A fault diagnosis method for planetary gear under multi-operating conditions based on adaptive extended bag-of-words model

The traditional bag-of-words (BoW) model needs to set the number of basic words manually, and is not suitable for the fault diagnosis under multi-operating conditions. To solve these problems, a fault diagnosis method for planetary gear under multi-operating conditions based on adaptive extended bag-of-words (AEBoW) model is proposed. Defining the sum of squared errors (SSE) which describes intra-class dispersion, the optimal number of basic words is obtained by the gradient descent of SSE. The BoW model is extended to three layers, and the codebooks in each layer are constructed by extracting local kurtosis (LK), local 2-dimensional information entropy (L2DIE) and Unoriented-SIFT features respectively. The final diagnostic result is obtained by screening the classification results with higher probability layer-by-layer. The overall recognition rate is 99.5% for 25 fault states (5 rotating speeds and 5 gear fault types), and the accurate recognition rate of gear fault type is 100%. (C) 2020 Elsevier Ltd. All rights reserved.