Research On The Model Of Mechanical System Reliability Evaluation Based On ANN

With the rapid development of science and technology, product structures have become more and more complicated, while product working environment is becoming worse, which lead to people’s higher and higher requirements for the reliability of products. To these systems of high reliability requirements, the redundant system is often designed in order to improve the reliability,however the existence of dependent failure greatly reduce redundant system safety. Therefore, more and more people pay attention to the research on the dependent failure.Most dependent failure models are empiric models based on the historical events of multiple failures. Due to the rarity of dependent failure events, precision, scope and other problems can not be avoided if the empiric models are used to predict dependent failure probability.This paper fully utilizes powerful function approximation function, non-linear mapping function and tolerance of the neural network technology, discretizes system failure probability model, establishes the non-linear relationship of the system failure probability and the components failure data, and then constructs the parametric model of mechanical system reliability evaluation based on neural network, provides a new way for evaluating the system reliability with dependent failure considered.This paper is placed on the following topics:(1) Generating the K/N system failure data by Monte-Carlo method. We mainly consider the static K/N system reliability simulation of different strength-different stress and the dynamic K/N system reliability simulation of only considering strength degradation;(2) Taking K/N system as research object, this paper analyses the main influence factors of system failure on the point of stress-strength interfere in system level;it’s believed that system components number,system failure order, components strength and working load are the main factors of influencing K/N system failure(3) According to the main factors of influencing system failure, extracting dependent failure information from given low-fold failure data,this paper uses the BP neural network algorithm for constructing the the static and dynamic neural network model of K/N system, establishs the non-linear relationship of the system failure probability and the components failure data,and presents the implementation process of the network model by BP toolbox; (4) In this paper, we also utilize the Monte-Carlo method and engineering example analysis to validate the feasibility of the ANN model on the predictation of K/N system reliability.