← All articles

· Crossover Software

Stellantis Goiana: AI on welding equipment wins trust because engineers can verify it

According to an article published on 11 November 2026, anomaly detection prevented 14 failures at Goiana in six months. The most useful takeaway for plant managers, though, is how the engineers' trust was earned.

Industrial AISmart factoryPredictive maintenanceProduction dataOEE

On 11 November 2026, IIoT World published a write-up of a session at the Industrial AI Summit 2026 titled "From Months to Days: Accelerate Industrial AI with a Modern Data Architecture". The speakers were Aron Semle (HighByte) and Denis Leite (Mekatronik), and the session covers the Stellantis plant in Goiana, in the state of Pernambuco, Brazil.

One caveat is necessary: the article is sponsored content from HighByte, whose product is a data contextualization layer. The account should therefore be read with that framing in mind.

The case and the reported results

The system uses AI-based anomaly detection to help maintenance teams identify likely failures of welding equipment before they stop production. After six months, the following results are reported:

  • 14 failures avoided;
  • 15 hours of production recovered in a single month;
  • an expected capacity increase of 810 vehicles per year.

According to the speakers, results have improved further since then, but no more recent figures are provided. The data are self-reported: there is no independent verification and no detail on methodology, and the article specifies neither the type of model nor the costs.

Data first, then the model

The welding data came from heterogeneous sources: TXT files, SQL databases and operational systems. The team did not send it raw to the data lake; instead, it organized and contextualized it around individual welding operations. Anyone who has worked on MES projects knows this step well: a process value only makes sense when tied to what was produced, when and where.

The project was also designed for production from the start, with no pilot phase, with attention to reliability, maintainability and scalability. Stellantis involved its own engineers throughout, so as to tap their first-hand knowledge of the process.

Trust comes from verifiable evidence

The most instructive part concerns adoption. Engineers used dashboards, time series and distribution charts to compare normal welds with the anomalies flagged by the model. The first visualizations did not clearly separate the two behaviors: technicians saw little reason to intervene and questioned every alert.

After the visualizations were refined, the engineers found the flagged cases worth investigating.

When the flagged cases turned out to be real problems, trust grew. Today the model is considered reliable and is used in 24/7 operations. Stellantis is evaluating extending the approach to other plants and applications; according to the article, the open question is where to start.

What a line manager can take away

The case is a concrete example of a sequence: contextualized data, then the model, then evidence that technicians can check. It shows that adoption depends on explainability for maintenance staff, not just on model accuracy. An alert that cannot be verified gets ignored or challenged.

The lesson also applies to MES and OEE projects on lines such as poultry processing, where unplanned stops and capacity are directly linked. A few practical pointers:

  • structure data around recognizable process units (order, machine, cycle) before applying any analytics;
  • involve maintenance technicians and process engineers from the design stage;
  • show the comparison between normal and anomalous behavior with every alert;
  • measure impact in downtime hours avoided and capacity recovered, not just in model accuracy.

Caution is warranted when assessing figures like those from Goiana, given the sponsored and self-reported nature of the source. The method, however, remains replicable: reliable data, transparent evidence and trust built case by case.