Scholarlink

<p>Online weld joint inspection by non-destructive testing is necessary for modern joining industries. Nondestructive testing gained popularity through its dominance in examinations and reliability in confirming the part’s excellence. Joining dissimilar metals is preferable in industries due to reduction in the mass of components and less cost of manufacturing using the safety and structural requirements in various applications ranging from automotive to railway and naval trades. The weld joint imperfection examination plays a significant role in the manufacturing industry. A setup of Gas Tungsten Arc Welding (GTAW) has been proposed for joining stainless steel grades of 316L, 310 and 410 thick sheets of 150 × 60 × 3&#xa0;mm using variable process parameters. An autonomous technique known as Computer Aided Graphical User Interface (CAGUI) has been proposed for online detection and classification of multiform weld joint imperfections precisely comprising of crack, undercut, gas pores, porosity, tungsten inclusion, wormholes, lack of penetration, and non-defects in radiographic images using Support Vector Machine (SVM) and Recurrent Neural Network (RNN) developed using a MATLAB workbench. The support vector machine classifier has classified the weld images by finding the best hyperplane that separates all the weld joint images into defects and non-defect classes. SVM has classified the weld joint defects and non-defect images and confirmed their accuracy performance as 97.50% using the confusion matrix. It confirmed the lack of penetration defects are erroneous for gas pores. A RNN classifier handles the nonlinear weld joint images along with the parallel processing of information and flexibility in system. The feedforward neural network classified weld joint defects and non-defect and confirmed their accuracy performance as 98.75% using a confusion matrix. The confusion matrix confirmed that the lack of penetration defects is erroneous for undercut. The proposed CAGUI improved the computation period without disturbing the correctness of features selection.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Detection and Classification of Engine Exhaust Weld Joint Defects Using RNN and SVM on SS316L–SS410 and SS310–SS410

  • 期刊论文

  • 发表日期：

    2025年10月2日

查看全文
  • Rajesh Patil,
  • Magnus Löfstrand

摘要

Online weld joint inspection by non-destructive testing is necessary for modern joining industries. Nondestructive testing gained popularity through its dominance in examinations and reliability in confirming the part’s excellence. Joining dissimilar metals is preferable in industries due to reduction in the mass of components and less cost of manufacturing using the safety and structural requirements in various applications ranging from automotive to railway and naval trades. The weld joint imperfection examination plays a significant role in the manufacturing industry. A setup of Gas Tungsten Arc Welding (GTAW) has been proposed for joining stainless steel grades of 316L, 310 and 410 thick sheets of 150 × 60 × 3 mm using variable process parameters. An autonomous technique known as Computer Aided Graphical User Interface (CAGUI) has been proposed for online detection and classification of multiform weld joint imperfections precisely comprising of crack, undercut, gas pores, porosity, tungsten inclusion, wormholes, lack of penetration, and non-defects in radiographic images using Support Vector Machine (SVM) and Recurrent Neural Network (RNN) developed using a MATLAB workbench. The support vector machine classifier has classified the weld images by finding the best hyperplane that separates all the weld joint images into defects and non-defect classes. SVM has classified the weld joint defects and non-defect images and confirmed their accuracy performance as 97.50% using the confusion matrix. It confirmed the lack of penetration defects are erroneous for gas pores. A RNN classifier handles the nonlinear weld joint images along with the parallel processing of information and flexibility in system. The feedforward neural network classified weld joint defects and non-defect and confirmed their accuracy performance as 98.75% using a confusion matrix. The confusion matrix confirmed that the lack of penetration defects is erroneous for undercut. The proposed CAGUI improved the computation period without disturbing the correctness of features selection.