Research On Billet Quality Analysis Algorithm Based On Neural Networks

With the control technology of the steel rolling increasingly stable,the level of the steel making has gradually become an important measurement standard of the final product quality.But the control of steel making is more difficult than the steel rolling process control: how to control the composition and temperature in the process accurately,how to improve the quality of billet product and reduce the incidence rate of the billet internal defects,that becomes a current problem of the manager of iron and steel enterprises need to solve.This paper based on the software and hardware platform of the big data analysis project of the NO.5 steel-making line of ANGANG Steel Company Limited,and it establishes a decision model of strand quality based on neural network BP algorithm.It used to predicate the billet defects and reduce the final product scrap rate as a reference for the following rolling process.The main contents of this paper are as follows:(1)Firstly,this paper analyzed the classification,the causes and possible affected factors of billet quality,and then it makes sure the data range which is needed to be collected.After that,we also establish data acquisition interface with all levels of automation and information system,and collects the process data of the steelmaking procedure related to billet quality.(2)This paper further determines the important affected factors,which relates to the billet quality with correlation analysis technology,and confirms Key Input Value(KIV).After a certain amount of preprocessing and standardization,samples are formed for analysis and modeling.(3)Using neural network BP algorithm as the mathematical basis of the system and IBM SPSS data mining software,this paper develops the quality decision model of the billet quality to predict the billet quality,especially the internal quality.According to the characteristics and advantages of BP neural network model based on the metallurgical mechanism,it establishes a prediction model based on BP neural network algorithm and preliminary determines the structural parameters of BP network and learning algorithm,Finally the model training carried out by using 80% of the test set samples,and the model test carried out by another 20%.It proves that the prediction accuracy of the model can reach more than 90%.The samples in this system from the perspective of the metallurgical mechanism comprehensively consider the various factors which affect the quality of billet from steel-making to continuous casting the coverage is relatively comprehensive and the conclusions have the guiding significance and the actual function.The neural network BP algorithm is the most widely useful in this algorithm,and this paper puts forward some improvement measures,using variable step size learning method and adding the momentum in order to prevent network oscillations,to accelerate the convergence effect of the network and meet the requirements of field application.