Stellantis Goiana: A Two-Layer Data Architecture for 30+ AI Applications

On October 26, 2026, IIoT World published an article summarizing a session from the Industrial AI Summit 2026 titled "From Months to Days: Accelerate Industrial AI with a Modern Data Architecture". The speakers were Aron Semle (CTO of HighByte) and Denis Leite (Director of Digital Transformation at Mekatronik), moderated by Hamish Mackenzie of IIoT World.

A caveat up front: the article is labeled "Sponsored by HighByte | Editorially Independent". The figures come from the vendor and the integrator, so they should be treated as vendor-reported data, without independent verification.

The Goiana Case

The Stellantis plant in Goiana, in Pernambuco (Brazil), built a scalable industrial data architecture on top of its existing Siemens infrastructure. It now supports more than 30 AI applications. The team standardized how data is collected, contextualized and shared, and integration times dropped from months to weeks.

The reason for the change is clear: PLC tags were already feeding dashboards and OEE, but machine learning and digital twins also require SQL databases, CSV files, images and video. With hundreds of machines, case-by-case integration was not sustainable; a single approach was needed.

The Two-Layer Model

The architecture described is structured as follows:

• Lower layer (DataOps): handles the protocols and formats specific to each machine.
• Upper layer: organizes data into a common model, so that machines with the same function on different lines produce the same data structure.

Once the upper model works for one line, replicating it on the next is much faster. It was, however, the most demanding part. In discrete manufacturing, the model must represent the parts moving through the line, with operations, sequences and materials arriving from different paths. In process industries, modeling mainly means associating tags with their meaning, such as speed or product.

PLC data accounts for roughly 80% of the volume, while the bill of materials, product details and cycle data provide the context.

UNS and DataOps

According to Semle, Industrial DataOps covers about 80% of the data preparation work for AI. A Unified Namespace (UNS) can be implemented in an MQTT broker, a data lake, a historian or an OPC UA server: what matters is that contextualized data is available consistently. The raw value of a tag means little if you don't know which machine, line or product it belongs to.

Other Timelines and Figures Cited

The session surfaced other examples, again vendor-reported:

• A plastics plant connected its machines and started statistical process control within a week.
• A process site with thousands of existing tags had its first applications running within a few weeks; the full migration took several months.
• HighByte Intelligence Hub is in use at more than 400 sites in 25 countries.
• Gousto reportedly raised mechanical availability from 70% to 95% within six months of adopting the architecture.

Neither the HighByte nor the Gousto figures are independently verified in the article.

Why It Matters for Automation Professionals

Scaling AI depends more on data modeling and contextualization than on the models.

A reusable line-level data model, sitting above PLC and protocol details, is a practical pattern for consultants and integrators. The topic ties in with IIoT World's companion article "Why AI Data Architecture Breaks at Scale", which addresses inconsistent tag names across MES systems.

Practical Takeaways

Several lessons can be drawn from this case:

• Separate machine access (protocols, formats) from the shared semantic model.
• Define the model on a pilot line and reuse it on the others, measuring replication time.
• In discrete manufacturing, include parts, operations, sequences and materials in the model, not just tags.
• Enrich PLC data with the bill of materials, product and cycle data.
• Treat vendor figures with caution until they are independently confirmed.