Research on Ultrasonic Nondestructive Testing Method Based on Multimodal Sensor Fusion and Deep Learning

In this paper, an innovative ultrasonic nondestructive testing (NDT) method is proposed, which combines multimodal sensor fusion technology with deep learning algorithms to address the limitations of traditional single-modal ultrasonic testing in complex materials and structures. A multimodal detection system integrating ultrasonic, infrared thermal imaging and vibration sensors is constructed, and a deep learning architecture based on improved residual network is designed, which includes feature extraction branches optimized for each modal characteristic and an innovative cross-modal attention fusion mechanism. Experimental results show that compared with the traditional ultrasonic testing method, the proposed method improves the defect detection accuracy by 15.7^% and reduces the missed detection rate by 12.3%. It shows significant advantages in the detection of four typical defects, namely cracks, delamination, pores and inclusions, with an average accuracy of 93.6%. In particular, the method shows excellent anti-interference ability in a strong noise environment, and can maintain an effective detection rate of 59.2 % even under extremely harsh conditions of -5dB, which is due to the complementary redundancy characteristics of multimodal data and the adaptive weight adjustment ability of the attention fusion mechanism. This study provides a new technical path for high-reliability nondestructive testing in complex industrial environments, which is of great significance for improving the detection efficiency and reliability in practical application scenarios.