ANN diagnosis for defect detection and classification in two-layer printed circuit boards using supervised back-propagation algorithm

Abstract

In this work, the proponent makes use of Artificial Neural Network (ANN) to visually inspect and classify the defect found in two-layer Printed Circuit Boards (PCBs). The proponent trained and tested the data for pattern recognition using C language. The supervised back-propagation learning algorithm was used for training and testing of PCB patterns. This learning algorithm is suitable for training multi-layered neural network and for generating the deltas of all output and hidden neurons. Considering that training and testing the data would only provide outputs with respect to generated weights, the proponent makes use of another program for defect detection. Excel VBA macro program was used for commonality testing of actual versus expected outputs. Also, it was used in making PCB defect detection possible by marking each defective unit. The proponent modeled a bare PCB circuit with 80 × 44 dimensions. The PCB board was further divided into 32 panel sides, each with 10 × 11 dimensions. There were five defective units modeled in the first layer and there were 14 classified defects used in the second layer. These data were trained and tested successfully, accurately and reliably using ANN.