Training an Artificial Neural Network for an Effective PCB Defect Detection

The printed circuit boards (PCBs) are crucial components of most electronic devices. In the last decades, the PCBs' manufacturing process was significantly improved, mainly by surface mounted technology (SMT) and automatic optical inspection (AOI) implementation. The real data as an output from the AOI device used for our analysis have been composed in a real manufacturing company. The currently used AOI solution achieves an accuracy of 95.82%. The goal of our study was to train an artificial neural network (ANN) to detect the defect PCBs with the highest possible accuracy. Different approaches have been used for ANN training, such as the experimental approach, regression, and Taguchi method. The resulted PCA-ANN model combines principal components analysis (PCA) method for data dimensionality reduction and ANN for low quality products detection. Our proposed model increases the AOI accuracy rate by 3.95%.