Detection is step zero. Now you can ask what comes next.

The job only starts at detection. Now you can ask what to do next.

Predictive maintenance is a workflow, not a technology. Acquire the data, decide which machines need attention, figure out what is wrong, decide what to do, plan the work, do it, confirm the fix. Thirty years of investment have gone into the front of that workflow — sensors, thresholds, models that notice when something changes. The steps closest to a decision, troubleshooting and assessment, have remained some of the least supported by tools. They depend on an engineer's concentration and on whatever they can remember about a similar machine three years ago.

That is why I keep making an uncomfortable point about this market. Much of what is sold as a "predictive maintenance system" focuses on detection. It tells you something is abnormal. For the team investigating the problem, that is step zero. The job only starts there.

Rotomate has always been built for what comes after step zero. It continuously analyses every monitored machine in the plant, weighs the measurements against the machine's own history, filters the noise, and points the team at the machines that need them today. It provides a current, reasoned assessment of every machine, kept fresh around the clock.

Today we are changing how you reach that assessment. As of now, you can ask. In plain words. What is wrong with this machine, why, and what should we do about it? Rotomate's AI colleague answers, shows the evidence, and keeps the conversation going for as long as you need. It can even send you emails, bringing its findings into your inbox.

I want to look at the steps detection leaves to the reliability team, and what changes when the analysis is already done.

Detection hands you a question

A flagged machine is not an answer. It is a question. Vibration has changed at one measurement point on a fan — now what? Is it the bearing, the alignment, a loose foundation bolt? How quickly is the condition changing? Is it the same pattern we saw on the sister fan last winter? Does it need someone this week, or can it wait for the planned stop?

When a system stops at detection, it hands the reliability team that question. The expensive part — answering it — is left to people already struggling to keep up with the work. That is the gap where predictive maintenance can quietly turn back into reactive maintenance: the sensors picked up the change, but nobody had time to get from the alarm to the decision before the machine failed.

Starting where detection ends

Rotomate starts where detection ends, and it can do that for one reason: it has already done the reading before you ask. Every monitored machine has a fresh condition assessment, built from trends, spectra, fault frequencies, component data and history. So when you open a conversation about a fan, you are starting with an existing analysis and asking for the part of it you need.

This is the difference between a search box on top of a database and a colleague who has been watching the plant. The colleague already knows which machines are showing signs of degradation, what changed, and when. The question you ask determines what they tell you first.

Troubleshooting, by asking

Troubleshooting often means an engineer sifting through spectra, sideband patterns and trend graphs. Those measurements may sit together for one machine, while other machines are monitored in separate systems. Maintenance records and process historians hold more of the context needed to understand the problem, but these systems rarely communicate. The engineer has to bring the picture together.

Now you can ask. "Which machines are degrading right now? Show me the evidence." Or, one level down: "What is wrong with the fan where the vibrations changed?" The answer comes back with the likely cause and the evidence behind it — which frequencies moved, when the trend started, what in the machine's history makes this diagnosis more likely than the alternatives.

Nobody has skipped the analysis. The engineer can examine the result and investigate further without first having to assemble the whole diagnosis by hand.

The conversation is the way in. The method is the intelligence.

I want to be precise about what is answering you, because it matters to anyone who has sat through a demo of a language model confidently misreading a spectrum.

The conversation is the way in. Underneath it sits a real analyst's method: trends, fault frequencies, component knowledge, the machine's own history, and the rules for weighing them against each other. We built that method by studying how the best reliability engineers diagnose — what they look at first, what they rule out, what makes them confident enough to recommend maintenance. That analysis engine is what has been watching the plant, and that is what answers the question.

Rotomate approaches condition monitoring the way a human specialist does: examining the measurements, considering the machine's context, weighing possible causes and reasoning about what to do. It detects, diagnoses and proposes maintenance. The conversation makes that work accessible to anyone responsible for reliability.

"Why do you think so?"

The question that decides whether a plant trusts a system is not just "what did you find". It is "why do you think so".

Ask it, and Rotomate answers with the spectrum, the trend and the history it used to reach the conclusion. A confidence score alone cannot explain a diagnosis to a team deciding what to do with a machine. They need the actual evidence, presented so a specialist can check it and a maintenance manager can follow it.

This is what makes the answers verifiable. An explanation that carries its evidence is something you can check, disagree with, or act on. An alarm alone leaves that work on your desk.

A discussion, not a report

A report is one-way. A colleague is not.

You can push back. "That looks like a resonance to me, not a bearing defect." You can ask for the underlying data. "Show me the envelope spectrum from last week against this one." You can widen the question. "Has anything else on this line changed at the same time?" You can work through the analysis in the conversation, examining the evidence and testing the interpretation as you go.

That is what turns a monitoring tool into a colleague. You can work a problem through with it and arrive somewhere together.

It recommends. You decide.

The goal is a recommended next step and the reasoning behind it. Inspect this bearing at the next planned stop. Check the alignment on this coupling. Leave this one alone, the change is explained by the process.

Then the team decides. That is deliberate. Decision authority belongs to the people accountable for reliability, the production plan and the people on the floor. Rotomate's job is to make that decision well-informed and fast — to bring the evidence, the diagnosis and a proposed action to the table, so the team can move forward without waiting for someone to find time for the analysis.

Back to the gaps

The workflow is the same one plants with leading reliability practices already follow: find the machines that need attention, understand what is wrong, decide what to do, do it, confirm the fix. The diagnostic principles remain. What changes is that AI performs analysis that previously required manual work, and the team can question and use the results through a conversation.

Unplanned stops cost the most when they are the most surprising. Surprise lives in the gaps — between what is really happening inside the machine and what gets detected, and between what gets detected and what somebody decides to do about it. When the demand for analysis exceeds the team's available time, machine problems can go uninvestigated until they stop the line.

Rotomate's AI colleague helps close those gaps by keeping the analysis current, the evidence accessible, and recommendations ready to discuss. That reduces the manual investigation competing for time the plant's personnel do not have.

The job only starts at detection. Now the next step is a question away.

See it live on 8 October

On 8 October we're doing this live. Sampo and I will open Rotomate on Metsä Group's real data, work through a gearbox fault from the first question to the recommended action, and take yours as we go. If you've ever watched an AI demo and wanted to ask "show me the spectrum", this is the one where you can.


Save your seat →

And if you'd rather try it on your own machines first, the trial is open.

Start your free trial →