Damage Detection and Identification on Elevator Systems Using Deep Learning Algorithms and Multibody Dynamics Models.

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Structural Health Monitoring

Background:

• Timely damage detection in mechanical systems is crucial for preventing catastrophic failures and optimizing maintenance schedules.
• Existing signal analysis methods for anomaly detection often struggle with scarce or non-existent data.
• Elevator systems require robust structural health monitoring to ensure safety and operational efficiency.

Purpose of the Study:

• To develop a novel approach for damage detection and identification in elevator systems.
• To address the challenge of limited vibration data in real-world applications.
• To create an effective structural health monitoring tool for elevators.

Main Methods:

• Combining physical vibration measurements with high-fidelity multibody dynamics models.
• Utilizing deep learning algorithms, including an ensemble of autoencoders and convolutional neural networks.
• Generating high-quality training data through multibody dynamics simulations.
• Developing and integrating a dedicated data acquisition system with an elevator cabin.

Main Results:

• The developed framework accurately identifies damages within the elevator system.
• The methodology demonstrates the potential for effective condition monitoring.
• The approach successfully trains deep learning models using a combination of simulated and real-world data.

Conclusions:

• The novel approach offers a powerful tool for structural health monitoring in elevator systems.
• This methodology can significantly reduce the time and effort required for manual damage localization.
• The integration of simulation data and deep learning enhances damage detection capabilities in data-scarce environments.