Transformer model improves cross machine fault diagnosis accuracy

FreqCondNorm: Towards Cross-domain Predictive Maintenance through a Frequency-Conditioned Transformer Foundation Model

Artificial Intelligence

Summary

Predictive maintenance models usually struggle to work well when applied to different machines or conditions, especially with limited labeled data and very different signal frequencies. The authors developed a new transformer-based model called FreqCondNorm that uses a special frequency-aware normalization to handle varied time-series data together. They trained it on multiple datasets and found it greatly improved fault detection accuracy and could recognize faults in new machine data without extra training. However, the model did not improve predictions of how long a machine would last before failure, suggesting further work is needed there.

What this means in practice

  • For maintenance teams: Detect machine faults reliably across different machines and operating conditions using a single pretrained model.
  • For industrial machine operators: Monitor equipment health with fewer labeled examples by applying one model across various sensor frequencies and setups.

Authors

Zaynab Raounak, Camille LHermine, Zhiguo Zeng

Abstract

Deep learning predictive maintenance models suffer from poor transferability across machines and operating conditions, especially when labelled data are scarce and signals span five orders of magnitude in sampling frequency (1 Hz to ~100 kHz). We propose FreqCondNorm, a Transformer-based architecture that introduces a FiLM-style frequency-conditioned normalization layer to unify heterogeneous time-series within a single model. The architecture is pretrained on five public predictive maintenance datasets (CWRU, MFPT, UOC18, PRONOSTIA, CMAPSS) using masked auto-encoding and contrastive learning with balanced domain sampling. On fault diagnosis, the model achieves 99.2% accuracy on CWRU (+6.4 pp over CNN) and 82.1% zero-shot accuracy on MFPT, demonstrating strong transfer across sampling frequencies. However, the approach does not improve remaining useful life prediction, suggesting a mismatch between pretraining and RUL objectives that warrants future investigation.
View PDFOpen arXiv