Descent log
From spreadsheet
to ship
-
− 0600 M
Chapter 01 · The origin
The expert who saw it coming
A senior engine expert could call a critical engine failure days before any alarm tripped. The tools were humble: exported spreadsheets, charts read by hand, morning by morning. The skill was not. Where everyone else saw noise, the expert saw the trend starting to bend.
Those days are the whole prize. Time to plan parts, people and a berth, instead of dealing with a stopped engine at sea. But the method lived in spreadsheets, manual chart-reading and one person's head, applied to as many engines as one person can watch. That gap is the seed of the live failure-prediction layer built on mtu NautIQ Foresight.
The head start: the expert reads the bend, the alarm waits for the threshold. The shaded window is the time won. Illustrative. -
− 1300 M
Chapter 02 · The build
From spreadsheet to signal
I saw what it could become, and what it must not become. Not a research project, not a black box: a faithful copy. The same signal, the same thresholds, the same judgement the expert applied by hand, made mechanical.
The method stays deliberately plain. It watches the drift the way he did: not one reading but the rate of change, read over several horizons at once, so a slow bleed and a sudden break both show themselves. Plain is the point. The people who have to trust it can follow every step, because every step is one the expert already took.
So this is not an AI standing in for a person. It is the expert's eye at fleet scale, running every hour of every day, on engines nobody had time to watch by hand. Working prototype in a week. Engineering team and OEM stakeholders aligned behind it, because everyone could see whose judgement it encoded.
The fit, sliding: windows of operating days move along the signal, each yields a gradient. The expert's chart-reading, made mechanical. Illustrative. -
− 2000 M
Chapter 03 · The answer
Time to alarm, in operating days
Crews do not act on probabilities. They plan in operating days: what must happen before the next crossing, the next port call, the next scheduled stop. So the product answers the one question that matters on board: how long have I got?
The current gradient, projected forward, becomes exactly that. Time to alarm, in operating days. Not a score, not a confidence value. A number a chief engineer can put into a plan.
The answer has consequences because it lands on a state ladder: NOMINAL, ATTENTION, ALERT. Nominal means carry on. Attention means start planning. Alert means act now. A number on its own is a curiosity. A state with an agreed meaning is an instruction.
The projection cone: the current trend, played forward. Crews plan in days, not probabilities. Illustrative figures. -
− 2700 M
Chapter 04 · The real work
Used, not just deployed
The hard part was never the mathematics. A gradient over a pressure trend is not a breakthrough; any engineer can fit a line. The hard part was everything around the line.
A prediction only matters if it changes what somebody does on a Tuesday. That meant connecting the service organisation with development, development with product, product with design, and all of it down to the workshop where repairs actually happen. Who sees the warning first. Who decides it is real. Who opens the work order, who orders the parts, who stands at the engine. Every handover had to be designed and agreed, or the warning would have died in an inbox.
So the page the prediction lives on sits inside the daily watch, not next to it. The people who watch the fleet meet it inside the screens they already trust, and when the state changes the path to a work order is short and known.
Deployment is a milestone. Daily use is the product. Most prediction projects end at the first and call it success. This one is measured by the second: a warning that becomes a work order that becomes a fixed engine, week after week.
The integration hub: the page in the middle, the organisation around it, down to the workshop. Deployment is a milestone; daily use is the product. -
− 3400 M
Chapter 05 · The road ahead
The loop that learns
Running on real vessels three weeks after the prototype, a pace the industry's normal processes don't allow. Presented in person to the service organisation that watches the fleet. In production now.
Production is where the loop closes. Every warning the service organisation confirms or dismisses becomes a label: this one was real, that one was not. Labels are the road from an expert's encoded judgement toward machine-learned models the organisation can still trust, because they would be trained on its own verdicts.
That road is ahead, not behind. The transfer underway is trust: it used to live with one person, and step by step it is moving into the product, where every confirmed warning will deepen it.
The labelling flywheel: every verdict becomes a label, labels are the road to models the organisation can trust. The road ahead. Illustrative.