采用残差结构和卷积神经网络的铣刀磨损研究
Research on Milling Cutter Wear Using Residual Structure and Convolution Neural Network
程胜明 1王雅君 1张昕晨 1冷峻宇1
作者信息
- 1. 大连工业大学机械工程与自动化学院,大连 116034
- 折叠
摘要
对于传统的卷积神经网络准确率低的问题,提高刀具磨损监测的准确性,提出一种具有残差结构的一维卷积神经网络.模型中采用两个残差块,残差结构能够跳跃一部分卷积层减少模型的训练时间,并将信息保留下来与下一层输出连接起来.采集的信息具有多维性,卷积神经网络能够自适应地提取相关特征,比传统机器学习需要人工依靠经验来提取特征更具有可靠性.实验结果表明,具有残差结构的卷积神经网络比传统的卷积神经网络不仅有较低的损失函数值,在准确度方面也有很好表现,提高了刀具磨损分类的精度.
Abstract
To solve the problem of low accuracy of traditional convolutional neural network and improve the accuracy of tool wear monitoring,a one-dimensional convolutional neural network based on residual structure was proposed in this paper.Two residual blocks were used in the model,and the residual structure can skip a part of convolution layer to reduce the training time of the model,and keep the information to connect with the output of the next layer.The collected information is multidimensional,and the convolutional neural network can adaptively extract relevant features,which is more reliable than traditional machine learning methods depending on manual experience to extract features.The experimental results show that the convolutional neural network with residual structure has lower loss function value than the traditional convolutional neural network,and it also has a good performance in accuracy,which improves the classification accuracy of tool wear monitoring.
关键词
一维卷积/残差结构/刀具磨损监测/机器学习Key words
one-dimensional convolutional/residual structure/tool wear monitoring/machine learning引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据