The use o f component reliability estimates (CRE) in both the design phase o f new systems and the operational phase o f existing systems is described here. The importance o f these estimates is highlighted. Several complementary sources o f CRE are available to engineers. The particular strength o f CRE based on internal field experience is discussed. 1.2.1 Estimates in the Design Phase
As described previously, component reliability distribution estimates can be used in the system design (and development) phase to predict the reliability o f a new system design that uses some existing components in a new configuration, especially when no prototypes are available. These predictions provide critical input in decisions regarding [Usher and Hodgson, 1990; Usher et al, 1991]:
• Modification to the design o f the system,
• Tests and stress screening to perform on the new design and required sample sizes, • Strategies for bum-in o f its electronic components,
• Marketability, i.e. reliability performance claimed in marketing the product, • Warranty period offered and at which price (based on warranty cost estimates), • Liability exposure,
• Number and type o f spare parts to produce, or (recommend to) acquire,
5
• Pricing o f the product, and
• Whether to produce the system at all.
The earlier such predictions are made in the system development stage, the sooner these critical decisions and consequential actions can be made, and the more likely it is that the system will be a commercial success. Usher et al. [1991] comment on the timing o f system reliability predictions during the design phase that “unfortunately, accurate (system) reliability predictions are extremely difficult to make, especially in the earliest stages o f system development, i.e., when few prototypes are available for testing.”
1.2.2 Estimates in the Operational Phase
Component reliability estimates are also commonly made during the operational phase of a system. Reliability engineers use warranty claims data to make reliability estimates to revise/update warranty claim budget estimates and to obtain an early warning on troublesome parts. This can lead to system design changes, a resulting improvement of the reliability o f the system and reduced warranty costs. It might also leads to a recall of equipment that has already been sold, if a safety concern exists. A production-planning department could use the updated reliability estimates to plan production o f the right amount o f spare parts and ask the purchasing department to acquire the needed amount and type o f raw materials. Alternatively, the purchasing department could use the updated reliability estimates to purchase spare parts manufactured by a third party.
1.2.3 Importance o f Reliability Estimates
It follows from the previous subsections that the accuracy and timing o f component and system reliability estimates are important ingredients for the economical success o f the
6
organization that produces and sells, or plans to produce and sell, the systems to which they pertain. This can be gleaned from the long list o f possible uses of component and system reliability distribution estimates during different phases in its life cycle. It can also be determined from the financial impact o f the decisions based on them. Now that the need for CRE is clearly established, the question emerges what these estimates should be based upon.
1.2.4 Sources o f Component Reliability Estimates
Component reliability information can be obtained from different sources. The following sources are discussed here in some detail (see also [Usher et al, 1990; Usher and Hodgson, 1990]):
1. Reliability test data from the component manufacturer or vendor, 2. Life tests on the components,
3. Reliability databases,
4. Field experience o f the organization with these components functioning in systems operating in similar environments.
1.2.4.1 Test Data from the Manufacturer or Vendor
The component manufacturer or vendor generally can provide reliability estimates to the purchasing organization. These data are commonly available and accepted at face value. However, it is often unclear how (under which operating conditions) these estimates have been obtained and how rigorous the components were tested. Their accuracy is questionable. Furthermore, the exact conditions under which the components were tested likely differ from the conditions under which the components will operate in the intended application, limiting their value for reliability prediction.
7
1.2.4.2 Life Tests
Components can be tested under normal operating conditions or subjected to ALT. For highly reliable components, such tests require large samples and can be expensive and time-consuming. Though less time-consuming, ALT require costly special facilities to accelerate the life o f the component as well as special analytical skills. For highly reliable components, a large sample size is required to obtain a small number o f failure observations used to draw reliability estimates on. Availability or costs o f these components might be prohibitive. In addition, individual component life tests might not reflect true component operating conditions [Usher, 1996]. On the other hand, such tests allow for total control o f the operating conditions and accurate assessment o f the distribution type to use in modeling the reliability behavior.
1.2.4.3 Reliability Databases
The m ost commonly used sources for reliability prediction are reliability databases [Usher et al, 1990]. Next to proprietary databases that large industrial organizations might have for their own specific needs, publicly available database for electronic components like MIL-HDBK 217D or E, GIDEP (Government — Industry Data Exchange Program) or INS PEC might be used. GIDEP also contains some estimates for mechanical components.
These databases have some drawbacks o f their own. Although many such databases are available, they contain a fairly limited variety o f components and consider a limited set o f operating conditions [Usher and Hodgson, 1990]. The (instantaneous) failure rates under various operating conditions that are listed in the military handbooks assume these electronic components exhibit a constant failure rate. This exponential distribution
8
assumption has led to misapplication and criticism o f the values listed in the handbooks. Failure rates o f newly designed systems could be up to one order o f magnitude larger than predicted using the handbooks [Evans, 1983]. Other publications also criticize the handbooks [Yellman, 1985; Evans, 1988].
The accuracy o f system reliability predictions based on component reliability estimates from reliability databases is limited. This is caused by the conditions under which the components will be operating as compared to those o n which the estimates were based, the assumed type o f reliability distribution and the small range o f component types considered in the databases.
1.2.4.4 Field Experience of the Organization
Unlike the previously discussed sources o f component reliability estimates, those based on experience o f the organization with these components operating under field conditions generate estimates after assembly o f the components into operational systems. Such estimates include the effect such assembly might have on the degradation o f the components, as well as any possible effects resulting from the machining, soldering, distribution and handling processes that preceded assembly. In addition, human error can contribute to component failure. As a result o f using non-field data, the previously discussed approaches can result in gross overestimation o f system reliability [Usher and Hodgson, 1990]
Real-life data exhibit the reliability o f components in the non-ideal operating environment where failures result from inherent failure processes as well as human errors in distribution, handling, assembly or use o f the product. These errors are hard to model and their effects are difficult to predict, but they need to be considered to accurately
9
predict the reliability o f the assembled system. Because o f the described drawbacks o f the other sources for component reliability estimation, this research focuses on component reliability estimation from internal field experience, i.e. historical system life data.