Use Digital Twins to Support Fault Diagnosis From System-level Condition-monitoring Data

TL;DR

This paper introduces a digital twin-based framework to enhance fault diagnosis in systems, reducing the need for extensive failure data and enabling component-level failure detection from system-level data.

Contribution

It presents a novel approach using digital twins to support data-efficient fault diagnosis and component failure localization from system-level condition-monitoring data.

Findings

Deep learning models trained with digital twins can identify 9 faults in a robot system.

The digital twin approach reduces the need for large labeled failure datasets.

Performance can be improved when digital twin models closely match real systems.

Abstract

Deep learning models have created great opportunities for data-driven fault diagnosis but they require large amount of labeled failure data for training. In this paper, we propose to use a digital twin to support developing data-driven fault diagnosis model to reduce the amount of failure data used in the training process. The developed fault diagnosis models are also able to diagnose component-level failures based on system-level condition-monitoring data. The proposed framework is evaluated on a real-world robot system. The results showed that the deep learning model trained by digital twins is able to diagnose the locations and modes of 9 faults/failure from $4$ different motors. However, the performance of the model trained by a digital twin can still be improved, especially when the digital twin model has some discrepancy with the real system.