Cleaning Profile Classification Using Convolutional Neural Network in Stencil Printing

This research proposes a novel framework to classify the stencil cleaning profile in the stencil printing process (SPP) on a real-time basis. The stencil cleaning operation is necessary to reduce printing defects. The proper control of the cleaning profiles selection process determines the quality and efficiency of the SPP. The wet profile provides high-quality cleaning but requires more time and materials. Dry cleaning is more efficient but cannot provide adequate cleaning quality with high levels of residue. This research aims to develop an intelligent model to guide the stencil cleaning profiles' decision-making and improve the SPP performance. Stencil cleaning is considered a sequential detection problem. Based on the historical printed boards' quality measures, a novel feature space is proposed by encoding the time-series printing data into images to better understand the abnormal and trend patterns. The resultant images are classified into the proper cleaning profile through a transfer learning-based convolutional neural network (CNN) model, denoted as cleaning profile classification (CPC) model. Transfer learning is adopted to overcome the limited data problem and enhance the model's generalization capabilities. The experimental results show that the proposed CNN architecture outperforms other complex state-of-the-art CNN structures in accurately classifying the cleaning profiles, which enhances the overall SPP quality and productivity. The CPC's actual implementation to control the cleaning profile indicates that the standard deviation of the printing results and the process capability has improved by 34% and 10%, respectively, compared to the best-known cleaning parameters.