Helical fault diagnosis model based on data-driven incremental mergence

With the improving capability for acquiring real-time data in the field of intelligent manufacturing, the data driven machine learning approach has been an effective means for equipment fault diagnosis. Although incremental learning can make up for the shortcoming of machine learning that newly generated data must be combined with the original data for retraining, it cannot be carried out directly and effectively in the face of problems caused by fault data streams of massive-volume, imbalance, strong noise, and strong causality. In this paper, a helical fault diagnosis model based on data-driven incremental mergence is proposed to tackle this problem. Each helical cycle includes four procedures to handle incremental data blocks for imbalanced data processing, feature extraction and classification, effective example selection, and dynamic evaluation of features and examples. The effective features and examples are then transmitted to the next helical cycle to merge for preserving the fault information. The experimental results of bearing operation data demonstrate that the proposed model could efficiently solve the problem of incremental learning with massive and imbalanced fault data, significantly improve the recognition rate of minority faults, and reduce the time cost, thus contributing to meeting the specific requirements of equipment fault data.