Fuzzy Reliability Study

Reliability is one of the most important indices to evaluate the quality of product. It has become a most popular research direction. With the development of research, people begin to realize that there exist not only randomness but also fuzziness in practical problems. So people begin to study fuzzy reliability. The theory and method of fuzzy reliability are not perfect. The paper has studied several aspects of fuzzy reliability as follows:For the definition of fuzzy safety state is different, the calculation of fuzzy reliability is different under different condition. A simple and practical method, to define membership function of fuzzy event denoted by generalize fuzzy strength, is developed in the paper. Then fuzzy reliability optimization model is developed. In the model, fuzzy reliability is regarded as constraint and genetic algorithm method is introduced to resolve it.The most efficient means to analyse the reliability of small sample is Bayes method because it takes full use of priori information. The theory of fuzzy Bayes reliability prediction is not perfect. Based on current research results, a systematic method of fuzzy Bayes reliability prediction is developed. Neural network and genetic algorithm are introduced into the method.Censored test plan cannot meet with actual producer risk and customer risk because the censored number must be integer. Producer risk and customer risk are defined as fuzzy numbers and the optimum censored test plan is determined according to the membership grade after determining the feasible region. Fuzzy censored numbers described with linguistic variables are transformed into completed numbers, then fuzzy Bayes reliability method is introduced to analyse them.Fuzzy S-N curve is introduced in the paper. How to create the ideal P-S-N curve with the least test numbers is an important problem to be resolved. A maximum likelihood method to determine P-S-N curve is introduced and a method with neural network simulating P-S-N curve is developed in the paper. The method can avoid the problem that the P-S-N curve doesn't meet reality when the samples are scattered.