Getting Failure Rate Data

It should be well understood that estimating failure rate data is complicated. In general for probabilistic SIF verification purposes, the less one knows, the more conservative one must be. If little data is available, then conservative choices must be made when choosing data. Fortunately, the quality of data is improving (see Chapter 8). 

A software bug causes a logic solver to fail in an unpredictable and apparently random manner. Will this failure be considered a random failure or systematic failure  

A system has a probability of failure (aU modes) for each one-year mission time of 0.15. What is the probability of a failure for a ten-year mission time (No wear out, etc.) 

Unreliability for a system with one failure mode is given as 0.002. What is the reliability  

A module has an MTTF of 75 years for all failure modes. Assuming a constant failure rate, what is the total failure rate for aU failure modes  

---

Several problems exist with this method of getting failure rate data. Often needed information about a failure is not collected. This includes total time in operation, failure confirmation, technology class, failure cause and stress conditions. The results are usually a significantly higher failure rate than the number needed for probabilistic SIF verification. This is due to ... 

Any random number simulation requires a seed value to get it started. If the same modef using the same failure rate data and the same time scale is run twice with a different seed value in each case, then the results of the two simulations will be different. Many managers who have difficulty grasping this point—they expect that the answer to a problem should always be the same, given the same input data. 

Fortunately for the safety instrumented function designer, many valve manufacturers are getting their products and their design processes assessed to lEC 61508 (Ref. 1). These manufacturers supply a safety manual and all failure rate data (typically assuming a well designed final element) needed for SIF probabilistic verification. Some manufacturers are... 

So, what do these numbers actually mean This simple question actually gets to the core issue. In older, analog systems with discrete components it was possible, in principle, to take failure rate data for each component and thereby determine an overall system reliability, using basic arithmetic. Now that most new systems are digital (microprocessor-based), although it is stiU possible to do a failure rate calculation for the hardware parts of the system, the reliability of the software cannot be determined by numerical means. And, furthermore, the reliability of the software is likely to dominate the overall system reliability. 

This chapter has discussed some of the factors that may affect equipment reliability and necessitate data adjustment. At this time, little documented assistance is available to help develop these data adjustments. It may be necessary to get help from experts in some situations. Lastly, failure rates are often reported to several decimal places, a precision frequently unwarranted by the data. It is suggested that only the failure rate s first significant number and associated exponential power be used. 

As we discussed in introduction, the field return data can not be used for analysis of the wear out region since electronic boards get into this region long after their warranties expire. Therefore, we only deal with early failure and useful time periods. We develop our hazard rate function with phases having Decreasing Failure/Hazard Rate (DFR) and Constant Failure Rate (CFR) like those proposed by Yuan et al. (2010) and Chen et al. (1999). Here, DFR and CFR correspond to early failure and useful hfe regions, respectively. The proposed overall hazard rate function h it) is presented in Equation 1 where h i) = hazard rate function in early failure period, h ii) = hazard rate function in useful-life period, t = time (TTF), and X = change point from DFR to CFR. 

---