Deep learning for non-contact machinery fault diagnosis: A review.

Abstract

In industrial production settings, mechanical equipment may suffer from wear, cracking, and fracture due to harsh conditions such as variable loads, high impacts, and strong disturbances, leading to performance degradation, production halts, and even catastrophic accidents. With the rapid advancement of deep learning, neural network-based approaches have emerged as the dominant paradigm for intelligent fault diagnosis, enabling automatic hierarchical feature extraction and complex pattern recognition from multi-source non-contact sensing data. Traditional contact based diagnostic methods, despite their high precision, are constrained by deployment limitations in extreme environments. Non-contact fault diagnosis technology, centered on acoustic sensors, infrared thermal imagers, and laser Doppler vibrometers, fundamentally eliminates operational risks associated with physical connections while providing innovative solutions for equipment monitoring in hazardous conditions. This review first introduces the basic background and current development status of non-contact fault diagnosis. It then classifies and discusses non-contact fault diagnosis methods, including acoustic emission, vibration and thermal imaging. Finally, existing challenges in current methods and potential future development directions are analyzed.