Reliability prediction stands at the center of many reliability programs across government and industry.
The basic principle of reliability prediction is to define a rate of failure for all key components in a
system and then add them together to obtain an overall system failure rate. This process explicitly
considers all components to be in series, which means that if one component fails, the entire system goes
down. The result gives a conservative estimate of when a system will most likely fail.
Over the past few decades, several standards have been developed to assist in conducting this type of analysis. The standards define models for different component types based on test and/or field data. With few exceptions, the models assume a failure rate that is constant with time, addressing the useful life of a component where failures are regarded as random.
The major factors in reliability prediction models that contribute to predicting component failure consist of:
The first decision you must make before choosing a reliability prediction model is whether to use a parts count prediction or a parts stress prediction. The parts count prediction is generally used for the early design stage of a project when parts and part parameters have not been exactly identified. It uses generic failure rates for various part types given an operating environment and temperature, multiplies them by a quality factor, and then adds them up to obtain a system failure rate. This methodology is specifically defined in MIL-HDBK-217, Telcordia, and GJB/z 299B.
Parts stress prediction is normally used later in the development stage when most of the components and operating conditions have been identified. In a parts stress prediction, temperature and electrical stress become important factors in predicting the part failure rate. Temperature can be set at the system level, the assembly level, and the component level. A junction temperature rise per component may also be considered, depending on the depth of the parts stress prediction. The electrical stress usually assumes the form of a ratio of operating value to rated value. For instance, the defining stress factor for capacitors is voltage. Consequently, operating voltage and rated voltage are used in the failure calculation model. These factors are generally consistent across the different reliability prediction standards.
Part type identification in any prediction model is the major factor affecting failure rates as well as the model inputs that must be considered. As an example, Table 1 compares the device parameters required for microprocessor and silicon field effects transistor parts stress models in MIL-HDBK-217 and Telcordia. Each of these parameters has a dramatic effect on the failure rate prediction of the device. It is also very important to consider the different part types supported by a particular standard when deciding which one to choose. For example, MIL-HDBK-217 includes a model for laser diodes and Telcordia includes a model for batteries, but the reverse is not true.
| Model | MIL-HDBK-217 | Telcordia |
|---|---|---|
| Microprocessor |
|
|
| Si FET |
|
|
The values for these parameters establish Pi (π) factors, which are the variables used in the failure rate
equation for the part type. The figure below shows a MIL-HDBK-217 equation for an integrated circuit. The
data for the variables in this equation is located in data tables in different sections of MIL-HDBK-217,
Military Reliability Prediction of Electronic Equipment.
Part quality level is a measure of a manufacturer's production and test procedures and the quality controls in place. Quality level scales vary significantly from standard to standard and from part type to part type for some standards. For example, Telcordia defines a single quality scale for all part types while MIL-HDBK-217 has different scales for different part types. When assessing which model to use, you may want to consider which quality rating you use in your company and see which model closely matches it.
Sometimes, you may have data obtained from actual fielded units or laboratory tested units. This information can be very useful in adjusting the predicted failure rates to more accurately reflect what has been experienced. Some models specifically support this feature, including Telcordia and PRISM. Though this may limit your selection if you are doing prediction calculations manually, software prediction packages such as Windchill Prediction extend this capability to all prediction models.
There are several major factors in reliability prediction models that contribute to predicting component failure. These include the type of prediction (parts count versus parts stress), the types of parts in the system, the quality of parts, the environment in which the system operates, and the availability of life data. Part 2 of this three-part article will explore, in more detail, the various standards that are available, including MIL-HDBK-217, Telcordia, 217Plus, PRISM, IEC62380 and FIDES. Additional information about Windchill Prediction, supported models, and predictive modeling in general can be found by visiting our Windchill Prediction product page.
Sources:
"Reliability Prediction Models: Use and Evaluation", Reliability eFlash, 2009