optimization. The domain size seem to influence
the result most. This is expected since some variables are on the boundary for the
optimized linkages, see Appendix C.
That the all-round linkage is lighter than the bucket version is surprising and
reduces the trust in the software, which seems to be sensitive to changes in the
constraint function. However, to analyze the result more thoroughly a CAD model of
the result must be done, were all conditions can be verified and the mass calculations
confirmed.
41
5.2. FUTURE WORK
5.2 Future work

For a more complete optimization of the TPC linkage must a help system for the par-
allel alignment should be implemented, preferably a system with hydraulic coupling
between lift and tilt. This can be done by adding extra cylinders, mechanically at-
tached between the boom and front frame and coupled to the tilt cylinders hydraulic
system. Since the final loading unit must include such a system, it would be of interest
to include it in the optimization directly instead of correcting for it afterwards.
The parallel alignment is of high importance for some applications, e.g. fork
handling, and it may be interesting to include this metric in the optimization. This
could easily be implemented by just evaluating and comparing the attachment angle
at different hinge pin heights. For this to make any sense on the TPC linkage must
the above mentioned tilt alignment system be implemented in advance.
It could also be interesting to include the standard Volvo TP linkage in the op-
timization. That way an even more thorough comparison could be done. There are
probably other linkages that also could be of interest, for the same reason.
Another interesting implementation is to expand the software to consider the full
machine. Then the optimization could use real drive cycles to optimize the linkage
for use in a certain application (e.g. short load cycles), material (e.g. large rocks
or fine sand) or with a certain tool (e.g. fork). It makes it also possible to change
the objective function to minimize the fuel consumption, productivity or some other
machine related physical property. More information on the subject can be found in
[18].
It could also be interesting to implement automatic differentiation (as suggested
in section 3.1) to decrease the execution time and increase the reliability of the result.
Changing to a non gradient based optimization method, i.e. genetic method, could
also increase the performance of the software, if the rough constraint functions are
causing problems.
42
Bibliography
[1] T. Ekevid. Optimering av 1 ett lyftramverks utlägg; tillämpning på ett z-länkage
för en L90G. Technical report, Volvo Construction Equipment, 2009. Internal
Volvo document.
[2] J. Unnebäck. How to make a loader linkage. Technical report, Volvo Construction
Equipment, 2003. Internal Volvo document.
[3] R. Filla. Optimisation of lift cylinder size and placement in the conceptual design
of wheel loaders. Technical report, Volvo Construction Equipment, 2003. Internal
Volvo document.
[4] H. Shin Y. Yoo J. Kim K. Shin, S. Lee. Coupled linkage system optimization for
minimum power consumption. Journal of Mechanical Science and Technology,
26(2), 2011.
[5] L. Ipek. Optimization of backhoe-loader mechanisms. Master thesis, Middle

East Technical University, 2006.
[6] Q. Bi J. Shen, G. Wang and J. Qu. A comprehensive genetic algorithm for design
optimization of z-bar loader. Journal of Mechanical Science and Technology,
27(11), 2013.
[7] L. Shen Y. Yu and M. Li. Optimum design of working device of wheel loader.
Mechanic Automation and Control Engineering (MACE), 2010.
[8] L. Deming. Strengt analysis for executive machanism of excavator. 3rd Inter-
national Conference on System Science, Engineering Design and Manufacturing
Information, 2012.
[9] J. Österman C. Nordling. Physics Handbook for Science and Engineering. Stu-
dentlitteratur, eighth edition, 2006.
[10] A. Tucker H. Kuhn. Nonlinear programming. Proceedings of 2nd Berkeley Sym-

posium. Berkeley: University of California Press, 1951.
[11] S. Wright J. Nocedal. Numerical Optimization. Springer, second edition, 2006.
[12] K. Miettinen. Nonlinear Multiobjective Optimization. Kluwer Academic Pub-

lishers, 1999.
43
BIBLIOGRAPHY
[13] A. Vehreschild C. Bischof, B. Lang. Automatic differentiation for matlab pro-

grams. Proceedings in Applied Mathematics and Mechanics (PAMM), 2:50–53,
2003.
[14] A. Walther A. Griewank. Evaluating Derivatives: Principles and Techniques

of Algorithmic Differentiation. Society for Industrial and Applied Mathematics
(SIAM), second edition, 2008.
[15] S. Timoshenko. History of Strength of Materials. McGraw Hill Book Company,

1953.
[16] B. Torstenfelt. Finite Elements - from the early beginning to the very end. LiU-
IEI-S–08/535–SE, 2008.
[17] Load case catalogue. edition 4.0, Volvo Construction Equipment, 2003. Internal
Volvo document.
[18] R. Filla. Using Dynamic Simulations in the Development of Construction Equip-

ment. PhD dissertation, Linköpings University, Department of mechanical En-
gineering, 2005.
44
APPENDIX A. DESIGN VARIABLES
Appendix A
Design Variables
45
Table A.1: Design variables for Z-bar linkage.
No Description Direction Coordinate System State

1 Lower waist x Front frame Fix
2 Lower waist z Front frame Fix
3 Upper waist x Front frame Fix
4 Upper waist z Front frame Fix
5 Front axle x Front frame Fix
6 Front axle z Front frame Fix
7 P x Front frame Variable
8 P z Front frame Variable
9 E x Front frame Variable
10 E z Front frame Variable
11 O x Front frame Variable
12 O z Front frame Variable
13 Width y All Fix
14 Stabilizer x Boom Variable
15 Stabilizer z Boom Variable
16 A x Boom Variable
17 A z Boom Variable
18 C x Boom Variable
19 C z Boom Variable
20 D x Boom Variable
21 D z Boom Variable
22 GDF F x GDF Variable
23 GDF G x GDF Variable
24 GDF G z GDF Variable
25 GJ J x GJ Variable
26 Bucket A x Bucket Variable
27 Bucket A z Bucket Variable
28 Bucket J x Bucket Variable
29 Bucket J z Bucket Variable
30 Design pt 1 x Boom Variable
31 Design pt 1 z Boom Variable
32 Design pt 2 x Boom Variable
33 Design pt 2 z Boom Variable
34 Rear axle x Rear frame Fix
35 Rear axle z Rear frame Fix
36 Width D y Boom Fix
37 Lift cylinder min Variable
38 Lift cylinder max Variable
39 Tilt cylinder min Variable
40 Tilt cylinder max Variable
41 Counter Weight Variable
46
APPENDIX A. DESIGN VARIABLES
Table A.2: Design variables for TPC linkage.
No Description Direction Coordinate System State

1 Lower waist x Front frame Fix
2 Lower waist z Front frame Fix
3 Upper waist x Front frame Fix
4 Upper waist z Front frame Fix
5 Front axle x Front frame Fix
6 Front axle z Front frame Fix
7 O x Front frame Variable
8 O z Front frame Variable
9 P x Front frame Variable
10 P z Front frame Variable
11 A x Boom Variable
12 A z Boom Variable
13 C x Boom Variable
14 C z Boom Variable
15 D x Boom Variable
16 D z Boom Variable
17 F x Boom Variable
18 F z Boom Variable
19 DEG E x DEG Variable
20 DEG G x DEG Variable
21 DEG G z DEG Variable
22 GJ J x GJ Variable
23 Bucket A x Bucket Variable
24 Bucket A z Bucket Variable
25 Bucket J x Bucket Variable
26 Bucket J z Bucket Variable
27 Rear axle x Rear frame Fix
28 Rear axle z Rear frame Fix
29 Width Att. y Boom Fix
30 Width D y Boom Fix
31 Lift cylinder min Variable
32 Lift cylinder max Variable
33 Tilt cylinder min Variable
34 Tilt cylinder max Variable
35 Counter Weight Variable
47
48
APPENDIX B. FINITE DIFFERENCES
Appendix B
Finite Differences
Here are some reference figures, showing the result after optimizing with an interior
point method using a forward finite difference scheme
f (x + h) − f (x)
ff0 (x) = (B.1)
h
and a centre finite difference scheme
f (x + h2 ) − f (x − h2 )
fc0 (x) = (B.2)
h
for the gradient calculations.
3 3
2 2
1 1
0 0
−3 −2 −1 0 1 2 3 4 5 −3 −2 −1 0 1 2 3 4 5
Figure B-1: Optimized TPC linkage with all-round conditions.
The left linkage in Figure B-1 are showing the result from the forward scheme,
right linkage shows the centre scheme.
49
50
APPENDIX C. ADDITIONAL RESULTS
Appendix C
Additional Results
Here is figures showing the position of each variables position on the interval [xlb , xub ]
for the optimized linkages (left graph in the Figure C-1, C-2, C-3 and C-4). In the
same figures are also a graph showing the convergence of the constraint functions, i.e.
the largest constraint violation at each iteration.
Variable Domain Positions Constraint convergence
Joint Positions 2 Force

0.4 Upper Boundary Cylinder Geometrical
1.8
Counter Weight Cylinder
0.3 Element Sizes Stress
1.6
0.2 1.4
0.1 1.2
0 1
−0.1 0.8
0.6
−0.2
0.4
−0.3
0.2
−0.4 Lower Boundary
0
20 40 60 80 100 120 10 20 30 40 50 60 70 80 90 100

Variable Iteration
Figure C-1: All-round optimized Z-bar linkage.
51
Joint Positions 2 Force

0.4 Upper Boundary Cylinder Geometrical
1.8
0.3 Element Sizes Stress
1.6
0.2 1.4
0.1 1.2
0 1
−0.1 0.8
0.6
−0.2
0.4
−0.3
0.2
−0.4 Lower Boundary
0
20 40 60 80 100 120 10 20 30 40 50 60
Variable Iteration
Figure C-2: Bucket optimized Z-bar linkage.

0.4 0.5
Joint Positions Force
0.3 Cylinder Geometrical
Counter Weight 0.4 Cylinder
Upper Boundary
Element Sizes Stress
0.2
0.3
0.1
0 0.2
−0.1
0.1
−0.2
Lower Boundary
0
−0.3
−0.4 −0.1
10 20 30 40 50 60 70 80 90 5 10 15 20 25 30 35 40 45 50
Variable Iteration
Figure C-3: All-round optimized TPC linkage.

0.4
Joint Positions Force
0.3 Cylinder 6 Geometrical
Upper Boundary Element Sizes Stress
0.2 5
0.1
4
0
3
−0.1
2
−0.2
Lower Boundary 1
−0.3
−0.4 0
10 20 30 40 50 60 70 80 90 10 20 30 40 50 60
Variable Iteration
Figure C-4: Bucket optimized TPC linkage.
52
APPENDIX D. FUNCTION ANALYSIS
Appendix D
Function Analysis
Here are the complete result from the constraint analysis of the Z-bar linkage, Figure
D-1 and D-2. For more information about the data, see section 4.4.1.
dv: 10 dv: 16
20 measurements 100 measurements 20 measurements 100 measurements
4 4 200 250
3 3
200
150
2 2
150
Constraint Value
Constraint Value
100
1 1
100
0 0
50
50
−1 −1
0
0
−2 −2
−3 −3 −50 −50
x_lb x_ub x_lb x_ub x_lb x_ub x_lb x_ub x_lb x_ub x_lb x_ub x_lb x_ub x_lb x_ub
dv: 17 dv: 20
120 200 120 120
100 100 100

150
80 80 80
Constraint Value
Constraint Value
100
60 60 60
40 40 40
50
20 20 20
0
0 0 0
−20 −50 −20 −20

Figure D-1: Z-bar constraint functions for problematic design variables.
53
dv: 25 dv: 26
90 120 3 5
80 2
100
0
70 1
60 80 0
−5
Constraint Value
Constraint Value
50 −1
60
40 −2 −10
40
30 −3
−15
20 20 −4
10 −5
−20
0
0 −6
−10 −20 −7 −25

dv: 27 dv: 28
70 100 3 20
60 2
80 10
1
50
60 0
0
40
Constraint Value
Constraint Value
40 −10
−1
30
−2
20 −20
20
−3
0 −30
10
−4
−20 −40
0 −5
−10 −40 −6 −50

dv: 40
20 measurements 100 measurements
20 50
15 40
10
30
5
20
Constraint Value
0
10
−5
0
−10
−10
−15
−20
−20
−25 −30
−30 −40
x_lb x_ub x_lb x_ub x_lb x_ub x_lb x_ub
54
APPENDIX D. FUNCTION ANALYSIS
The data collected for the analysis of the TPC linkage are assembled here. Figure
D-3 shows the objective function evaluations and Figure D-4 shows the constraint
functions for the problematic design variables.
5120 9000 6000
5100 8000
5500
5080 7000
f(DV)
f(DV)
f(DV)
5000
5060 6000
4500
5040 5000
5020 4000 4000

20 20 20
15 40 15 100 15 100
10 30 10 80 10 80
20 60 60
5 10 5 40 5 40
DV − position 0 0 DV − position 0 20 DV − position 0 20
DV # DV # DV #
Figure D-3: Evaluation of TPC objective function.
dv: 8 dv: 31
3 3 10 10
2 2 8 8
1 1
6 6
0 0
4 4
Constraint Value
Constraint Value
−1 −1
2 2
−2 −2
0 0
−3 −3
−2 −2
−4 −4
−4 −4
−5 −5
−6 −6 −6 −6
−7 −7 −8 −8
dv: 32
20 measurements 100 measurements
3 3
2 2
1 1
0 0
Constraint Value
−1 −1
−2 −2
−3 −3
−4 −4
−5 −5
−6 −6
−7 −7
x_lb x_ub x_lb x_ub x_lb x_ub x_lb x_ub
55

FULLTEXT01 Hydraulique

Uploaded by

Copyright:

Available Formats

FULLTEXT01 Hydraulique

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

FULLTEXT01 Hydraulique

Uploaded by

Copyright:

Available Formats

Load Unit Geometry Optimization

for Heavy Duty Machinery

4 Result and Analysis 25

1-1 Schematic figure of a wheel loader. . . . . . . . . . . . . . . . . . . . 3

3-1 Simplified schematic of the optimization routine. . . . . . . . . . . . . 20

4-1 Initial state of Z-bar linkage. . . . . . . . . . . . . . . . . . . . . . . . 27

B-1 Optimized TPC linkage with all-round conditions. . . . . . . . . . . . 49

C-1 All-round optimized Z-bar linkage. . . . . . . . . . . . . . . . . . . . 51

D-1 Z-bar constraint functions for problematic design variables. . . . . . . 53

4.1 Resulting masses for optimized Z-bar. . . . . . . . . . . . . . . . . . 26

A.1 Design variables for Z-bar linkage. . . . . . . . . . . . . . . . . . . . . 46

1.1 Problem Definition

1.2 Related Work

function are presented in section 2.4.

1.3 Machine Design

Front Frame Hinge Rear Frame

Figure 1-1: Schematic figure of a wheel loader.

1.4 Linkage Designs

Figure 1-2: The common Z-bar linkage set-up.

Figure 1-3: Volvo invented TP linkage.

Figure 1-4: Common backhoe loader linkage.

2.1 Mathematical formulation

ceq (x) = 0 (2.2)

Most gradient based optimization algorithms for constrained non-linear problems

c(x) ≤ 0, ceq (x) = 0 (2.5)

λTlb (xlb − x + slb ) = 0 (2.6)

λTub (x − xub + sub ) = 0

must be satisfied [11].

2.2 Design Variables

The geometrical constraints consist of three subgroups: Attachment angles, collisions

No Description Condition [m]

No Description Condition [m]

Measure Value [m]

Figure 2-3: Geometric performance constraints.

Figure 2-4: Break-out torque definition.

The tipping force is also of interest, in addition to of the constraint above. A

HPH [m] Angle [°] Torque

HPH [m] Lift force

2.4 Objective Function

2.4.1 Preliminary Functions

Figure 2-5: 0Optimization

Figure 2-6: Optimization of Z-bar with respect to tipping force.

2.4.2 Final Function

THIS PAGE INTENTIONALLY LEFT BLANK

3.1 Optimization Algorithm

Design Terminate? Optimized

Figure 3-1: Simplified schematic of the optimization routine.

was unfortunately not tested.

3.2 Stress Calculations

Figure 3-2: Showing elements of initial and optimized Z-bar linkage.

The elements are approximated by Euler-Bernoulli beam elements [15]. It is a

With the use of the bending moment

the normal stress can be calculated as

The shear stress is calculated with

where Wx is the section modulus [9].

(c) The machine is crowded

Figure 3-3: The three load cases used in the optimization.

THIS PAGE INTENTIONALLY LEFT BLANK

Result and Analysis