A Novel Integrated Fault Diagnosis Method Based on Digital Twin

Abstract Fault diagnosis plays a crucial role in the actual production activities of enterprises. In recent years, with the development and popularization of Internet of Things (IoT) technology, the efficient acquisition and storage of actual production data have become possible, enabling data-driven methods based on deep learning to achieve remarkable results in the field of fault diagnosis. However, existing technologies still have issues, such as less consideration of the temporal information of fault occurrences and the imbalance between normal and fault data in production activities, which can affect the performance of fault diagnosis. To address these problems, this paper proposes a novel integrated fault diagnosis method, comprehensively considering data balance, feature extraction, and temporal information at the time of fault occurrence.This method is established based on two key processes: the creation of a dataset using Digital Twin technology and the development of an integrated fault diagnosis model (CNN-BLSTM-Attention). The virtual production data generated under various operating conditions through Digital Twin technology provide us with a rich set of sample data. The integrated fault diagnosis model processes the input data using a sliding window to consolidate feature and temporal information, enabling precise fault diagnosis. This paper addresses the issue of small sample fault diagnosis for screw press faults and validates the effectiveness of the proposed method in practical applications. Experimental results demonstrate that, compared to existing fault diagnosis methods, the proposed method reduces noise sensitivity and significantly improves fault diagnosis accuracy, highlighting its superiority.