The Application of Machine Learning for Elevator Fault Diagnostic in Changsha, China

Abstract

Elevator guideway vibration fault diagnosis is critical to ensure the safety and stable operation of elevators. However, vibration signals often exhibit complex non-stationary characteristics, while the number of samples of abnormal vibrations is small. Deep learning enables automatic feature extraction from raw data, while transfer learning addresses the challenge of limited target domain samples, making them key techniques in fault diagnosis. This paper proposes a novel fault diagnosis method combining Multi-Channel (MC) One-Dimensional Convolutional Neural Networks (1D-CNN) with transfer learning for elevator guideway. 1D-CNN is adopted as the core framework due to its ability to extract local temporal correlation in sequential vibration signals. First, Empirical Mode Decomposition (EMD) decomposes vibration signals into multiple Intrinsic Mode Functions (IMFs), offering multi-frequency features as multi-channel inputs to improve the learning performance of 1D-CNN. Second, the Multi-Channel One-Dimensional Convolutional Neural Networks (MC-1DCNN) is pre-trained on the Case Western Reserve University (CWRU) bearing fault dataset to learn universal mechanical fault features. Finally, the pre-trained MC-1DCNN is transferred to the elevator guideway vibration dataset by freezing some lower convolutional layers and fine-tuning the rest higher convolutional layers, achieving high classification accuracy in small-sample scenarios. Experimental results indicate that the proposed approach achieves excellent fault classification accuracy and convergence, validating its effectiveness in application scenarios.

-