Pragmatism in industrial modelling, applied to "ladle lifetime in the steel industry"

TL;DR

This paper demonstrates a pragmatic physics-based modelling approach to predict ladle lining erosion in the steel industry, emphasizing iterative development, industry collaboration, and potential integration with machine learning and cognitive methods.

Contribution

It adapts the PPBM methodology to a steel industry case, highlighting effective development practices and discussing future integration with ML and cognitive techniques.

Findings

Successful application of PPBM in steel ladle erosion prediction

Importance of iterative development and industry interaction

Potential for integrating ML and cognitive methods

Abstract

A methodology for building pragmatic physics based models (Zoric et al., 2015b) is here adapted to a use-case in the steel industry. The challenge is to predict the erosion of steel ladle linings, such that the model can support operators to decide if the lade lining can be used one more time or not. If the ladle has too thin lining 140 tons of hot liquid steel may escape out of the ladle, with huge consequences for workers and plant. The development was done with a very small core team (two developers), which is typical for many industrial developments. The adopted workflow for the development, challenges that were faced, and some model results are presented. One key learning is that development of models should allow time for maturing the process understanding, and time should be given for many iterations by "questions-responses and actions" at the various levels in the model development. The good interactions between development team and industry case owner is an important success factor. In this case the results of using the PPBM (Pragmatism in physics-based modelling) were good thanks to very successful interaction between development team and industry case owner. Combining or extending the model with use of ML methods and cognition-related methods, such as knowledge graphs and self-adaptive algorithms is discussed.