Steel Producer Achieves Significant Savings With Ignition & Custom Machine Learning Model

Project Overview

ECON Tech’s project addressed the critical challenge of controlling molten steel superheat temperature in a steelmaking plant by implementing a unique solution that integrated Ignition with a custom Machine Learning (ML) model. Developed for Gerdau Corsa’s facility in Tultitlan, Mexico, the system used Ignition as a central hub to integrate disparate signals, track ladle movements, and provide real-time predictive guidance via an intuitive Perspective interface, significantly improving temperature consistency and process efficiency.

Challenges

The primary problem was the significant inconsistency and excessively high temperature of molten steel arriving at the tundish, frequently deviating from the target range of 1525−1545 °C and varying widely and far from desired (1515−1560°C). This common industry challenge directly impacted final product quality, leading to defects like non-metallic inclusions, cracking, and segregation, potentially requiring reprocessing or compromising the steel's mechanical properties. It also caused operational inefficiencies, including process stoppages and reduced productivity. Gerdau Corsa determined that the root causes were inadequate integration and control of thermal data across process stages (EAF, LF, CCM), siloed information often managed manually, and insufficient dynamic prediction to account for heat losses during transfers and ladle residence times.

To successfully implement this solution, ECON Tech needed to overcome significant hurdles in regards to technical complexity:

• Integrating and synchronizing real-time data from disparate control systems (PLCs) across the EAF and CCM stages presented substantial technical difficulty.
• Accurately tracking ladle movements and timings was crucial yet challenging to implement reliably for accurate heat loss calculation.
• Developing and validating a predictive model capable of handling the complex thermodynamics and process variability (steel grades, weights, processing times, additions, ambient factors) across multiple stages was a major undertaking.
• Establishing robust, real-time communication between the Ignition platform and the external Python-based ML model required careful API design and implementation.

Beyond the technical aspects of the project, there were also operational and human factors to overcome:

• A period of validation and trust-building to quell initial operator skepticism toward the ML model's predictions, which sometimes differed from traditional practices.
• Essential coordination efforts between automation, operations, IT, and data science teams proved complex.

Solution

ECON Tech leveraged Ignition as the core platform of this project to deliver an innovative solution:

• Unique Architecture: The system employs a novel architecture where Ignition acts as the central data and communication hub. It collects real-time process data from PLCs across the plant floor (EAF, CCM). Critically, Ignition interfaces via a custom API with an external, sophisticated Machine Learning model developed in Python. This model dynamically computes the optimal EAF tapping temperature needed to achieve the target tundish superheat, considering ladle tracking and process parameters. This integration of OT data via Ignition with advanced AI/ML computation represents a unique approach in steelmaking process control.

Ignition Modules & Functionality:

• Perspective Module: Provided an intuitive, web-based UI/UX, delivering a holistic, real-time view of critical process data and the ML model's temperature predictions/targets to operators and management. The interface incorporates the ApexCharts library to render rich, interactive, and visually appealing charts, enhancing data exploration and understanding for users. This replaced siloed, manual data management and facilitated informed, timely decisions.
• Tag Historian & Alarming: Used for collecting time-series data for model training/validation and monitoring process conditions.
• Scripting/Web Dev Modules: Utilized to build the necessary API communication link between Ignition and the Python ML environment.
• Overcoming Silos: Ignition successfully broke down data silos, integrating previously fragmented information (temperatures, timings, ladle status) into a unified operational view.

Result

This project resulted in several key upgrades to Gerdau Corsa’s operations:

• Improved Process Control & Quality: Gerdau Corsa achieved consistent tundish temperature control within the target range (1525−1545°C), reducing variability and leading to a 10% increase in heats meeting the target temperature. This directly translated to enhanced final product quality with fewer defects. For certain steel grades requiring VD processing, the model allowed superheat to be reduced by approximately 13°C compared to previous operator practice, while still meeting targets.
• Increased Efficiency & Productivity: The optimized temperature control and reduced variability led to smoother casting operations and energy savings in the ladle furnace of up to 1.86%. Overall LF productivity increased by up to 8.8% due to optimized heating times.
• Enhanced Decision-Making & Trust: The real-time data availability and reliable predictions from the Ignition-ML system empowered operators, overcoming initial skepticism and building significant confidence in the data-driven approach.
• Operational Benefits: Contributed to better refractory performance/lifespan due to more stable thermal cycling.

Further details on the innovative Machine Learning methodology employed in this project were published in the January 2023 issue of the magazine Iron & Steel Technology (an AIST publication), in an article titled “Energy Savings and Quality Reliability by Superheat Control Prediction to the Continuous Casting Through Machine Learning.” This paper, which elaborates on the technical foundation of the predictive models integrated with Ignition, received significant industry recognition. It was awarded the prestigious 2023 Digitalization Applications Best Paper Award by the Association for Iron & Steel Technology (AIST), highlighting the significance and successful application of advanced digital techniques in the steel sector enabled by this project.

Project Team

This project was brought to life through a collaborative effort leveraging diverse expertise. Sergio Feria served as the Senior Project Engineer, guiding the crucial design definitions. The core Ignition implementation and development were expertly handled by Arturo Rivera and Angel Rios. The sophisticated Machine Learning model, central to the project's success, was developed by Data Engineers and Specialists Esnardo Morales and Richard Marquez. This blend of process understanding, Ignition platform mastery, and data science specialization was key to achieving the project's goals.

Start Date: August 2023

Deploy Date: December 2024