Research on Fault Prediction Method of Elevator Door System Based on Transfer Learning

Area of Science:

• Engineering
• Artificial Intelligence
• Acoustics

Background:

• Elevator door system safety is paramount for accident prevention.
• Fault prediction in elevator systems is critical for proactive maintenance.
• Operational sound analysis offers a non-invasive method for fault detection.

Purpose of the Study:

• To develop a deep learning model for predicting the remaining useful life (RUL) of elevator door systems.
• To address variations in elevator operating environments and sound acquisition methods.
• To leverage historical sound data for accurate fault prediction in target elevator systems.

Main Methods:

• Collected opening and closing sounds from various elevators.
• Extracted acoustic features: A-weighted sound pressure level, loudness, sharpness, and roughness.
• Employed a Graph Neural Network (GNN)-Long Short-Term Memory (LSTM)-Bhattacharyya Distance domain adversarial neural network (BDANN) model with transfer learning.

Main Results:

• The GNN-LSTM-BDANN model effectively extracted deep features from transformed graph data.
• Transfer learning enabled knowledge transfer from historical data to predict RUL for target systems.
• Experimental results validated the model's capability in predicting potential failure timeframes.

Conclusions:

• The proposed deep learning approach accurately predicts elevator door system faults.
• This method enhances elevator safety through reliable remaining useful life prediction.
• The model's adaptability to different environments and acquisition methods makes it a valuable tool for predictive maintenance.