Design And Analysis Of Experiments For Reliability Assessment And Improvement

Global competition and increasing customer expectations dictate that products must be highly reliable. This is especially true in hightechnology and safety-critical applications. There is also a great deal of emphasis on reducing costs and product development cycle time and getting the product to market as quickly as possible. In this environment, we need efficient and economical methods for assessing and improving product reliability.In this dissertation, based on the practice, for specific problems, the theoretical ananlysis methods and techniques for grouped data are systematically studied using unbiasing factor method, two-stage method, and bootstrap method. These studies have a very important significance for manufacturers to assess and improve product reliability. The main contents are summarized as follows:1. We prove that there are two pivotal quantities with shape parameter and scale parameters of Weibull distributions for complete data or type II censored data, and then reduce the biases of maximum likelihood estimators using unbiasing factor method. We compare the proposed method with the modified maximum likelihood method for relative bias(RB) and mean square error(MSE) in Monte Carlo simulations. The results show that the unbiasing factor method is better in most cases.2. We focus on the analysis technologies for assessing product reliability. Firstly, we analyze lifetime data for accelerated life test based on improved two-stage approach and bootstrap approach; secondly, we analyze accelerated lifetime data with a nonconstant shape parameter; finally, we analyze accelerated lifetime data with subsampling using two-stage bootstrap method. We illustrate our proposed methods by examples, and compare with other methods in simulation.3. We focus on the analysis technologies for improving product reliability. Firstly, we provide the model that describe the relationship between response and experiment factors; then we select significant factors considering biger is better for percentile or mean time to failure; finally, we illustrate our propsed method using two examples: one contains censored data, the other one does not contain censoring.