A Physics AI living digital twin for primary medium-voltage underground distribution.
Every segment scored. Every score auditable back to the inputs behind it — measured or imputed. Earthflow Underground is a solution for Underground Cable Health.
A health score on every segment, with remaining useful life and a confidence band.
Where headroom runs out — before the new load arrives to find it.
Rank segments for partial discharge and tan δ, instead of testing on the calendar.
Which corridors buy the most risk reduction per mile of spend.
~1.1 million miles of underground distribution cable — and utilities cannot see any of it.
Replaced on age, not evidence — and an underground fault costs 3–4× an overhead one to restore.
Headroom judged on nameplate ratings — while electrification and new load keep arriving.
VLF, tan δ and sheath integrity all need the segment de-energized. Most utilities test a fraction and run the rest to failure.
+5,000–10,000 miles a year are going underground. The blind spot in the power grid is growing.
Earthflow powers granular geographic site intelligence — environmental perils on infrastructure. Acre by acre.
70+ data sources · 200+ KPIs per location · 1,100+ data points per lat/long · 1,300 sites · 1M+ acres analysed
- Asset registry · route & vintage
- Outage history · 5–10 yr, the ground truth
- Feeder loading · ≥1 yr, 15-minute
- Meter intervals · de-identified load shape
Underground
Physics AI
Engine
- Soil chemistry · 26
- Soil moisture · 44
- Flood exposure · 49
- Precipitation · 36
- Ground movement · 24
- Surface erosion · 60
Fused per segment — it resolves to your cable route, not a regional average.
How does Earthflow Underground compare to incumbent modeling approaches?
Every model demo looks good. So we ran a head-to-head test. Earthflow Underground has been benchmarked against every publicly implementable alternative that exists today — two regulator methods, the academic peer-reviewed state of the art, and the age baseline. We lead all of them, across every dimension.
At the job that matters most: detecting the top-10% worst-ranked cables in the fleet.
We analyzed two distribution territories — segment by segment.
Real-pattern load profiles for one full year across both networks (DOE / NLR SMART-DS) — 15-minute profiles per feeder — establishing the year's duty profile.
A 10-year stress-tested simulated outage log — timestamps, cause codes, replacements — with its realism checked against 16 years of real UK (Ofgem) fault data and New Zealand's network disclosures.
Lightly loaded fleets — median duty 5–8% of planned rating — yet 500 segments run above 80% of rating, and 264 above 100%. This is the baseline every method was graded on, head to head.
76% of the cables that failed were already on Earthflow's top-10% list.
Earthflow captured 76% of all cables that went on to fail in the two years after the prediction date — across both territories, Greensboro NC and Austin TX.
135–200% better than CNAIM · ~70% better than Pattabi 2023
What we are — and what we are not.
Everything else is the Earthflow Underground Physics Engine.
- Not pre-trained on anyone's data — no utility data, no public datasets, no cross-utility bleed.
- Not training on your data — tuning seven parameters is not learning a model from your records.
- Not an LLM — no foundation model anywhere in the engine.
- Not sending your data anywhere — nothing leaves for a third party or another AI at inference.
- Not a fine-tune of someone else's model — nothing borrowed, nothing wrapped.
- Not a SaaS — we deliver answers, not another login.
- A Physics AI engine — constrained by governing equations, never free-fitted.
- First-principles physics — thermal, electrical, chemical, mechanical, hydrological — with cited constants; the statistical layer is the only fitted layer.
- Tuned per utility — seven parameters fitted to your outages; zero cross-utility anything.
- Built for imperfect data — degraded inputs widen the uncertainty band, never produce silent guesses.
- A data product — CSV · Parquet · GeoJSON · JSON API · rendered PDF/HTML — monthly, resolved per segment.
- Our own architecture — every module purpose-built, our custom code end to end.
The same four questions — now answered on every feeder segment.
Vintage, your outage history, thermal duty, soil corrosivity and flood exposure — with remaining useful life.
A thermal rating from soil properties and burial method, against your measured loading.
The worst-segment list your partial-discharge and tan δ crews work from.
Corridors ranked by risk reduction per mile of spend.
Produced by the Earthflow underground physics module family. Every score carries a value, a confidence interval and its provenance — each input tagged measured or imputed, so a number can be defended rather than just quoted.
Up to 10 years of deferred capital, and up to 10% off underground O&M (modelled)