Research on Fuel Nozzle Defect Classification Based on Ultrasonic Flaw Detection and Time-Frequency Fusion Deep Learning

Jian-Ting Lai, Ruo-Yu Yang, Yan-Xiao Jia, Jia-Yi Hu, Wei Yan, Bo-Liang Zhang

Ultrasonic flaw detection, Fuel nozzle, Wavelet transform, Fourier transform, Time-frequency fusion, Deep learning, Defect classification

As one of the core precision components of power equipment, fuel nozzles are prone to concealed defects invisible to the naked eye, including pores, cracks, and structural corrosion, due to long-term operation under high-temperature and high-pressure working conditions. These defects affect the combustion efficiency and operational safety of the equipment. Traditional ultrasonic flaw detection has certain limitations when facing complex structures and tiny defects. Therefore, this paper proposes a frequency-domain extraction method integrating Fourier transform and wavelet transform, and completes model training and verification on the ultrasonic flaw detection dataset based on the PyTorch deep learning framework. Experiments show that the proposed method outperforms single-signal processing methods in terms of accuracy, recall, and F1-score. This method can provide a reference for intelligent non-destructive testing and industrial automatic flaw detection.

Received: 13 Sep 2026, Received in revised form: 11 Oct 2026, Accepted: 15 Oct 2026, Available online: 20 Oct 2026

10.22161/ijaers.1310.1