Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Data-Efficient Motor Condition Monitoring with Time Series Foundation Models

Deyu Li, Xinyuan Liao, Shaowei Chen, Shuai Zhao

TL;DR

The paper tackles data scarcity and class imbalance in motor fault diagnosis by leveraging pre-trained time-series foundation models (MOMENT and Mantis). It introduces a two-stage workflow that uses LogME to select transferable models and then applies either LoRA or full fine-tuning for classification, achieving high diagnostic accuracy with minimal labeled data. Empirical results show Mantis consistently delivers superior transferability and accuracy under scarce supervision, while MOMENT provides strong gains under data-limited conditions; scaling analyses reveal favorable gains with larger models and efficient fine-tuning. The approach demonstrates data efficiency, generalization, and CPU-friendly deployment, offering scalable solutions for intelligent motor condition monitoring in industrial settings.

Abstract

Motor condition monitoring is essential for ensuring system reliability and preventing catastrophic failures. However, data-driven diagnostic methods often suffer from sparse fault labels and severe class imbalance, which limit their effectiveness in real-world applications. This paper proposes a motor condition monitoring framework that leverages the general features learned during pre-training of two time series foundation models, MOMENT and Mantis, to address these challenges. By transferring broad temporal representations from large-scale pre-training, the proposed approach significantly reduces dependence on labeled data while maintaining high diagnostic accuracy. Experimental results show that MOMENT achieves nearly twice the performance of conventional deep learning models using only 1% of the training data, whereas Mantis surpasses state-of-the-art baselines by 22%, reaching 90% accuracy with the same data ratio. These results demonstrate the strong generalization and data efficiency of time series foundation models in fault diagnosis, providing new insights into scalable and adaptive frameworks for intelligent motor condition monitoring.

Data-Efficient Motor Condition Monitoring with Time Series Foundation Models

TL;DR

The paper tackles data scarcity and class imbalance in motor fault diagnosis by leveraging pre-trained time-series foundation models (MOMENT and Mantis). It introduces a two-stage workflow that uses LogME to select transferable models and then applies either LoRA or full fine-tuning for classification, achieving high diagnostic accuracy with minimal labeled data. Empirical results show Mantis consistently delivers superior transferability and accuracy under scarce supervision, while MOMENT provides strong gains under data-limited conditions; scaling analyses reveal favorable gains with larger models and efficient fine-tuning. The approach demonstrates data efficiency, generalization, and CPU-friendly deployment, offering scalable solutions for intelligent motor condition monitoring in industrial settings.

Abstract

Motor condition monitoring is essential for ensuring system reliability and preventing catastrophic failures. However, data-driven diagnostic methods often suffer from sparse fault labels and severe class imbalance, which limit their effectiveness in real-world applications. This paper proposes a motor condition monitoring framework that leverages the general features learned during pre-training of two time series foundation models, MOMENT and Mantis, to address these challenges. By transferring broad temporal representations from large-scale pre-training, the proposed approach significantly reduces dependence on labeled data while maintaining high diagnostic accuracy. Experimental results show that MOMENT achieves nearly twice the performance of conventional deep learning models using only 1% of the training data, whereas Mantis surpasses state-of-the-art baselines by 22%, reaching 90% accuracy with the same data ratio. These results demonstrate the strong generalization and data efficiency of time series foundation models in fault diagnosis, providing new insights into scalable and adaptive frameworks for intelligent motor condition monitoring.

Paper Structure

This paper contains 18 sections, 8 equations, 11 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Experimental Setting for Motor Fault Diagnosis
  3. The Occurrence of Stator Winding Faults
  4. Dataset Description and Preprocessing
  5. Proposed Motor Fault Diagnosis Framework
  6. Overview of the Process
  7. Model Assessment
  8. Supervised Fine-Tuning
  9. Parameter-Efficient Fine-Tuning
  10. Full Fine-Tuning
  11. Results and Discussion
  12. Selection of Foundation Models
  13. Model Assessment
  14. Foundation Model Fine-Tuning Results
  15. Fine-Tuning MOMENT
  16. ...and 3 more sections

Figures (11)

  • Figure 1: Experimental setup for PMSM fault data acquisition.
  • Figure 2: Stator winding of short circuit fault scheme where $R_\text{bypass}$ is bypassing resistance, and $R$ is stator circuit resistance.
  • Figure 3: Example of aligned PMSM fault signals: three-phase currents (Channels 1–3) and one vibration signal (Channel 4).
  • Figure 4: Label distribution of the 1% training subset used for model evaluation. Yellow denotes normal samples, while blue indicates fault samples with varying degrees of defect severity.
  • Figure 5: Proposed workflow for adapting pre-trained TSFMs to motor fault diagnosis, enabling efficient model selection and data-efficient fine-tuning.
  • ...and 6 more figures