AGC has selected Cognite’s industrial data platform as the manufacturing-data integration layer for its Chemicals Company, with a phased rollout planned across major domestic production bases following a proof of concept. The programme is intended to bring plant and equipment information into a common structure so that operating teams can access and reuse it across maintenance, production and decision-making workflows.
The immediate task is data integration rather than the installation of another isolated analytics application. Chemical plants accumulate information in process historians, maintenance systems, engineering documents, equipment databases and local applications over decades of operation.
Those sources can contain valuable operational information while remaining difficult to combine. Asset names, tags and equipment relationships are often represented differently across systems or sites, leaving engineers to reconstruct context manually whenever a new application or analysis project is started.
AGC and Cognite intend to consolidate and structure manufacturing and equipment data distributed across the Chemicals Company’s production bases. The contextual layer is designed to reduce the time required to locate relevant information and make the same underlying data usable by people, software applications and AI tools.
The companies have not disclosed the number of plants included in the first deployment phase or the value of the agreement. The significance instead lies in moving from a proof of concept into a multi-site programme, where differences in legacy systems and local working practices become much harder to avoid.
Industrial-data projects can perform well in a limited trial because the relevant tags, documents and assets are selected in advance. Scaling that model requires the platform to deal with different control systems, equipment structures, naming conventions and maintenance processes without rebuilding the information model for every site.
AGC’s objectives include reducing dependence on individual expertise, standardising work processes, accelerating decisions and improving productivity through greater use of data and AI. None of those outcomes follows automatically from centralising information.
Plant data has to be mapped correctly to the physical equipment and processes it describes. Users also need confidence in units, timestamps, quality and provenance before the information can be used for maintenance or operating decisions.
Cognite’s architecture is designed around that contextualisation problem. Its Industrial Knowledge Graph links operational data to assets, equipment and relationships so that a user or application can move between a physical item and associated time-series information, events, documents or other records.
For a chemical producer, the same contextual layer can support several functions. Operators may need current process conditions and historical trends, maintenance teams may need alarms and work histories, and engineers may need drawings, inspection records or process models.
AI applications introduce another requirement. Models used for anomaly detection, forecasting, maintenance support or operator assistance need more than disconnected measurements. A temperature value has limited meaning without knowing which asset produced it, how that asset was operating and what other process conditions applied at the same time.
Creating consistent relationships between data and physical equipment is therefore less visible than the AI layer but often determines whether an application can be scaled beyond a single demonstration. Once context has been rebuilt repeatedly for separate projects, much of the claimed efficiency of industrial digitalisation disappears.
The new rollout also fits into a longer digitalisation programme at AGC. The group has previously developed chemical-plant digital twins, private LTE networks and integrated operations systems, including a process digital twin placed into full-scale operation at a vinyl chloride monomer plant operated by P.T. Asahimas Chemical in Indonesia.
That work illustrates the difference between collecting operational data and using it to change plant behaviour. Digital twins combine live information with process models to reproduce operating conditions and estimate variables that may not be measured directly.
AGC has described its wider digitalisation approach in stages moving from visualising operations towards understanding and transforming them. A shared data architecture provides the foundation needed for those later stages because models and applications can draw on information that has already been structured rather than beginning again with each project.
The workforce argument is equally important. Chemical plants depend heavily on experienced operators and engineers who understand how equipment behaves during start-ups, shutdowns, disturbances, maintenance events and other conditions that are difficult to capture in formal procedures.
AGC identifies knowledge transfer and future workforce pressure among the reasons for the investment. Structuring historical information can preserve part of that experience, but tacit knowledge still has to be converted into records, models or procedures before software can reproduce it reliably.
The company has not attached quantified productivity or cost-saving targets to the programme, so progress will have to be judged by deployment rather than headline software claims. Useful measures will include the number of plants connected, the proportion of relevant data sources contextualised and the time required to deploy new operational applications.
The harder test will be maintaining those relationships as plants change. Equipment is replaced, tags are modified, documentation is revised and production processes evolve throughout the operating life of a chemical site.
AGC is therefore moving its digitalisation programme towards a common information architecture rather than another stand-alone plant trial. The value will come from making existing data reusable across operations, maintenance and AI projects; keeping that data accurately connected to the physical plant will determine whether the platform continues to deliver after the initial rollout.
