← SangMil / Research
Active · 2026 research programme

Thermal power plant
shutdown planning.

A planned outage is the highest-stakes schedule in industry: thousands of interdependent activities, a unit earning nothing while it's offline, and a return-to-service date the whole grid is counting on. We're building AI that plans it, watches it, and re-baselines it in real time.

~$1M
PER DAY OFFLINE
10k+
ACTIVITIES / OUTAGE
14–35
DAYS TYPICAL
600 MW
UNIT SCALE
The problem

Why a shutdown is different from any other schedule.

A turnaround compresses years of deferred work into a fixed window. The unit is de-pressurised, cooled, opened, inspected — and only then does the real scope reveal itself. Findings on the boiler drive turbine work; a failed hydrotest ripples through every downstream trade.

Traditional planning tools treat this as a static Gantt. But an outage is a living system: hold-points, permits and clearances gate progress hour by hour, and the return-to-service date moves the instant one predecessor slips. Miss it and the cost is measured in millions per day — plus grid penalties.

Discovery scope
Half the work isn't known until the unit is open.
Hard hold-points
Permits and clearances gate every phase transition.
Resource contention
Cranes, scaffolding and crews shared across trades.
Fixed RTS date
The grid is counting on the megawatts coming back.
Our approach

Four research threads, one engine.

R.01

Predictive risk on the RTS date

Monte Carlo simulation over the outage network gives P10–P90 return-to-service dates and the live probability of hitting the target — updated as findings land.

R.02

Live hold-point & permit tracking

Every clearance, isolation and permit modelled as a gate in the schedule — so the plan reflects what's actually cleared to start, not what was planned weeks ago.

R.03

Discovery-scope re-baselining

When inspection reveals new work, agents propose an updated sequence and recovery options in minutes — grounded in CPM, with the RTS impact quantified before you commit.

R.04

Cross-trade conflict detection

Crew, scaffolding, crane and parts contention surfaced across boiler, turbine and HRSG work before two trades collide on the same platform.

Anatomy of an outage

Six phases, one moving finish line.

UNIT 3 · 21-DAY WINDOW · ILLUSTRATIVE RTS ▸ DAY 21
Cooldown & isolation
Open & inspect
Boiler overhaul
Turbine & HRSG
Reassembly & hydrotest
Return to service
Critical path Driving activities RTS milestone
Return-to-service risk

The finish date, as a probability — not a promise.

We run Monte Carlo over the live outage network to produce a confidence band on the return-to-service date. As findings land and hold-points clear, the distribution tightens — and management sees the odds of hitting the target date update in real time.

Day 20.1
P10 · BEST CASE
Day 23.4
P90 · +2.4 DAYS
61%
HIT TARGET
RTS PROBABILITY DISTRIBUTION · 10,000 RUNS
Day 19Target · Day 21Day 25
Collaboration

Built with industry and academia.

This programme runs in partnership with an operating power utility and a university research group. Details disclosed on request.

industry partner logo
Industry partner
Operating power utility · disclosed on request
academic partner logo
Academic collaborator
University research group

Running an outage, or researching one?

We're looking for utilities, EPCs and researchers to pilot the engine on real turnaround data. Reach out to join the 2026 programme.

Contact the research team ← Back to SangMil