Vibration prediction for abnormal elevator door system faults based on attention mechanism and neural networks with time-frequency domain features

Abstract

Frequent issues such as accelerated door opening and closing, belt bearing obstruction, obstruction at the upper sill, friction between door and door frame, belt tooth surface wear, and slider wear are primary causes of elevator faults. To prevent these failures, it is essential to assess the mechanical condition of elevator door system in real-time, monitor the development of their mechanical state, and achieve early warning of potential malfunctions. This paper proposes a CNN-LSTM-A-TF neural network-based method for predicting elevator door system faults. By deriving relevant features from both the time and frequency domains, the original one-dimensional vibration acceleration data is significantly enhanced, incorporating four distinct time-domain features and two frequency-domain characteristics that collectively provide a more comprehensive representation of the signal dynamics. Integrating convolutional neural networks (CNN), long short-term memory (LSTM) networks, and an attention mechanism, a predictive model is developed to capture spatial and temporal patterns, thereby improving prediction accuracy. Historical data is fed into the CNN-LSTM-A-TF model, where the attention mechanism allows the model to automatically prioritize the most significant features in the input data. The CNN-LSTM-A-TF model demonstrates a substantial improvement over the conventional LSTM model in terms of predictive performance, with reductions in RMSE of 0.9822, 0.005356, and 0.00183, as well as decreases in the standard deviation of MAE by 0.009206, 0.004146, and 0.00216 for three common fault types. Furthermore, comparisons with the CNN-LSTM model show even greater reductions in RMSE and MAE standard deviations, highlighting the enhanced stability and precision of the proposed model in vibration forecasting tasks. This study validates the model’s feasibility and precision, providing a theoretical foundation for elevator door system fault prediction.