Non-contact intelligent diagnosis method for key components in energy equipment based on acoustic signal and deep learning

With the development of clean energy, the deployment of energy equipment such as natural gas compressors and wind turbines has been steadily increasing. Non-contact monitoring and fault diagnosis of key components in energy systems are critical for preventing operational failures and ensuring safety. Acoustic signals, as a prominent non-contact sensing modality, are highly sensitive to fault features. This paper conducted corresponding research and proposed a novel MF-IDCL (2D Feature Mapping - Improved Deep Convolution Lightweight Framework) method based on noisy acoustic signals. Non-contact acoustic monitoring experiments were designed and conducted on energy equipment gearboxes, and the collecting data were used to perform a comparative validation between the proposed framework and conventional methods. Furthermore, cross-domain validation was conducted to analyze the method's generalization capability. The results demonstrate that the proposed 2D feature mapping mechanism exhibits advantages in accuracy, computational efficiency, and adaptability over traditional feature extraction methods. The MF-IDCL framework achieves superior performance, with an accuracy of 87.9 % at SNR5 and 85.2 % at SNR0, while maintaining lower training costs and memory consumption. In cross-domain validation, the framework also showed promising applicability to non-acoustic data such as vibration signals. It is of positive significance in ensuring safety and stability of energy equipment.