Construction of 3D solder paste surfaces using multi-projection images

This paper aims to develop the 3D surface construction applications for the solder pastes using multi-projection images and the neural network approach. The proposed solution uses the image features of multi-projection angles as the inputs and the laser surface scanning results as the outputs of the neural network model to perform precise 3D solder paste surface construction. In this manner, the proposed methodology can measure the 3D solder paste surfaces in a precise way like the laser scanning results. The advantages of this work is to use a low cost and high speed image solution to overcome the disadvantages of high cost and slow speed laser solution while the inspection accuracy is maintained. The multi-projection images are captured from the multi-channel light source and the coaxial light source to perform precise and efficient inspections, respectively. On the other hand, the back-propagation (BP) neural network approach is used to construct the 3D solder paste surface models for various solder pad geometries. Finally, the proposed system was experimentally verified. The experimental results showed that the multi-channel light source solution with pad based learning achieves 95% volumetric accuracy in average, and the coaxial light source with sub-area based learning just achieves 80% volumetric accuracy in average when compared to the actual laser surface scanning.