Research On Reliability Assessment Methods Based On Performance Degradation Data

Reliability index is an important information source to establish the maintenance plan and update cycle for products, it has important engineering significance to obtain accurate product reliability index. At present, a large number of long-life products never be failure in life or accelerated life test subject to reasonable test time, so traditional reliability assessment methods based on life data become unuseful, reliability assessment methods based on performance degradation data has become the best way to evaluate the reliability index of long-life products.This paper first conducts the research on reliability assessment methods based on regression model, Wiener process model and Wiener process model with measurement error, and the specific algorithms are given in detail. For model selection methods, based on the characteristics of each model, this paper presents a new method based on prior knowledge such as measurement error, then demonstrated the model selection method based on statistical test of degradation data is undesirable under the engineering backgrounds like insufficient sampling times and candidate models are simplified which ignore factors such as environment and load.Secondly, this paper studied the applicability of neural networks in reliability assessment, pointed out the two basic prerequisites for effective application of neural network should have, namely generalization input and training input similarity condition and the network input and output mapping regularity condition. According to the characteristics of the performance degradation process, conclude that neural networks is not a good tool in off-line reliability assessment based on performance degradation data. Then introduces the basic concept of realtime life prediction, and proposed a real-time life prediction method based on neural networks and similarity measurement of degradation trajectory.Finally, the basic concepts of accelerated degradation test, acceleration condition and acceleration model are introduced, the commonly used accelerating models and the accelerating stress type they applicable respectively are summarized. Then the reliability assessment methods based on constant stress and step stress accelerated degradation test data are given, the problem of possible sensitive parameters in reliability assessment and the importance of identifying sensitive parameters are discussed.