Accurate prediction model of bead geometry in crimping butt of the laser brazing using generalized regression neural network

• Published in: IOP conference series. Materials Science and Engineering Vol. 103; no. 1; pp. 12036 - 12044
• Main Authors: Rong, Y M, Chang, Y, Huang, Y, Zhang, G J, Shao, X Y
• Format: Journal Article
• Language: English
• Published: Bristol IOP Publishing 09.12.2015
• Subjects: Algorithms; Artificial neural networks; Back propagation; Back propagation networks; Beads; Crimping; Errors; Folding; Laser beam brazing; Mathematical models; Mean square errors; Network reliability; Neural networks; Prediction models; Predictions; Regression; Root-mean-square errors; Stability
• ISSN: 1757-8981; 1757-899X
• DOI: 10.1088/1757-899X/103/1/012036

There are few researches that concentrate on the prediction of the bead geometry for laser brazing with crimping butt. This paper addressed the accurate prediction of the bead profile by developing a generalized regression neural network (GRNN) algorithm. Firstly GRNN model was developed and trained to decrease the prediction error that may be influenced by the sample size. Then the prediction accuracy was demonstrated by comparing with other articles and back propagation artificial neural network (BPNN) algorithm. Eventually the reliability and stability of GRNN model were discussed from the points of average relative error (ARE), mean square error (MSE) and root mean square error (RMSE), while the maximum ARE and MSE were 6.94% and 0.0303 that were clearly less than those (14.28% and 0.0832) predicted by BPNN. Obviously, it was proved that the prediction accuracy was improved at least 2 times, and the stability was also increased much more.