A novel multichannel feature extraction method for ship bearings: composite multiscale weighted cross improved diversity entropy

The multichannel monitoring signals for ship bearings have the advantage of richer information compared with single-channel signals, but how to fully extract the feature information is still a major difficulty in realizing the intelligent diagnosis of bearings. The existing advanced cross-diversity entropy does not give enough consideration to the feature weights and fusion strategy of each channel and interchannel, which makes the accuracy and stability of the related diagnostic methods improved. To solve this problem, this paper proposes composite multiscale weighted cross-improved diversity entropy (CMWCross-IDE) by introducing a novel two-channel feature extraction strategy. Furthermore, the bearing diagnosis method is constructed by combining CMWCross-IDE and Least Squares Support Vector Machine,T-distributed stochastic neighbor embedding (LSSVM) classifier, and the collected fault data of ship rolling bearings and sliding bearings are used as experimental verification. Comparison with four advanced entropy methods shows that CMWCross-IDE has the best feature extraction effect and the highest diagnostic stability, and the average diagnostic accuracy is improved by 4–6% in comparison to existing advanced entropy methods. The experiments have fully proved that the multichannel feature extraction strategy CMWCross-IDE proposed in this paper has good superiority and versatility, which can provide effective support for multichannel signal monitoring and intelligent fault diagnosis of ship bearings and even other mechanical equipment.