Deep Learning-Based Conversion of Phased Array Ultrasonic Imaging using U-Net

Ultrasonic testing is an important technique for non-destructive evaluation of internal defects in materials. Sectorial scan (S-scan) and total focusing method (TFM) are imaging algorithms employed in phased array systems. The S-scan allows fast image acquisition but has limitations in the image quality and resolution. The TFM provides high resolution and accuracy for defect detection, but with a high computational cost. In this study, we propose a deep learning-based approach for converting S-scan images into TFM images. First, we acquired both S-scan and TFM images through numerical simulations based on finite element methods. Using the obtained data, we used a U-Net-based model to convert images and evaluate the predictive performance of the model through evaluation metrics. The results indicated that the proposed model demonstrated high accuracy in high-resolution image prediction. This approach can improve the efficiency and accuracy of ultrasonic inspection, contributing to the assessment of component integrity in industrial and manufacturing fields. Furthermore, this is expected to enhance the development and application potential of ultrasonic image conversion technology.