PCB DEFECT DETECTION USING CONVOLUTIONAL AUTO ENCODERS

Abstract

The manufacture of printed circuit boards (PCBs) is one of the most important stages in electronic development. A very small defect on a PCB can lead to serious problems in the final product. Consequently, detecting and locating these defects are essential. An approach based on denoising convolutional auto encoders were used to detect defective PCBs and to locate PCBs it is used. Denoising auto encoders take a corrupted image and attempt to recover the intact image. By training this model with defective PCBs and forcing it to repair the defective parts, this model help us not only to detect all kinds of PCB defects and locate them, but also to fix them. The defective parts can then be located by subtracting the repaired output from the input. Experimental results indicate that this model is highly accurate in detecting the defective PCBs when compared to state of the art manual methods.