Equipment Health State Assessment Based on MIC-XGBoost

Abstract

Accurate health assessment is the premise and core to intelligent operation and maintenance of equipment. A health assessment method based on Principal Component Analysis(PCA) and Maximum Information Coefficient(MIC)-XGBoost is proposed in this paper. Firstly, Savitsky-Golay filter algorithm is conducted to denoise the original monitoring parameters. Secondly, PCA is introduced to extract health index(HI) from the denoised data automatically. Health state labels is set manually according to the characteristics of HI curves. Thirdly, MIC is used to describe the complex functional relationship between monitoring parameters(also regarded as features) and HIs, and the correlation coefficients are calculated. Features with high correlation coefficients are selected as sensitive features to form the preliminary optimal feature set. Finally, the XGBoost model is trained with optimal feature set and its important parameters are optimized by grid searching. The output of the model can achieve the health state prediction of equipment. The method is verified by using NASA turbofan engine dataset. The results show that the method can extract HIs automatically and effiectively when health state is unknown, and the accuracy is generally above 97%, which can meet the needs of applications.