A fault diagnosis method of rotating machinery based on improved multiscale attention entropy and random forests

In order to precisely diagnose the fault type of rotating machinery, a fault diagnosis method for rotating machinery based on improved multiscale attention entropy and random forests is proposed in this study. Firstly, a nonlinear dynamics technique without hyperparameters namely multiscale attention entropy is proposed for measuring signal complexity by extending attention entropy to multiple time scales. Secondly, aiming at the insufficient coarse-graining of multiscale attention entropy, composite multiscale attention entropy is exploited to extract the features of rotating machinery faults. Then, t-distributed stochastic neighbor embedding is used to overcome the feature redundancy problem by reducing the dimension of the extracted features. Finally, the reduced-dimensional features are inputted into the random forests model to complete fault pattern recognition of rotating machinery. The results of the experiment indicate that the proposed method achieves the optimal diagnostic performance on two different fault datasets respectively, showing an extremely competitive advantage in comparison with conventional diagnosis models. Meanwhile, the proposed method is adopted to the actual hydropower unit without misjudgment, which verifies its strong adaptability. The research proposes a novel method for detecting faults in rotating machinery such as hydropower units.