图谱数据深度学习融合模型及焊缝缺陷识别方法

A Deep Learning Fusion Model of Wave and Image Data for Weld Defect Recognition

针对当前利用超声衍射时差法(TOFD)图谱数据进行缺陷类型判定时以人工判别为主,主观性大、效率低以及缺乏从波形特征及图像特征进行集成分析的问题,提出了一种深度学习融合模型及焊缝缺陷识别方法。通过对TOFD检测原理及缺陷检测图谱数据特点进行分析,建立了综合考虑波形数据和图像数据的缺陷特征表征方法,实现了缺陷标准数据集构建;通过构建基于时间卷积网络(TCN)的波形序列数据分析模块、基于卷积神经网络(CNN)的图像数据分析模块以及特征自适应融合分类模块,建立了一种可以实现波形序列特征与图像特征综合分析的深度学习融合网络模型(DLFM)及模式分类方法。以企业实际TOFD检测焊缝缺陷数据对所提方法进行了验证,结果表明所提DLFM方法对缺陷类型的识别率明显高于单独使用基于TCN、CNN以及CNN-TCN的方法;所提方法拓展了现有深度学习模型的构建方法,并可以推广应用到其他模式识别领域,具有较强的通用性。

Aiming at the problems of manual identification,subjectivity,low efficiency,and lack of integrated analysis from waveform features and image features in using the time-of-flight,diffraction(TOFD)data for weld defect recognition,a deep learning fusion model(DLFM)and a weld defect recognition method are proposed.Based on the analyses of TOFD detection principle and weld defect detection data characteristics,a defect feature representation method considering waveform data and image data is set up,and the defect standard data set is established.Combining waveform sequence data analysis module based on time convolution network(TCN),image data analysis module based on convolutional neural network(CNN)and feature adaptive fusion classification module,the DLFM with pattern classification is constructed.A case study of TOFD weld defect recognition is conducted to illustrate the work.The results show that the proposed DLFM has higher defect recognition rate than the method based on CNN,TCN or CNN-TCN.The proposed method improves the traditional deep learning models,and can be applied to the other pattern recognition fields with stronger universality.