Fan bearing fault diagnosis method based on multivariate feature fusion

The invention discloses a fan bearing fault diagnosis method based on multivariate feature fusion and aims to solve a problem that a single method is difficult to accurately judge a bearing fault. Themethod comprises the following steps of 1) collecting bearing data in normal and fault states, and preprocessing the bearing data; 2) calculating a time domain index of a bearing signal, and calculating Mahalanobis distances with different faults to serve as a feature I; 3) processing the bearing signal as an audio, extracting a feature vector by using Hilbert-Huang transform, taking the featurevector as input data of an LSSVM, calculating probability output, and taking the probability output as a feature II; 4) extracting frequency domain features of the bearing signals by using a short-time Fourier transform technology, training a convolutional neural network to learn features of different fault signals, and obtaining a classification result as a feature 3; and 5) fusing the three features by adopting a DS evid