基于LSTM算法的冷连轧机架振动动态预警  

Dynamic early warning of cold continuous rolling stand vibration based on LSTM algorithm

作　　者：张海辉[1] ZHANG Haihui(School of Automobile and Electromechanical Engineering,Zhoukou Institute of Vocational Technology,Zhoukou 466000,Henan China)

机构地区：[1]周口职业技术学院汽车与机电工程学院,河南周口466000

出　　处：《锻压装备与制造技术》2024年第5期144-146,共3页China Metalforming Equipment & Manufacturing Technology

摘　　要：为了克服传统研究方法只根据工艺反应机制开展理论建模的缺陷,建立一种基于长短期记忆网络(LSTM)算法的冷连轧机架振动预警模型。网络预警结果准确性与效率也受到超参数直接影响,选择验证集均方误差作为目标函数,通过网格搜索方式寻优计算获得最佳超参数组合结果,构建最佳振动预警模型。研究结果表明:第一卷在310s形成了剧烈振动,第二卷位于开轧后615s形成了比第一卷更大的峰值,表现为剧烈振动特征。随着报警阈值的降低,第一卷和第二卷的提前报警时间均表现出单调增加的变化规律,符合实际情况。该研究对提高冷连轧机工作稳定性具有很好的实际指导意义。In order to overcome the shortcomings of the traditional research method,which is only based on the theoretical modeling of the process response mechanism,a vibration early warning model of cold rolling stand based on the long short-term memory network(LSTM)algorithm is established.The accuracy and efficiency of the network warning results are also directly affected by the hyperparameters,and the validation set mean square error is selected as the objective function,and the optimal hyperparameter combination results are obtained through the grid search method of optimization calculation to construct the optimal vibration warning model.The research results show that the first volume formed intense vibration at 310s,and the second volume was located at 615s after the opening roll formed a larger peak than the first volume,which was characterized by intense vibration.With the reduction of the alarm threshold,the first volume and the second volume of the early alarm time show a monotonous increase in the law of change,in line with the actual situation.The study has good practical significance for improving the stability of the cold rolling mill.

关 键 词：冷连轧 轧机振动 LSTM神经网络 预报 模型 

分 类 号：TG333.13[金属学及工艺—金属压力加工]