95 / 2025-04-15 10:20:28

Integrated Adversarial Fine-Tuning for Robust Intelligent Fault Diagnosis in Industrial Systems

fault diagnosis,adversarial defense,adversarial attack,deep neural networks

In the era of Industry 4.0, intelligent fault diagnosis (IFD) systems based on deep learning have become essential for maintaining the safety, reliability, and efficiency of industrial operations. However, these data-driven models are highly vulnerable to adversarial attacks, where subtle and often imperceptible perturbations to input signals can cause incorrect predictions, posing significant safety and economic risks. To address this challenge, we propose an Integrated Adversarial Fine-tuning (IAF) method that enhances model robustness without sacrificing diagnostic accuracy on clean data. Unlike conventional adversarial training that involves retraining the entire model and may degrade generalization, IAF introduces a lightweight fine-tuning mechanism. Specifically, it generates a diverse and representative set of adversarial samples by combining multiple attack methods and uses them to fine-tune pre-trained fault diagnosis models. This targeted adaptation allows the model to learn adversarial patterns while preserving its original feature representations and generalization ability. Extensive experiments on publicly available fault diagnosis datasets demonstrate that IAF significantly improves robustness against both white-box and black-box attacks, while maintaining or even enhancing performance on clean data. This study provides a practical and effective defense strategy for deploying reliable IFD systems in safety-critical industrial environments.