Global aircraft manufacturer turns factory-floor data into real-time insights for AI

Challenge

Building faster, using data that arrives too late

Aircraft manufacturers are under pressure from every direction. Order books stretch years out, airlines expect on-time deliveries, and every part must meet exacting quality standards and stay traceable for certification. Defects caught late mean rework or delayed deliveries, and experienced inspectors are hard to hire.

As a leading global aircraft manufacturer, the company launched an executive-sponsored smart factory program to keep pace. The plan was to connect every shop-floor machine to one data platform, then use that data to catch quality issues earlier and automate routine work, scaling from 10 to 20 connected tools today to more than 10,000 per site.

However, the legacy data architecture couldn’t keep up. Each site, tool, and system moved data its own way, and machine readings reached analysts in two daily batches, up to 12 hours late. By then, a problem on the line could already have affected the next station in the assembly process and the next batch of parts. The cloud service in the middle was scaled and priced for only a handful of machines, so costs would climb with every new connection. The platform also had to run inside air-gapped plants with strict data residency rules, because production can’t stop when the network does.

Then AI raised the stakes even further. The team had five AI projects lined up, from robots that photograph panels to spot imperfections to assistants that help engineers with documentation and reports. Without a shared foundation, each project would need custom data pipelines, every app would be locked to one model, and no one in management could easily see who was using AI or what it cost. 

The goal was one secure platform, simple enough for a two-person team to run, that delivers live shop-floor data across their entire multi-site physical plant, and gives every AI project a safe, governed way to work.

Why Redpanda

Data streaming that runs inside the plant, with a governed path for AI

After validating Redpanda internally and running a formal vendor comparison against Apache Kafka^®, the team chose the Redpanda Data Platform, specifically for Redpanda Enterprise and Redpanda Connect. As one integrated system with no extra components to manage, Redpanda is simple enough for a small team to run, and it can be deployed anywhere, from fully offline plants to individual sites and central analytics, with data staying in company-controlled environments. The team also chose Redpanda for:

• Full compatibility with Kafka APIs, so existing apps and tools work unchanged, with no rewrites and no vendor lock-in
• Ready-made connectors that aggregate and stream Industrial Internet of Things (IIoT) machine data straight from the factory floor into the company’s cloud data lake, with no custom code
• A single, governed entry point for AI, where every request and the app sending it are checked before any automated workflow runs
• Freedom to choose AI models, so teams can use cloud-hosted models today and move to in-house hardware later without rebuilding their apps
• A path to the Redpanda Agentic Data Plane on the same platform, adding access controls, usage tracking, and a full record of every AI agent’s actions

Day to day, the team deploys and updates the platform with the same automation tools it uses for the rest of its infrastructure, and built-in resilience stores every message three times so data survives a hardware failure. Throughout the proof of concept, Redpanda engineers provided hands-on support, from architecture reviews to weekly check-ins.

Results

Live shop-floor data today, and a safe foundation for AI tomorrow

Redpanda passed a six-week proof of concept against the company’s own requirements, and the company standardized on Redpanda Enterprise and Redpanda Connect. 

With Redpanda Connect streaming machine data straight into the company’s data lake, the team removed the twice-daily batch processing and the costly cloud service from the middle of its data integration path. Adopting data streaming cleared the way for quality issues to surface while parts are still on the line, not half a day later, and costs no longer climb with every machine connected. The company is now able to scale to support up to 20,000 connected tools up from a few dozen previously.

Because Redpanda runs inside the plant and the deployment is fully automated:

• Data keeps flowing during an internet outage, so production continues uninterrupted.
• Data never leaves company-controlled environments, so the team meets its security requirements.
• The two-person platform team spends less time maintaining infrastructure and more time on new projects.

A review with Redpanda’s engineers also confirmed the company needs three servers in production, not the five first planned, cutting the infrastructure to run by 40%.

With machines, cameras, and AI requests all running through one platform, each new AI project builds on a shared, governed foundation instead of custom plumbing, so ideas reach the shop floor faster and with less risk. As AI use grows, the team is evaluating the Redpanda Agentic Data Plane to give every AI agent clear permissions, tracked costs, and a complete audit trail.