Weld quality monitoring research in small scale resistance spot welding by dynamic resistance and neural network. (March 2017)

• Record Type: Journal Article
• Authors: Wan, Xiaodong; Wang, Yuanxun; Zhao, Dawei; Huang, YongAn; Yin, Zhouping
• Is Part Of: Measurement. Volume 99(2017)
• Page Start: 120
• Page End: 127
• Publication Date: 2017-03
• Subjects: Small scale resistance spot welding -- Titanium alloy -- Quality monitoring -- Dynamic resistance -- Multiple linear regression analysis -- Back propagation neural network; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2016.12.010
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: Features extracted from dynamic resistance are obviously affected by welding current. Weld quality shows the highest correlation with end resistance. Regression analysis could be used for a preliminary quality estimation. The BPNN could be adopted to further improve quality estimation accuracy. Abstract: Our study aims at developing an efficient quality monitoring system in small scale resistance spot welding based on dynamic resistance. The dynamic resistance variation was related to weld nugget formation process. An initial resistance peak caused by asperity heating was detected. The second peak in dynamic resistance and single peak in dynamic voltage could be attributed to bulk material heating. An obvious interrelationship could be found between end resistance and weld quality. The features extracted from dynamic resistance curve were mainly influenced by welding current. The overall resistance level was dropped as welding current enlarged. The multiple linear regression analysis and back propagation neural network were then used to estimate the weld quality in combination with extracted features. Result of the regression analysis for quality prediction was basically satisfactory. The proposed neural network model showed a better performance than regression analysis regarding the maximum estimation error and root mean square error. Accuracy of the neural network based quality estimation could be further improved combining quality level classification strategy. Combination of the dynamic resistance measurement with neural network model was supposed effective to achieve the quality monitoring purpose in small scale resistance spot welding.