Development of a Mechanical Component Failure Database

William M. Goble, Ph. D., exida

Julia V. Bukowski, Ph. D., Villanova University

Key Words: FMEDA, mechanical failure rates, mechanical failure modes, safety, safety instrumented systems

SUMMARY & CONCLUSIONS a vessel and closes feed valves to prevent rupture if pressure
becomes too high.
In this paper, we present a methodology to derive
component failure rate and failure mode data for mechanical
components used in automation systems based on warranty Programmable Logic Controller Logic
and field failure data as well as expert opinion. We describe a Solver
CPU
process for incorporating new component information into the Subsystem
Inputs Outputs
database as it becomes available. The method emphasizes
random mechanical component failures of importance in the
world of safety analysis as opposed to the wear-out and aging Sensor
Sensor PT
mechanical failures that have dominated mechanical reliability PT Pressure
analysis. The method provides a level of accuracy Sensor Sensors Final
Remote
significantly better than warranty failure data analysis alone. Subsystem Activated Element
Valves
The derived database has the same form as that for Subsystem
electrical/electronics databases used in FMEDA analyses used Process
to show compliance with international performance-based
safety standards. Thus, the mechanical database can be used
in conjunction with existing electrical/electronics databases to
perform required probabilistic safety analysis on automation
systems comprised of both electrical and mechanical
Figure 1 - Safety Instrumented Function Example – High
components.
Pressure Protection
1 INTRODUCTION
To perform the safety analysis for the SIF, the designer
needs to estimate or predict (in the case of a design not yet
Safety instrumented systems (SIS) are automatic systems
constructed and/or fielded) the failure rates and failure modes
designed for the purpose of taking action to avoid danger or to of each of the products used in each subsystem.
reduce the consequences of a potentially dangerous event. For the electrical/electronics equipment, Failure Modes,
International performance-based standards [1,2] require that
Effects and Diagnostic Analysis (FMEDA) techniques [3,4,5]
designers of these systems use probabilistic analysis for have been used to provide failure rates, failure mode
equipment failures classified as “dangerous” to determine if distributions, and diagnostic self-test capability measures for
any given design meets risk reduction goals. The analysis products based on extensive component failure rate and failure
must incorporate all equipment needed for the automation mode databases [6,7,8]. In essence, these techniques compute
system to protect against pre-identified hazards. Typical
a product failure rate based on the failure rates and failure
equipment includes mechanical/electronic sensors, electronic modes of the components which comprise the product. These
signal conditioning modules, microcomputer controllers, techniques rely on the existence of, and regulatory authorities'
relays, solenoids, pneumatic actuators and valves. The
acceptance of, these component-level databases of failure rate
probability of failure analysis requires, at a minimum, the and failure mode data that have been collected (over many
failure rates and failure modes data for all subsystems. years) from field failure data for a wide variety of electrical/
SIS typically consists of one or more safety instrumented
electronics components.
functions (SIF). Each SIF has three subsystems. Each For products that contain only electrical/electronic
subsystem consists of one or more instrumentation products. components these techniques provide the required subsystem
A sensor is used to detect the potentially dangerous condition.
failure rates and failure mode information. However, when
A “logic solver” is used to perform filtering, timing, the subsystem contains mechanical components (either alone
comparisons and other functions required to generate a trip or in conjunction with electrical/electronics components) these
signal. A “final element” is used to actually execute the
techniques cannot be applied because of a lack of mechanical
required action. An example typical of the petro-chemical component failure data. No database, similar to the
industries is shown in Figure 1. This SIF measures pressure in

From the Proceedings of the Annual Reliability and Maintainability Symposium (RAMS) 2007, Presesnted on February 1, 2007
Copyright IEEE, reprinted with author permission for educational purposes. Page 1 of 6
electronics component failure database, yet exists for are intended to illustrate the process incorporating expert
mechanical components. If such a database of failure rate and opinion and are not to be taken as actual data for the product.
failure mode information for mechanical components existed,
it could readily be used in conjunction with FMEDA
techniques to generate the information required at the product INITIAL
level to comply with the new standards [1,2]. Warranty
MECHANICAL
While it could be argued that a variety of mechanical Failure Product λ Expert Opinion
COMPONENT
failure models to predict failure rates exist, it is also true that Data Estimate
DATABASE
most work on mechanical component failure models is
% of λ
focused on mechanical failure due to aging or wear-out. Usage Assumptions
Due to each
However, in the highly-maintained safety systems of Based on Expert Opinion Draft
Component
relevance to IEC 61508 [1], mechanical failures due to aging Component
and wear-out should be infrequent occurrences due to λ’ s
Initial
preventive maintenance. The failures that are principally of Study Function of Each Failure
concern are random failures during the useful life of the Mechanical Component Modes
Failure Mode
equipment. Thus, there is a need for techniques to estimate
Distribution
the rates of random mechanical component failures in order to Failure Mode Database
perform the probabilistic failure analysis required.
In this paper we describe the methods we used to obtain Figure 2 - Initial database creation process
failure rates and frequency of failure modes for a variety of
mechanical components. These components include springs,
3.2 Estimating λprod and Relative Frequency of Component
plungers, seats, o-rings, diaphragms, bearings, structural
Failures
members and many others. These are components used to
build solenoid valves, pneumatic actuators, hydraulic The initial database of mechanical component failure
actuators, many different types of valves and other mechanical rates was created during the analysis of a valve positioner, an
control products. electro-mechanical product that implements microprocessor-
based control of a pneumatic subsystem. The product has
2 NOTATION been in production for several years and the company has a
good database of warranty failures including failure analysis
FMEDA Failure Mode, Effect, and Diagnostic Analysis to identify the specific component failure which caused the
SIF safety instrumented function(s) product failure. Warranty failures are a subset of all field
SIS safety instrumented system(s) failures. They include only the failures reported during the
λi initial product failure rate in failures/hr warranty period. Generally a warranty failure database will
λprod product failure rate in failures/hr contain no information about actual operating hours prior to
failure.
3 INITIAL DATABASE To create the initial database for component failure rates
and failure mode distributions, we first estimated an initial
3.1 Overview of Creation of Initial Database product failure rate, λi, as…

FMEDA analysis techniques use electrical/electronic λi = estimated total # of warranty failures / total
component failure rate and failure mode data to calculate operational hours during warranty period (1)
product failure rate and failure mode data. In order to create a
relevant mechanical component failure rate and failure mode The warranty data clearly contained the total number of
database, we essentially "reverse engineered" existing product reported failures. However, not all failures are reported by
level mechanical failure data to estimate component level users even during the warranty period. Based on the expert
information. opinion of the service and reliability group of the product
First we estimated the total product failure rate, λprod, and manufacturer, we believe about 70% of all warranty failures
estimated the relative frequency of failure attributable to each were reported while the remaining 30% were unreported.
component. Individual component failure rates were then Consequently, we estimated the actual total number of failures
estimated by multiplying λprod by the relative frequency of as…
failure attributable to that component. Finally, expert opinion
was used to estimate the relative frequency of each of the estimated total # of warranty failures = total # of reported
failure modes for each component. failures during warranty period / 0.7 (2)
The overall process used to produce the initial mechanical
component database is shown in Figure 2. Below we provide The estimated total operational hours during the warranty
a detailed description of the process. Note that due to the period were computed based on the operating hours during
proprietary nature of the manufacturer's data, the actual warranty of all fielded product units (whether or not a failure
numbers used in the example are purely hypothetical. They occurred in a given unit) assuming that the product unit

From the Proceedings of the Annual Reliability and Maintainability Symposium (RAMS) 2007, Presesnted on February 1, 2007
Copyright IEEE, reprinted with author permission for educational purposes. Page 2 of 6
operates 24 hours a day and was put in service 3 months after electrical/electronics databases to estimate the product failure
its shipping date. The 3 month figure is typical of the average rate as a function of product failure mode. These results were
delay between shipment, and installation and startup. checked for reasonability by the expert team.
Next, the relative frequency of failure was computed for
each component as… 4 DATABASE UPDATES AND VERIFICATION

relative frequency of failure of a component = # of times 4.1 Database Augmentation Based on Another Product
a particular component failed/total # reported failures (3)
Analysis was required for a second mechanical product.
This resulted in the fraction (or percentage) of product failures This product was of similar functionality to the product used
attributable to a given component. At this point, a group of to create the initial component database but was produced by a
three experienced (20+ years each) mechanical, service and different manufacturer. When we tried to use the initial
reliability engineers from the product manufacturer were component database to perform a FMEDA on this new
asked to review the relative component failure rates. Based on product, we discovered that some different design principles
their expertise, adjustments were made. were used in the new product and components existed in this
Say a particular component contributed to 25% of the product that did not appear in the initial database.
reported product failures, but the experts felt that the 25% In this case there was no good field failure data to use to
figure was too high. Their reasoning was that the 25% figure estimate the component failure rates for the new components.
included a number of component failures due to a However, the mechanical designers and service experts for
manufacturing problem which had since been corrected. this product were available so the initial database was
Based on their expert opinion, the 25% relative failure rate augmented with expert estimates of component failure rates
was reduced to 20%, and the 5% removed from that and relative frequency of component failure modes for the
component's relative failure rate was apportioned (again using new parts. Using this new, augmented component database, a
expert opinion) over the remaining components with reported FMEDA analysis was performed to estimate product failure
failures. Thus, the relative component failure rates summed to rate. The results were reviewed by the second set of experts
100%. for reasonability and adjusted by changing the new component
However, a significant number of product components failure data based on their experience.
(about 30) had no failures in the warranty data. Yet clearly
they could fail. This implies that λi underestimated the actual 4.2 Verification of Mechanical Component Database
product failure rate, λprod. Rather than assign these
Clearly, to verify the correctness of the above described
components a relative failure rate of zero, initially each was
assigned a modest relative failure rate of 0.1% adding an techniques for building the database of mechanical component
additional 3% to the relative failure frequency. Thus the total failure rates and frequencies of failure modes, we should
ideally perform a FMEDA analysis to estimate product failure
relative failure frequency summed to 1.03 instead of 1. λprod
rates for a product (not used to build the database) with good
was estimated as…
field failure data. We then could compare the FMEDA
estimate of product failure rates to the product failure rates
λprod = λi * 1.03. (4) computed from the field failure data.
Detailed end-user field failure data, based on periodic
This final estimate of product failure rate, λprod, was inspection and testing, became available for a third mechanical
compared with product failure rate data from other databases instrumentation product. Such end-user field failure data is
[9,10,11] in order to check for reasonability. We then considered to be of the best kind because it inherently
estimated individual component failure rates as λprod times the incorporates field application conditions. Unfortunately, it
relative failure frequency for each component. contained some components not currently in the database.
Therefore, the database was once again augmented with new
3.3 Estimating Relative Frequency of Component Failure component information based on expert opinion. Then a
Modes
FMEDA analysis using the newest mechanical component
database was performed for this product and the results were
Our warranty data did not include any failure mode
information. To obtain estimates for relative frequency of compared to the field data. The FMEDA-estimated product
failure modes, our expert group then studied the function of failure rate was about 50% larger than that computed from the
field data. This was considered quite acceptable. The
each mechanical component and created an initial list of
failure modes for each component. Using their expert opinion, database should produce conservative estimates because it is
they assigned a relative frequency to each of the failure difficult to know or predict field application conditions when
performing a FMEDA analysis.
modes. This completed our initial component failure database
that consisted of a list of components, the failure rate of each
component, the failure modes of each component and the
relative frequencies of each failure mode. The FMEDA
analysis technique was then applied using the data in the
initial mechanical component database [12] and the accepted

From the Proceedings of the Annual Reliability and Maintainability Symposium (RAMS) 2007, Presesnted on February 1, 2007
Copyright IEEE, reprinted with author permission for educational purposes. Page 3 of 6
 based on ratios by these experts.
4.3 Review, Feedback, Update One example is a coil spring. The key variables identified
by the mechanical designers for a coil spring included initial
The steps described above for augmenting the mechanical spring set, cycle rates in normal operation, cycle stroke length,
component failure database amount to a continuous review, maximum load versus spring rate, ratio of spring length to coil
feedback, and update process which is illustrated in Figure 3. diameter and wire size.
Each design variable was categorized into several “order
Field MECHANICAL of magnitude” type buckets. The relevant combinations were
FMEDA
Failure Product λ Compare COMPONENT defined and for each combination a new part entry was created
Product λ
Data DATABASE
in the database with failure rate and failure mode distribution.
In addition to specific design variables, end-user
Industry application conditions were considered. It was recognized that
Database YES Update
Significant
Component
the database should include random failures due to:
Difference?
Database
a. random environmental stress
NO b. material defects
Finish c. probability that the product designer did not predict
all application conditions
Figure 3 - Mechanical component database update process d. accidental misapplication and
e. accidental abuse in the end application.
The feedback loop has been performed many times now
over a wide variety of mechanical instrumentation products No consideration was made to account for failures due to
including: wear-out (end of useful life) conditions. No attempt was made
to create an analytical equation to calculate the component
a. Floating Ball Valve (2) failure rate as a function of the design variables. This was not
b. Trunion Mount Ball Valve (3) needed to achieve results sufficient for safety analysis and it
c. Spool Solenoid Valve (3) was agreed that not enough data existed for this purpose.
d. Pilot Operated Spool Solenoid Valve (2) An example set of components is shown in Table 1. The
e. Hydraulic Solenoid Valve first five columns of Table 1 are part descriptor fields. The
f. Quick Exhaust Valve failure rate is the sixth column of Table 1 (in FITS). The
g. Pneumatic Booster Relay failure modes and the distribution of each mode are presented
h. Poppet Solenoid Valve in subsequent columns in Table 1.
i. Pilot Operated Poppet Solenoid Valve
j. Pressure Relief Valve ID PGroup PPrimaryPSecondary Application
any design,
Lambda FModeA DistributioFModeB Distributio

k. Gate Valve 73 spring coil tension

formed hook
200 settle 0.3 break 0.7
on end of
l. Butterfly Valve spring
low cycle
m. Triple Offset Butterfly Valve 46 spring coil compression
design, full
100
settle, loss of
0.3 break 0.7
stroke, low force <20%
n. Globe Valve cycle
low cycle
o. Valve Positioner (4) design, partial settle, loss of
60 spring coil compression 25 0.3 break 0.7
stroke, low force <20%
p. Piston Pneumatic Actuator cycle
q. Rack and Pinion Pneumatic Actuator (3) high cycle
design, full settle, loss of
59 spring coil compression 60 0.3 break 0.7
r. Diaphram Pneumatic Actuator stroke, high
cycle
force <20%

high cycle
design, full settle, loss of
58 spring coil compression 25 0.3 break 0.7
Each of these product types has different design stroke, low force <20%
cycle
characteristics and each review has added knowledge to the high cycle
design, partial
database. The component descriptions and application 47 spring wave compression
stroke, low
10 settle 0.8 break 0.2
cycle
conditions have been refined with each review.
Fixed load,
disc, very restricted
48 spring compression 5 settle 0.1 break 0.9
bellville stroke and
5 APPLICATION CONDITIONS cycle
any design,
position
One of the consistent issues highlighted in the feedback 71 spring coil compression
retention, no
significant
5
settle, loss of
force <20%
0.3 break 0.7
process was the need for new components based on different cycles or
stroke
application conditions. When this became apparent, new high cycle
design, partial settle, loss of
components were created with a description that included the 56 spring coil compression
stroke, high
25
force <20%
0.3 break 0.7
cycle
new application conditions. In order to attempt to make the high cycle
design, partial
set of application conditions comprehensive, mechanical 49 spring leaf flex
stroke, high
25 settle 0.8 break 0.2
cycle
design experts from different backgrounds reviewed each part.
The key design variables were identified and documented. Table 1 – Spring Failure Data
Then a failure rate for each set of conditions was estimated

From the Proceedings of the Annual Reliability and Maintainability Symposium (RAMS) 2007, Presesnted on February 1, 2007
Copyright IEEE, reprinted with author permission for educational purposes. Page 4 of 6
 6 COMPREHENSIVE REVIEW 9 ACKNOWLEDGEMENTS

Once the database included all components from the The work for this paper was performed by several people.
mechanical instrumentation products listed above, a The authors wish to thank Steve Close, Lindsey Bredemeyer,
comprehensive review was conducted. Two independent Rachel Amkreutz and Chris O’Brien for expertise, detailed
experts, each with 30 or more years of experience, and two review and analysis work. The authors also thank the many
exida engineers, who had participated in the development of mechanical design and service engineers who reviewed the
the database, re-examined each component entry performing a data.
relative comparison of all database numbers. Some entries in
the database were modified to make the database more REFERENCES
conservative.
1. IEC 61508, Functional safety of electrical/electronic/
7 RESULTS programmable electronic safety-related systems, Geneva,
Switzerland, 2000.
The results of the first task (initial database construction) 2. ANSI/ISA SP84.00.01 – 2004 (IEC 61511 Mod.),
clearly show the feasibility of constructing a database for Application of Safety Instrumented Systems for the
mechanical component random failure rates and failure modes Process Industries, Raleigh, NC, ISA, 2004.
using existing field failure data. 3. W. M. Goble, J. V. Bukowski, and A. C. Brombacher.
The “analyze and compare” approach has been used on “How diagnostic coverage improves safety in
nearly thirty different applications with comparison results programmable electronic systems," ISA Transactions,
used to update and enhance the original database. The Vol. 36, No. 4, Amsterdam, The Netherlands: Elsevier
feedback of experienced mechanical designers has been Science B. V. , 1998.
invaluable as good field failure data is rarely available. The 4. W. M. Goble and A. C. Brombacher, “Using a Failure
result is now a set of mechanical component failure rate and Modes, Effects and Diagnostic Analysis (FMEDA) to
failure mode numbers that produces results comparable to and Measure Diagnostic Coverage in Programmable
agreeable with expert opinion as well as some limited field Elecronic Systems,” Reliability Engineering and System
failure data. Safety, Vol. 66, No. 2, November 1999.
We believe that the method provides a higher level of 5. W. M. Goble and Harry Cheddie, Safety Instrumented
accuracy than limited field failure analysis alone. Not only is Systems Verification: Practical Probabilistic
field failure data limited in quantity but it is often of Calculations, Appendix E, ISA, Research Triangle Park,
questionable quality. One important variable is the number of NC, 2005.
failures actually recorded. It is generally acknowledged that 6. IEC 62380, Reliability data handbook - Universal model
field failures of less expensive equipment are very likely to be for reliability prediction of electronics components, PCBs
underreported compared to the reporting of field failures for and equipment, Geneva, Switzerland, 2004.
more expensive products. 7. FMD – 91: Failure Mode / Mechanism Distributions,
The mechanical FMEDA method in conjunction with the Rome, NY, Reliability Analysis Center, 1991.
mechanical component database is now being used for 8. Faller, R., Functional Failure Assumptions, Revision 0.3,
analysis of safety protection functions and has been accepted TÜV Product Services Gmbh., August 1995.
by regulatory authorities in various parts of the world. The 9. Handbook of Reliability Prediction Procedures for
results of the mechanical component database have been Mechanical Equipment, Naval Surface Warfare Center,
published [13]. West Bethesda, MA, 1998.
10. Offshore Reliability Data – OREDA-97, DNV Industry,
8 FINAL NOTES 1997.
11. Guidelines for Process Equipment Reliability Data,
The methods developed represent an ongoing work. We Center for Chemical Process Safety, Americaon Institute
continue to gather data on specific products in different of Chemical Engineers, 1989.
applications. We are expanding the data to include end of life 12. W. M. Goble, “Accurate Failure Metrics for Mechanical
constraints in terms of operating time or cumulative cycles. Instruments,” Proceedings of IEC 61508 Conference,
Additional operating profiles are also being added as new Augsberg, Germany, RWTUV, January 2003.
applications, such as automotive, are considered. 13. Electrical & Mechanical Component Reliability
Additionally, we recently received four additional years of Handbook, exida, Sellersville, PA, 2006.
field failure data on the original mechanical product. We plan
to perform a new FMEDA analysis on this product and
compare the results to the product failure rates estimated from
this new field data.

From the Proceedings of the Annual Reliability and Maintainability Symposium (RAMS) 2007, Presesnted on February 1, 2007
Copyright IEEE, reprinted with author permission for educational purposes. Page 5 of 6
 BIOGRAPHIES

William M. Goble, PhD Julia V. Bukowski, PhD

exida Department of Electrical & Computer Engineering
Sellersville, PA 18960 USA Villanova University
Villanova, PA 19085 USA
e-mail: [email protected]
e-mail: [email protected]
William M. Goble is currently Principal Partner and co-founder of
exida, an engineering consulting company focused on automation Julia V. Bukowski is currently an Associate Professor of Electrical
system safety and reliability. He has over 30 years of experience in and Computer Engineering at Villanova University. She has
electronic design, software, reliability analysis and management. He approximately 25 years experience in the field of reliability and has
has a BSEE from Penn State, an MSEE from Villanova and a PhD conducted research in the areas of hardware, software, and network
from Eindhoven University of Technology in Reliability Engineering. reliability. She received her BSEE and Ph.D. (Systems Engineering)
He is a registered professional engineer in the State of Pennsylvania from the University of Pennsylvania, and her DIC in Electronics
and a Certified Functional Safety Expert (CFSE). He is a fellow Engineering from Imperial College of Science and Technology,
member of ISA and author/co-author of three ISA books. University of London. She has been a Fulbright Senior Lecturer and
Visiting Associate Professor with the Faculty of Industrial
Engineering and Management at the Technion Israel Institute of
Technology in Haifa, Israel. She is a senior member of the IEEE and
has been a guest editor for a special issue of the IEEE Transactions
on Reliability.

From the Proceedings of the Annual Reliability and Maintainability Symposium (RAMS) 2007, Presesnted on February 1, 2007
Copyright IEEE, reprinted with author permission for educational purposes. Page 6 of 6