Big Data Analysis Of Strip Mill Based On Machine Learning

With the rapid development of computer software and hardware,big data technology has been widely used in the fields of commerce,logistics and transportation,and has achieved remarkable results.In industrial production,the data collected by sensors can reflect the various status information of the equipment,and the amount of data is much higher than that in other fields.Due to its complexity,relying solely on manual labor has not fully and effectively tapped its intrinsic value.This subject takes the massive data collected during the production process of the double cold rolling mill of a certain factory as the research object,and realizes the big data analysis of the strip rolling mill through the method of machine learning.Firstly,it introduces the double cold rolling mill under study,the data object source form,mainly rolling mill PDA(Process Data Acquisition)data,rolling schedule information and rolling mill report data,etc.In order to achieve efficient data analysis of a certain working condition of the rolling mill,the production process of the rolling mill is refined into five working conditions: threading,acceleration,constant speed,deceleration,and shutdown.The combined signal time-frequency diagram is used as a sample and convolutional nerves are applied.The network(CNN)recognizes its working conditions.Secondly,wavelet time-frequency transformation is performed on the rolling mill production process data through data visualization,and the signal changes under different working conditions can be intuitively understood.In order to measure the correlation between the rolling process data,in view of the shortcomings of the traditional correlation coefficient method in the non-periodic time series data analysis,the maximum mutual information coefficient is proposed to calculate the rolling force,tension,The correlation between the two signals such as roll bending force,and the stepwise regression analysis of the rolling force,obtain the key factors that affect the rolling force under the condition of multi-signal coupling.Finally,for each coil of strip steel,only the corresponding deformation resistance value is measured at the set measurement point,it is difficult to grasp the distribution of deformation resistance of the entire coil,and its value is predicted.Calculate the friction coefficient based on the actual measured data on the spot,obtain the theoretical value of the deformation resistance by the inverse calculation method of the rolling force formula,take the error between the theoretical value and the measured value as the prediction target,and predict the deformation resistance of the strip through a two-layer feedforward neural network.Compared with directly using the actual measured value of deformation resistance as the neural network prediction target,the prediction accuracy is improved.Strip deformation resistance reflects its physical characteristics and can be used as a strip quality evaluation index,combining rolling mill production data and predicted deformation resistance value for fuzzy evaluation of strip quality.