在鋼鐵產業的產業鏈中,鋼板是廣泛被下游廠商用來製作各式鋼鐵製品應用的基材。鋼板表面若有瑕疵,輕則影響產品價格,嚴重則可能導致商譽的巨大損失,因此鋼板生產過程中的品質管理一直都是業者十分注重的議題。然而,傳統製造流程中只能以人工目測方式來檢測,不僅耗時也難以做到全面進行表面瑕疵檢驗。因此,透過自動光學檢驗來達到高效自動檢測,無疑是各家廠商在工業4.0時代,改善製程的重點項目。隨著深度學習技術的快速發展,大幅提升影像辨識相關應用的效果,本研究提出基於深度學習技術的自動光學影像表面檢測系統並與後端資訊系統整合。首先透過光學感測器拍攝製程的鋼板表面影像,再透過影像分析技術進行疑似有瑕疵鋼板表面影像的即時偵測(Real-Time Defect Detection),並針對疑似有瑕疵鋼板表面影像以卷積神經網路(Convolutional Neural Networks,CNN)技術進行缺陷分類,以控制鋼板的品質。後續並整合現有資訊系統達到及時偵測與提出嚴重缺陷警報,並保留追蹤原始缺陷影像與位置。研究結果不僅在學術上能對於深度學習技術相關應用的實務效果有更好的理解,也有助於相關業者在改善製程與提升品質上的參考.
In the industrial chain of the iron and steel industry, steel plates are widely used by downstream manufacturers to make various types of steel products. If there are defects on the surface of the steel plate, it will affect the price of the product in light, and may lead to a huge loss of goodwill in serious cases. Therefore, the quality management of the steel plate production process has always been a topic that the industry pays great attention to. However, in the traditional manufacturing process, it can only be detected by manual visual inspection, which is not only time-consuming but also difficult to perform comprehensive surface defect inspection. Therefore, achieving efficient automatic inspection through automatic optical inspection is undoubtedly a key project for manufacturers to improve their processes in the era of Industry 4.0. With the rapid development of deep learning technology, the effect of image recognition related applications has been greatly improved. This study proposes an automatic optical image surface detection system based on deep learning technology and integrates it with the back-end information system. First, the optical sensor is used to capture the surface image of the steel plate in the process, and then the image analysis technology is used to perform Real-Time Defect Detection of the surface image of the suspected defective steel plate, and the convolutional neural network is used for the surface image of the suspected defective steel plate Convolutional Neural Networks (CNN) technology is used to classify defects to control the quality of steel plates. Follow-up and integration of existing information systems to detect and issue critical defect alerts in a timely manner,and keep track of original defect images and locations. The research results not only provide a better understanding of the practical effects of deep learning technology related applications in academics, but also help relevant industry reference in improving process and quality.