Rolling bearing residual life prediction method considering model and data uncertainty

Abstract

The invention discloses a rolling bearing residual life prediction method considering model and data uncertainty. The method comprises the steps of collecting a rolling bearing full life cycle vibration acceleration signal; extracting a morlet wavelet transform time-frequency diagram of the vibration acceleration signal; constructing health factor data by utilizing a multi-scale deep convolutionalnetwork, and meanwhile, obtaining a model uncertainty quantitative analysis result by adopting a variational inference method; performing regression prediction analysis on the health factor data by utilizing an improved relevance vector machine, predicting the residual life, and quantitatively analyzing the data uncertainty at the same time; and comprehensively considering model uncertainty and data uncertainty quantitative analysis results to obtain a prediction result confidence interval. Improvements related to the prior art are as follows: a polynomial regression prediction model is fusedinto a relevance vector machine, so that the residual life prediction precision is improved; uncertain factors in residual life prediction are comprehensively considered, model uncertainty and data uncertainty are quantitatively analyzed, and the reliability of a prediction result confidence interval is improved.

Description

technical field [0001] The invention relates to the technical field of prediction of the remaining life of mechanical equipment, in particular to a method for predicting the remaining life of a rolling bearing considering the uncertainty of models and data. [0002] technical background [0003] Rolling bearings are one of the core components of the mechanical transmission system of artillery, tanks, helicopters, ships and other weapons and equipment, and their performance directly affects the reliability and safety of weapons and equipment. Due to long-term continuous work under high load, high speed, high impact and variable working conditions, rolling bearings are extremely prone to damage and failure. Therefore, it is of great significance to study the performance status of rolling bearings at the current running time and predict the remaining life to avoid major accidents and maintain the integrity of weapons and equipment. [0004] During the operation of rolling beari...

AI Technical Summary

Problems solved by technology

The monotonicity and trend of the traditional health factors that characterize the performance degradation characteristics of rolling bearings are not good enough, which affects the prediction accuracy of remaining life

Deep learning technology can effectively use historical monitoring data, and the performance of the constructed health factors has been significantly improved. However, when applied to the prediction of remaining life, it can only provide point estimates and cannot obtain the confidence interval of the prediction results, which is not conducive to the formulation of maintenance plans.

Correlation vector machines are widely used in the field of remaining life prediction, which can provide confidence intervals for prediction results, but have the disadvantage of weak long-term trend prediction ability

Images