Selection of Heating Lines in the Line Heating Process for Steel Plates Using Faster R−CNN

In the process of surface forming through line heating of a steel plate, accurately determining the position of the heating lines is crucial to achieving the desired curvature. In this study, we propose a Faster R−CNN based model to predict the positions of heating lines from images of the desired curved surface. Initially, a finite element model for line heating on a steel plate is employed to obtain deformation analysis results based on the positions of the heating lines. The shape resulting from the combination of heating lines is acquired by superimposing the deformation analysis results of each heating line. Color map images of curved surfaces and corresponding information about heating lines are utilized as training data for the proposed model. The model is subjected to training through backpropagation to minimize the total output error. Testing the trained model revealed that the model accurately predicted the positions of the heating lines to achieve the desired surface. Furthermore, validation of the model on a surface with arbitrary curvature confirmed that heating the plate based on the predicted positions of the heating lines resulted in obtaining a curved surface that was very similar to the arbitrary target surface. Consequently, it was determined that the model could be effectively utilized to predict the positions of the heating lines in the line heating process.