An improved mechanical fault diagnosis algorithm based on weighted entropy fusion and modified DS theory

With the rapid rise of modern industrial technology, how to ensure the safe operation of mechanical equipment has become increasingly important. Accurate and effective mechanical fault diagnosis approach can ensure the timely and individualized processing of mechanical fault, which plays a significant role in the safe operation of mechanical equipment. In the field of mechanical fault diagnosis, the noise and interfere of measurement environment and the limited measurement precision of sensors will cause the diversity and complexity of fault signals, which increases the uncertainty of fault diagnosis system. In order to manage the uncertainty of mechanical fault diagnosis system, we adopt the idea of information fusion and information theory to establish a multi-sensor monitoring system based on entropy fusion and DS theory. The proposed fault diagnosis algorithm consists of two steps. Firstly, single sensor utilizes a weighted fusion method to combine four entropies of fault signal (Singular spectrum entropy, power spectrum entropy, wavelet energy spectrum entropy and wavelet space state feature entropy). The weighted fusion method comprehensively fuses fault signal's characteristics from the time domain, frequency domain and time-frequency domain, which can build an accurate evidence for single sensor. Then, a modified DS combination rule based on Lance distance function is put forward in multi-sensor monitoring system to fuse multiple evidences from multi-sensor signals. The modified DS combination rule fully considers the similarity of fault signals from multiple sensors and the difference of multi-sensor monitoring environment, which can obtain a reliable diagnosis result. The improved fault diagnosis algorithm can not only fuse multiple entropies of fault signal reliably under noisy environment, but also combine multi-sensor signals from different sensors. Experimental results and analyses reveal that compared to the contrast methods, the proposed algorithm identifies the correct mechanical fault accurately even with a faulty sensor. Therefore, the proposed algorithm can monitor the running state of mechanical equipment well, and further ensure its safe operation.