Challenges in setting up and analyzing a multi-stress accelerated test

Planning a viable Multi-stress Accelerated Test can be a good challenge. It should start with the objective of the accelerated test being identified. It might be to estimate life, identify the main failure modes or mechanisms, or perhaps the purpose of the test is to find as many design flaws as possible and correct them before release to production. Once the purpose of the test is identified then the selection of a best set of stresses may occur. Knowledge of prior related field data or field experience would be helpful. Any root cause(s) of field failures might lead to good stresses. When four or five simultaneous operating stresses in the field are present, only two or three major ones might be selected for accelerated testing. If time and resources permit, a DOE, FMEA or Physics of Failure investigation would be helpful. For mobile systems, perhaps mechanical loads combined with temperature extremes might represent the best combination. Then temperature combined with vibration might be a good combination. Sometimes acids or salt air combined with temperature might be selected to simulate field corrosion situations. Functional test parameters or degradation measures can be key to meaningful test results, no matter what stresses are selected. Multiple stresses may also result in non-linear behavior or results, so this must be anticipated. Data analysis with only a small number (one or two) of hard failures increases difficulty of analysis. Then degradation measures become indispensable. Data collection times during the accelerated test may also impact the analysis or mask non-linear behavior. The test data collection points are often set for convenience of data collection and not to yield the best spread of information. Add intermittent or soft failures to the test results with occasional non-repeatable events to this mix and reliability challenges increase. This paper will present two detailed examples, representing the search for best stresses for accelerated testing. The discussion shows ways to limit the myriad of possible test conditions. These examples employ practical sample size and test time collection points and so do not represent worst case possibilities. The examples also document some data handling and noise issues. Asking questions and planning for the accelerated test can be more important than the execution of the accelerated test itself.