基于机器学习的铝合金铸锭夹杂物的定量检测方法  

Quantitative Detection Method of Inclusion in Aluminum Alloy Ingots Based on Machine Learning

作　　者：刘锦林 杜群贵[1] LIU Jinlin;DU Qungui(School of Mechanical&Automotive Engineering,South China University of Technology,Guangzhou 510640,China)

机构地区：[1]华南理工大学机械与汽车工程学院,广东广州510640

出　　处：《铸造技术》2021年第10期845-851,共7页Foundry Technology

摘　　要：为解决人工检测铸造铝合金锭含渣量效率低、误差大的问题,在K模检测技术的基础上,提出基于机器学习的利用卷积神经网络和误差逆传播网络双网络结合的方式检测方法,实现对铸造铝合金锭含渣量定量化检测。首先利用卷积神经网络定性分析K模断口面是否含渣,再利用误差逆传播网络和滑动窗口对K模断口面进行特征提取和定量计算含渣量,最后利用非极大值抑制算法对检测结果进行优化和后处理。结果表明,利用卷积神经网络判断K模断口面含渣与否的准确率为85.88%,能够较为准确地判断K模断口面是否含渣以及计算K模断口面的含渣量,满足生产实际的基本需求。In order to solve the problems of low efficiency and large error in manual detection of Inclusion content in cast aluminum alloy ingot,based on k-mode detection technology,a detection method combining convolutional neural network and error reverse propagation network based on machine learning was proposed to realize quantitative detection of Inclusion content in cast aluminum alloy ingot.Firstly,the convolution neural network was used to qualitatively analyze whether the K-Die fracture surface contained Inclusion.Then,the backpropagation network and sliding window were used to extract features and calculate the Inclusion content quantitatively.Finally,the non-maximum suppression algorithm was used to optimize and post-process the detection results.The results show that the accuracy of the convolutional neural network is 85.88%,which can accurately judge whether the fracture surface of K-Die contains Inclusion and calculate the Inclusion content of the fracture surface of K-Die,so as to meet the basic requirements of actual production.

关 键 词：铝合金锭 含渣量检测 K模 卷积神经网络 误差逆传播算法 

分 类 号：TG247[金属学及工艺—铸造]