Application of improved genetic programming for feature extraction in the evaluation of bearing performance degradation

Abstract

In the evaluation of bearing performance degradation, discovering a good HI (Health Indicator) is one of the most crucial parts, because it determines whether a precise result can be obtained in the prediction of remaining useful life. In this paper, GP (Genetic Programming), which is a heuristic iterative search algorithm inspired by the theory of biological evolution, is improved in genetic operation and fitness function, and a feature weighted matrix is used in GP innovatively. The improved GP is applied to discover a HI by fusing multiple features, which is very close to linearity. Furthermore, by optimizing the discovered HIs, an optimization HI is obtained, which has a higher fitness and can get a more precise result in the prediction of RUL. The proposed approach is verified in the experimental data for the entire life of the bearing provided by 2012 IEEE PHM challenge, and a total of three bearings are used in the verification.