基于深度学习的TC4钛合金零件微小缺陷超声相控阵检测图像降噪方法研究  

Research on Ultrasonic Phased Array Images Denoising Method for Micro Defect Detection of TC4 Titanium Alloy Parts Based on Deep Learning

作　　者：汪小凯[1,2,3] 蒋秋月 关山月 华林[1,2,3] WANG Xiaokai;JIANG Qiuyue;GUAN Shanyue;HUA Lin(Hubei Key Laboratory of Advanced Technology of Automobile Parts,Wuhan University of Technology,Wuhan 430070,China;Hubei Collaborative Innovation Center of Automotive Components Technology,Wuhan University of Technology,Wuhan 430070,China;Hubei Engineering Center of Material Green Precision Forming Technology and Equipment,Wuhan University of Technology,Wuhan 430070,China)

机构地区：[1]武汉理工大学现代汽车零部件技术湖北省重点实验室,武汉430070 [2]武汉理工大学汽车零部件技术湖北省协同创新中心,武汉430070 [3]武汉理工大学材料绿色精密成形技术与装备湖北省工程中心,武汉430070

出　　处：《航空制造技术》2023年第22期46-52,共7页Aeronautical Manufacturing Technology

基　　金：国家自然科学基金(U2037204,52175362);武汉市东湖新技术开发区“揭榜挂帅”项目(2022KJB128)。

摘　　要：钛合金因具有强度高、耐蚀性好、耐热性高等特点被广泛用于航空航天等领域,针对其内部微小缺陷超声相控阵检测过程中存在信噪比低、易漏检等问题,提出一种基于深度学习的微小缺陷超声相控阵检测图像降噪方法。首先通过钛合金试块相控阵检测试验获得缺陷含噪原始图像,采用Mask RCNN模型训练并构建高噪–低噪数据集,进而基于变分自编码器设计微小缺陷检测图像降噪模型,通过与传统滤波降噪、时频域降噪算法对比,证明所提出的算法可保留原始图像缺陷细节信息,与含噪原图对比,其峰值信噪比优化了11.35%,结构相似性提升154.17%。最后开展了某钛合金航空机匣环件超声相控阵检测试验,采用所提方法对环件内部φ0.2 mm平底孔缺陷检测图像进行降噪处理,有效降低了散射噪声对微小缺陷检测的影响,同时也证明所提降噪算法具有良好的泛化性能。Titanium alloy has the characteristics of high strength,good corrosion resistance and high heat resistance,and is widely used in aerospace and other fields.Aiming at the problems such as low signal-to-noise ratio,easy omission during ultrasonic phased array detection of internal micro defects,a deep learning based ultrasonic phased array detection image noise reduction method for micro defects was proposed.Firstly,the original images with defects and noise are obtained through the phased array detection experiments of titanium alloy test block,and the Mask RCNN model is trained to construct high–low noise data sets.Then,the noise reduction model of micro defects detection images is designed based on the variational autoencoder.By comparing with the traditional noise reduction algorithms,it is proved that the proposed algorithm can retain the defect details of the original image.Compared with the original image with noise,the peak signalto-noise ratio is optimized by 11.35%and the structural similarity is improved by 154.17%.Finally,the ultrasonic phased array testing experiment of a titanium alloy aviation casing ring was carried out.The proposed method was used to reduce the noise of the image with aφ0.2 mm flat bottom hole inside the ring,effectively reducing the influence of scattered noise on the detection of small defects,it’s also proved that the proposed noise reduction algorithm has good generalization performance.

关 键 词：超声相控阵 微小缺陷检测 图像降噪 深度学习 自编码器(AE) 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术] TG146.23[一般工业技术—材料科学与工程]