Research On Variable Dimension Reliability Evaluation Method Based On Radial Basis Function Neural Network

Reliability is a key indicator of the performance of CNC machine tools.For many years,it has affected the development of China’s CNC machine tool industry and seriously affected the market competitiveness of domestic CNC machine tools in the international market.Therefore,how to improve the reliability of CNC machine tools has become a major problem faced by China’s equipment manufacturing industry,and has been repeatedly proposed in major national projects.This paper studies the reliability index of a type of censored life test of CNC machine tools,namely MTBF(mean time between failures).The study found that most of the failure forms of CNC machine tools obey the Weibull distribution,and the failure data includes some failure data and censored data.For reliability evaluation of censored data obeying the Weibull distribution,there is currently no mature evaluation method and standard in China.The main contents of this paper are as follows:(1)This paper studies the censoring method for the reliability test of CNC machine tools,and obtains the machine tool failure data and censoring data in accordance with the replacement timing censoring.This method provides fault data for the reliability research of CNC machine tools,which is of great significance to the research of CNC machine tools.(2)When the censored test with replacement timing is performed on the CNC machine tool,the censored data obtained is variable-dimensional data.For the standardized censored data with replacement timing,that is,reliability test data with a certain dimension of input and output data This paper,for the first time,uses the error back-propagation network and radial basis neural network for reliability evaluation,and obtains better evaluation results.(3)For the experimental data with variable timing truncation with variable dimensions,based on the radial basis network function,for the first time,the center method is selected for self-organization,and the minimum center is solved by identifying and controlling the input data dimension.Difficulties in distance and validation in programming practices with a limited amount of data.