Research On Conversion Methods Of Reliability Data Under Different Distributions

We Problems on the reliability data analyzing of products under different circumstances usually occur when we are working on some projects. Due to the limitation of the actual environment, it is difficult to get the experimental data under the complex circumstances while it is comparatively easier to get them under the simple circumstances. Therefore we need to convert the easy ones to the difficult ones. This method is usually applied when we need the data in some special environments, especially in the aerospace field, in order to expand the sample volume and increase the reliability analyzing.For the problem of data conversion between different environments, we usually use the method of environment factor and regression. The former has the defect of poor function ability and big errors because its environment factor is unitary. The regression, which usually applies the linear model, is also poor in achieving a high degree of accuracy due to the limitation of sample volume and the model itself.The thesis presents a new conversion method based on the geometrical mean of the ratio of sample factors. It makes pairs by analyzing the sample factors after separating them into different groups. Afterwards, it gets the environment factor by calculating the geometrical mean of the ratio of matched sample factors. This method can not only guarantee the wave properties of the data but also effectively increase the degree of accuracy of parametric estimation.The thesis presents the way to find the best method for the conversion of reliability data under different distributions and different parameters by comparing the conversion effects of the three conversion methods- the method of environment factor, the method of regression and the method of geometrical mean-under Normal Distribution, Lognormal Distribution, Exponential Distribution and Gamma Distribution with massive value experiments. Not only can the conclusion help increase the accuracy of parametric estimation in the actual environment, but also it is of high utility of the effects that the change in sample data volume and the change in variable coefficients have on the three methods.