基于深度学习的焊缝PAUT数据智能化分析方法

超声相控阵检测技术(PAUT)凭借其突出的技术优势被广泛应用在船舶、铁路、石油石化和航空航天等诸多领域。在焊缝超声相控阵检测(PAUT)中，对检测数据缺陷的识别定位目前多采用传统的人工判读方式，判读效率较低，对检测人员的判读经验有较高要求，难以满足自动化超声检测的要求。基于深度学习中的目标检测和跟踪算法构建智能识别模型，通过对焊缝超声相控阵检测的S、B扫图特征进行融合，并结合焊缝的三维结构信息，识别并定位出缺陷在焊缝中的三维空间位置。实验结果显示: 缺陷框的平均三维IOU(预测三维缺陷框和实际三维缺陷框的平均交并比)达到0.644 9，较为接近缺陷的真实空间位置，可以实现焊缝超声相控阵检测成像结果智能识别和定位。 

PAUT data intelligent analysis method of welding seams based on deep learning

In the welding seam phased array ultrasonic testing (PAUT), the traditional manual judgment method is used to identify and locate the defects in the inspection data. However, this method has lower interpretation efficiency and higher requirements for the experience of the inspectors, and it has difficulty for meeting the requirements of automated ultrasound inspection. In this paper, combined with the features of S and B scan images of welding seam PAUT and 3D structure of weld, an intelligent recognition model based on target detection and tracking algorithm in deep learning is proposed to identify and locate the weld defect automatically. The experimental results show that the average value of 3D IOU of the defects (the average intersection ratio of the predicted and the actual 3D defect frame) reaches 0.644 9, which is close to the real defects' location. This method can realize the intelligent recognition and positioning from PAUT imaging data in welding seam.