Compound-Fault Diagnosis of Rotating Machinery: A Fused Imbalance Learning Method

Rotating machinery plays an important role in large-scale equipment. The fault diagnosis of rotating machinery is of great significance and can increase industrial safety. Up until now, most existing fault diagnosis techniques have been proposed under the condition that only a single fault will occur at the same time. However, in industrial applications, compound faults are more common to take place due to the tight coupling of different components. To diagnosis compound faults accurately is of great significance to the safe operation of industrial equipment. A fused imbalance learning method is proposed in this article exploiting the nonlinear-mapping ability of neural networks. The dimensionless parameterization combined with time-frequency transformation method is utilized to extract data features and construct different evidence sources. Basic probability assignment with nested structure is generated from a novel weighted extreme learning machine based on sensitivity analysis. Evidence combination is implemented to obtain a final inference about the compound-fault class. Experiments are conducted on a large rotating machinery fault diagnosis experimental platform. Both single faults and compound faults in bearings and wheel gears of the large rotating machinery are considered. Experimental results illustrate the effectiveness of the proposed method.