Purpose This study addresses the challenge of extracting effective state features for intelligent fault diagnosis in elevator door systems, aiming to enhance model representational capability for reliable decision-making and accident prevention in real-world engineering. Design/methodology/approach The proposed discriminative multi-scale residual network (DMSRN) uses a one-dimensional multi-scale architecture to capture rich and hierarchical features, combined with a novel feature discriminative enhancement strategy that enhances intra-class compactness and inter-class separability, thereby improving feature discriminability. Findings Comprehensive experimental results validate the superiority of the proposed approach. The DMSRN achieves an average diagnostic accuracy that exceeds the best baseline method by more than 18.16%, demonstrating a substantial competitive advantage in fault diagnosis for elevator door systems. Originality/value The key contribution of this research lies in the novel integration of multi-scale feature learning with a specialized feature discriminative enhancement mechanism. This combination offers significant practical value by improving diagnostic reliability in safety-critical elevator applications.
Li et al. (Mon,) studied this question.