Elevator door machine systems are critical for safe and efficient elevator operation, yet traditional diagnostic methods often suffer from limited accuracy and predictive capability. This study introduces an innovative artificial intelligence framework for fault prediction and diagnosis in elevator door machines, addressing these limitations. A hybrid TCN-BiLSTM-ECA model is proposed, integrating Temporal Convolutional Network (TCN), Bidirectional Long Short-Term Memory (BiLSTM), and Efficient Channel Attention (ECA) mechanisms, to achieve precise real-time vibration pattern prediction by capturing temporal dependencies and enhancing feature extraction. For fault diagnosis, a novel Wavelet Packet Transform (WPT) and Multi-scale Convolutional Neural Network (MCNN) model is developed, utilizing two-level WPT decomposition to process multi-channel vibration signals into distinct subbands, followed by MCNN-based feature fusion for accurate fault identification. An experimental platform was designed to collect data on six typical fault cases, overcoming dataset scarcity. The WPT-MCNN model achieved an average fault diagnosis accuracy of 99.05% across 10 trials, a 40% improvement over non-signal-processed methods, with two-level WPT outperforming three-level decomposition by up to 20% in small-sample scenarios. Comparative analyses demonstrate superior accuracy, robustness, and generalization compared to traditional machine learning approaches. Key contributions include the TCN-BiLSTM-ECA prediction model, the WPT-MCNN diagnostic framework, and comprehensive experimental validation. These advancements provide a robust foundation for intelligent, predictive maintenance systems, transitioning elevator fault diagnosis from reactive to proactive strategies, enhancing safety, and operational efficiency.
Wan et al. (Tue,) studied this question.