Similarity Remaining Life Prediction Method Based on Multiscale Feature Fusion

To solve the problem of low prediction accuracy of the traditional similarity matching method, a similarity matching remaining life prediction method combining monotonicity and Spearman correlation is proposed. Firstly, monotonicity is used to screen the time domain and frequency domain features of the bearings, and then Spearman correlation is used to extract the nonlinear relationship of the screened features. Next, the extracted features are fused using principal component analysis to construct the health indexes that characterize the degradation trend of the bearings. An autoregressive moving model is then used to construct a similarity model, in which the Manhattan distance is used to measure the degree of similarity between the two curves and to make the remaining life estimation. Finally, the validation is carried out on the public dataset PHM2012 dataset and compared with Poly2 and Exp1. The experimental results show that the proposed method can effectively improve the remaining life prediction accuracy, which proves the effectiveness and superiority of the proposed method.