Research On Intelligent Defect Detection Technology In Touch LCD Module Manufacturing

With the increasing demand for high-precision touch screen,the manufacturing process of touch LCD modules has become more and more complicated,which makes it a challenging prob-lem to perform efficient quality inspection.The integration and preparation of the touch panel are the key processes in the manufacturing,which determine the quality of touch function.However,the inspection of two processes are mainly made manually,which not only lacks efficiency,but also cannot guarantee the yield rate.What's worse,it is hard to satisfiy the complex production requirements.Therefore,an automatic inspection system is crucial.The integration process is to make interconnections between the electrodes of the touch panel and the peripheral driving circuit with Anisotropic Conductive Film(ACF).The quality of the pro-cess is determined by the amount of ACF conductive particles,which is an indicator of conduc-tivity.Aiming at the difficulties such as shape,distribution and contrast of conductive particles on each bump,we propose U-ResNet,a deep neural network that can automatically learn features from labeled data.The full detection system of conductive particles is then proposed based on U-ResNet while a dataset of conductive particles is also made for model training.The experimental results show that compared with the handcrafted-feature based algorithm,our learning-from-data algorithm exceedingly outperforms the previous work,which meets the production demand as well.During the preparation process,a thin-film of Indium Tin Oxide(ITO)is coated on the glass to make touch-sensing circuits,which is essential to the touch panel.However,the electrical defects of the ITO circuit result in poor quality and even failure,especially the short defect.As the short defect is not distinct on the ITO circuit pattern,it is redefined with the end points of black isolation segments,which is expressed as an endpoint detection problem.Then the proposed U-ResNet is applied to learning the endpoint features which is the cornerstone of short-circuit defect detection algorithm.A dataset is also created for model training.The experimental results show that the U-ResNet can also learn the endpoint features well,while the whole algorithm meets the actual production requirements.In this paper,we propose a detection method based on deep learning,targeting to the inspetion of two essential processes in the touch LCD module manufacturing process,which is a preliminary exploration of artificial intelligent technology applied in the LCD manufacturing field.We believe it will make positive significance in China's LCD industry upgrade.