Design of Defect Detection Tools in Through Hole Technology (THT) Solder Joints with Classification Based on Deep Learning

Abstract

A tool has been created for soldering defect detection to reduce the processing time for testing the quality of solder joints for THT PCBs in soldering practical activities at the Padang State Polytechnic. This tool is equipped with the help of a Logitech c920 Webcam and a Jetson Nano microprocessor which is used to store and run programs that have been created in Python programming software so that this tool can be used portably. The research method starts with creating a dataset, labeling, training data, and making tools. In this research, the YOLO Convolutional Neural Network method was used to help determine soldering defects with four classifications, namely hole soldering, bridge soldering, void on solder pad, and good soldering. The test results showed that the tool was able to detect these four classifications with mAP@0.5 values, Bridge 97.20, Good 54.50%, Hole 83.90%, and padless 57.20%. Overall, the tool can function well.