A Semi-Supervised Enhanced Fault Diagnosis Algorithm for Complex Equipment Assisted by Digital Multitwins

Abstract

The accuracy of fault diagnosis technology is crucial for the reliable operation of complex machinery. However, traditional diagnostic methods often rely on large amounts of labeled data, making it difficult to address the challenge of scarce labeled data in real industrial environments. To tackle this issue, this article proposes a three-stage semi-supervised fault diagnosis method that combines digital multitwins and lightweight multiscale attention (MSA) mechanisms. By leveraging digital multitwins technology, we build a triplex pump mechanism simulation model in Simscape to obtain operational data for various typical fault modes. Additionally, a deep data twin (DDT) approach is employed for self-supervised data augmentation, effectively expanding the sample space and enhancing the model’s generalization capabilities. Furthermore, we design a lightweight multiscale attention network (LMAN), which utilizes multiscale convolution and channel attention mechanisms to enhance the extraction of fault features, thereby improving diagnostic accuracy. Under the framework of a three-stage semi-supervised strategy, labeled and unlabeled data are gradually integrated to boost the accuracy of the fault diagnosis model. Experimental results demonstrate that this method exhibits excellent classification capability across different labeling ratios, achieving a significant performance improvement, particularly in scenarios with limited labeled data. This study provides an efficient semi-supervised learning solution for fault diagnosis of complex machinery, offering strong potential for industrial applications.