Doosan Corporation - Electric Furnace Steel Capacity Prediction
Country: Korea South
Organization: Doosan Corporation
Doosan Group is a South Korean multinational conglomerate corporation. It is the oldest running company in South Korea and is ranked as one of the world’s top 10 largest heavy equipment manufacturers. Within Doosan, our department, the Head of Digital(HoD), performs various roles to build and execute digital transformation strategies with the goal of transforming Doosan Group into an Intelligent Enterprise that innovates business and gains strategic insights centered on data. In particular, the AI Strategy Team under the HoD focuses on AI transformation to help company members utilize AI to innovate existing businesses and gain business insights.
Best Positive Impact Use Case
Best MLOps Use Case
Doosan Enerbility's steel mills produce various types of steel to meet the diverse needs of its customers. As the number of customers has increased and diversified, the steel mills have had to produce more types of steel using more raw materials. As a result, there are more than 150 different types of raw materials used in steel production.
In order to sustain and upgrade the steelmaking operations, which were originally operated efficiently, the company proactively introduced Smart Factory, and as one of the measures, the company carried out the 'Electric Furnace Steel Capacity Prediction Project' to improve the accuracy of steel capacity prediction by changing from the 'work order manual' and 'expert operation know-how' method to an AI steel capacity prediction model based on accumulated operation data.
The 'Electric Furnace Steel Capacity Prediction Project' aims to support AI model services with 98% accuracy and build a culture that can make data-driven decisions. This requires not only building an AI model with high performance, but also long-term performance management of the AI model and stable service provision.
Therefore, we selected Dataiku, a solution that supports everything from building a machine learning model to maintaining it after service deployment, and built a lifecycle management system for the AI model of electric furnace steel capacity based on it.
1. Data connection, preprocessing, and exploratory data analysis
Data generated in the electric furnace steelmaking process was stored and managed in the DB through the MES system, but in this project, the Dataiku platform's GUI was used to easily link and integrate DB data and gain visibility using Node. This enabled smooth and fast communication with the field, and domain knowledge and insights were acquired through this. In addition, the complex data preprocessing process was templated by adding visibility to the MLOps platform, making it easier for project performers involved in electric furnaces to use.
2. Building AI models
Through good communication with Dataiku, we were able to understand the meaning of the furnace operation and design a suitable model, and then select the most suitable model by conducting a quick performance test using the AI model function. As a result, we were able to build an AI model with 98% accuracy in predicting furnace steel capacity. This is a 21% improvement in accuracy compared to the previous model.
The entire analysis process of the electric furnace steel capacity project was assetized, and the lifecycle monitoring and management of AI models was automated using the scenario function to provide performance continuity for AI models. In addition, a weekly analysis report was published to compare and analyze the predicted results of the AI model for electric furnace molten steel production with the actual amount of molten steel produced through the dashboard, and contributed to the formation of a data-based decision-making culture by enabling continuous improvement of the manufacturing process.
Business Area Enhanced: Manufacturing
Use Case Stage: In Production
1. Financial Benefits, Environmental and social impact
The optimization of electric furnace molten steel production is expected to reduce electric energy costs and molten steel production costs, while reducing greenhouse gas emissions through energy savings.
2. Employee Benefits
Furnace Operations Planners: The AI Molten Steel Prediction Model enables them to plan weekly furnace molten steel production operations while predicting the molten steel output according to various raw material inputs in real time, enabling accurate operation planning compared to the previous molten steel output prediction formula.
Furnace site workers: Depending on the inventory status of raw materials, it may be necessary to input raw materials differently from the furnace operation plan, and at this time, the AI molten steel capacity prediction model can be used to check the molten steel production amount according to the input raw materials in real time, which helps the site workers make decisions on the use of raw materials.
Electric furnace operation manager: Provides reports and dashboards to compare AI prediction statistics and production performance data, providing insights for improving operations.
By integrating the AI model into the MES so that furnace operation planners, furnace field workers, and furnace operation managers can see the AI steel capacity predictions on each of the existing MES system screens, we made it easy for employees to use the AI steel capacity predictions with as little change as possible to the way they use the MES.
In addition, the MLOps configuration automated as much as possible the tasks for maintaining the accuracy of the AI steel capacity prediction model, so that no additional effort was required to manage the AI model.
A weekly report was issued to compare the AI predicted statistics of the furnace molten steel with the production performance data, providing a basis for furnace operation planners and managers to gain insights to improve operations.
Value Brought by Dataiku:
Spreading data analysis culture and decision-making by providing a data analysis environment
By introducing the Dataiku solution within the group, which is easy to use and can facilitate collaboration between departments in data analysis, we supported the spread of data analysis culture and decision-making through data analysis.
By providing a data analysis environment that can be used for practice and training, we have improved individual data analysis capabilities and contributed to the establishment of a data analysis culture.
Supported no-code/low-code data analysis to lower the barrier to entry for data analysis skills, making it easier for the data analysis culture to spread within the group.
By making the entire data analysis project visible, we were able to facilitate collaboration between different organizations working on data analysis.
Creating new opportunities through Doosan X Dataiku Co-innovation
On March 7, 2023, Doosan and Dataiku co-hosted an AI Strategy Seminar. The seminar was aimed at AI representatives from Doosan Group companies and third-party AI companies in key industries to introduce AI trends in manufacturing, examples of manufacturing innovation using AI technology, and ways to accelerate digitalization, and to share MLOps strategies and insights from the strategic partnership with Dataiku.
Through this, we promoted the strategic partnership and secured potential customers, and identified opportunities for Doosan and Dataiku to co-innovate to create rapid and meaningful AI outcomes in the future.