Physics-Informed Neural Network-Based Elevator Degradation Diagnosis and Early Warning.

Area of Science:

• Engineering
• Artificial Intelligence
• Predictive Maintenance

Background:

• Urbanization increases elevator density, raising concerns about system reliability and maintenance.
• Conventional monitoring methods struggle with complex conditions, noise, and detecting progressive degradation.

Purpose of the Study:

• To develop an advanced method for elevator health monitoring and early warning using physics-informed neural networks (PINNs).
• To improve the accuracy and reliability of degradation detection in elevator systems.

Main Methods:

• Multi-sensor data processing with time alignment and feature reconstruction.
• A dual-path acceleration estimation for stable dynamic state calculation.
• Embedding a simplified elevator dynamic model into PINN for parameter identification.
• Constructing electrical and dynamic residual indicators for system condition assessment.
• Implementing a time-accumulated risk model to detect progressive degradation.

Main Results:

• Achieved stable parameter convergence and effective system condition assessment.
• Demonstrated earlier detection of degradation trends compared to threshold-based methods.
• Reduced false alarms caused by transient disturbances.

Conclusions:

• The PINN-based approach offers an interpretable and practical solution for elevator predictive maintenance.
• This method enhances the intelligent operation and safety of elevator systems.
• It provides a more robust alternative to conventional monitoring techniques.