Automated PCB Damage Detection in Early-Stage Maintenance Using Digital Image Processing Techniques Dianthika Puteri Andini, Endang Sukarna, Muhammad Yusuf Fadhlan, Farchan Aditya Johar
Politeknik Negeri Bandung
Abstract
Checking PCB conductor path damage physically in the electronic system maintenance process is still performed manually so that it requires relatively long time and high accuracy. Checking damage manually can cause failure due to human error, especially on PCBs that have complexity in the conductor path. This research aims to create a system that can detect damage to PCB conductor lines by utilizing digital image processing technology that can be used for electronic system maintenance processes. The system uses an external webcam to take pictures of the PCB. The image is processed and detected damage to the conductor line using the Faster R-CNN method. The detection results are displayed on the laptop screen with information in the form of a bounding box that marks the location of the conductor line damage and the type of damage. The types of damage that can be detected by the system are three types of conductor line damage, namely missing hole, mouse bite, and open circuit. The system uses a dataset of 375 PCB images with 10 different types of paths. The dataset used is divided into two parts, namely training images and test images with a ratio of 4: The system uses four training parameters, namely batch size, learning rate, IoU threshold, and step with values of 1, 0.0002, 0.7, 200,000, and one testing parameter, namely the minimum score threshold of 0.90. System testing is carried out on PCBs with masking as many as 15 pieces that have different positions, numbers, and types of conductor line damage. The system has a detection accuracy rate of 96.80% and an average detection time of 10.95 seconds.
Keywords: Training Model, CNN, Pre-processing, Hyperparameter
