Sharing pattern feature selection using multiple improved genetic algorithms and its application in bearing fault diagnosis

Abstract

In order to select the effective features or feature subsets and realize an intelligent diagnosis of aero engine rolling bearing faults, this paper presents a sharing pattern feature selection method using multiple improved genetic algorithms. Based on the simple genetic algorithm, a multiple-population improved genetic algorithm was proposed, which improves the speed and effect of algorithm and overcomes the shortcomings of local optima that simple genetic algorithm is easy to fall into. Because all populations regularly share and exchange their selecting features, the proposed algorithms can quickly dig up the current effective feature patterns, and then analyze and deal with the strong correlation between the feature patterns. This will not only give clear directions for the descendant evolution, but also help to achieve high accuracy feature selection, for, the features are highly distinctive. This multiple-population improved genetic algorithm was applied to rolling bearing fault feature selection and comparisons with other methods are carried out, which demonstrates the validity of sharing pattern feature selection method proposed.