基于深度神经网络的活套与带钢打滑研究  

Research on Looper and Strip Slipping Based on Deep Neural Network

作　　者：王俊怡[1] 韦晓[1] 雷思雨 Wang Junyi;Wei Xiao;Lei Siyu(Baoshan Iron&Steel Co.,Ltd.,Shanghai 200941;School of Mechanical Engineering,University of Science and Technology,Beijing 100083)

机构地区：[1]宝山钢铁股份有限公司,上海200941 [2]北京科技大学机械工程学院,北京100083

出　　处：《冶金设备》2022年第3期9-14,18,共7页Metallurgical Equipment

摘　　要：某钢厂生产的镀锡板存在后续深加工时在剪切机组矫直机活套打滑现象,本文通过摩擦磨损实验对镀锡板生产工艺中的打滑问题展开研究。首先根据实际工艺参数为确定实验参数提供依据,并根据实际生产产品的表面微观形貌范围制备多种磨损试样,进行摩擦磨损实验记录对比不同形貌下摩擦副之间的摩擦系数。之后建立表面微观形貌参数与摩擦系数之间的统计模型,计算两两特征之间的皮尔逊相关性系数并进行低方差特征过滤,剔除相关性较高和影响不显著的参数,利用决策树方法分析特征的重要性,采用深度神经网络的方法建立以试样表面微观形貌参数为特征的摩擦系数预测模型,设定摩擦系数阈值提出重要表面微观形貌参数的控制范围。最后通过工艺参数的设定控制带钢表面微观形,实现对摩擦副之间摩擦系数的控制,最终减少打滑现象。The tin plate produced by a steel mill has the slippage phenomenon in the shear unit leveler lob during the subsequent deep processing. This paper studies the slippage problem in the tin plate production process through friction and wear experiments. Firstly, the contact pressure was estimated according to the actual process parameters to provide the basis for determining the experimental parameters, and various wear samples were prepared according to the surface microscopic morphology of the actual products. The friction and wear experiment records were made to compare the friction coefficients between friction pairs under different morphologies. Then, a statistical model was established between the surface microscopic morphology parameters and the friction coefficient. Pearson correlation coefficient between the two features was calculated and low-variance feature filtering was carried out to eliminate the parameters with high correlation and insignificant influence. The importance of features was analyzed by decision tree method. A prediction model of friction coefficient was established by deep neural network, and the control range of important surface morphology parameters was proposed by setting the friction coefficient threshold.Finally, the micro-morphology of strip steel surface is controlled by setting process parameters so as to realize the control of friction coefficient between friction pairs and finally reduce the occurrence of slipping phenomenon.

关 键 词：打滑 微观形貌 摩擦系数 深度神经网络 

分 类 号：TG580.6[金属学及工艺—金属切削加工及机床] TH117.1[机械工程—机械设计及理论]