Determination of Joint Defects in Copper Tube Induction Heating Brazing Area Using Infrared Thermal Image Based on CNN Algorithm

Due to its excellent processability, thermal conductivity and high corrosion resistance, the copper tube applied to the heat exchanger is joined by the brazing process. In order to improve the performance of heat exchangers, it is essential to inspect the joint quality of copper tubes, but it is difficult to identify defects in tube-shaped joints without cutting. To solve this problem, this study proposes a new detection method based on the Convolutional Neural Network (CNN) model to detect joint defects that occur in brazing joints of copper tubes. In the experiment, a brazing joint using high-frequency induction heating was performed on a 12.71 mm diameter copper tube, which is mainly used in heat exchangers, and the joint failure was judged based on the penetration depth of the filler material measured by vertically cutting the joined copper tube. In addition, thermal image data having a data structure of 80 × 80 pixels per frame was collected to be used as data to determine whether the brazing joint is defective in real time. Finally, using the collected thermal image data, we developed a CNN model with a structure that applies different hyperparameters to determine whether or not the joint of copper tube is defective. The selected CNN model produced an f1 score of 0.991 and a recall of 99.73%, and formed the basis for developing a system that identifies defects in brazing joints of copper tubes through thermal image data obtained in real time.