Scholarlink

<p>Robotic welding demonstrates considerable potential in the automation of metal joining processes, resulting in enhanced consistency. This study proposes a methodology for evaluating weld quality by utilizing a laser sensor in conjunction with a hybrid neuro-fuzzy model. The system, designed for AISI 1030 mild steel, utilizes a Design of Experimentation (DOE) methodology to collect empirical data and train the model. A MOTOMAN MA1440 robotic arm, integrated with an AccuFast-II laser sensor, was utilized to acquire real-time weld characteristics. The proposed model integrates fuzzy logic with artificial neural networks (ANNs) for predicting weld quality and is subsequently optimized using the Class Topper Optimization (CTO) algorithm. The model exhibited a high level of prediction accuracy, as indicated by R-squared values of 1.0, 0.99677, 0.99851, and 0.97561 for the training, testing, validation, and overall WQCI datasets, respectively. The process parameters obtained from the CTO analysis yielded a WQCI of 0.824, exceeding the highest experimental value of 0.808, which reflects a 1.98% enhancement in weld quality. The system demonstrated strong performance on both straight and curved weld paths, achieving a positional error of less than 0.29 mm, which falls within the acceptable weld gap range of 1–1.6 mm. This study emphasizes the practical implementation of a neuro-fuzzy prediction system integrated with an innovative metaheuristic for quality control in robotic arc welding. The integration improves weld consistency, minimizes defects, and increases production efficiency, representing a notable advancement in intelligent manufacturing.</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

Neuro-fuzzy model optimization for laser sensor-based quality control for robotic welding of AISI 1030 steel

  • 期刊论文

  • 发表日期：

    2026年4月16日

查看全文
  • Amruta Rout,
  • Golak Bihari Mahanta,
  • Chiranjibi Champatiray,
  • B. B. V. L. Deepak,
  • Bibhuti Bhusan Biswal

摘要

Robotic welding demonstrates considerable potential in the automation of metal joining processes, resulting in enhanced consistency. This study proposes a methodology for evaluating weld quality by utilizing a laser sensor in conjunction with a hybrid neuro-fuzzy model. The system, designed for AISI 1030 mild steel, utilizes a Design of Experimentation (DOE) methodology to collect empirical data and train the model. A MOTOMAN MA1440 robotic arm, integrated with an AccuFast-II laser sensor, was utilized to acquire real-time weld characteristics. The proposed model integrates fuzzy logic with artificial neural networks (ANNs) for predicting weld quality and is subsequently optimized using the Class Topper Optimization (CTO) algorithm. The model exhibited a high level of prediction accuracy, as indicated by R-squared values of 1.0, 0.99677, 0.99851, and 0.97561 for the training, testing, validation, and overall WQCI datasets, respectively. The process parameters obtained from the CTO analysis yielded a WQCI of 0.824, exceeding the highest experimental value of 0.808, which reflects a 1.98% enhancement in weld quality. The system demonstrated strong performance on both straight and curved weld paths, achieving a positional error of less than 0.29 mm, which falls within the acceptable weld gap range of 1–1.6 mm. This study emphasizes the practical implementation of a neuro-fuzzy prediction system integrated with an innovative metaheuristic for quality control in robotic arc welding. The integration improves weld consistency, minimizes defects, and increases production efficiency, representing a notable advancement in intelligent manufacturing.