Abstract: Aiming at the problems of missing detection,wrong detection and low recognition efficiency in traditional visual detection methods,a product surface quality detection method based on deep learning was proposed to accelerate the detection efficiency of production line and improve the intelligent level of quality control.Starting from the basic process of product surface quality inspection,the product surface quality problem modeling was carried out.On this basis,an improved vgg16 network model was constructed for image recognition,which used Support Vector Machine(SVM) to replace the softmax classifier in VGG16 network model,and applied Sparrow Search Algorithm(SSA) to further optimize the super parameters of SVM,so as to enhance the classification accuracy of the model.At the same time,the image defect feature knowledge base was built and the surface defect data system of standard products was developed.Finally,the quality detection system of industrial cloud platform based on deep learning was designed and developed to realize the efficient interactive connection between production line,equipment and personnel,as well as the real-time collection,transmission,intelligent detection and data management of product surface quality data.The feasibility of the proposed model and method proposed was verified by the case of cast impeller.

Key words: quality inspection, VGG16 network model, support vector machine, sparrow search algorithm, industrial cloud platform

摘要： 针对传统视觉检测方法容易出现漏检、错检、识别效率低等问题,提出一种基于深度学习的产品表面质量检测方法,加速产线检测效率和提高质量控制智能化水平。首先,从产品表面质量检测基本流程出发,进行产品表面质量问题建模。在此基础上,构建改进的VGG16网络模型进行图像识别,该模型采用支持向量机(SVM)代替VGG16网络模型中的softmax分类器,并引用麻雀搜索算法(SSA)进一步优化SVM超参数,从而增强模型分类精度。同时搭建图像缺陷特征知识库,完善标准产品表面缺陷数据体系。最后,设计开发了基于深度学习的工业云平台质量检测系统,实现产线、设备、人员之间的高效交互联通,以及产品表面质量数据的实时采集、传输、智能检测和数据管理,采用铸造叶轮案例验证了所提模型和方法的可行性。

关键词: 质量检测, VGG16网络模型, 支持向量机, 麻雀搜索算法, 工业云平台