Evaluation of the probability of failure using Bayesian theorem for real-time condition monitoring
Abstract »This study is about the method for numerical evaluation of probability of failure using Bayesian theorem, from diagnostic result of real-time condition monitoring. When performing maintenance from the result of real-time monitoring, diagnostic result without inspection error is ideal. However, the failure is not caused even if the monitoring method overestimates sufficiently small damage for causing the failure. And the failure is not caused even if it slightly underestimation large enough damage, too. In other words, for the reduction of probability of failure, improvement of damage estimation accuracy of the specific damage level is important. In this study, the method of reducing the risk by improving the diagnostic accuracy of the specific damage level, by control sampling ratio of the training data for learning using weight function, is proposed. The consequences caused by the overestimation and the underestimation of damage differ. The risk caused by the underestimation is defined as failure risk. And the risk caused by the overestimation is defined as economic risk (inspection expense risk). In this paper, shape of the weight function to reduce the economic risk is discussed. And for the validation of the method, proposed method is applied to the delamination identification problem of CFRP beam using the electric potential method.