This paper describes the application of several imagingtechnologies available at the Center for solder joint inspection. X-raylaminography was combined with artificial neural networks to classifysolder joints. Components with ball grid array, gull-wing and J-leadjoints were imaged and several neural network methods were used toidentify different classes of defects particularly significant to eachtype of joint. A novel probabilistic neural network approach for 2-Dimage classification has been developed which performs as well as orbetter than a conventional backpropagation network. The smear caused bythe laminographic process poses a great challenge to accuratereconstruction and subsequent evaluation of the object. An improvedmethod of accurately reconstructing the solder joint shape from thelaminographic images has been developed as part of this research. Themethod removes artifacts caused by out-of-plane contributions, noise,and smear due to rotation of the source around the object while formingeach laminograph, and can be adapted to consider the finite size of theaperture and X-ray scattering. Preliminary application of the method hasproduced dramatic improvements in the visual quality and signal-to-noiseratio for laminographs of experimental objects. More importantly, theability to accurately measure the dimensions of the objects being imagedhas been made possible by this approach. The possible extension of thiswork by using more X-ray projections and mathematically intensiveroutines brings this research into the realm of microtomography, whichcan help achieve more precise reconstruction at a much smaller scale. Anew method of microtomography has been developed that can exceedprevious limits in image resolution
展开▼
