Improved Multi-scale Divergence Entropy Combined with Extreme Learning Machine Classifier for Rotating Machinery Fault Recognition

Yang Shi^*

• Shanxi Engineering Vocational College, Taiyuan, China

As the core equipment of industrial production, the prevention of bearing failures in rotating machinery is of great significance. Aiming at traditional feature extraction algorithms being susceptible to noise interference and inaccurate complex feature extraction, an improved multi-scale divergence entropy method is proposed for feature extraction of vibration signals. By combining multi-scale sample entropy and divergence entropy, the discrimination of signal features can be improved. In response to the high requirements for feature extraction and insufficient generalization ability of traditional fault classification algorithms, a regularized extreme learning machine model is proposed to identify faults in rotating machinery. The model introduced with regularization constraints avoids pathological matrices. Refined composite multi-scale divergence entropy is taken to extract features. When the scale was set to 20, the entropy value was minimized and the classification accuracy was highest, reaching 98.79%. When taking the Softplus function as the activation function and setting the neuron to 17, the model had the lowest loss rate and the highest average classification accuracy of 93.98% ± 0.94%. The running time of the model was relatively short, only 400ms. The proposed method improved the classification accuracy of rotating machinery faults and could provide new technical support for machine fault prevention work.