Research On Prediction And Analysis Of Typical Internal Quality Defects Of Continuous Casting Billets

The internal quality of continuous casting billet has a decisive influence on the rolling process and the properties of the products.The main internal quality defects of billets are crack,segregation,porosity,and shrinkage cavity,and the common quality defects of different carbon billet internal will be different.In view of the above problems,65#high carbon billet and Q235 low medium carbon billet will be used as the research object.The target is to forecast and analyze the typical internal quality defects of billet with different carbon content.In order to achieve the target,the prediction model based on improved BP neural network algorithm and the quality diagnosis model based on dynamic kernel principal component analysis method are presented.Finally,the quality control system is established by combining the two models and the verification is carried out.The main research contents are as follows:(1)The quality prediction model is established based on the improved BP neural network algorithm.The typical internal quality defects of different carbon content square billet are analyzed based on the actual production data of steel mills.The causes of the defects are analyzed,and the production parameters that will be used for the modeling are put forward considering of the cause analysis and production experience.The segregation part is determined by 21 parameters,and the crack part is determined by 12 production parameters.In view of the traditional BP neural network model is unstable with a low accuracy,its input function is improved and its initial weights and thresholds are optimized by genetic algorithm based on theoretical analysis and practical operation effect,to ensure the stability and accuracy of the model.(2)The quality diagnosis and analysis model is established based on the dynamic kernel principal component analysis method.The kernel principal component analysis method has several problems:it has a bad tolerance for fault data;it will be significantly affected by nuclear parameters;its kernel function has a large amount of computation;and It can not consider the association between adjacent data.The following method is used to solve these problems:the method of outlier data processing,wavelet denoising,normalization processing and modeling data selection is used to preprocess a large number of data in the field to ensure its accuracy;the kernel parameters are determined by matrix similarity measure theory and genetic algorithm,in order to guarantee the classification effect of kernel function in the mapping space;feature vector selection method is used to reduce the computation of the kernel function;the parallel analysis method is used to calculate the length of delay,then construct dynamic augmented matrix to account for correlation between the adjacent sample data.Thus,the model of typical internal quality defects of square billet is built by using dynamic kernel principal component analysis method.For defective billet samples,CT2,new,i and CSPE,new,j statistical variables for each parameter is calculated,and the contribution diagram represented by them is drawn in order to display the parameters that may not be reasonable.(3)The control system of internal quality for continuous casting billets is implemented based on the prediction and diagnosis model and its accuracy is verified.When using two sets of models to monitor the internal quality,the operation is complex,and the output is not intuitive.In order to solve this problem,the internal quality control system is established.The system effectively combine the two groups of models,and realize their visualization.The prediction model of two parts is verified by two different sets of low power sample data.The prediction accuracy of the central segregation is 90%,the internal crack part reaches 87%,and the expected target is achieved.