基于多过程信号的轧辊磨削表面粗糙度智能预测

Intelligent prediction of surface roughness in roller grinding based on multi-process signals

由于轧辊磨削表面粗糙度预测困难,且其预测精度不足,为此,笔者提出了一种基于多过程信号的轧辊磨削表面粗糙度智能预测方法。首先,以砂轮转速、磨削深度、拖板速度和头架转速为变量,对轧辊进行了全因素磨削实验,采集了磨削过程中的多过程信号,即声发射信号、振动信号和主轴电流信号,测量了磨后轧辊的表面粗糙度;对信号进行了分段处理,强化了信号与粗糙度的关联,并对粗糙度进行了离散化处理,将回归问题转化为分类问题;然后,提取了各类信号在时域和频域上的众多特征值,并利用主成分分析法(PCA)对其进行了特征降维融合,构建了多种类型的特征输入;最后,利用网格搜索法优化了多层感知机(MLP)网络,得到了粗糙度的预测模型,实现了对轧辊磨削表面粗糙度的智能预测。研究结果表明:相较于单信号方案,多信号方案能够提供更全面、准确的信息;基于PCA的降维融合特征能进一步提高MLP网络的预测效果,其准确率为78.16%,F1值为0.7776,平均偏离距离为0.29。

Due to the difficulty in predicting the surface roughness of roll grinding and its insufficient prediction accuracy,an intelligent prediction method of rolling surface roughness based on multi-process signals was proposed.Firstly,with the wheel speed,grinding depth,carriage speed and headstock speed as variables,a full-factor grinding experiment was carried out on the roll,and multi-process signals during the grinding process were collected,namely acoustic emission,vibration and spindle current signals,the surface roughness of the roll after grinding was measured.Then,many eigenvalues of various signals in the time domain and frequency domain were extracted,and feature dimension reduction and fusion were carried out by using principal component analysis(PCA),and various types of feature inputs were constructed.Finally,the multi-layer perceptron(MLP)network was optimized by grid search method,and the prediction model of roughness was obtained,which realizes the intelligent prediction of the surface roughness of roll grinding.The results show that the multi-signal scheme can provide more comprehensive and accurate information than single-signal scheme;the reduced dimensional fusion feature based on PCA can further improve the prediction effect of MLP networks,and the accuracy,F1-score and mean deviation distance are 78.16%,0.7776 and 0.29 respectively.