A digital twin-assisted fault diagnosis method for stage lifts based on lightweight multi-scale convolution and broadcast self-attention mechanisms

To address the challenges of scarce abnormal data and limited model applicability in stage lift fault diagnosis, this study proposes a Digital Twin (DT)-assisted method that integrates virtual modeling with a lightweight deep learning framework. A high-fidelity virtual model is constructed via geometric modeling in SolidWorks and finite element analysis in ANSYS. This model facilitates virtual fault injection, generating essential fault data—such as unbalanced loads and overloads—thereby alleviating the scarcity of real fault samples. Subsequently, a Lightweight Multi-Scale Convolutional Broadcast Attention Network (L-MSBA-Net) is presented for effective diagnosis. Experimental results demonstrate that the accuracy of L-MSBA-Net achieves 97.12% under ideal conditions, while also exhibiting robust performance in noisy environments, thus fulfilling the real-time diagnostic requirements for stage equipment. This research provides both a theoretical framework and practical tools for ensuring the safe operation of stage equipment, offering an effective pathway for the engineering application of digital twin technology and aligning with the roadmap for intelligent fault diagnosis driven by digital twin innovations.