1:00pm - 1:30pm10 min Introduction + 20 min PresentationAL10 - Development, Testing and Deployment of the METRICS Pot Control System
Sylvain Fardeau1, Anne Resch2, Finn Breidenbach2, Kay Kessel2, Roman Düssel2, Valentin Olbricht2, Wim Toebes3, Thomas Bertram3, Matthias Dechent4
1Trimet France, Saint-Jean-de-Maurienne, France; 2Trimet Aluminium SE, Essen, Germany; 3Trimet Aluminium SE, Voerde, Germany; 4Trimet Aluminium SE, Hamburg, Germany
TRIMET Aluminium SE operates four different smelter technologies, all of which are facing the obsolescence of their respective process control systems. After analysing the solutions available on the market, TRIMET decided to develop its own system, METRICS, in order to retain its core business know-how and to be able to adapt quickly to new challenges in an ever-changing energy landscape.
This paper gives a brief history of the project and shows how some of the features of METRICS helped us to speed up development and encourage innovation: Off-the-shelf hardware, model-based design, automated testing, continuous integration and deployment. It also gives an overview of the rollout in our German plants: The Essen aluminium smelter was the cradle of the first trials and has now installed its first potline with 120 pots. The Voerde aluminium smelter completed the hardware rollout last year, while two thirds of its production was down due to the energy crisis, and restarted its pots in early 2024 using METRICS.
1:30pm - 1:50pmAL11 - An Explainable AI Approach for Predictive Bath Temperature Regulation
Prateek Kumar Lath1, Anish Das1, Amit Gupta2, Kamal Kant Pandey1, Anshu Mangal1, Kapil Kumar1
1Hindalco Industries Limited, India; 2Aditya Birla Science and Technology Company Pvt. Ltd
Consistent bath temperature regulation in aluminium smelting pots remains challenging despite technological advancements. In potlines, bath temperature plays a crucial role as it has a direct impact on the productivity of the pots. In the past, a few Artificial intelligence/Machine learning (AI/ML) models have been developed that offer predictive capabilities, but these models lack practical interpretability for control actions. This paper introduces a novel approach using SHapley Additive exPlanations (SHAP), an explainable AI framework, where this AI model not only does the prediction but also uses the same data to prescribe actions. This method, not only predicts temperature variations but also uncovers factors causing these fluctuations, filling the gap between continuous thermal measurements for real-time monitoring and control. It autonomously generates actionable insights 16 hours before measurement, enabling potroom engineers to implement effective control strategies. This has resulted in decreasing the number of pots with temperatures greater than 970 °C per day.
1:50pm - 2:10pmAL13 - EGA Pot Feed Systems - Challenges and Solutions
Budoor Ali1, Mohamad Abdulghafor Hussein2, Ajay Salian3, Auhood Aljasmi4
1Emirates Global Aluminium (EGA), United Arab Emirates; 2Emirates Global Aluminium (EGA), United Arab Emirates; 3Emirates Global Aluminium (EGA), United Arab Emirates; 4Emirates Global Aluminium (EGA), United Arab Emirates
Emirates Global Aluminium (EGA) has adopted several pot feeding systems (PFS) through the past 44 years. Each system has its advantages and disadvantages. This paper explores the PFS installed in EGA to feed secondary alumina for standard purity (SP) metal, the challenges with each system. The results of previous trials to overcome these challenges and to optimise the feeding system for consistent feeding are discussed. There are eight different PFS currently used in EGA, grouped in either continuous-filling system or time-based filling system. In this paper, all these systems will be compared, based on certain criteria and solutions to overcome some of their problems.
|