A Rolling Bearing Life Prediction Model Based on Adaptive Multi-Core Combined Correlation Vector Machine

Abstract

A rolling bearing life prediction model based on an adaptive multi-core combined correlation vector machine. First, the combined kernel function weight matrix is ​​initialized by particle filter to obtain a combined kernel function set, and then a multi-core combined correlation vector machine set is established. The resampling process adaptively obtains the optimal multi-core combined correlation vector machine model, and finally uses it to predict the running state and remaining life of the rolling bearing. The multi-core combined correlation vector machine model obtained by the present invention adaptively combines the excellent characteristics of multiple single kernel functions , reducing the dependence of the single kernel function correlation vector machine model on parameters, improving the prediction accuracy, better prediction stability, stronger model robustness, and better engineering application value.

Description

Technical field [0001] The invention relates to the technical field of the operating state and life prediction of rolling bearings, in particular to a rolling bearing life prediction model of an adaptive multi-core combined correlation vector machine. Background technique [0002] Rolling bearings are widely used in rotating machinery and other equipment, and their health is directly related to the safe operation of mechanical equipment. Because rolling bearings often work in harsh environments such as high speed and heavy loads, they are prone to failure or even failure. Once the rolling bearing fails or fails, it will inevitably pose a serious threat to the safe service of mechanical equipment, ranging from production accidents that cause mechanical equipment to shutdown, and at worst, leading to major disasters such as machine destruction and death. Because the effective life of each rolling bearing varies greatly, the traditional regular maintenance strategy is not only time...

AI Technical Summary

Problems solved by technology

However, the current kernel function selection of correlation vector machine mainly relies on experience, and mostly uses a single kernel function, which leads to a great increase in the dependence of the prediction accuracy of the correlation vector machine model on parameters.

Similarly, since different single kernel functions have different characteristics and have different sensitivities to different trend data, the single kernel function correlation vector machine model established by using a single kernel function has low prediction accuracy stability and weak robustness

Even in a small number of combined kernel function research, there are shortcomings, that is, two kernel functions are artificially selected for combination, the number of combined kernel functions cannot be automatically selected, and the performance of the fused kernel function is limited.

Images