Adaptive weighting strategy for fault detection and diagnosis of rotating machinery components incorporating multiple operational conditions

Driven by the increasing needs for production safety in mass varieties production process, an adaptive weighting strategy is proposed for fault detection and diagnosis to weaken the influence brought by the change of operational conditions. To this end, considering the correlation between features and operational conditions, representative features extracted from multi-domain are selected using max-relevance and min-redundancy (mRMR) to establish the connection between features and operational condition. On this basis, a health index (HI) is fused for fault detection based on adaptive weight coefficients calculated according to the difference of features. Moreover, the input weight matrix of extreme learning machine (ELM) is redesigned based on the adaptive weight coefficients and the historical information to curb the impact of random factors. The effectiveness of the proposed method is demonstrated by rolling bearing test rig and industrial reciprocating pump. The results show that the rate of detection using the HI designed in this paper can reach 100% and the average diagnosis accuracy can reach 94.3% and 98.8%.