When a plant goes quiet, a phone rings
Plant telemetry lands in a store we own, a physics based model says what the site should be producing in the conditions it is actually in, and when the difference means a plant has lost power the platform telephones the person on duty and speaks to them in Turkish. Every number said on that call traces back to a measured value.
- DomainSolar and wind generation, EPC and independent power production
- EngagementJoint venture, our own platform
- StageDemonstrated live, from fault to answered call
Alarms are cheap. Reaching a person is the hard part.
The platform is a joint venture between Koca Ventures Ltd (UK) and 88Miles XR B.V. (Leiden, NL), built for solar and wind developers, for the contractors who build their sites and for the operators who run them. Plant telemetry arrives through vendor specific plugins into a timeseries store the company itself owns, and a physics based expectation runs beside it: what the site should be producing given the conditions it is in. A deviation that means a plant has lost power becomes a spoken call to whoever is on duty.
Everything else follows from where the data sits. It never leaves the company's own infrastructure. Every number an operator hears on a call traces to a measured value, and where nothing was measured the platform says so out loud. Investor reporting for project companies, storage predictions and market operations in Turkey and the Netherlands run on the same fabric, under role based access and an audit trail that cannot be edited afterwards.
The alarm fired. Nobody was looking.
Generating assets fail at night, in the rain, on a public holiday. The gap is never detection alone: it is the distance between a detection and a human who acts on it.
Alarms nobody hears
A fault raised on a control screen in the middle of the night waits for the next person to open that screen. The asset stands idle for the whole of that wait, and the alarm itself knows neither who was on duty nor whether they were reachable.
Numbers nobody can vouch for
Reports circulate carrying figures that no longer trace to a measurement. Once a gap has been filled with an estimate, a plausible average or a value carried forward, nothing in the report separates the numbers that were witnessed from the ones that were invented.
A night with no story
In the morning someone asks what happened. The answer gets assembled from memory, a photograph on a phone and a chat message, because nowhere was the sequence recorded: what was detected, who was reached, what was decided, who approved it.
Detection that ends in a conversation, not a notification
Telemetry the company owns, an incident agent that speaks, a second voice that answers from the company's own documents, and field work closed with evidence.
Plant telemetry in a store the company owns
Site control systems land through vendor specific plugins into a timeseries store, and a physics based expectation runs on top of it: what this site should be producing in the conditions it is in. Deviation is judged against that expectation.
- Vendor plugins, so supporting a new site type means adding a plugin
- A physics based expectation, so the comparison is against what the conditions allow
- Data held on the company's own infrastructure, never handed to an outside service
An incident agent that places a real phone call
When a plant loses power the platform telephones whoever is on duty at that moment and speaks to them in Turkish. It works down a configurable duty ladder, it knows when it is calling outside business hours, and it asks for spoken confirmation that the person is taking the incident.
- A configurable duty ladder, so an unanswered call reaches the next person
- Business hours awareness, so behaviour out of hours differs from working hours
- Spoken confirmation of authority, and a morning digest of what happened overnight
A second voice that answers from the company's documents
Engineers ask their questions out loud and get answers drawn from the company's own document vault, with the page each answer came from cited beside it. It answers only from what the company holds.
- Answers grounded in the company's own document vault
- Page level citation on every answer, so a claim can be opened and read
- Nothing leaves the company's infrastructure in order to answer a question
Field work closed with evidence
A mobile application carries the incident to the person in the field and closes it there, with photographs attached as evidence. Every change the platform proposes ends at a human approval before anything is committed.
- Field closure from a phone, with photographic evidence attached
- Role based access, and an audit trail that cannot be rewritten afterwards
- Every change the platform proposes ends at a human approval
What it does when there is nothing to measure
The decisions worth describing here are about restraint: what the platform refuses to say, and where it stops and asks a person.
Unmeasured is visibly unavailable
Where nothing was measured, the platform shows the gap. No carried forward value, no plausible average, no quiet substitution that would let a missing reading pass for a real one.
Every number traces to a measurement
Anything an operator hears on a call or reads in a report can be followed back to the measured value behind it. A figure that cannot be traced is not published.
Root cause runs as deterministic stages
Fault analysis is a fixed sequence of deterministic stages that runs the same way every time. A single bounded advisory slot inside that sequence is served by a model running on the company's own hardware. That slot advises. It does not decide.
The honesty rules are tested adversarially
The rules about absent measurement are held by a test suite written to attack them, so a change that lets an unmeasured value quietly become a number fails before it reaches anyone.
What has actually been demonstrated
No availability or revenue figure appears here, because none has been measured on a running site. What follows is what the platform has been seen to do.
Proven live
Fault to answered call, in front of people
The recording of a real turbine fault was replayed into the platform. Its own detector latched the fault, the incident agent placed the call, and a human answered a phone that actually rang and confirmed the incident out loud.
Stays home
Where the data lives
Telemetry, documents and the model that advises on root cause all sit on the company's own infrastructure. Nothing is sent to an outside service in order to understand an incident.
The whole night
A recoverable account
What was detected, who was called, who confirmed, what was decided and who approved it are all recorded, so the morning question about what happened has an answer that does not depend on anyone's memory.
Do your alarms reach a person?
If your assets can sit idle because a detection never turned into a conversation, that is exactly the gap this platform was built to close. Tell us what your operation looks like.
Start a conversation