Traditionally, a reliability growth test was performed at the levels of various operational and environmental stresses, often at a level equal to that expected in use. Other than formal failure modes and effects analysis, reliability growth tests were often the only practical means for identification and mitigation of failure modes of a newly designed product.
With the present high reliability requirements and long product useful life, the length of reliability growth tests may become cost and schedule prohibitive; therefore, accelerated testing is taking the place of prior testing at the use levels. This practice, however, does not address the dilemma of possibly unrealistically skewed test results highly dependent on the sequence of individually applied stresses as, unfortunately, it is often difficult to impossible to apply all of the environmental and operational stresses simultaneously. An example of this problem would be a case where the majority of failure modes in a product are a result of or are related to a specific stress, and this test was performed early in the program. These early failures would then produce a high growth rate and an incorrect estimate of the product achieved reliability if the analysis were done by standard analytical models. The test data also may be skewed in the opposite way, producing little or no reliability growth. This concern has been addressed as a serious caution in Edition 2 of the International Electrotechnical Commission (IEC) 61014 Programmes for reliability growth.
This study shows how data analysis applied to an accelerated life test based on reliability growth methodology may produce a viable solution to the calculation problems. The stresses applied in this test are an accelerated application of most of the stresses expected to take place during product use. Each of the tests represents a lifetime exposure to an individual stress. If those stresses are applied individually and in sequence, they are considered to be equivalent to being applied in parallel with one another, as the duration of each stress is calculated to represent life of the product.
Time to failure in each test is re-calculated to represent time to failure in real life. Failure occurrences are then sorted in their increasing order and analyzed using one of the reliability growth test analytical methods.