muono
Platform
Prediction

The alarm is not the first thing the machine tells you.

A threshold tells you a reading crossed a limit. It does not tell you how long it has been heading there, how fast, or whether the same shape has been seen on this pump class before. Most alarms arrive with none of that attached, which is part of why so many of them get acknowledged and left.

What the model is looking at

One reading says little. Two moving together say a great deal.

Vibration inside its limit is unremarkable. Vibration climbing while bearing temperature climbs, at steady load, is a mode developing.

VIBRATION
BEARING TEMP
LOAD, STEADY
NORMAL BAND
ACT HERE
Intervention window
Weeks
Trend · pattern · asset class

Every individual reading on this pump stayed inside its alarm limit for the whole fortnight, so a threshold alarm would have said nothing. What matters is the shape: a steady departure from the band the machine normally holds, in two variables at once, while load stayed flat.

That pattern matches early bearing wear recorded on the same pump class elsewhere on site. Naming it now buys the P-F interval: the weeks between a fault you can detect and a failure that takes the machine off you. That is long enough to put a bearing swap on a shutdown you already have booked.

How conditions develop
Not the single reading, but the shape it has traced over days and weeks.
How variables move together
Rising vibration at steady load and rising bearing temperature are one finding, not two readings.
Whether the pattern is known
The shape is matched against known failure modes for that class of asset, and against what this site has seen before.
What arrives on your desk

A recommendation, not another unexplained alert.

An alert that says a number moved leaves the engineer to do all the work. Muono names the asset, the failure mode it believes is developing, the evidence behind that belief and the window in which intervening is still cheap. Then it tells the engineer responsible before the next reporting cycle, not after it.

The asset
Named, with its record and its open work already attached.
The probable mode
What appears to be developing, stated as a mode rather than a score.
The evidence
The readings and the pattern that led to the conclusion.
The window
How much P-F interval is left before the planned repair becomes an unplanned one.
A worked example

How a developing failure reads.

A feed pump on a two-week trend. Each line is something the model can point at, so the reliability engineer can argue with it instead of taking a score on trust.

The signal
Bearing vibration rising steadily, still inside its alarm limit on every individual reading.
The correlation
Rise tracks a climb in bearing temperature and holds through a period of steady load, so load does not explain it.
The pattern
Sequence matches the early stage of bearing wear recorded on the same pump class elsewhere on the site.
The conclusion
Probable bearing degradation rather than an instrument fault. The two readings would not move together if it were.
The window
Weeks rather than days, while the repair is still a planned swap instead of an unplanned outage.
Who is told
The reliability engineer who owns the pump, with the trend and the open work attached.

Your questions, answered

Is this just threshold alarms with a nicer interface?
No. A threshold fires when a single reading crosses a line. Prediction looks at how conditions develop over time and how variables behave together, so it can flag a machine whose every individual reading is still in range.
How do we know it is not guessing?
Because every recommendation shows the readings and the pattern behind it. If the evidence does not convince the engineer who owns the asset, it should not convince anyone. They can see it rather than take a score on trust.
What happens to alerts nobody acts on?
They stay on the asset's record with what was known at the time. When the same mode develops again the history is there, which is how the model and your team both get better at the call.
Does it need years of failure history to start?
It needs enough signal to establish how the asset normally behaves. Known failure modes for a class of equipment carry over, and the site-specific patterns build as it runs.

Start with one machine that keeps surprising you.

Request a pilotBook a walkthrough