Combination of inverse problem and neural network for thermal behaviour calculation of mould process based on temperature measurements in plant trial

Abstract

Heat transfer between mould and strand has a critical influence on billet quality, caster productivity and operating safety. It is very important to obtain the correct distributions of temperature and heat flux, and many studies are made on the calculation methods of heat transfer between strand and mould, aiming to reduce the computation time and improve the calculation accuracy. In the present paper, based on measured data of temperature and heat flux during round billet continuous casting, the calculation method which combines the online measurement data and numerical simulation was investigated. Through identifying the local thermal resistance and its distribution between the mould and the strand by an inverse heat transfer model, the heat flux and shell thickness profiles were calculated. To avoid the iterative solution by inverse model, a faster alternative model using an artificial neural network was developed to predict the thermal resistance from the measured temperature. After training, there is an exact correspondence between the observed temperature values and the thermal resistance. The calculation results obtained by the combination of neural network and numerical simulation can correctly reflect the characteristics of non-uniform heat transfer around the mould circumference, which provides a worthwhile and applicable method for online calculation and visual technology of heat transfer and solidification in continuous casting mould.