North America has roughly ~1.1 million miles of underground medium-voltage distribution cable — and utilities can't see any of it. As resilience programs (FPL SSUP, PG&E's $15–30B undergrounding plan, Dominion's Strategic Underground Program) push toward 50% underground by 2040, the data layer for these invisible assets has become the binding constraint on grid reliability.
Earthflow's Physics AI™ engine stands up the underground analytics layer in as little as 7 days for a first-look risk map — and a full 90-day pilot using only a utility's existing GIS, OMS, and SCADA exports. No new sensors are required for v1. Sensor pilots come in weeks 9–12 to validate predictions before scaling. This report documents the data sources, the 9 production Physics AI™ models, sensor hardware reference, KPI math, deployment roadmap, and ROI model that make rapid implementation feasible today.
Overhead distribution failures are visually obvious: a downed wire, a blown fuse, a tree on the line. Underground failures are invisible. Crews can spend hours with thumpers and TDR reflectometers narrowing down a fault. As a result, underground outage restoration is typically 3–4× more expensive than overhead and lasts longer. The trade-off is fewer outages, but each one costs more.
The grid is going underground faster than the data layer is keeping up. Utilities have SCADA at the substation and AMI at the meter — but everything in between is dark unless physically inspected. With aggressive undergrounding programs converting 5,000–10,000 miles per year nationally, the volume of "dark" infrastructure compounds.
| Overhead | Underground | |
|---|---|---|
| Failure rate | Higher (storm, vegetation, animal) | Lower (3–5× less frequent) |
| Locate time | Minutes (visual) | Hours–days (TDR, thumper) |
| Repair cost | $5–10K typical | $20–50K+ typical |
| Routine O&M | High (vegetation, pole inspections) | Low (75–80% lower than overhead) |
| Public safety | Live wires after storms | Manhole gas / explosion / stray V |
| Visibility | Drone / line patrol | None without sensors |
An underground distribution feeder is not a single cable — it's a system of interconnected components, each with distinct failure modes and inspection regimes. A digital twin must model all six classes:
Primary feeders and laterals at 5–35 kV. XLPE (cross-linked polyethylene) in modern installs; PILC (paper-insulated lead-covered) in legacy systems.
Green metal cabinets in suburban yards or sealed vaults in urban areas. Step down 13.2 kV primary to 240/480 V service. Often the first sign of trouble: oil temperature.
SF6 or vacuum-bottle switchgear at loop ties. Network protectors on secondary spot networks (NYC, Chicago, etc.) auto-isolate faulted feeders.
Concrete access points containing splices, joints, and sometimes transformers. ConEd has ~267,000 manholes in NYC alone — with ~2,000 smoke/explosion events annually.
The structural weak point. Tens of millions of splices US-wide. PD activity (partial discharge) is the strongest leading indicator of joint failure.
Smart line sensors clamped at branch points. Modern devices (Sentient UM3+, SEL) capture waveforms, geolocate faults, and stream over cellular.
Every field rendered in the live demo maps to a real production data source. The table below is the master integration spec for a utility deployment. Status badges: Off-The-Shelf = utility almost certainly has it today; Sensor Deploy = requires a hardware install for full coverage (partial coverage from existing pilots is usually available).
| Demo synthetic field | Production source | Format | Status |
|---|---|---|---|
cableSegments[].polyline |
Esri ArcFM Conduit Manager / ESRI Utility Network | GeoJSON LineString | Off-The-Shelf |
installYear, material, kV, lengthM |
GIS attribute table | CSV / shapefile | Off-The-Shelf |
loadPctMean |
SCADA via OSI PI historian | OPC-UA / CSV | Off-The-Shelf |
pdActivity |
Online PD monitors (Doble, Omicron, IPEC) | API / CSV | Sensor Deploy |
dtsAnomalyC |
DTS controller (Sumitomo, Bandweaver) | Vendor API | Sensor Deploy |
lastFault, saidiContribMin |
OMS (Schneider EcoStruxure ADMS, GE PowerOn) | Event log export | Off-The-Shelf |
nodes[manhole].iot (gas, temp, humidity, stray V) |
CNIGuard / ConEd-style multi-sensor | MQTT / vendor cloud API | Sensor Deploy |
nodes[padmount] (oil temp, load) |
AMI MDM (Itron, Landis+Gyr) + transformer monitors | API | Off-The-Shelf |
nodes[fci] (fault sensors) |
Sentient UM3+ smart line sensor cloud | REST API | Sensor Deploy |
meterClusters[].status |
AMI “last gasp” / no-ping | AMI MDM | Off-The-Shelf |
customersDownstream |
CIS join through GIS topology trace | Periodic batch | Off-The-Shelf |
| Soil corrosivity zones | NRCS SSURGO + utility internal map | GeoJSON polygons (already in repo via extract_ssurgo_results.sh) |
Off-The-Shelf |
| Flood risk zones | FEMA NFHL (National Flood Hazard Layer) | GeoJSON tile | Off-The-Shelf |
The first nine rows above are the foundation. Most utilities export GIS via Esri ArcFM (or modern Utility Network), keep OMS event logs in a SQL warehouse, and have at least 12–24 months of SCADA-historian data and AMI meter status. Earthflow ingests these as CSV / GeoJSON / SQL connectors and produces an AHI score for every cable segment within 2–3 weeks of data delivery.
The four amber rows require hardware. The recommended pilot footprint:
Two map-derived layers are essential and Earthflow already has both in production:
extract_ssurgo_results.sh, upload_spatial_to_firebase.js) ingests NRCS soil polygons and joins to cable segments by intersection.A pragmatic survey of the devices a utility would actually purchase. The platform is sensor-agnostic — we ingest from any of these.
Submersible clamp-on sensor for pad-mount cabinets, junction boxes, and submersible vaults. Up to 12 phases per unit. 256 samples/cycle waveform capture, GPS-synced.
Earthflow ingests REST · produces FCI status, fault waveforms, load currents
Multi-sensor (gas, temp, humidity, stray V, IR camera, accelerometer) for underground vaults. Field-proven by Con Edison NYC at scale (~thousands of installs). Cellular telemetry.
Earthflow ingests MQTT · produces vault explosion-risk index
Capacitive or inductive coupling at cable terminations to listen continuously for partial discharge. Time-of-flight localization within ~1–5 m on long cables.
Earthflow ingests vendor API · produces PD trend, location, severity
Optical fiber co-installed with the cable acts as a thermometer every meter. Detects splice hotspots, thermal overload, and conduit blockages. Common in HV, growing in MV with new conduit installs.
Earthflow ingests vendor API · produces hot-spot map, thermal anomaly
Battery-powered RF sensors that attach to elbows and bolted connections inside switchgear. Detect loose / hot connections that would otherwise fail silently.
Earthflow ingests gateway data · produces connection-health alerts
Stand-alone sensors at branch points. Modern intelligent FCIs report fault passage (and sometimes load) over cellular. Older devices are flag-only.
Earthflow ingests REST or DNP3 · produces fault-direction map
The composite KPI — details in Chapter 6.
Survival-analysis estimate of years remaining. Anchored to AHI, modulated by trend in PD / load / thermal data when available. Output as years with confidence bands.
Earthflow's Physics AI™ engine is the analytical core of the underground twin. It is not one model — it is a coordinated portfolio of nine production models that together turn raw utility data into prioritized, actionable risk intelligence. Each model is grounded in either a physics-based equation (corrosion kinetics, thermal aging, hydrology) or in an empirically-validated statistical method (survival analysis, time-of-arrival fault localization), then fused via the AHI composite. The result: explainable, defensible risk scores for every segment in the network.
| # | Model | Type | Inputs | Output · Use |
|---|---|---|---|---|
| 1 | AHI Composite | Weighted Fusion | Age · material · load · PD · faults · environment | 1–5 health score per segment. The headline KPI surfaced in the 3D twin. |
| 2 | RUL Survival Model | Statistical | Failure history + AHI trend | Years-remaining estimate with 80% confidence bands. Cox proportional-hazards backbone. |
| 3 | PD Trend Forecaster | Time Series ML | Online PD monitor stream (pC vs time) | 30 / 90 / 365-day PD forecast. Triggers alerts when accelerating. |
| 4 | Fault Triangulation | Physics | Sentient FCI time-of-arrival + GPS | Geolocates fault to ~200 ft section. Cuts patrol time 65%. |
| 5 | Soil Corrosion Kinetics | Physics | SSURGO chloride/sulfate, soil resistivity, pH, cable jacket type | Lead-sheath corrosion rate (mm/yr). Drives EnvironmentalRisk sub-score. |
| 6 | Flood Hydrology Spatial Join | Physics + GIS | FEMA NFHL polygons, cable elevation profile | Water-tree degradation multiplier per segment. AE = +20%, X-shaded = +10%. |
| 7 | Manhole Explosion Risk | Multivariate Classifier | Gas (% LEL), temp, humidity, stray V, IR camera, history | Real-time vault hazard score. Triggers immediate dispatch when crossing threshold. |
| 8 | Capex Deferral Optimizer | Optimization | AHI distribution, RUL bands, $/mi replacement cost, budget constraint | Annual replacement schedule that maximizes SAIDI improvement per $ spent. |
| 9 | AMI Dark-Sector Correlator | Pattern Detection | AMI “last gasp” / no-ping clusters + GIS topology | Localizes outage to 1–2 cable segments before crews are dispatched. |
Models #1 (AHI), #2 (RUL), #5 (Soil), #6 (Flood), #8 (Optimizer), and #9 (AMI Correlator) all run from day-1 off-the-shelf data — no new sensors required. Models #3 (PD), #4 (Fault Triangulation), and #7 (Manhole Risk) require sensor data and come online in weeks 9–12 of the pilot. You get six of nine models running in your first week.
The AHI is the most important of the nine because it surfaces the highest-impact, easiest-to-act-on signal: a 1 (excellent) to 5 (end-of-life) composite score for every cable segment. It blends static attributes (age, material) with dynamic indicators (PD, load, fault history, environment) using a transparent, weighted-sum formula.
| Sub-score | 1 (Healthy) | 3 (Watch) | 5 (Replace) |
|---|---|---|---|
| AgeScore | < 25% of design life | 50–75% | > 100% |
| MaterialRisk | Modern XLPE / EPR | 1990s XLPE | PILC, pre-1985 XLPE |
| LoadStress | < 60% rated | 60–80% | > 80% sustained |
| PDSeverity | < 100 pC | 200–400 pC, trending up | > 500 pC, accelerating |
| FailureHistory | Zero faults | 1 fault in 10 yr | 2+ faults in 10 yr |
| EnvironmentalRisk | Dry sandy soil, no flood | Mixed loam, X-shaded zone | Sulfate clay + AE flood |
Earthflow is engineered for frictionless onboarding. There is no on-prem install, no infrastructure procurement, no 12-month enterprise rollout. A utility can be looking at Physics AI™ risk scores for their actual network in one week, and have a full sensor-validated pilot in three months.
Drop a GIS shapefile / GeoJSON of underground cable + an OMS event log CSV into the Earthflow ingest portal. That's it for the utility's day-1 effort.
Utility: 2 hrs · Earthflow: 0 hrsEarthflow's auto-mapper proposes column ↔ schema mappings. A utility analyst reviews and approves in a single session.
Utility: 4 hrs · Earthflow: 4 hrsAHI Composite, RUL, Soil Corrosion, Flood Hydrology, Capex Deferral Optimizer, AMI Dark-Sector Correlator — all running on the utility's actual data. Top-50 worst-cable list generated.
Earthflow: 8 hrs (mostly compute)The utility receives a private demo URL. Asset Management, T&D Engineering, and the CIO can all see the underground network in 3D, color-coded by Physics AI™ AHI, with click-through to per-segment detail. Decision-quality output by end of week one.
Utility: review session 1 hr · Earthflow: 2 hrs hand-offAfter the 7-day quick-start, the full pilot extends to validate predictions with sensors and refine the models. Each step below assumes one utility data engineer + one Earthflow integration engineer in parallel.
Pull GIS export (Esri ArcFM / Utility Network), OMS event log (last 5 yr), SCADA daily aggregates, AMI meter status, customer-feeder join from CIS.
Utility: 1 FTE-wk · Earthflow: 0.5 FTE-wkMap utility-specific column names and code lists (material types, voltage classes, fault codes) to Earthflow's internal schema. Manually validate 10% of records for quality.
Utility: 1 FTE-wk · Earthflow: 1 FTE-wkRun baseline model on full inventory using only off-the-shelf data. Output: AHI for every segment, ranked replacement list, top-50 worst segments report.
Earthflow: 1.5 FTE-wkStand up the Earthflow underground twin with the utility's actual GIS geometry. First demo to Asset Management & Operations.
Earthflow: 1 FTE-wkDeploy 5–10 Sentient UM3+ on the model's worst-flagged feeders. Add 2–3 manhole IoT in highest-risk vaults. Optional 1–2 PD monitors.
Utility: 2 FTE-wk (field crew) · Vendor lead time 2–4 wkPipe live sensor data into Earthflow. Re-score all segments with sensor evidence. Validate that the 30-day sensor record corroborates the model's risk ranking.
Earthflow: 1 FTE-wk · Utility: 0.5 FTE-wkFindings deck for utility leadership. Recommended capex prioritization. Scope for full-fleet rollout. Decision point: scale or stop.
Earthflow: 0.5 FTE-wk| Driver | Mechanism | Magnitude |
|---|---|---|
| Avoided outages | Preempt failures by replacing high-AHI cable | $20K avg per underground fault avoided |
| Faster restoration | Sentient + AI narrows fault location | 20%+ CMI reduction, 65% patrol-time reduction |
| O&M efficiency | Condition-based vs time-based maintenance | ~11% O&M cost reduction |
| Capex deferral | Don't replace healthy cable too early | Up to 10-yr deferral on segments AHI≤2 (Siemens claim) |
| Safety / liability | Manhole gas alarms before incidents | Avoid 1 explosion = pay for the program |
| Regulatory | SAIDI/CAIDI improvement → performance incentives | Varies; can offset rate cases entirely |
Larger utilities see proportionally larger absolute savings. The Dominion / FPL / PG&E case studies cited in industry reports show 20%+ SAIDI improvements and capex deferrals north of $1.76 per $1 spent on the analytics layer.
Four major vendors offer overlapping but differentiated solutions. Earthflow's positioning: the open, geospatial-first platform that complements rather than replaces existing OT/EAM stacks.
AI/ML on existing utility data. Claims 10× failure reduction, 10-yr capex deferral. Strong sensor portfolio (SICAM EFI fault indicators).
Differentiator vs Earthflow: Siemens bundles consulting; Earthflow ships as software-first.
Unified APM across all asset classes. AI-driven prognostics, prescriptive recommendations, scenario simulation. Strong ABB heritage in grid hardware.
Differentiator vs Earthflow: Lumada is an enterprise platform; Earthflow targets a single use case with deep geospatial integration.
Industrial-grade APM with strong T&D track record. Reports 20% reduction in reactive maintenance at customer deployments.
Differentiator vs Earthflow: GE optimizes for ADMS-integrated workflows; Earthflow optimizes for visual decision-making.
Owns ArcFM (the dominant utility GIS) plus Conduit Manager. Strongest position on the data side; lighter on advanced AI.
Differentiator vs Earthflow: Schneider integrates with their own GIS; Earthflow is GIS-agnostic and works on top of any feed.
Every chapter of the live demo corresponds to one or more data sources and one or more user personas. The mapping below is the customer-facing rosetta stone for stakeholder briefings.
| Demo Chapter | Data Source(s) | Primary Persona | Decision Enabled |
|---|---|---|---|
| 3D underground twin (X-ray) | GIS + OMS | T&D VP, CIO | "What's actually under our streets?" |
| Cable Health Index colorant | GIS + OMS + SCADA | Asset Management | Replacement prioritization |
| Load / Stress colorant | SCADA + AMI | System Planning | Capacity expansion targeting |
| PD Activity (joints) | Online PD monitors | Reliability Engineer | Schedule splice repair before failure |
| DTS Temperature | DTS fiber controllers | Operations | Detect splice hotspots |
| Soil Corrosivity heat-map | NRCS SSURGO | Asset Management | Long-term replacement strategy |
| Flood Risk heat-map | FEMA NFHL | Resilience & Storm Hardening | Storm prep + post-event triage |
| Manhole IoT live data | CNIGuard / equiv | Field Operations, Safety | Prevent manhole explosions |
| AMI dark-sector overlay | AMI MDM | OMS Operator | Narrow fault location |
| FCI fault sensors | Sentient UM3+ | Reliability Engineer | Fast fault triangulation |
| Replacement scenario | All of the above | CFO, Capex Planning | Justify replacement spending |
| Today vs Earthflow split-screen | (narrative) | Executive | Visualize the value gap |
| Term | Meaning |
|---|---|
| ADMS | Advanced Distribution Management System (Schneider EcoStruxure, GE PowerOn) |
| AHI | Asset Health Index, 1 (best) – 5 (worst) |
| AMI | Advanced Metering Infrastructure (smart meters) |
| APM | Asset Performance Management software |
| CAIDI | Customer Average Interruption Duration Index (restoration speed) |
| CIS | Customer Information System (billing & meter location) |
| DGA | Dissolved Gas Analysis (transformer oil testing) |
| DTS | Distributed Temperature Sensing (fiber optic thermometer along cable) |
| EAM | Enterprise Asset Management (SAP, Maximo) |
| FCI | Faulted Circuit Indicator (smart line sensor) |
| FEMA NFHL | FEMA National Flood Hazard Layer (flood-zone polygons) |
| MV | Medium Voltage (5–35 kV distribution class) |
| MDM | Meter Data Management system (Itron, Landis+Gyr) |
| OMS | Outage Management System |
| PD | Partial Discharge (insulation degradation indicator) |
| PILC | Paper-Insulated Lead-Covered (legacy MV cable) |
| RUL | Remaining Useful Life |
| SAIDI | System Average Interruption Duration Index |
| SCADA | Supervisory Control and Data Acquisition |
| SSUP | Storm Secure Underground Program (FPL's flagship undergrounding initiative) |
| SSURGO | Soil Survey Geographic Database (USDA NRCS) |
| XLPE | Cross-Linked Polyethylene (modern MV cable insulation) |