A comparison of two types of neural network for weld quality prediction in small scale resistance spot welding. (1st September 2017)

• Record Type: Journal Article
• Authors: Wan, Xiaodong; Wang, Yuanxun; Zhao, Dawei; Huang, YongAn
• Is Part Of: Mechanical systems and signal processing. Volume 93(2017)
• Page Start: 634
• Page End: 644
• Publication Date: 2017-09-01
• Subjects: Small scale resistance spot welding -- Titanium alloy -- Quality monitoring -- Electrode voltage -- Dynamic resistance -- Neural network; Structural dynamics -- Periodicals; Vibration -- Periodicals; Constructions -- Dynamique -- Périodiques; Vibration -- Périodiques; Structural dynamics; Vibration; Periodicals; 621
• DOI: 10.1016/j.ymssp.2017.01.028
• Languages: English
• ISSNs: 0888-3270

Abstract

Highlights: The second resistance peak and single voltage peak is caused by bulk metal heating. A higher welding current indicates a smaller dynamic resistance. A higher welding current leads to larger voltage peak and higher voltage change rate. Variation of extracted features is more sensitive to change of the welding current. Combination of both neural network may be useful for a better quality monitoring. Abstract: Our study aims at developing an effective quality monitoring system in small scale resistance spot welding of titanium alloy. The measured electrical signals were interpreted in combination with the nugget development. Features were extracted from the dynamic resistance and electrode voltage curve. A higher welding current generally indicated a lower overall dynamic resistance level. A larger electrode voltage peak and higher change rate of electrode voltage could be detected under a smaller electrode force or higher welding current condition. Variation of the extracted features and weld quality was found more sensitive to the change of welding current than electrode force. Different neural network model were proposed for weld quality prediction. The back propagation neural network was more proper in failure load estimation. The probabilistic neural network model was more appropriate to be applied in quality level classification. A real-time and on-line weld quality monitoring system may be developed by taking advantages of both methods.