A neural network approach to Gaussian image based solder joint inspection

Abstract

An optical solder joint inspection system (OSJIS) has been developed for the automatic visual inspection of soldered parts on printed circuit boards. Its advantages over existing techniques include the detection of three-dimensional shape of specular objects with high reliability and speed. In this paper, a solder joint inspection scheme for a prototype of the OSJIS is proposed. The inspection scheme is composed of two steps: feature extraction and classification. In the feature extraction step, the Extended Gaussian Image (EGI) was adopted to extract the feature values of objects. The EGI is obtained by placing at each point on a Gaussian sphere a mass proportional to the area of elements on the object surface that have a specific orientation. Features of the estimated EGI including areas, center of mass, radius of gyration of mass in the zenith angle, and polygonality thus obtained are used as the basis for classification and inspection. In the classification step, a neural network classifies the solder joint according to the application requirements by using the features. Experiments were performed for SOP (small outline package)s and QFP (quad flat package)s in insufficient, normal and excess soldering conditions. Based upon observation of the experimental results, the proposed inspection scheme shows excellent consistency with those of visual inspection and a good accuracy of classification performance of 98.3%. (C) 1997 Elsevier Science Ltd. All rights reserved.