Advanced Process Models for Analysis and Process Control of Continuous Casting of Iron

The systemic nature of the present research work is pivoted around the digitalisation of the horizontal continuous casting manufacturing process at a cast iron foundry, Tasso A/S, from an Industry 4.0 perspective. The context of digitalisation for this study is streamlined to strengthen the understanding of the casting process and develop a strategy for automation of the process.

The aim of the project is concretised through the formulation and development of advanced process models. The models for the casting process, in a broad sense, were effectuated from physics-based models, production data analytics, and artificial intelligence.

More specifically, the analysis of the process was engineered through 1) Numerical models that included fluid flow, heat transfer, and solidification physics; 2) Statistical models that deployed design of experiments (DoE), time-series modelling, and machine learning algorithms; 3) Application of sensor technology and experimentation, which facilitated the successful development of the process models.

The process models were divided into four major kinds; the first two were versions of numerical models, while the final two were categories of statistical models. The outcome of the models is integrated, and built upon each other, to achieve the overall objective of process knowledge and automation. The four models are as follows:

Numerical Model, Version I: A 3-dimensional steady-state numerical model simulated to study the effect of metal-mould air gap on solidification of the molten melt. The outlet shell thickness decreased quadratically when the length of the metal-mould air gaps was increased from 0 mm to 150 mm in increments of 25mm. The overall result of this model was a working model validated against experimental results used for further analysis.

Numerical Model, Version II: A 3-dimensional steady-state numerical model, built on Version I, with an extended metal strand exposed to the atmosphere outside the mould. The outcome of this model was to establish the output space from the input design space materialising from the category I model.

Statistical Model, Category I: A statistical meta-model, built upon the results of version II, for establishing a steady-state causal relationship between the relevant input process and output parameters. The meta-model was based on the concepts of Latin Hypercube Sampling and Gaussian Process Regression. The overall result was that the pulling speed was the most influential input process parameter in the casting process.

Statistical Model, Category II: A combination of statistical models resulting in a customised Engineering Process Control (EPC) scheme or control strategy. The two components of the category II model are as follows:

• Predictive Modelling: A machine learning model that predicts the inherent operational behaviour of the response process parameter emerging from producing different sizes of products. The response process parameter was temperature drop per pouring cycle for the round grey cast iron product family.

• Transfer Function-Noise Modelling: A time series model that estimates a real-life and dynamic (time-dependent) relationship between the input pulling speed and output bar temperature. The relationship was developed for the round ductile cast iron of cross-section 140 mm in diameter.

The procedure of the customised EPC scheme has been formalised. It has been developed for a part of the product family produced at Tasso A/S. The complete customised EPC scheme has to be developed in future. Moreover, for a robust control strategy, the complete EPC scheme should be integrated with Statistical Process Control, in future, for smoother operations.

A qualitative assessment of the Sustainable Development Goals (SDGs) for the collaborative Industrial PhD project between DTU and Tasso suggested that the most impacted were SDG 12 (Sustainable Consumption and Production), SDG 17 (Partnerships for the Goals), SDG 9 (Industry Innovation and Infrastructure), and SDG 8 (Decent Work and economic growth).

A graphical abstract depicting the overall workflow and results of this work is presented after the Danish abstract.