Experimental and mathematical analysis on spring-back and bowing defects in cold roll forming process

Failure analysis and material behavior in the cold roll forming process play an important role to optimize and modify the final product. Therefore, the effect of material behavior especially plastic anisotropy and process parameters on the final geometry of channel section, which is produced by cold roll forming technology was theoretically and experimentally investigated in the present research. Based on the experimental results, mathematical modelling of the spring back and longitudinal bowing was performed by regression and artificial neural network (ANN) for different values of the input parameters such as plastic anisotropy, strip thickness, angle increment or flower pattern, the width of section and inter distance between forming stands. DC03 cold-rolled steel because of widely use in this process was used to perform the experimental tests. The strip thickness, the angle increment and web width have a significant effect on the profile bowing. Subsequently, the influence of inter-distance between successive stands and the plastic anisotropy on the bowing defect was neglected. On the other side, the spring-back was intensively influenced by angle increment, web width and plastic anisotropy. In addition, the spring-back was not significantly affected due to the variation of inter distance and the strip thickness. In order to validate and suggest a suitable tool for predicting the characteristics of the spring-back and longitudinal bowing of the products, ANN model along with various multivariate regression methods were applied. It is concluded that ANN model with 10 neurons has the best performance for modeling the longitudinal bowing with the 96% accuracy and ANN model for spring back modeling has 81% accuracy with 5 neurons. The comparison of the proposed ANN with the regression methods indicates good accuracy of the 3-factor regression model in predicting the spring-back of the products. According to the results, the accuracy of the ANN model for estimating the durability parameters did not significantly follow the number of hidden nodes.