Using Object Detection Technology to Measure the Accuracy of the TFT-LCD Printing Process by Using Deep Learning

Currently, manufacturing industries have begun to employ Artificial Intelligence (AI) technology to enhance production efficiency, and reduce management costs and other problems encountered in their manufacturing processes, e.g., TFT-LCD manufacturing companies are typical examples. During their manufacturing processes, if a dedicated measurement machine is used, it takes time, also lengthening the production time and increasing the production cost. This study would like to measure the printing accuracy of the PI (Polyimide film) process in the cell segment process of a TFT-LCD panel manufacturing factory in central Taiwan by proposing a object detection architecture called Object Detection Measurement (ODM) system which measures PI Coater accuracy by using a deep learning object detection technology. In the ODM, an AOI (automatic optical inspection) machine is utilized to take photos for PI Coater. After that, Yolo v3, an object Identify algorithm, is used to evaluate accuracy of the PI Coater. When PI Coater’s printing accuracy is abnormal it notifies users to check the parameters of the proposed system. It also builds SPC (statistical process control) control graph for PI Coater accuracy. With the SPC, process engineers can query and monitor whether the accuracy has trended to abnormal.