Understanding accelerated degradation

Accelerated degradation testing can be employed when there are measures that relate system health to function or operation. This approach may also be needed when the system life is much longer than the test time available. Degradation might be applied to any component or system that shows change, wear, corrosion, crack growth or fatigue as a failure mechanism that leads to eventual system failure. Thus metals, plastics, batteries, adhesives, semiconductors, insulators, bearings, chemicals, paints, asphalt, solar radiation, drugs, one shot devices, planning for maintenance and even shelf life situations may need degradation approaches. The theory has been described by Meeker (1), Nelson (2) and Lawless (3), but these sources have used simple examples that fit with theory. When sample sizes are small and measurements show variability, the simple theory may be hard to apply. For a single stress at a constant level, being able to measure the degradation in the presence of "noise" can be a challenge. When multilevel stresses are employed, the stress-life relationship may change from a simple linear model at low stress to a non-linear model at high stress. Some degradation tests can involve multiple stresses at discrete levels. In this situation non-linear effects can enter. Analysis becomes more difficult when identifying the causes of any failure modes. Treating multiple stresses as a Design of Experiments can help set up and analysis. A good grounding in Physics of Failure (POF) can help sort out the failure modes and causes as the degradation is monitored. There are limits to the application of accelerated degradation testing and at least seven assumptions for successful use of degradation methods have been identified by papers. A broader list includes: · Observed degradation (change) is not reversible when the stress is removed. · There is a single or dominate degradation process that can be studied (POF). · Any degradation before the start of the accelerated test is negligible. · There should be a low error in the ability to measure the degradation (i.e. low variability). · The stress remains constant at each stress level of a test. · Samples selected are typical of the population before the start of the test. · There are no outliers in the sample that lead to unusual behavior. · Many processes follow a simple cumulative damage model. · The single degradation mechanism leads to unique (single) failure mode. · The failure mode observed is a key measure of the health of the sample. · When multiple stresses are present behavior is probably non-linear. Basic degradation models include linear degradation, Arrhenius degradation, exponential degradation, power law degradation and the Eyring degradation. The detailed examples of the paper will include an adhesive representing a continuous process, a mechanical actuator as a cyclic stress example and a two stress degradation examples. The paper will cover single and multilevel stress tests as well as one multiple stress example combined with discussion of data analysis and modeling. Several different models will be used in the analysis. Selection of sample size and time of testing will be briefly discussed as will different stress life models. One example will include the Monte Carlo method as part of the analysis when dealing with small sample size.