Stellantis Goiana: A Two-Tier Data Model Behind 30+ AI Applications
Stellantis' Goiana plant in Brazil has built an industrial data architecture that supports more than 30 AI applications and cuts integration times from months to weeks. A useful case for anyone who already has OEE dashboards and wants to reuse the same data.
On 26 October 2026, IIoT World published a report on an Industrial AI Summit 2026 session devoted to Stellantis' Goiana plant (Pernambuco, Brazil). According to the speakers, the plant has built a scalable industrial data architecture on top of its existing Siemens infrastructure. The architecture now supports more than 30 AI applications, and integration times have dropped from months to weeks.
A note on sources: the article is sponsored by HighByte, although it is labeled editorially independent, and it reports talks by Aron Semle (CTO of HighByte) and Denis Leite (Director of Digital Transformation at Mekatronik). The figures therefore come from a vendor and an integrator and have not been verified elsewhere.
Why change approach
PLC data was already feeding dashboards and OEE analysis. But machine learning and digital twins need more: SQL databases, CSV files, images and video. Doing this across hundreds of machines the traditional way, with application-specific calculations and manual integration for every line and every plant, is not sustainable.
The answer was to contextualize raw data once, let each application consume it, and follow the principle "model once, propagate across lines and plants".
The two-tier DataOps model
The architecture described is organized as follows:
- Lower tier: handles the protocols and formats specific to each machine.
- Upper tier: organizes data into a common model, so that machines with the same function on different lines produce consistent data structures.
Once the upper tier works on one line, rolling it out to the next is much faster. It was, however, the tier that demanded the most effort. In discrete manufacturing it is not enough to attach meaning to tags, as is often the case in process industries: you must model the parts moving through the line, with operations, sequences and materials arriving via different routes.
PLC data accounts for roughly 80% of the volume. Bill of materials, product details and cycle data provide the context.
The role of the Unified Namespace
According to the speakers, Industrial DataOps does about 80% of the work required to build a Unified Namespace (UNS). The UNS can sit on MQTT brokers, data lakes, historians or OPC UA servers: what matters is making contextualized data available in a consistent way.
Other implementation timelines cited
The session reported further examples:
- a plastics plant connected its machines and started statistical process control within one week;
- a process site with thousands of existing tags launched its first applications within a few weeks, completing the migration in a few months.
Timelines depend on each site's processes, systems and documentation. The commercial figures as well (HighByte's Intelligence Hub at more than 400 sites in 25 countries; Gousto going from 70% to 95% mechanical availability in six months) are vendor-reported.
What plant managers can take away
The case points to a practical path for those who already have OEE dashboards and want to reuse the same data for AI: standardize the data model once and replicate it across lines, instead of integrating each application separately. The same applies to MES and line-monitoring platforms: a common model per line is what makes rollouts repeatable across multiple lines and sites.
Model once, propagate across lines and plants.
The speakers' advice is to start small and build on what works. In practice: pick one line, define the model of machines and product flows, validate it with a first application, and only then extend it.