JFE Steel was formed in 2003 through the merger of the steel divisions of NKK and Kawasaki Steel Corporation, and, as of FY2023, it is the second largest crude steel producer in Japan and the 13th largest in the world. The company owns two integrated steelworks, the East Japan Works and the West Japan Works, and two manufacturing plants. It produces various steel products, including steel sheets and plates, shaped steel, steel bars, steel sheet piles, rails, electromagnetic steel sheets, wire rods, and steel pipes.
Steel products are made from coal and iron ore and go through numerous manufacturing processes before becoming finished products. A blast furnace, which reduces iron ore to produce iron, is a massive facility standing 100 meters tall with a capacity of 5,000 m³ and an internal temperature exceeding 2,000°C. Meanwhile, in a hot rolling mill, where steel is thinned and stretched, precision manufacturing is essential—for instance, controlling the thickness of a 2-kilometer-long steel sheet moving at 100 km/h with micron-level accuracy.
JFE Steel has a long history of production activities, with its accumulated expertise in high-grade steel manufacturing, measures for aging equipment, and advanced utilization of data for predictive and preventive maintenance serving as key sources of competitiveness. To further drive data utilization and digital transformation, the company established the DX Strategy Headquarters in the 2024 fiscal year.
Challenges Before Implementing Cognite Data Fusion
JFE Steel has been working toward an intelligent steel plant, aiming to optimize overall plant operations and enhance the efficiency of quality management. In an intelligent steel plant, rather than merely constructing a virtual model like a digital twin, real-world manufacturing process data—such as sensor and operational data—is collected and used to understand real-time conditions, predict future states, and provide real-time feedback to the physical manufacturing process. To achieve this, JFE Steel has been progressively building its own Cyber-Physical System (CPS).
For example, a blast furnace operates at extremely high temperatures, making direct observation of its interior impossible. Traditionally, operations relied on the experience of skilled operators to infer production and maintenance needs. With the advancement of CPS, a cyber model of the blast furnace will enable real-time visualization of its internal state. This allows for operational guidance based on real-time data, leading to greater stability and efficiency in operations. However, JFE Steel faced several challenges developing a real-time, trustworthy CPS.
Challenge 1: Managing Large-Scale Data
The sheer scale of data handled was the first challenge. Across JFE Steel's six manufacturing sites, there are more than 100 major production lines. Each line generates thousands to tens of thousands of data points, including operational conditions, quality data, production records, and manufacturing parameters. In total, the data volume reaches millions of data points, necessitating a unified platform capable of managing massive amounts of time-series and product data.
Challenge 2: Linking Complex Manufacturing Process Data
The second challenge was establishing a mechanism to link the highly complex manufacturing process data unique to steelmaking. The production process begins with melting and solidifying iron, followed by hot rolling, coating, and additional steps such as hot rolling, cold rolling, and cutting before the final product is completed. During this process, steel sheets are elongated in the longitudinal direction, and product positions and orientations change due to defect trimming and coiling. If defects are discovered in the final stage, quality analysis requires integrating quality and operational data from multiple production lines for root cause analysis. However, because each product undergoes different processes, including various treatments like trimming and splitting, collecting, preprocessing, and linking data across different production lines and IT systems, this data collection and analysis requires extensive time and effort. A standardized and efficient system for these processes was essential.
Challenge 3: Real-Time Data Processing
The third challenge was ensuring real-time data processing. JFE Steel's CPS is designed not only to detect and predict anomalies but also to provide feedback to actual processes, such as operator guidance and automation. Therefore, any delay between detecting an anomaly and taking action would render the system ineffective. The ability to analyze vast amounts of sensor data using models and respond instantaneously was a critical requirement.
JFE Steel sought a data integration platform that seamlessly links complex and massive datasets while enabling real-time analysis and feedback.
Background: From Evaluation to Decision-Making
JFE Steel first evaluated Cognite's technology for its focus on contextualizing data. During a visit to Cognite’s headquarters in Oslo to learn more about the capabilities and use cases of Cognite Data Fusion®, JFE Steel engaged in direct discussions with Cognite’s engineering team, including the CTO. The discussions centered on the feasibility of integrating multi-process longitudinal data using Cognite’s Data Model, one of its key data-linking frameworks.
At the same time, JFE Steel was also considering building a custom DIY system by combining native services of public cloud platforms to collect large volumes of time-series data and data from existing internal systems. Cognite Data Fusion was ultimately selected due to several key advantages:
- Its ability to flexibly design data models and natively link manufacturing data across multiple processes
- Capability to process large-scale time-series data in real time
- The availability of up-to-date extractors, APIs, and SDKs that enable seamless integration with various systems.
PoC implementation and company-wide deployment
A Proof of Concept (PoC) was then conducted across five production lines. Four lines dedicated to manufacturing thin sheets for automotive applications were used to verify the linkage of multi-process longitudinal data, while one process line was used to evaluate real-time data processing capabilities.
To verify the multi-process longitudinal data linkage, thousands of sensor data points collected from multiple processes were linked with IT data related to product quality and specifications. The PoC also included integration with an AI model for product quality analysis. As a result, it was confirmed that Cognite Data Fusion’s Data Model could standardize data linkage and facilitate AI-driven output generation.
For the real-time processing verification, the PoC demonstrated that the entire workflow—from sensor data transmission on a single process line to CPS model execution and result output—could be completed within one minute.
Through the PoC, it has been demonstrated that operational data, quality data, defect/defect data, and longitudinal position data from each process can be integrated and operated in a standardized manner. This enables AI-based estimation of the causes of quality defects and speeds up operational improvements. It was also confirmed that CPS can provide real-time operational guidance to process lines. After completing the PoC, JFE Steel decided to deploy Cognite Data Fusion company-wide.
Just six months later, JFE Steel released J-DNexus® with the support from JFE Systems as a platform to integrate IT data, such as production performance and product quality data, and OT data, such as operation data obtained from sensors, with Cognite Data Fusion as the data integration platform, and to centralize the development and execution of CPS on the cloud. J-DNexus®” can shorten the time required to build a new CPS system by 30% compared to conventional systems.
Future Outlook
JFE Steel has been implementing various CPSs for a series of manufacturing processes, including blast furnaces. It aims to establish a consistent CPS for the entire manufacturing process as well as evolve CPSs for individual manufacturing processes by utilizing J-DNexus®, which uses CDF as its database. As part of this effort, the company plans to complete data integration of 100 major lines into Cognite Data Fusion by the end of FY2025.
The integration of Cognite Data Fusion will also create an environment where anyone can view and analyze plant data in real time, regardless of location, using Charts, Industrial Canvas, and other tools within Cognite Data Fusion. JFE Steel plans to expand the use of Cognite Data Fusion with JFE Systems beyond the CPS construction platform to a broader range of internal data utilization platforms.