Lin Szu-Yin, Li Hao-Yu
Department of Computer Science and Information Engineering, National Ilan University, Yilan City, Taiwan.
Department of Information Management, Chung Yuan Christian University, Taoyuan City, Taiwan.
Front Neurorobot. 2021 Nov 11;15:762702. doi: 10.3389/fnbot.2021.762702. eCollection 2021.
Industry 4.0 has been a hot topic in recent years. The process of integrating Cyber-Physical Systems (CPS), Artificial Intelligence (AI), and Internet of Things (IoT) technology, will become the trend in future construction of smart factories. In the past, smart factories were developed around the concept of the Flexible Manufacturing System (FMS). Most parts of the quality management process still needed to be implemented by Automated Optical Inspection (AOI) methods which required human resources and time to perform second stage testing. Screening standards also resulted in the elimination of about 30% of the products. In this study, we sort and analyze several Region-based Convolutional Neural Network (R-CNN) and YOLO models that are currently more advanced and widely used, analyze the methods and development problems of the various models, and propose a suitable real-time image recognition model and architecture suitable for Integrated Circuit Board (ICB) in manufacturing process. The goal of the first stage of this study is to collect and use different types of ICBs as model training data sets, and establish a preliminary image recognition model that can classify and predict different types of ICBs based on different feature points. The second stage explores image augmentation fusion and optimization methods. The data augmentation method used in this study can reach an average accuracy of 96.53%. In the final stage, there is discussion of the applicability of the model to detect and recognize the ICB directionality in <1 s with a 98% accuracy rate to meet the real-time requirements of smart manufacturing. Accurate and instant object image recognition in the smart manufacturing process can save manpower required for testing, improve equipment effectiveness, and increase both the production capacity and the yield rate of the production line. The proposed model improves the overall manufacturing process.
工业4.0近年来一直是热门话题。将网络物理系统（CPS）、人工智能（AI）和物联网（IoT）技术整合的过程，将成为未来智能工厂建设的趋势。过去，智能工厂是围绕柔性制造系统（FMS）的概念发展起来的。质量管理过程的大部分环节仍需通过自动光学检测（AOI）方法来实施，这需要人力和时间来进行第二阶段测试。筛选标准还导致约30%的产品被淘汰。在本研究中，我们对目前更先进且应用广泛的几种基于区域的卷积神经网络（R-CNN）和YOLO模型进行分类和分析，剖析各种模型的方法及发展问题，并提出一种适用于制造过程中集成电路板（ICB）的实时图像识别模型及架构。本研究第一阶段的目标是收集并使用不同类型的ICB作为模型训练数据集，建立一个能基于不同特征点对不同类型ICB进行分类和预测的初步图像识别模型。第二阶段探索图像增强融合及优化方法。本研究中使用的数据增强方法平均准确率可达96.53%。在最后阶段，讨论了该模型在<1秒内以98%的准确率检测和识别ICB方向性的适用性，以满足智能制造的实时需求。智能制造过程中准确且即时的物体图像识别可节省测试所需人力，提高设备效率，并提高生产线的产能和良品率。所提出的模型改进了整体制造过程。
