PRISM takes in time-series telemetry and returns time to failure with the probable fault named. It learns each asset's whole operating signature, so it catches modes nobody wrote a rule for.
One prediction as the operator sees it. Site names anonymized.
Not a health score to interpret. Every monitored asset laid out against the clock, so the work sorts itself.
Each bar is one asset's predicted days-to-failure window. Where it sits is when to act. How dark it is is how certain and how severe. Sorted soonest first, so a planner works top to bottom.
A threshold watches one channel and fires when it crosses a line. The information that would have warned you lived in the relationship between channels, months earlier.
Pressure is fine. Current is fine. Temperature is fine. The three moving together in a way they never have is the signal, and no single line can represent it.
When efficiency fades, the controller compensates to hold the same output. The result looks healthy for months. The effort spent to hold it does not.
Every alarm encodes a failure someone had already seen. A model that learns the whole signature can flag drift that has no rule attached to it.
PRISM came out of clinical prediction work, turning a messy medical record into a scientific instrument. The techniques that let you see a patient decline read a pumping station just as well.
A healthy system holds a steady state under changing load. Failure is the loss of that regulation, and it shows up well before any single reading goes out of range.
Telemetry is projected into a high-dimensional space. Once behavior is a position in that space you can do arithmetic on it, and see the jump across many dimensions at once.
The first layer judges each sensor alone. The second combines them into a verdict for the asset, which is why a struggling lead pump with a healthy standby does not become an alarm.
A station drops off the network for a week then transmits it all in two minutes. A tag is named for the contractor who installed it in 2009. Half the assets have no failure history. That is the expected input, not a blocker.
As heat transfer or hydraulic efficiency fades, the control system raises flow or speed to hold the same result. PRISM watches the compensation.
Live means running in production on a customer's assets. Deploying means built, with evaluations underway. Validating means designed and being proven before we ship it.
Per-asset drift caught across correlated channels, before any threshold is crossed.
The probable fault named from the shape of the deviation, not just an anomaly score.
A window in days, ranked across the fleet, that a planner can schedule against.
Severity-routed into the CMMS, dispatch queue and phones your team already uses.
Including sensor classification and calibration for site-specific naming.
So the return is measured rather than modeled, in a monthly one-page summary.
Prediction delivered through a pump and SCADA manufacturer's installed base.
Expansion of the reference deployment beyond the original station set.
Chillers, towers, CRAHs, CDUs, UPS and switchgear. Models are built; we need your data to calibrate and score them.
Taking the existing alarm stream as an input and judging what actually needs action.
Every documented mode for an asset class classified by how it is covered today. Designed, not yet shipped.
Midstream pumping, mining dewatering, chemical and pharmaceutical plant.
We will tell you the same thing on a call. Anything marked Validating has no measured results behind it yet, and we will not present it as if it does.
| Layer | Typical interface | Direction |
|---|---|---|
| Water & wastewater | Historian export or API, Modbus over the existing gateway | Read only |
| Data center cooling | BACnet via the BMS; VFDs over BACnet or Modbus | Read only |
| Liquid cooling | Redfish REST over JSON; Modbus or SNMP on older units | Read only |
| Data center power | SNMP via DCIM, Modbus via EPMS | Read only |
| Maintenance history | Flat export, or API where available | Read only |
| Dispatch | API, webhook, or the channel you already use | Write, to your workflow, never to your plant |
Where BRICK tagging or ASHRAE Guideline 36 naming is in place, calibration is faster. Neither is a prerequisite. A site with sixty CRAHs is a better fit than a site with one chiller.
That is not a policy we adopted. It is how the product is built, which is a different and more durable thing.
PRISM cannot change a setpoint, close a valve or start a machine. Not because it is disabled, but because the capability is not built. Ingest is one-directional under a read-only credential.
Six to twelve months of history as flat files under NDA. No connection to your environment, no credential issued, no firewall change. If your security team never opens a ticket, that is the design working.
We state our current SOC 2 and ISO 27001 position in writing at the scoping call, specific about what is in place and what is in progress, rather than putting a badge on a web page. An architecture one-pager covering the data path, encryption and retention is available before you commit to anything.
No to both. Mixed, gapped and inconsistent telemetry is the expected input, and the models were developed against data of exactly that character. Where work-order history exists we parse it ourselves, free text included. Where it does not, we surface anomalous periods for a short human review and bootstrap from there.
No, and it is not a configuration option. There is no write path to plant controls in the product, and no agent, controller or gateway is installed on your network. The only thing PRISM writes to is your workflow: a work order, a text, a call, a webhook.
Threshold and anomaly tools fire when one channel crosses a line, which by definition happens once the failure is underway, and they only cover faults someone already encoded. PRISM models the joint signature across correlated channels, so it can flag a developing fault that has no rule attached to it. It also reduces alert volume rather than adding to it, because every flag is triaged before a person sees it.
Four to eight weeks, at no cost, and in the data center case with no live connection at all. One call of about forty-five minutes to scope the export. You send six to twelve months of history as flat files under NDA. We calibrate, run a retrospective against failures you already remember, and score it with misses counted as loudly as hits. Then you decide.
We build a model on your telemetry, then review the predictions against your real failure history.
No cost to evaluate. No live connection required. No obligation.
