Fault diagnosis and inference of hoist main bearing based on transfer learning and ontology

To overcome the challenges still faced by data-driven hoist main bearing fault diagnosis methods, including data imbalance due to a lack of fault samples under real operating conditions, diagnostic performance degradation of fault diagnosis models caused by significant differences in data sample distribution under varying conditions, single fault diagnosis function, and a lack of reasoning analysis and localization for the causes of hoist main bearing system failures, a new fault diagnosis and reasoning method for hoist main bearing systems is studied, which includes two aspects: ① Bearing fault diagnosis based on convolutional neural network transfer learning and domain adaptation. Firstly, the collected bearing vibration signals are processed, and it uses continuous wavelet transform to decompose the original vibration signals and extract two-dimensional time-frequency images; secondly, to adaptively extract deep features from both the source and target domains, the pre-trained and fine-tuning approaches are employed for convolutional neural networks transfer learning, and it constructs a transfer convolutional neural network model to achieve adaptive extraction of deep features from both domains; then, a new supervised domain adaptation method is proposed, integrating dynamic joint distribution adaptation with enhanced separability, to further reduce the distribution differences between deep feature sets of the source and target domains while effectively enhancing the separability of feature data; finally, a patterns recognition classifier is trained based on the source domain features after domain adaptation, which achieves fault patterns recognition in the target domain. ② Hoist main bearing fault knowledge reasoning based on ontology. Initially, by analyzing hoist main bearing system fault knowledge, professional fault knowledge and expert experience are standardized and formalized to construct an ontology knowledge base for hoist main bearing faults; then, by using semantic mapping methods, fault phenomenon instances in the fault ontology knowledge base and data-driven fault status recognition results are associated, completing the entire process from fault patterns recognition to fault knowledge reasoning analysis; finally, based on the above research, a prototype system for hoist main bearing fault diagnosis and health management services has been designed and developed. To verify the effectiveness of the proposed methods, experiments analysis of cross-domain fault diagnosis under balanced and unbalanced training samples using two types of bearing fault datasets were conducted. The experimental results show that the proposed fault diagnosis model can achieve the average fault diagnosis accuracies of 98.71% and 91.00% under the two datasets, respectively, which significantly outperforms comparative models; additionally, experiments in fault knowledge reasoning and application verification were carried out, and it further demonstrates that the proposed hoist fault knowledge reasoning method can efficiently perform fault cause reasoning analysis and localization.