Research On Remaining Life Prediction Of Multi-component System Clustering With Time-varying Kernel Density

With the rapid development of high-end manufacturing,electronic information,artificial intelligence and other technologies,all kinds of equipment are also developing in the direction of intelligence.High-performance sensors collect massive amounts of data on the operation of the device and can reflect the health of the device.The prediction of the remaining life of a system has received increasing attention from scholars.Remaining useful life prediction can estimate the future operating status of a system or component before a component fails.It can give health management information such as maintenance strategies to facilitate the development of maintenance programs.This not only reduce maintenance costs,but also avoid incurring major catastrophic losses.In this paper,we model the remaining useful life prediction of different components in the system based on the data collected by sensors during the operation of the equipment.Experimental results show an approximate 3% improvement in the accuracy of the remaining useful life prediction,and the main research work is as follows:(1)For the problem of random correlation of degradation processes of different components in the system,a remaining useful life prediction method considering multi-component random correlation sub-cluster time-varying kernel density estimation is proposed.Firstly,multiple components in the system with random correlation are studied in clusters,considering that different components in the same cluster are affected by random correlation and components between different clusters are mutually independent.The correlation characteristics between components are judged,and the correlation characteristics of different components are used to model the cluster degradation.Secondly,the parameters of the time-varying system are changing with time,and considering that the monitoring data near the current moment has more influence on the lifetime prediction than the historical data,the data near the current moment is given a larger weight to establish a time-varying kernel density estimation prediction model,in which the window width adopts the nearest neighbor adaptive window width selection method.Finally,the effectiveness and practicality of the proposed method is verified by selecting the sensor data of turbofan engines as an example.(2)For the dimensional catastrophe problem in the solution process of highdimensional Copula function,a remaining useful life prediction method with multi-part reduced-dimensional time-varying kernel density estimation is proposed.Firstly,a high-dimensional Copula function method is established for the stochastic correlation multi-component,the multi-dimensional joint probability density function are then decomposed into the product of the edge probability density function and the bivariate Copula function to achieve multivariate correlation modeling,reducing the multi-dimensional data to two dimensions,the accuracy of multi-dimensional remaining useful life prediction is improved.The edge probability density function and bivariate Copula functions are estimated by using the time-varying kernel density method.Finally,the accuracy and reliability of the remaining useful life prediction model was verified experimentally.