由于机械零件在生产过程中许多因素的影响,金属工件的表面会出现不同类别的缺陷,进而减少零件寿命,威胁工作人员的生命.为解决机械零件表面缺陷图像模糊和缺陷种类较多的问题,研究首先对采集的图像进行预处理,然后设计一种粒子群优化算法改进支持向量机模型(Particle Swarm Optimization Support Vector Machine,PSO-SVM),最后构建自制的金属插头数据集进行应用实验.结果显示,在训练过程中,PSO-SVM模型在迭代136次时损失即可收敛,且平均准确率为0.989.在实际应用中,PSO-SVM模型在迭代22次后就可到达目标损失值,且分类准确率最高为0.996,分类识别延时为47 ms.综上所述,PSO-SVM模型有较好的性能与适用性.
Application of Intelligent Classification for Surface Defects of Mechanical Parts Based on PSO-SVM
Due to the influence of many factors in the production process of mechanical parts,different types of defects may appear on the surface of metal workpieces,which will reduce the life of the parts and threaten the lives of workers.In order to solve the problems of blurred images and multiple types of defects on the surface of mechanical parts,the research first preprocesses the collected images,then designs a particle swarm optimization algorithm to improve the support vector machine(PSO-SVM)model,and finally constructs a self-made metal plug dataset for application experiments.The results show that in the training process,the PSO-SVM model converges with a loss of 136 iterations,and the average accuracy rate is 0.989.In practical applications,the PSO-SVM model can reach the target loss value after 22 iterations,with a maximum classification accuracy of 0.996 and a classification recognition delay of 47 ms.Therefore,it can be seen that PSO-SVM model has better performance and applicability.
NETL
NSTL
万方数据
