Automated inspection of solder joints-a neural network approach

Abstract

This paper describes a PC-based system for automated inspection of solder joints using neural networks. It presents extensive application of neural networks to solder joint quality data in the form of visual images. Numerous methods for data compression and feature extraction have been applied to enhance the performance of the neural networks. Up to 92 per cent accuracy in identifying solder joint defects was achieved using visual images. This discussion deals with visible light images only but all techniques may be extended equally to X-ray laminographic images as preliminary results from such applications indicate.