In the development of complex products reliability is becoming a more and more important aspect. From the engineering perspective the aim is to increase the reliability of a complex system by testing it in various situations which cover the range of customer use appropriately. During the tests the failures occurring are recorded and monitored over the course of time. In a successful test programme, the number of failures over time decreases due to corrections implemented into the system, and thus reliability growth can be monitored. Developing tests in order to detect failure causes and modes and verify system reliability is an important objective in reliability engineering. The focus of the thesis is not on reliability engineering, but on monitoring reliability growth by statistical methods. Statistical methods encompass assessing the current reliability, determining whether there is an increase in reliability and predicting the reliability of the system at a desired time point in the future. The task of the thesis was to develop a monitoring algorithm which allows the user to assess reliability and make predictions at arbitrary time points. Furthermore the algorithm should be able to determine whether a previously specified reliability target will be reached.
In order to fulfill these requirements, the Crow-AMSAA reliability growth tracking model was incorporated as the model frame into the monitoring algorithm. Estimations are made by using the Maximum Likelihood method, and confidence bounds for the model are computed by applying the Fisher matrix approach. The reliability growth monitoring algorithm may be applied to both exact failure time data and grouped data, in which case the failure counts in time intervals are known only. The algorithm was developed such that it can be used for data in which no information of failure modes is available, and incorporates the information resulting in more detailed evaluations otherwise. Application of the algorithm is shown in several examples with real-life datasets. Analyses and predictions are performed retrospectively and the goodness-of-fit of the Crow-AMSAA model is evaluated. The results of the analyses are displayed in both tables and graphs, which are produced by the algorithm automatically.