Understanding Data Correlations in Continuous Casting Systems for Autonomous Fixed Weight Cutting

Abstract

Continuous casting is the process whereby molten metal is solidified and cut into fixed weight billets. The key requirement is to cut the billets into fixed weight, so that the subsequent rolling steps can roll the billets into high quality fixed diameter, fixed length mills while avoiding wasting or insufficiency of the metal materials. To accomplish this goal, existing casting systems exploit camera systems to measure the cutting length and to control the flame cutter to cut the hot billet, which is called Length-based Cutting using Weight Feedback control (LCWF) approach. However, LCWF approach still provide unsatisfactory cutting performances in production, because the billet weight depends not only on the cutting length, but also on the billet temperature, density, cutting errors, and the billet dragging speed etc. To further improve the cutting weight accuracy, a data driven approach is necessary to investigate how the various features in the continuous casting system impact the cutting errors. In this paper, data mining on real datasets collected from Tangshan Iron company is conducted. We mine data features and data correlations with the cutting errors. Suggestions on how to improve the cutting accuracy using online learning approach are also provided.