An optimized support vector regression for prediction of bearing degradation

Abstract

Machine learning and deep learning models are gradually being applied to predict the remaining useful life of bearings. But, still extracting effective features from the bearing signals and enhancing the prediction accuracy is a major concern. Therefore, an optimized support vector regression (SVR) model for prediction of bearing degradation is proposed. Firstly, the original time domain and frequency domain features extracted from the bearing vibration signal are learned by using the deep neural network (DNN) to improve the quality of degradation features. Secondly, a novel multi-population fruit fly optimization algorithm (MPFOA) is proposed by introducing multi-population mechanism. Thirdly, MPFOA is employed to choose the parameters of SVR, then we use MPFOA-SVR to predict the bearing remaining useful life through the degradation features learned by DNN. At last, CEC 2017 unconstrained benchmark functions and a real bearing dataset (IEEE 2012 PHM) are used to verify the performance of MPFOA and optimized SVR models respectively. Numerical experimental results show that MPFOA has a better optimization ability than the compared meta-heuristic algorithms. The optimized SVR model has a higher prediction accuracy in predicting the bearing remaining useful life.