A Review of Machine Learning-based Input Features for Rotating Machinery Fault Identification

The continuous application of machine learning theory has promoted the in-depth development of fault diagnosis. There are various types of machine learning-based fault diagnosis algorithms for rotating machinery with various input feature forms. In order to deeply understand the effects of various feature forms, the existing research on the input feature forms of machine learning algorithms is reviewed in the light of the current research status in this field. The basic generation principles, application status, advantages and disadvantages of statistical features, information entropy, time-frequency map feature parameters and grayscale map, Gramian angular field image, spectral kurtosis map, wavelet coefficient matrix, and time-frequency map feature forms are discussed, and finally, the challenges and future development directions of machine learning-based rotating machinery fault diagnosis are summarized. Finally, the challenges and development prospects of machine learning-based fault diagnosis of rotating machinery are summarized and prospected.