--- title: "HiveMQ at ICC 2026: bringing industrial data into context with Ignition" description: "See how HiveMQ and Ignition put plant data into context at ICC 2026, from shared KPIs in Ignition dashboards to AI agent diagnostics that a person approves." date: 2026-10-02 author: Shashank Sharma source: https://www.hivemq.com/blog/hivemq-ignition-icc-industrial-data-context --- # HiveMQ at ICC 2026: bringing industrial data into context with Ignition An industrial agent network traced a capper fault in Packaging to its real cause, a low air header in Utilities set off by a cleaning cycle two areas away. The point: connected data isn't the same as usable data, and the fix starts with context at the edge. That’s what our CTO Magnus McCune showed as he put the newly launched HiveMQ Platform to work alongside Ignition, as HiveMQ returned to the Ignition Community Conference (ICC) 2026. ## Industrial data and AI take center stage at ICC 2026 More than 1,800 industrial professionals came to the SAFE Credit Union Convention Center in Sacramento on Sept. 22-24, 2026, for ICC's "Innovation Unleashed" edition. Across 60+ sessions, the conversation kept coming back to one question: now that the plant is connected, what do we do with the data? Inductive Automation used the stage to preview Ignition 2027, including Ignition Catalyst, a secure AI framework, and new multi-site management tools. That set the scene well for our showcase: AI on the plant floor is only as good as the data and context underneath it. ## The showcase: when the machine that's failing isn't the machine that's broken Magnus' Technology Ecosystem Showcase session, *Agentic MTBF/MTTR diagnostics with HiveMQ + Ignition*, opened with a problem most attendees knew well. You've connected the machines but connected data isn't necessarily usable data. The numbers back that up, with [70% of manufacturers saying data quality, contextualization and validation are major obstacles to AI](https://action.deloitte.com/insight/4246/manufacturing-orgs-get-real-about-ai-data-strategy). One of our manufacturing customers told us they don't trust 60% of the data they collect. Magnus framed the answer to this problem as two architectural shifts: 1. **Contextualize at the edge.** Validate, bind to the information model and reconcile as each value is produced, instead of rebuilding context in the warehouse hours later. 2. **Act at the edge.** Detect the deviation, decide and recommend or actuate in the same loop, instead of waiting on a cloud round-trip. Those shifts map directly to the four layers of the [new HiveMQ Platform](/blog/the-new-hivemq-platform/): - **Connect** (a secure, real-time MQTT backbone across OT and IT), - **Contextualize** (data models and governance for trusted, reusable data), - **Analyze** (KPIs calculated near the source), - **Act** (intelligence deployed into governed, supervised workflows). The showcase was a chance to see all four working together on a real Ignition architecture. ## The demo: a cleaning cycle breaks a capper ![ICC 2026 demo architecture: Ignition and OT systems at a site feed a HiveMQ local namespace, connected to a HiveMQ global namespace with an agent network, across the Connect, Contextualize, Analyze and Act layers](https://a.storyblok.com/f/243938/2566fa6817/2.webp) This demo plant had one site, three areas and one shared air header: - **Packaging:** Line A fills and caps 60 bottles a minute. The capper, cap-01, runs on compressed air. - **Processing:** Blend 1 mixes and pasteurizes product, then runs a routine clean-in-place (CIP) cycle that uses a lot of air. - **Utilities:** Compressed air, chilled water, and steam, serving every area and owned by none. Operations started a CIP cycle on Blend 1. On the packaging line, MTBF collapsed from 3.1 minutes to under 30 seconds. OEE dropped from 67% to 45%, and every stop read *E-217 Cap Misfeed*. The line-level view said the capper was broken. The agent network saw something different. MTTR was falling while MTBF collapsed, so stops were getting shorter and more frequent. Neighboring machines weren't starving or blocking, and closure torque was in spec. Then it found the cause: header pressure fell from 6.5 to 5.5 bar about 40 seconds after CIP went active. The capper faulted, the air compressor was at fault, and the agent recommended a setpoint change, a person approved it and OEE stabilized at about 60%. ### How HiveMQ and Ignition turn industrial data into context Ignition stayed exactly where it was. Ignition Edge brought in [OPC UA](/blog/opc-ua-mqtt-bridge-ot-protocols-industrial-data/) tags and published [Sparkplug B](/mqtt/mqtt-sparkplug-essentials/). HiveMQ decoded Sparkplug once and fanned it out as plain MQTT into a browsable global namespace, so nothing downstream needs a Sparkplug library. OEE, MTBF, and MTTR were computed in the platform from ordinary tags, and governance caught a malformed lab result on the way in. On top of that, an agent network (orchestrator, monitor, and analyst) could read every area of the namespace, not just one dashboard. It drew on Ignition logs, MQTT, SQL, and REST, and used a pluggable LLM per agent, so you can bring your own model. **The answer was already in the plant.** One namespace so everything downstream can read the plant without a decoder. KPIs computed where the data already lives. The agent recommended, a person decided, and the capper was never touched. ## How Ignition teams can build on their existing architecture with HiveMQ You don't need to rip anything out to get here. The showcase ran on the stack many ICC attendees already have: Ignition at the edge, Sparkplug B over MQTT, and HiveMQ as the backbone. The only thing that changes is where the work happens: - **Put context where the data is born.** Contextualize and validate at the site, so everything downstream gets trusted data. - **Compute KPIs at the source.** [OEE, MTBF, and MTTR](/blog/real-time-reliability-kpis-mtbf-mttr-availability/) from ordinary tags, available to every consumer in the namespace. - **Look across areas, not just lines.** The costliest faults often start somewhere no single screen is watching. - **Keep people in charge.** Agents that recommend, with auditing and guardrails, earn trust faster than agents that act alone. - **Build once, reuse everywhere.** Central models and policy mean each new site inherits the work of the last one. ## Continue the conversation Thank you to Inductive Automation and the Ignition community for another great ICC. To explore how this architecture could support your operations, [learn more about the new HiveMQ Platform](/blog/the-new-hivemq-platform/) or follow our [Ignition integration guide](/blog/a-step-by-step-guide-connecting-ignition-mqtt-hivemq/).