Reactive
The failure triggers the work. The asset sets the schedule, and the cost lands as unplanned downtime.
Prevantec AI reads the historian data you already collect (OSIsoft PI, DCS exports) and flags an asset before it trips the line, giving you the lead time to schedule the repair into a planned window instead of reacting to a shutdown. No new sensors, no rip-and-replace, no CMMS migration.
Validated on a Canadian ethanol producer's dryer system: 79 hours mean lead time ahead of 7 of 7 confirmed mechanical shutdown events, using their existing PI historian data.
The failure triggers the work. The asset sets the schedule, and the cost lands as unplanned downtime.
The calendar triggers the work. Healthy assets get serviced early, and assets that fail between intervals still reach the trip.
The asset's own trend triggers the work. You still repair on a planned schedule, but the schedule follows actual condition instead of the calendar.
Prevantec AI operates hereMost plants run two modes: repair the asset after it fails, or service it on a fixed schedule whether it needs it or not. The problem in both cases is the trigger. Time-based PM services every asset on the same interval, regardless of actual condition. Fixed high-high alarms only fire once a value has already crossed a static limit. Neither mode reads the trend. Your historian has years of tag data that already shows it.
An asset serviced on a 12-month interval that has 18 months of life left gets pulled anyway. One that wears out in 6 months does not get caught until the next scheduled interval or a trip, whichever comes first. The interval is fixed. The equipment is not.
A fixed setpoint tells you a value has already crossed a limit. It does not tell you a bearing, seal, or filter is trending toward that limit over the preceding days. By the time the alarm fires, the deviation has already reached its terminal stage.
An experienced millwright can recognize early degradation by sound or feel. That judgment is not logged in the CMMS or captured in a work order. When that person leaves, the plant loses the only leading indicator it had for that asset.
You run a plant, not a data science group. Prevantec AI sits on top of the systems you already pay for, and almost all of the lifting is on our side. The output is specific: this asset, this tag, this trend, this much lead time before it becomes a shutdown.
OSIsoft PI, a DCS export, a scheduled historian drop. We work with the tags your instrumentation already writes: motor current, vibration, differential pressure, bearing and outlet temperature, flow, speed. Whatever the asset already tells you.
Your part: point us at the tags and confirm the asset list.
Every machine has its own signature, and it moves with load, ambient and product. We build a statistical picture of normal for each asset in each operating state, so the model compares your dryer to itself instead of to a textbook.
Your part: nothing. Your historical data does the training.
When an asset drifts off its own baseline you get a notice that names the equipment, names the specific tag that moved and shows the trend that triggered it. Your reliability engineer opens it, reads the curve and makes the call, days before the asset reaches a trip condition.
Your part: plan the repair into a window that suits you.
Detection occurs while the asset is still running and still within spec. That interval is what lets you source the part, brief the crew, and decide whether the repair waits for the next planned outage. Mean lead time in validation was 79 hours.
A failure you catch early becomes a scheduled job at straight time with the right part already on the shelf. That is the whole difference between a four hour planned outage and a shift spent firefighting.
Once you can see which assets are actually drifting, PM intervals stop being a guess. Your crew spends its hours on the equipment that needs them instead of tearing down a pump that was running fine.
Every alert shows the tag and the deviation that drove it. Your reliability team can agree, disagree or dismiss it with a reason on the record. A tool your engineers cannot audit is a tool they will eventually ignore.
One plant, one asset, and no instrumentation added. Here is what the data was, what the system flagged, and what it produced.
We worked entirely from the producer's existing OSIsoft PI historian. The asset was the dust collector on the dryer system, and the primary signal was its differential pressure tag, at the sample rate the plant had already been logging for years. Nothing was instrumented for this work.
We baselined normal differential pressure behavior for that asset, then measured how far ahead of each confirmed shutdown the deviation from baseline was detectable. What preceded the failures was a progressive drift off the asset's own normal, well before any fixed limit was crossed.
Dust collector degradation was flagged ahead of 7 of 7 confirmed mechanical shutdown events, at a mean lead time of 79 hours. That figure is the interval between the flagged deviation and the shutdown itself.
The work was run against historical PI data from the producer's own tags, so every result above is measured against shutdowns that actually happened on that line. The client is under NDA, so we do not name them. We will walk your team through the full methodology, including how the baseline was built and how each event was confirmed.
An honest note: MASE AI is a young company, and this is one plant and one asset class. We are not going to tell you it generalises to every machine in your facility until we have looked at your data. That is exactly what the Phase 0 assessment is for, and why it ends in a written go or no-go instead of a contract.
The enterprise predictive vendors want an 18-month rollout and a number your board has to approve. That is why most plants do nothing. We start small enough that you can say yes this quarter, and honest enough that we will tell you if your data is not ready.
Your CMMS manages work: what is scheduled, what got done, what parts came off the shelf. It runs your corrective work orders and your time-based PM. It is a record of the past and a schedule for the future, and it has no opinion on whether a specific bearing is degrading right now.
Prevantec AI adds the condition-based layer. It reads process data, not work orders, and tells you which asset is heading for failure so the work order gets written before the trip instead of after it. It sits alongside Maximo, SAP PM or UpKeep. It does not replace them, and it does not require a migration.
Good, and they should not have to. Deployment options include running inside your own environment, on your infrastructure, with no inbound connection from us.
Wherever data does move, the path is scoped explicitly in Phase 0 and written into the deliverable, so your IT and OT people review the exact architecture before anything is connected. If your security posture rules an option out, we design around it or we tell you it is a no-go.
You are not supposed to. The modelling, the baselining and the tuning are our work, not a job we hand your maintenance planner.
What lands on your side is an alert that names an asset, names a tag and shows a trend. If your reliability engineer can read a trend in PI, they can read this. That is the whole bar.
Those modules are usually licensed at enterprise scale, priced for a corporate rollout, then handed to the site to configure. The software arrives, the engineering does not, and the module ends up switched off or ignored.
We come in on a fixed-fee assessment, do the asset-level engineering ourselves, and start with the handful of assets that actually hurt when they stop. If your historian vendor's module is already working for you, keep it. We would rather tell you that than sell against it.
You start with the Phase 0 assessment, which is a fixed fee agreed before we begin. There is no open-ended discovery bill and no six-figure commitment just to find out whether this fits your plant.
Phase 0 ends with a written scope that prices the monitoring work explicitly, alongside a go or no-go recommendation. If we do not think the return is there, that is what the document will say. Book the data review and we will give you the Phase 0 number on the call.
Twenty minutes on a call. You tell us what trips, we tell you what your historian would need to hold for us to see it coming. If the data is not there, you will hear that from us on the same call. No deck, no discovery process, no pressure.