Stencil Imaging and Defects Detection Using Artificial Neural Networks

Due to its relatively low cost stencil printing is one of the most widely used processes in industry for solder paste deposition. The stencil used within the solder paste printing process is a crucial element for obtaining high quality prints. There are many studies on how to improve solder paste stencil printing and what the impact of stencil features (such as aperture size, shape, and orientation) is on the successful printing. However, there are only a few studies on traditional stencil defects detection. These are based on human experience and rather complicated mathematical and statistical data processing methods. In this paper, deep Convolution Neural Networks (CNNs) are used for the first time to detect stencil defects and the results exceed our expectations. Our goals are to create an intelligent data-driven model regardless of how complicated the data analysis and work experience are. In addition, an optical system was designed to collect images of stencils which were then regarded as inputs to the CNN for training and detection. 1-CNN and 3-CNN were proposed to detect stencil defects and 3-CNN achieved a remarkable performance. The result of 3-CNN demonstrated that even tiny defects which were difficult to be recognized by human eyes or other traditional methods could be detected.