Development of Edge-based Deep Learning Prediction Model for Defect Prediction in Manufacturing Process

In a series of lens module manufacturing process for mobile cameras, many process parameters affect the final quality of the manufactured lens module. Defect prediction of lens module in manufacturing process using many of these process parameters can help design efficient manufacturing parameters at an early stage of production. In addition, at an early stage of the manufacturing process, predicted defect products can be discarded without going to the next stage, thereby avoiding unnecessary additional costs. Many existing approaches use shallow architectures in their prediction models that cannot learn features in multi-parameters sufficiently. In this paper, we propose a method to improve productivity and reduce the manufacturing cost by predicting the product quality using deep learning. We designed prediction models using SVM, DNN and CNN approaches for quality prediction where CNN prediction model showed the best performance. Furthermore, to enable real-time defect prediction on-device to improve productivity, we propose a low-cost and edge-based solution that does not rely on expensive server or cloud solution.