Casting Optimization For Surface Defect Reduction Using Machine Learning

In the realm of industrial manufacturing, particularly within the steel production sector, optimizing production efficiency holds paramount importance in sustaining competitiveness and driving profitability. Steel plants, characterized by their intricate processes and heavy reliance on energy-intensive operations, face a myriad of challenges ranging from fluctuating market demands to rising operational costs and stringent environmental regulations.

Steel production stands as a cornerstone of global infrastructure, serving as a fundamental building block for construction, automotive, machinery, and countless other industries. According to industry reports, the global steel production reached over 1.8 billion metric tons in 2020, underscoring its indispensable role in modern society’s infrastructure and economic development.

In the steel manufacturing sector, ensuring product quality and minimizing surface defects are paramount for maintaining customer satisfaction, meeting industry standards, and upholding brand reputation. Steel manufacturers operate in a highly competitive and demanding environment, producing a wide range of steel products crucial for various industries, including construction, automotive, and infrastructure. Steel manufacturers are constantly seeking ways to optimize their production processes to reduce surface defects and improve product quality.

Several steel surface defects, such as longitudinal and off-corner cracks, originate from the continuous casting process. Casting speeds and temperatures, cooling rates, and the chemical composition of different steel grades can contribute to the problem, leading to reworked or even scrapped production.

Problem

Controlling continuous casting to maintain stable production conditions while minimizing surface defects is challenging. The variability in the chemical composition of different grades of steel and equipment age, along with desired casting temperature and speed set points can lead to increased surface defect rates.

Process engineers and operators in charge of continuous casting processes define set points such as casting temperatures, speeds, mould water flows and pressures to maintain high quality and stable casting of steel into semi-finished products (blooms, ingots, billets, slabs, etc.). These are set based on solidification profiles, temperatures and other process parameters, and rarely reflect variations in chemistry.

The current solution involves physics-based models and following historical best practices. These do not capture the full properties of each batch being cast. This approach leads to:

  • Financial loss due to rework and scrap of products with too many surface defects,
  • Environmental cost of incurring Scope 1 and 2 emissions due to the unnecessary rework or scrapping of produced steel.

Solution

Continuous casting process operators can use Fero Labs software to minimize surface defects by adapting to each batch and learning over time. Operating limits for caster set points can be configured within the software to enable safe recommendations that operators can follow with confidence.

A Live Fero Analysis for this use case presents two screens:

  • Detailed View: For production engineers to monitor production and direct action plans at any moment.
  • Simplified View: For continuous casting process operators, with critical information clearly presented.

Process & Business Outcomes

Significant reduction of scrap rates: With Fero Labs providing optimal recommendations for continuous caster parameters, chemistry and caster temperature variability no longer translate to increased surface defects. Since each batch is optimized to safely minimize surface defect rates relative to its specific chemical composition and casting temperature, continuous casters can expect a 18% decrease in scrap rates.

Decreased severity of surface defects: Fero Labs software not only minimizes the probability of surface defects, but also recommends operational changes that reduce the length of surface defects. With a full adoption of Fero Labs software on the production line, continuous casters can see up to 40% reduction in average surface crack lengths. Knowing that production will reliably meet its specifications leads to smoother operations.

Measurable cost savings minimizing rework: Static casting operating procedures for product grades lead to unnecessarily high surface defect rates. Some of these products can be reworked, which incurs additional rework costs, both in personnel and in energy. With a full adoption of Fero on the production line, continuous casters can expect up to 15% reduction in rework of semi-finished products with mild surface defects.

Commensurate Scope 1 and 2 carbon footprint minimization: The minimization of scrapped and reworked product directly translates to a commensurate reduction of Scope 1 and 2 emissions. Since more than half of EAF steelmaking’s carbon footprint falls into Scope 1 and 2, a reduction here can reduce the carbon footprint of production by up to 10%. Fero can provide reporting capabilities that directly track and account for this reduction.

Bridging Gap Between Disconnected Goldmine of Production Data and Industrial Knowledge

Fero Labs helps factories work better together by bridging the gap between the disconnected goldmine of production data and industrial knowledge inside every plant. The Fero Labs Profitable Sustainability Platform collects data and knowledge, and augments it with powerful Fero Machine Learning (ML) so factories can make more confident changes that drive profit and sustainability.

Harnessing Fero Labs, a factory creates an augmented workflow which allows for better use of raw and recycled materials, production time, and energy utilization. Teams can work 90× faster, using Fero’s AI powered simulated predictions or live optimizations. They can run root cause analyses in minutes, and make continuous process improvements that drive Profitable Sustainability. Fero Labs’s white-box explainable ML makes decisions clearer by showing the context and confidence levels behind every prediction and recommendation. This expands a plant’s production knowledge and drives better production results for manufacturers, all while minimizing emissions.