17 September 2026

 

During a night shift, a conveyor motor begins vibrating differently. A wireless sensor detects the change and sends an alert to the maintenance team. Finding the anomaly is comparatively easy. Understanding it is not.

The vibration could indicate a worn bearing, a misaligned shaft or an unbalanced component. But it might also be caused by heavier material moving along the conveyor, a change in operating speed or even another machine nearby.

By the time the maintenance manager arrives in the morning, the alert is waiting alongside dozens of others. Someone must decide whether the motor needs immediate attention, can wait until the next planned shutdown or is behaving perfectly normally.

Finnish industrial technology company Treon believes artificial intelligence can take on more of that investigative work.

Its newly announced Treon IQ system analyses operational data, diagnoses potential problems and recommends what maintenance teams should do next. An early-access version is available to Treon Make and Treon Flow customers, with the technology being demonstrated at th International Manufacturing Technology Show in Chicago from 14 to 19 September.

Treon describes the result as “operational understanding”. The more familiar term is prescriptive maintenance.

 

What is prescriptive maintenance?

 

Industrial maintenance has traditionally been based on time. A component might be inspected every three months or replaced after a specified number of operating hours, whether or not it is showing signs of failure.

Condition monitoring made this more precise. Sensors attached to motors, pumps, conveyors and other equipment measure signals such as vibration, temperature and pressure. Changes in those signals can reveal imbalance, misalignment, lubrication problems or deteriorating components.

Predictive maintenance looks for patterns associated with future failure. Instead of waiting for a pump to break, a factory can arrange repairs during a planned shutdown.

Prescriptive maintenance adds another step. It uses equipment data and analytical models to identify the probable cause of a fault and recommend a corrective action.

Predictive maintenance says that a motor may fail. Prescriptive maintenance attempts to identify the damaged bearing, assess the urgency and recommend when and how it should be replaced.

That is a considerably more demanding task. A warning cannot be interpreted in isolation. The software may need to know what the equipment was doing at the time, whether its operating load had changed and whether similar behaviour preceded an earlier breakdown.

Factories already collect much of this information. The challenge is connecting it quickly enough to support a decision.

 

How Treon IQ combines industrial data and AI

 

Treon IQ is part of Treon Connect, the company’s cloud platform for industrial monitoring and maintenance. It draws on live and historical asset intelligence, along with information from computerised maintenance management systems, commonly known as CMMS, and enterprise resource planning software.

According to Treon, the AI layer combines its anomaly-detection technology, deep-learning diagnostic models and selected large language models.

Each has a different role. Diagnostic software searches for patterns associated with mechanical problems. Maintenance systems contain previous work orders and repair histories. A language model can organise this evidence into a readable briefing and allow users to question it in ordinary language.

Instead of beginning the day by working through multiple screens, a manager would receive a summary of site conditions, active problems, optimisation opportunities and priorities.

This is not the same as asking a general chatbot how to repair a motor. The large language model operates within a wider system that has access to specialist analytics and industrial information. On its own, it could produce a convincing answer without understanding the equipment at all.

 

Industrial knowledge is difficult to scale

 

The technology addresses a human as well as a technical problem.

Factories often depend on a relatively small group of experienced technicians who know the peculiarities of individual assets. They recognise that one motor always runs slightly hotter after a particular product change, that a recurring warning is usually harmless or that a subtle vibration has previously preceded a serious breakdown.

Much of this expertise is never formally recorded. It has been accumulated through years of listening to equipment, inspecting failed parts and seeing which repairs worked.

When an experienced engineer retires or is unavailable, the measurements may remain but some of their meaning disappears. Industrial AI vendors hope their systems can capture part of that context and make it available across shifts and sites.

That does not turn software into a veteran technician. It could, however, help a less experienced worker distinguish a routine irregularity from a situation requiring immediate attention.

 

How Siemens, ABB, IBM and Augury approach industrial AI

 

Treon is far from alone in attempting to turn industrial information into practical guidance, although vendors approach the problem from different directions.

Siemens is extending its Industrial Copilot into maintenance through the Senseye predictive-maintenance platform. Its offering combines fault forecasting, automated diagnostics and AI-assisted repair guidance. Siemens reported in 2025 that initial pilot projects had reduced time spent on reactive maintenance by an average of 25 per cent, although this was a company figure rather than an independently verified result.

The company is also developing an Operations Copilot designed to let shop-floor workers question machine information and receive error-resolution guidance in natural language.

ABB and IBM begin with large industrial and asset-management environments. ABB's Genix Copilot processes information from sources including factory operations, enterprise systems and work orders. Users can request summaries of plant events, asset health and possible root causes.

IBM's Maximo Condition Insight evaluates work orders, meter readings, alerts and Failure Mode and Effects Analysis records. It then links an asset’s condition to known failure modes and appropriate maintenance activities. IBM says future versions will be able to create or update work orders with minimal human involvement.

Machine-health specialist Augury and Treon start closer to the physical equipment. Both combine dedicated monitoring hardware with diagnostics and repair guidance. Augury adds another safeguard: it says certified reliability experts validate its AI findings before customers act on them.

The products differ in scope, but they reveal a common direction. The next generation of industrial AI is being designed not merely to identify trouble, but to assemble the evidence needed for a response.

 

Why factories do not need another dashboard

 

Industrial digitalisation has created an awkward problem: collecting more information does not necessarily make a factory easier to manage.

A maintenance department may receive readings from production equipment, work-order software, energy-management platforms and business systems. Each source may be valuable while still adding to the volume that someone must interpret.

The bottleneck is increasingly deciding which issue deserves attention first.

A useful industrial AI assistant must therefore account for circumstances. Did the change coincide with a heavier load? Was the equipment recently repaired? Is the affected line essential to today’s production schedule? A technically sound recommendation may still be impractical if the necessary spare part is unavailable.

This is where the promise of operational understanding encounters the untidiness of a real factory.

 

Can industrial AI be trusted with maintenance decisions?

 

Treon’s announcement makes ambitious claims about accurate diagnoses and “expert-level insight”, but provides no performance figures, named early-access users or results from deployments of Treon IQ.

It also leaves several practical questions unanswered.

Can an engineer inspect the vibration pattern behind a diagnosis? Does the system identify which maintenance records informed its conclusion? Will it acknowledge when several explanations are equally plausible? Can the completed repair be used to correct the model? And what happens when the sensor, rather than the equipment, is faulty?

The answers will determine whether maintenance teams regard industrial copilots as dependable tools or another source of noise.

A mistaken consumer-chatbot response may waste a few minutes. Incorrect industrial advice could lead to an unnecessary shutdown, damage expensive equipment or create a safety risk.

For now, the most credible role for these systems is decision support. They can gather relevant information, establish priorities and suggest an intervention, while a qualified person remains responsible for approving it.

Tomorrow’s maintenance manager may begin the day with a much shorter list of problems and a clearer account of what caused them. The software may even explain what it thinks should happen next.

But the decision to stop the production line, and the responsibility that comes with it, still belongs to a person.