Earthflow Underground
Industry White Paper · Underground Distribution Cable Analytics
Digital Solutions for Underground Distribution Cable Analytics
An industry analysis of underground distribution cable analytics across U.S. and Canadian electric utilities — covering grid undergrounding trends, current asset-management practice, data and sensor coverage, key performance indicators, AI/ML reference architecture, return on investment, and the competitive vendor landscape. With a reference implementation built on Orbyfy Earthflow Underground — a virtual-sensor physics engine for underground distribution cable.
Earthflow Underground · Today's GIS view vs the Earthflow Physics AI risk overlay
Report ID · ORB-2026-UCG-001
Published · 2026-09-30
Version · 2.0
Industry · Electric Utility · T&D
▷ Live Demo · earthflow.orbyfy.com/earthflow-underground
Authored by Orbyfy Earthflow Research
Orbyfy · Boston, MA

Executive Summary

North American utilities are burying their distribution networks faster than at any point in their history. About 20% of U.S. distribution line miles are already underground, and the largest investor-owned utilities — PG&E, FPL, Dominion, WEC, Eversource — are spending billions a year to push that share higher, with industry advocates calling for 50% underground by 2040. Every mile they bury becomes a mile they can no longer see. Utilities know less about the condition of their underground cable than about any overhead asset they own, yet an underground failure costs 3–4× more to restore and takes longer to find.

This paper argues that the binding constraint on the underground grid is no longer the cost of burying cable — it is knowing the condition and capacity of the cable already buried. It examines the data utilities already hold and why it falls short, the sensors and KPIs available today, the analytics architecture that turns those inputs into decisions, the business case, and the vendor landscape (Siemens, Hitachi Energy, GE Vernova, Schneider Electric). It then presents Orbyfy Earthflow Underground as a reference solution — a virtual-sensor physics engine that fuses 11 cable-specific Physics AI modules with a 19-module platform engine, running on a utility's existing GIS, OMS, SCADA, and AMI data with no new sensors. The reference solution answers two questions on every cable segment rather than one — Condition (how degraded is it, how likely is it to fail) as a 1–5 Health Index, and Capacity (how much load can it carry, how much headroom remains) as a Capacity Index built on an environmentally-adjusted in-situ line rating. Both are re-scored monthly, the pace at which buried cable and utility capital decisions actually move.

1.1M+ mi
US underground distribution cable12
~20%
Of US distribution line miles, already underground2
3–4×
Cost to restore an underground failure vs overhead
50%
Underground target by 2040 at major undergrounding programs
20%+
CMI reduction, instrumented feeders79
65%
Patrol-time reduction, line sensors58
11%
Distribution O&M cost reduction58
10×
Failure reduction, vendor-cited63

The second row reports published industry results from instrumented-feeder, line-sensor and vendor programs described later in this paper. They are cited to their sources and are not Earthflow results.

The central finding is that most of the information needed is already in hand. A utility's own GIS, outage, SCADA and AMI records, combined with an environmental layer describing the ground each cable sits in — soil chemistry, moisture, flood exposure, terrain, lightning — are enough to rank every cable segment by condition and by capacity before a single new sensor is installed. What the market has lacked is a solution that does this quickly, across every hazard at once, and with a record a regulator can audit. This paper sets out what such a solution requires and illustrates it on a synthetic demonstration feeder modeled on a coastal Florida service area.

Table of Contents

Part I
The Underground Problem

1 · Trend Toward Undergrounding — Utility Initiatives in US and Canada

Undergrounding momentum in the US. Roughly 20% of America's distribution lines are underground today, up from ~18% in 20092. For decades that growth came quietly, from new subdivisions built with buried service. It is now being driven deliberately, by some of the largest capital programs in the industry:

02,5005,0007,50010,000PG&E: originally announced goal of 10,000 milesPG&E (original goal)10,000 miPG&E ten-year undergrounding plan filed Oct 2026: ~5,000 miles, 2028–2037PG&E SB 884 plan, 2028–37~5,000 miDominion Strategic Underground Program: 4,000 miles of outage-prone tap lines by 2028Dominion SUP (by 2028)4,000 miWEC Energy Group: 2,000 miles undergrounded in 8 yearsWEC (done, 8 years)2,000 mi
Figure 1.1 · Miles of distribution line in major undergrounding programs PG&E’s first filed ten-year plan (highlighted) covers roughly half its originally announced goal. Sources: PG&E67127, Dominion15, WEC13.

1.1 Canadian Perspective

Canada's grid remains more overhead-centric overall, with roughly only 10–12% of distribution lines underground4. In Québec, for example, ~11% of distribution is underground, a rate "consistent with elsewhere in Canada"4. This lower penetration reflects Canada's vast rural networks and historically higher cost of underground construction. That said, urban centers and new subdivisions increasingly use underground distribution for aesthetics and reliability22. Provincial programs are generally less formalized than in the US (many decisions are left to municipalities)23, but the trend is slowly upward. For instance, Hydro Québec notes "more and more" towns choose to bury lines in new developments for resilience and beautification22. Toronto and other large cities already operate significant downtown underground networks. As ice storms and high winds continue to damage overhead lines, interest in strategic undergrounding is growing. However, costs remain a barrier; installing underground distribution in Canada can be up to 10× the cost of overhead in some cases24.

1.2 What the Programs Cost

Converting overhead line to underground is expensive, and the cost varies more with setting than with almost anything else. The Department of Energy’s 2024 resilience investment guide documents distribution conversions from about $0.16 million per mile in rural Wisconsin to nearly $3 million per mile for urban 15 kV work and PG&E’s wildfire program118 (Figure 1.2). Replacing cable that is already underground is a different and much smaller number — Chapter 3 returns to it.

$0M$0.5M$1M$1.5M$2M$2.5M$3M$3.5MWisconsin Public Service rural conversion 2012–2021: $0.16M per mileRural WI (WPS, 2012–21)$0.16MVEPCO rural single-phase 2016–2022: $0.48M per mileRural VA single-phase (2016–22)$0.48MNew Hampshire Electric Co-op 34.5 kV urban (2009): $1.43M per mileNH co-op 34.5 kV urban (2009)$1.43MNew Hampshire Electric Co-op 15 kV urban (2009): $2.88M per mileNH co-op 15 kV urban (2009)$2.88MPG&E 2023–2026 forecast: $2.97M per mile (2019 USD)PG&E 2023–26 (2019 $)$2.97MPG&E reported cost per mile in 2025: $3.1M (company-reported)PG&E actual, 2025$3.1M
Figure 1.2 · Cost per mile to convert distribution overhead to underground Figures are in the dollar years reported by each source and are not inflation-adjusted. Distribution conversions only; transmission costs are several times higher. Sources: DOE/LBNL118, PG&E 2025 (company-reported)128.

The sums are now large enough to show up as a line on customer bills. Florida’s investor-owned utilities recovered $2.47 billion of storm-protection costs in 2024, close to double the 2020 figure, and more than 95% of that spending now flows through the dedicated recovery clause129 (Figure 1.3). Every one of those dollars is reviewed for prudence — which is why the evidence behind each project matters as much as the project itself.

$0B$1B$2B$3BFlorida IOU storm protection cost recovery, 2020 actual: $1,350.0M2020 (actual)$1.35B2024 actual: $2,470.2M (FPL $1,549.0M; DEF $698.7M; TECO $203.2M; FPUC $19.2M)2024 (actual)$2.47B2025 estimate: $2,657.5M2025 (estimate)$2.66B
Figure 1.3 · Storm-protection cost recovery, Florida investor-owned utilities Annual spending recovered through the Storm Protection Plan Cost Recovery Clause, all Florida IOUs. Source: Florida PSC annual status report129.
Key Takeaway Most of the North American grid will stay overhead for years to come, but the direction of travel is unambiguous. Advocates call for ~50% of U.S. distribution to be underground by 204020. Each of the programs above adds thousands of miles of cable that no patrol will ever see again — which makes the ability to manage that cable digitally as important as the decision to bury it.

2 · Overhead vs Underground Infrastructure — Current Stats and Trends

RegionTotal Distribution Line LengthUnderground (%)Overhead (%)
United States (2023)~5.5 million miles1~20%2 (≈1.1 million miles)~80% (≈4.4 million miles)
Canada~572,000 km (≈355,000 miles)3~11%4~89%

Table 2.1 · Approximate Overhead vs. Underground Distribution Line Lengths. Sources: DOE Grid Deployment Office2, Hydro-Québec4. (Note: Transmission lines are overwhelmingly overhead in both countries, with <1% underground at high voltages5.)

0%25%50%75%100%Canada: ~11% of distribution undergroundCanada~11%United States: ~20% of distribution line miles underground (2023)United States~20%California: ~33% underground statewideCalifornia (statewide)~33%WEC Energy Group: 39% underground after its programWEC Energy Group39%Colorado Springs: ~77% undergroundColorado Springs~77%Fort Collins: ~99% undergroundFort Collins~99%
Figure 2.1 · Share of distribution line underground The national figure (highlighted) sits far below what long-running programs have reached. Sources: DOE2, Hydro-Québec4, California25, WEC13, Colorado municipal utilities18.

2.1 Distribution Systems

About one-fifth of U.S. distribution line mileage is underground, versus roughly one-tenth in Canada. In absolute terms, the U.S. has on the order of 1 million miles of underground distribution cables (out of ~5.5M total distribution line-miles)12. Canada's total distribution network is smaller (~572,000 km or 355,000 miles3) with perhaps ~60,000 km underground. States with long-running undergrounding policies run well above the national figure — California at ~33% statewide25 and Florida at ~40% and climbing. Meanwhile, many new residential developments across North America have been built with underground utilities since the 1970s, so the underground share grows steadily as old overhead lines are retired or converted26. DOE data confirms a modest upward trend (18% underground in 2009 to ~20% in 2023 nationally)2. This is expected to accelerate as the large programs (PG&E, FPL, Dominion, etc.) ramp up in the coming decade.

2.2 Transmission Systems

High-voltage transmission lines remain overwhelmingly overhead in both the US and Canada. Due to much higher costs and technical complexity, only special situations warrant underground transmission (e.g. dense urban cores, submarine cable links, or environmentally sensitive areas). As of 2009, as little as 0.5% of U.S. transmission line mileage ≥200 kV was underground5. Industry estimates for all transmission voltages put underground at well under 5% of mileage. Canada is similar – the vast majority of its ~160,000 km of transmission is overhead27, with just a few underground cable segments (for example, short runs in downtown Montreal or Toronto). The economics and repair difficulty of underground transmission mean it will remain rare, except for HVDC specialty projects or short urban sections. Thus, the focus for underground analytics is squarely on distribution networks, where underground circuits are far more common and growing.

2.3 Key Trends

In summary, North America's electric grid is still largely overhead, but the portion that is underground (millions of cable-miles) is significant and rising. Drivers for undergrounding include: resilience to storms (hurricanes, wildfires, ice), reduced routine maintenance (no tree trimming), improved public safety (no live wires falling), and aesthetics/property value gains2829. The trade-off is a higher upfront cost and a different set of failure modes — ones that are harder to see coming and slower to find once they occur. Chapter 3 looks at how utilities manage that trade-off today.

3 · Managing Underground Infrastructure — Organization, Budgets, Challenges

3.1 Utility Organizational Approach

Management of underground distribution infrastructure typically involves specialized teams within a utility's operations and engineering groups. Most utilities have a Distribution Operations department that oversees both overhead and underground networks, but they often assign underground-specific crews and engineers for certain tasks. For example, underground line crews / cable splicers handle cable installation, splicing, and repairs, distinct from overhead linemen. In asset management terms, some utilities have dedicated Underground Asset Managers or Engineers focusing on cable systems, while others fold this into broader distribution asset management roles. Functional groups commonly involved include:

3.2 Budget Profile

Budget-wise, underground infrastructure tends to have a different expense profile than overhead. Overhead lines incur substantial ongoing O&M costs (pole inspections, vegetation management, etc.), whereas underground lines avoid most of those but can be costly when failures occur. On average, maintenance costs for underground lines are 75–80% lower than for overhead (since there's no tree trimming or storm repair most years)28. Utilities save on vegetation management – studies show undergrounding saves about $7,000 per mile per year in avoided tree trimming costs on average30. Consequently, some utilities allocate a smaller routine O&M budget per mile for underground. However, they must budget for event-driven repairs and capital replacement. When an underground cable fails, repair costs can be 3–4 times higher than an overhead fix, due to the excavation, splicing, and often longer outage restoration time3132. Many utilities maintain contingency funds or higher capital reserves for cable replacement programs (e.g. replacing vintage PILC or early XLPE cables known to be failure-prone).

Routine maintenance cost (overhead = 100)
0255075100Overhead routine maintenance cost (index = 100)Overhead100Underground routine maintenance: 75–80% lower than overheadUnderground20–25
Cost to restore one failure (overhead = 1×)
0×1×2×3×4×Overhead: cost to restore a failure (= 1×)Overhead1×Underground: 3–4× the cost to restore a failureUnderground3–4×
Figure 3.1 · Fewer, but costlier, failures Underground saves on routine maintenance and pays it back when something breaks. The lighter extension shows the published range. Sources: maintenance28, restoration3132.

3.3 Challenges vs. Overhead

Underground assets are harder to manage than overhead ones in four specific ways:

3.4 Budget Implications

Due to the above challenges, utilities often end up spending a significant portion of distribution O&M and capital budgets on underground cable programs. U.S. utility capex data shows spending on underground distribution has grown slightly faster than overall distribution spending: from ~21.4% of distribution capex in 2010 up to ~22.2% by 201940. This indicates utilities are investing more in underground assets (new and existing) over time.

What replacement costs, and how often it is needed. The DOE-funded Cable Diagnostic Focused Initiative (CDFI) found a mean North American MV cable failure rate of approximately 12 failures per 100 miles per year, with some utilities above 100117. Its consensus cost ranges put installing a new single-phase run of cable at $15–25 per foot, rejuvenating one at $4–13 per foot, and a repair at $1,500–2,500 per event117. Underground cable also wears out sooner than overhead line: DOE’s 2024 guide puts underground lifetimes at 20–40 years against 30–50 for overhead118. Yet between 56% and 74% of utilities surveyed used no cable diagnostics at all117 — most of the fleet is replaced either on age or after it fails.

Where utilities do manage cable on condition, the budget splits deliberately between options. Alectra in Ontario budgeted C$27.2 million for cable injection and C$38.7 million for cable replacement in 2024119 (Figure 3.2). It asked the Ontario Energy Board for a further C$25.1 million of incremental capital to address deteriorated underground cable in 16 neighbourhoods; the Board approved C$17.3 million and declined C$7.9 million120.

C$0MC$10MC$20MC$30MC$40MAlectra 2024 budget (incl. ICM): cable injection C$27.2MCable injectionC$27.2MAlectra 2024 budget (incl. ICM): cable replacement C$38.7MCable replacementC$38.7M
Figure 3.2 · One utility’s 2024 cable renewal budget, by option Alectra Utilities, 2024 budget including incremental capital. Rejuvenation (injection) carries well over a third of the cable renewal spend. Source: OEB filing EB-2022-0013119.

3.5 What the Regulator Needs to See

Every dollar of cable spending eventually has to survive a prudency review, and regulators have become specific about the evidence they expect. Four jurisdictions show the pattern.

JurisdictionVehicleWhat the utility must show
OntarioDistribution System Plan (OEB filing requirements, Ch. 5)Asset condition by major asset type — "solely using asset age is not sufficient" — and the data used to "identify, select, prioritize, optimize and pace" investments121
FloridaStorm Protection Plan (s. 366.96 F.S.; Rule 25-6.030)Expected reduction in outage times and restoration costs, prioritising areas of lower reliability; feasibility; costs and benefits; three-year rate impact; and the criteria used to select and prioritise programs122123
VirginiaStrategic Underground Program (Va. Code § 56-585.1)Eligibility by outage history — tap lines averaging nine or more unplanned outage events per mile over ten years; costs within statutory caps are deemed prudent125
CaliforniaSB 884 ten-year undergrounding planRisk-based circuit prioritisation; annual unit-cost and mileage targets; comparison against covered conductor and other alternatives; cost targets below the utility’s three-year average136

Table 3.1 · What four regulators require before approving underground spending.

Two themes run through all four. First, regulators want the selection logic, not just the total — which segments, ranked how, on what evidence. The Ontario Board approved Alectra’s 2024 cable projects as "prudent investment in capital for cable injection/replacement based upon the current condition of the cable assets"120: the approval turned on condition. Second, the evidence most filings can offer is still outage history. FPL’s approved lateral ranking uses outage counts and recent hurricane experience, with no age or condition factor124, and Virginia’s eligibility test is an outage count125. Outage history finds cable that has already failed. It says little about cable that is about to — and that gap, between what regulators increasingly ask for and what most filings can show, is where condition analytics earns its place.

3.6 Capital versus Operating Expense

How a dollar is classified shapes which solutions a utility proposes. Investor-owned utilities earn a return on capital placed in rate base, but not on operating expense. As Lawrence Berkeley National Laboratory put it, utilities "typically have an incentive to make capital investments, but rarely to employ expense-based solutions, since utilities do not earn profits on expenses"130. For underground cable, that tilts the field toward replacement — a capital project — and away from testing, monitoring and analytics, which often sit in O&M.

Regulators have begun to correct for it. Britain’s Ofgem moved to a fixed capitalisation rate on total expenditure (“totex”), explicitly "to equalise the incentives on capex and opex and avoid distorting decision making"131. And classification can be settled program by program: Alectra books cable injection as capital alongside replacement, and the Ontario Board accepted it on that basis119120. The practical point for a utility building a cable-renewal case is that life extension and condition assessment need not be stranded in O&M. Framed as part of a capital renewal program, they compete with replacement on the same footing.

Underground distribution is a game of fewer, but costlier, failures. Utilities save on routine maintenance and pay for it when something breaks. When failures are rare, expensive and invisible until they happen, the value of knowing in advance which segment will fail next is unusually high — and that is the problem analytics exists to solve. Chapter 4 sizes the asset base involved.

4 · Key Underground Distribution Assets in the US and Canada

The underground distribution system is made up of five asset classes, each present in very large numbers across North America.

~1M mi
US medium-voltage underground cable2
60–80M
US distribution transformers in service4243
~264K
Manholes and service boxes, Con Edison NYC alone45
~12
Mean MV cable failures per 100 miles per year, North America117

4.1 Medium-Voltage Underground Cables

These are the primary distribution feeders and laterals in underground areas, typically carrying 5 kV to 35 kV. In the U.S., there are on the order of 1 million circuit-miles of MV underground cables (as noted, ~20% of ~5.5M total distribution miles)2. Canada likely has on the order of 60–70,000 km of MV underground cable (given ~11% of ~572K km)4. These cables may be directly buried or in conduit. Common types include cross-linked polyethylene (XLPE) insulated cables in newer installations, and paper-insulated lead-covered (PILC) or older polyethylene in legacy systems. Many urban utilities also have extensive secondary network cables (low-voltage) in downtown grids – e.g. Con Edison (NY) alone has ~90,000 miles of underground cable if one counts all secondary, service, and primary cables41. This number illustrates that the aggregate length of underground conductors is enormous, when secondary/service cables are included (each building service adds length). Managing this vast network is a data challenge without digital tools.

4.2 Underground Distribution Transformers

These include pad-mounted transformers in suburban systems (usually in a green metal cabinet on a concrete pad) and subsurface transformers in vaults for urban areas. The U.S. has an estimated 60–80 million distribution transformers in service4243, of which a significant fraction are pad-mounted or below-grade units. While exact breakdown is not readily available, one industry source notes "most of the estimated 45+ million U.S. distribution transformers are overhead single-phase", implying perhaps 10–20 million are pad-mounted (underground type)4443. Each underground residential development has one pad-mount per a cluster of homes; urban network transformers each serve a city block or building. Canada likely has on the order of a few million distribution transformers (roughly proportional to population). For example, Manitoba Hydro (a mid-size utility) alone has ~75,000 km of distribution lines and tens of thousands of transformers. The key point is that millions of transformers are in underground or pad-mount configuration, which means monitoring their loading and condition (oil temperature, etc.) is often more challenging (no pole for a visual inspection). New IoT sensors are emerging to monitor these sealed transformers remotely.

4.3 Underground Switchgear and Protective Devices

These include underground switches, sectionalizers, and network protectors. Pad-mounted switchgear (in cabinets) is common for 3-phase underground feeder loops in subdivisions – allowing utilities to sectionalize faults. In urban networks, network protectors are installed on secondary network transformers; a city like New York has hundreds of network protectors that require monitoring. These devices often have SCADA or remote telemetry in modern systems (for example, network protectors report status over fiber). Fault indicators are another asset – on long underground feeders, utilities install faulted circuit indicators (FCIs) in manholes or pad-mounted enclosures to help locate faults by showing a tripped flag or sending a signal. Digital sensors are replacing simple FCIs, as discussed later. The quantity: large utilities may have thousands of underground switching points and FCIs deployed.

4.4 Manholes and Vaults

These are the access points to underground cables and equipment, especially in urban areas with "spiderweb" secondary networks. The number of manholes/vaults is enormous in major cities. For example, Con Edison's underground system in New York City has ~264,000 manholes and service boxes45. Each contains cable splices, joints, or transformers. Other cities (Chicago, Boston, etc.) have tens of thousands of utility manholes each. Even smaller cities with downtown underground districts will have hundreds to thousands. These structures present maintenance challenges (water pumping, stray voltage hazards, etc.) but also opportunities for sensor installation (one can place environmental sensors in a manhole, as Con Ed has done). From an analytics perspective, tracking asset data by location (manhole ID) is important. Modern GIS-based asset systems (like Schneider's ArcFM Conduit/Structure Manager) explicitly model these structures and what's inside them46. This is crucial for map-based visualization of cable health – you need to know which segment is between which vaults.

4.5 Cable Accessories

The smallest components — splices / joints, terminations, and elbows (in pad-mount gear) — are where many underground failures begin (a bad splice or terminator can overheat or allow moisture). A typical underground feeder might have dozens of splices. Utilities keep inventories of splice types and ages, since certain older splice designs are known to be problematic. Any AI solution should consider accessory data, though utilities historically have fewer records on them. Modern digital partial discharge sensors can sometimes detect discharge at specific splices, identifying a failing joint. Quantitatively, if there are ~1 million miles of underground cable in the US, there are likely tens of millions of splices (assuming an average cable segment length of a few hundred feet between splices).

Scale of the Problem Millions of miles of cable, tens of millions of associated devices, hundreds of thousands of access points. No utility can track the condition of an asset base this size by hand or by inspection. Whatever tells a utility where to act has to be built from data — and Part II examines the data utilities actually have.
Part II
The Data Layer

5 · Data Collection and Monitoring of Underground Infrastructure Today

Side-by-side: utility GIS view (today) vs Earthflow Underground twin overlay
Figure 5.1 · Today vs Tomorrow The left side shows what a utility sees today — a flat GIS map of cable routes. The right side overlays the Earthflow Physics AI risk score on the same network. The data layer is the binding constraint, not the cable itself.  ·  See it live →

Utilities traditionally rely on a combination of SCADA systems, periodic testing, and reactive monitoring to manage underground equipment. Each sees part of the picture. None of them sees the condition of the cable between one failure and the next.

5.1 SCADA and Remote Monitoring

For the most part, distribution-level SCADA coverage on underground feeders is limited. Unlike transmission lines which have extensive metering, most distribution cables do not have real-time sensors along their length. However, SCADA is present at substations feeding underground circuits (measuring feeder currents, breaker status, etc.), and at some key points in the network: e.g., pad-mounted reclosers or switches may be SCADA-controlled, and network protectors in urban grids often report status remotely. SCADA gives macro-level data – if a feeder breaker trips, the control center knows it and can see the current/voltage to the fault point. But SCADA usually cannot pinpoint where along an underground feeder the fault is, nor give granular condition data. Some utilities have added smart fault indicators that communicate back to SCADA or an AMI network. For instance, there are underground fault sensors (by SEL, Sentient, etc.) that can wirelessly report fault passage and even load data from inside a pad-mounted transformer or vault. These are typically integrated with Distribution Automation systems (DA or ADMS). As an example, Sentient Energy's UM3+ smart underground sensor can monitor up to 12 cable phases in a vault or switch cabinet, measuring load and fault waveforms and sending data via cellular or mesh network4748. Such sensors effectively extend SCADA visibility into the underground domain, capturing high-resolution current and voltage data at points along the feeder. Currently, deployment of these is growing but not ubiquitous – usually part of targeted reliability projects.

5.2 Outage Detection and Customer Calls

In many distribution systems, the first "sensor" that an underground cable failed is the customers. Lights go out, customers call in or (with smart meters) the AMI system flags a loss of power. An Outage Management System (OMS) then deduces the likely failed component. Smart meters have improved this process by pinging meters – if an entire segment of meters is dark, the utility infers which cable section is out. This is a reactive approach and doesn't give condition info prior to failure, but it is how many distribution outages are initially identified. Some utilities have also deployed "last gasp" devices: certain pad-mounted transformers or network protectors send an alert when they lose power, aiding in outage localization. Again, this is event-driven data, not continuous monitoring.

5.3 Periodic Field Inspections

Because you cannot see an underground cable directly, utilities use testing technologies. Common practice includes infrared (IR) thermography on accessible terminations (e.g. open a transformer cabinet and IR scan the cable elbow for hot spots), cable insulation testing (offline tests such as VLF tan-delta or offline partial discharge on de-energized cables), and online partial discharge surveys (attaching sensors to detect PD while cable is live). Some utilities have proactive programs: for example, testing a sample of older cables each year to assess health. There are also cable monitoring vans that can inject signals to measure impedance changes, etc. These tests are labor-intensive and typically run on multi-year cycles that reach only a fraction of the fleet. IMCORP, a specialty firm, has technology to test new and service-aged cables to find defects and verify integrity, and they report that only ~60% of new cable installations meet factory standards upon commissioning (meaning 40% have some defect that could shorten life)4950. Regular testing can find such issues and "reset the clock" by fixing them51. Still, many utilities struggle to allocate enough budget and outages to test a large percentage of their cables, so these programs often cover only the highest-risk segments.

5.4 Environmental and Condition Sensors

A newer development is the installation of IoT sensors in underground structures to continuously monitor conditions. A prime example is Con Edison's deployment of manhole sensors in New York. Con Ed has worked with a vendor (CNIGuard) to put multi-sensor IoT devices in manholes, measuring parameters like combustible gas levels, stray voltage, humidity, flooding, and temperature52. These sensors report via wireless links to a cloud analytics platform, which can alert the utility if, say, a manhole is heating up or filled with gas – conditions that precede manhole fires or explosions53. The system even includes IR cameras that can detect cable "hot spots" so crews know exactly where to intervene54. Con Ed began deploying these in 2017 and has been expanding coverage across its 266,000 manholes55. Similarly, some utilities install temperature and oil-level sensors on underground transformers (especially large network transformers) to catch overload or oil leaks. Distributed Temperature Sensing (DTS) is another technology: a fiber run alongside a high-voltage cable can detect temperature changes along its length, effectively turning the cable into a sensor for hotspots – this is more used in transmission cable tunnels but could be applied to distribution feeders in long ducts.

5.5 Fault Indicators and Line Sensors

Many underground systems make use of faulted circuit indicators (FCIs) that are either manually checked or communicate remotely. Traditional FCIs are simple devices that clamp on a cable and pop a flag or light when a fault current passes. Newer intelligent FCIs (line sensors) do much more – for example, SEL's underground sensors or Sentient's UM3+ can capture high-resolution current waveforms, detect disturbances, and time-stamp and geo-locate fault events4856. They can differentiate transient vs permanent faults and help estimate distance to fault by comparing waveforms at multiple sensors. These are typically installed at cable sectionalizing points (switches, transformer feed-thrus). They significantly speed up fault isolation – rather than send a crew to manually dig up multiple splice pits, the control center can see which sensor saw the fault and which did not, narrowing down the fault section immediately. Utilities pairing such sensors with analytics report 20%+ improvements in Customer Minutes Interrupted (CMI) and up to 65% reduction in patrol time to locate faults5758. The gains show up twice: in reliability metrics, and in the crew hours no longer spent patrolling for a fault.

5.6 Which System Serves Which Objective

A useful way to organise the systems above is by the question each one helps answer. Underground analytics resolves to two objectives – Condition (how degraded is this asset, how likely is it to fail) and Capacity (how much load can it carry, how much headroom remains) – and most utility systems serve one or both. Below, C marks a system that serves Capacity and H one that serves Condition (health).

Source systemWhat it holdsServes
GIS / asset registryGeometry, age, insulation type, size, splices, connectivityC H
OMSFault and outage history, cause codes, restoration timesH
DMS / ADMSNetwork model, real-time state, switching, power flowC
SCADA + historianFeeder currents and voltages, breaker status, oscillographyC H
AMI / MDMInterval load, voltage, last-gasp signals, meter–segment topologyC H
EAM (Maximo and similar)Maintenance records, work orders, inspections, criticalityH
Protection / relay + DFRRelay events, fault current, digital fault recordsH
Test / PdM repositoryTan-δ, partial discharge, thermography – frequently as PDFsH
DERMSDER location and output, reverse-flow contextC
Weather / environmentalAmbient conditions, soil moisture, storm history, soil chemistryC H

GIS anchors both objectives – without route geometry neither question can be answered per segment. Loading history from SCADA and AMI is the bridge between them: it sets today's Capacity, and accumulated over years it is one of the strongest drivers of Condition.

5.7 Gaps and Needs

Despite the above tools, there are still gaps in data collection for underground assets:

The data that does exist on the underground side is also of materially worse quality than what utilities are used to on the transmission side, and the volume arriving from modern AMI and sensor deployments compounds the problem. Chapter 6 examines both, because any solution that ignores them will fail in production.

6 · Data Quality & Volume Realities

The biggest gap between an idealized AI/ML solution and a deployable one is not the algorithm — it is the state of the data the algorithm has to consume. This chapter names the specific data-quality problems on the distribution and especially the underground distribution side, the volume problems that emerge when you fuse modern AMI and sensor streams, and the design choices a credible solution must make.

6.1 The Premise — Distribution Data Is Materially Worse Than Transmission Data

Transmission grew up with SCADA-grade telemetry on every line. Synchrophasors, digital relays, line monitors, and decades of disciplined record-keeping have produced data assets that are largely complete, consistent, and recently calibrated. The distribution grid evolved bottom-up from CIS billing systems and paper switching diagrams, and its data layer reflects that history: heterogeneous, partial, and frequently stale.

Underground specifically suffers more than overhead distribution for one structural reason — you cannot inspect what you cannot see. Overhead lines get visual pole patrols and increasingly drone surveys; those inspections continuously correct the asset record. Underground cable runs are last visited when they were installed (in some cases, 60+ years ago) and the record from that installation may or may not have survived. The ground-truth feedback loop that keeps overhead GIS clean is largely absent below grade.

6.2 Data-Quality Issues by Source System

The problems are specific to each source. Below is what utility analytics teams typically encounter when they try to build a fused dataset for underground cable analytics:

Source systemTypical data-quality issues
GIS / asset recordsAs-built records often diverge from actual field state. Cables installed differently than documented. Missing splice records. Wrong duct assignments. Pre-1990s records often paper or lossy-scanned PDFs. Cable vintage, manufacturer, conductor specs frequently unknown for older segments. Schema heterogeneity across vendors (Esri ArcFM, Schneider Conduit Manager, in-house geodatabases).
OMS · Outage ManagementCause-of-failure often recorded as "unknown" or "underground equipment" without distinguishing cable vs splice vs transformer. Fault location often imprecise — the feeder is known, the exact segment is not. Timestamps for failure, customer report, and restoration are mis-aligned. Customer-driven outage reports and sensor-driven reports differ in granularity and arrive at different latencies.
SCADAFor most utilities, SCADA visibility ends at the substation feeder breaker — no granularity down the feeder. Sampling rates vary from 1-second to 1-hour depending on equipment vintage. Tag and label conventions are inconsistent across substations because each was commissioned at a different time. Communications outages create silent gaps in the historian. CT/PT calibration drift on older gear distorts long-term trends.
AMI · smart meters"No power" and "no comms" events are easily conflated. Topology mapping (meter → transformer → segment) is often stale or wrong, especially after switching operations. Last-gasp signals do not fire reliably across the fleet. Volume is high (15-min or 5-min interval data × millions of meters) but per-meter accuracy varies with hardware vintage.
Field-test resultsTan-delta and partial-discharge test results are often stored as PDFs, scanned reports, or paper logs in a maintenance management system — not as structured time-series. Test dates span decades. Older results were interpreted against different acceptance criteria than current results. Many cables in the fleet have never been tested at all.
Splice & accessory recordsAmong the worst-documented assets in the underground network. Splice type, age, manufacturer, and exact location are often unknown for older cables. A single "cable" record in GIS frequently hides 10+ splices, each with its own failure mode — but the GIS record does not differentiate them.
Environmental · ground-truth dataPublic soil-corrosivity mapping is statewide-grained; localized variation along a single duct run can be much larger than the mapped value. Groundwater levels depend on local conditions that may not match published maps. National lightning-strike history is good but only extends back to ~1989. Flood-hazard maps update infrequently and lag actual flood-history changes.

Table 6.1 · Data-quality issues typical of each source system feeding an underground cable analytics platform.

6.3 The Volume Problem — What Happens When You Fuse Everything

Set the quality problem aside for a moment. Even if every source were clean, the volume of data that a fused analytics platform has to manage is substantial — and growing fast as utilities deploy more sensors:

The volume problem is not just storage. It is making the join coherent at the cable-segment level, with confidence metadata about which source contributed what to each output. A modern utility analytics stack has to be designed for this from day one, not retro-fitted.

6.4 The Underground-Specific Compounding Factors

Four factors compound the data problem specifically for underground — they are less severe on the overhead side:

6.5 What a Realistic Solution Must Do

The implication for any AI/ML solution for underground cable analytics: a solution designed for clean data will not survive contact with a real utility. One that works in production makes five design choices up front:

  1. Tolerate missing data. Industry-default imputation when utility-specific data is missing or stale, with explicit confidence intervals on outputs. A health-index score with a wide confidence band is more useful than a precise number that pretends data it doesn't have.
  2. Provenance per output. Every numeric result traces to which source contributed what — so regulators, engineers, and rate-case reviewers can audit the basis for any specific recommendation. This is non-negotiable when the output of the system drives capex decisions defensible to FERC and state PUCs.
  3. Disclose data-quality caveats in answers. When the analytics layer (and especially an agentic-AI Q&A layer) answers a question, it should surface the data-quality caveats. "Health Index score 4.2 for segment X is based on GIS install date and soil-chemistry data — no test results on file in the last 8 years" is a more useful answer than "4.2" alone.
  4. Volume-aware architecture. Cloud-scale processing for the AMI, sensor, and historian volumes described above. Batch + stream patterns where each is appropriate. Segment-level join indices that don't require shuffling billions of rows for every query.
  5. Schema-flexible ingest. Auto-mapping across the common utility GIS, OMS, and SCADA schemas (Esri ArcFM, Schneider Conduit Manager, generic geodatabases) with manual override for the long tail of in-house schemas. The Schema Auto-Mapper is the single most important architecture decision for rapid deployment.
Bridge to Chapter 11 These five design choices are not abstract — they are exactly the architecture choices Earthflow Underground made. Chapter 11 documents the Schema Auto-Mapper, the citeable provenance JSON, the Cirra agentic layer's data-quality disclosure, the cloud-scale processing model, and how the platform handles the realities described in this chapter.

Imperfect data does not change what a utility needs to measure. Chapter 7 sets out the KPIs that matter for underground cable, and which data each one depends on.

7 · Key KPIs for Underground Cable Health and Risk

Evaluating the health and remaining life of underground cables requires a combination of reliability metrics and asset condition indicators. Utilities and industry research have developed several Key Performance Indicators (KPIs) and indices for this purpose:

7.1 Failure Rate and History

The most basic metric is the failure rate of cables, often expressed as failures per 100 miles per year. Utilities track how often their underground cable segments fail (e.g., 1 fault per 20 miles per year would be a rate of 5 per 100 miles/year). They also look at historical failure count per cable or per circuit. A cable that has failed multiple times (with splices each time) is at higher risk for future failure. Past failure history is a strong indicator in health indexing – if a particular vintage of cable shows, say, 0.2 failures/mi-year vs another type at 0.05, that informs replacement prioritization. Some utilities incorporate a "cable fault history score" into their asset health index — for example, scoring a cable at the top of the scale after more than two failures in ten years.

7.2 Cable Age vs. Expected Life

Age (time in service) is a simple but important KPI. Many utilities assume a nominal design life of ~40 years for underground distribution cables60, though actual life can be much longer if conditions are good (there are cables 60–70 years old still in service). Age by itself doesn't equal poor health, but age combined with certain insulation types is telling. For example, early-generation XLPE cables from the 1970s often had issues and many failed in the 20–30 year range due to insulation deterioration. Likewise, paper-lead cables might last a long time but become riskier if the lead sheath is degraded. Utilities often create an Age Score (older = higher risk) as part of a health index. Age is an easy KPI since utilities have installation dates in their databases for most cables.

7.3 Insulation Type and Condition Tests

The type of cable and its condition indicators form another KPI set. For instance:

7.4 Load / Stress Indicators

The operating conditions of the cable serve as KPIs for remaining life:

7.5 Reliability Impact Metrics

At a higher level, utilities also consider how a cable's failure would impact the system – effectively a criticality KPI. For example:

7.6 Asset Health Index (Health Index)

Many utilities synthesize the above factors into an Underground Cable Health Index, often a score from 1 (new/good) to 5 (end-of-life poor), or similar. This Health Index is a composite KPI, weighing age, type, test results, failure history, etc. For instance, a possible scheme:

Per-segment Decision Matrix showing Health Index, RUL, and recommended action
Figure 7.1 · Decision Matrix Per-segment Health Index 1-5, RUL bands, and recommended action: REPLACE / REPAIR / RETIRE / DEFER / MAINTAIN. Each cell carries provenance JSON pointing back to the source datasets used.

7.7 Remaining Useful Life (RUL) Estimation

Some advanced analytics attempt to directly predict remaining life in years. This often piggybacks on the health index – essentially translating the HI and degradation trends into a probabilistic RUL. For example, if PD activity is trending up, an algorithm might predict failure likely in <2 years, giving an estimated RUL distribution. While exact RUL prediction is tricky, it's an ultimate KPI the industry is interested in: knowing whether a given cable likely has 5, 10, or 20 years left can inform capital planning.

7.8 Thermal Rating and Capacity Headroom

A KPI class that receives considerably less attention than condition – but which increasingly drives the same replacement and reinforcement decisions – is capacity: how much current a given segment can actually carry, and how much of that capability is already consumed. Utilities have historically managed this with static nameplate ampacity drawn from published tables. Those ratings assume a reference installation: a specified soil thermal resistivity, burial depth, spacing, and ambient temperature. Real segments deviate from the reference substantially. Two runs of identical cable, one direct-buried in well-drained sandy soil and one in a congested duct bank at a substation exit, differ materially in the current they can carry without exceeding conductor temperature limits.

The underlying physics is long settled. The Neher–McGrath method – and the IEEE 835 ampacity tabulations derived from it – treat the cable and its surroundings as a thermal circuit, in which soil thermal resistivity is typically the dominant term. What has changed is the availability of the inputs: soil thermal properties, burial configuration and route-level ambient conditions can now be resolved per segment from geospatial data rather than assumed uniform across a feeder.

One soil, one temperature, a whole territory. In practice, the reference assumptions behind a rating are rarely set segment by segment. They are set once — for a service territory, and often for a country. The IEEE 835 ampacity tables used widely in North America are built on an earth ambient of 25 °C, a burial depth of 36 in and soil thermal resistivity tabulated at 60, 90 and 120 °C·cm/W114. The national practices catalogued in IEC TR 60287-3-1 tell the same story: a nominal 0.9 K·m/W with seasonal ground temperatures of 15–25 °C in the United States; 0.9 K·m/W in Canada absent measurement, raised to 1.2 where strong seasonality is expected; and 1.2 K·m/W across the United Kingdom, Australia and New Zealand115.

The ground does not hold to a single number. IEEE 442 places moist graded sand at 0.2–0.4 K·m/W and clay at 0.95–1.6 or more, while uniform sand that dries out climbs towards 3.2 — a sixteen-fold spread around a value the rating treats as constant. Below a critical moisture content, around 4.5% for uniform sand, resistivity rises several-fold as the soil dries116. A territory-wide value is therefore wrong in both directions at once. It overstates what cable can carry in dry, heat-trapping ground, where the rating is least safe, and understates it in moist, well-conducting ground, where capacity sits unused. Resolving soil and ambient conditions segment by segment is what turns a planning rating into a capacity figure a planner can act on.

00.511.522.533.50.9 · US, Canada1.2 · UK, AU, NZGraded sand: 0.2–0.4 K·m/WGraded sand (moist)0.2–0.4Sand: 0.4–0.65 K·m/WSand (moist)0.4–0.65Clay-silt: 0.65–0.95 K·m/WClay-silt (moist)0.65–0.95Clay: 0.95–1.6+ K·m/WClay (moist)0.95–1.6+Dry uniform sand: rises toward 3.2 K·m/WUniform sand, dried outup to 3.2
Figure 7.2 · Soil thermal resistivity: what the ground does vs what the rating assumes (K·m/W) Bars show the range by soil family; vertical lines show the single nominal values national practice applies across a territory. Sources: soil families116, national practice115.

The resulting KPI is a capacity headroom figure – the measured or forecast loading expressed as a percentage of the segment's site-specific in-situ rating, with the remainder available for growth. Its operational value is distinct from condition. A planner evaluating a data-centre interconnection, a fleet-charging depot, a rooftop-solar cluster reversing flow, or a storm-time switching transfer is asking a capacity question, not a health question. A segment can be in excellent condition and still be the binding constraint on the feeder; conversely a heavily degraded segment may have ample headroom. Utilities that track only one of the two are, in effect, planning with half the picture.

In practice, utilities combine leading indicators (like PD, test results) with lagging indicators (failure history) to manage cable risk. The KPIs above decide which cables are replaced first, which are monitored, and how renewals are budgeted. The constraint is the data behind them. Chapter 8 surveys the sensor hardware that can supply it; Part III turns to the analytics that convert it into decisions.

8 · Examples of Sensor Hardware Solutions and Their Data

Six classes of sensor now gather data from underground electrical infrastructure. Each is described below with the data it produces.

8.1 Smart Underground Line Sensors (Fault and Load Monitors)

One example is the Sentient Energy UM3+ underground line monitor. This device can be installed in pad-mounted switchgear, junction boxes, or even submersible vaults (it has a compact, submersible form factor for manholes)74. The UM3+ clamps around cable conductors (up to 12 phases with one unit) and measures:

Data Provided: The utility receives real-time alerts (trip signals), numerical load data (currents/voltages if equipped for voltage), waveform files for events, and periodic status. With analytics, one utility reported these sensors led to 20%+ reduction in feeder outage minutes and 65% reduction in patrol time, as noted earlier79. They also achieved an 11% O&M cost reduction due to efficiency gains from sensors58. These numbers show how valuable the data from such sensors can be.

8.2 Manhole / Vault Monitors

Con Edison's deployment, described in Chapter 5, uses multi-sensor IoT devices from CNIGuard. Units of this kind installed in manholes typically measure:

Data Provided: These monitors typically send continuous telemetry. Con Ed's system transmits data to a dashboard and sends alerts if any parameter goes out of bounds8182. For example, if a cable in a manhole starts arcing, the sensor could detect rapid temperature rise and a spike in carbon monoxide, triggering an immediate alert to dispatch crews before a major fire or outage occurs5283. The data can be integrated into the analytics platform to show a map of manhole conditions. Con Ed's initial deployment has prevented incidents by indicating cables "in need of attention or repair" before they fail55. A value proposition from their experience: the IoT sensors act as a "virtual inspection team," reducing the need for physical inspections of hundreds of thousands of structures83.

8.3 Partial Discharge (PD) Monitoring Systems

For higher-voltage cables (and applicable to distribution too), there are systems like OMICRON MonCablo or similar by other vendors. These install coupling sensors (capacitive or inductive) usually at cable ends (terminations or switchgear) to continuously listen for PD pulses. They can cover many kilometers of cable by using time-of-flight of pulses to localize PD sources along the cable.

Data Provided: They output PD magnitude, pulse count, phase-resolved patterns (to distinguish between different defect types), and trending over time59. These systems often have software that can alarm if PD exceeds thresholds. Utilities typically deploy these on critical feeders or after a suspicious test result. For an analytics platform, confirmed PD presence and its location are among the strongest signals for action — for example, scheduling a section replacement at the detected site.

8.4 Distributed Temperature Sensing (DTS)

If an optical fiber is embedded in or alongside a cable, DTS equipment can send pulses of light and measure backscatter to determine temperature every meter along the fiber.

Data Provided: Essentially a temperature profile of the cable route in real time. This can identify localized heating (perhaps due to a bad splice joint or an overloaded segment in a duct section with poor heat dissipation). While more common in transmission cables, distribution could use it where fiber is available (some newer cables or duct installations include fiber conduits). Real-time thermal data would feed into dynamic rating calculations (increasing utilization safely) and alert if something is running hotter than normal.

8.5 Smart Cable Bolts / Termination Sensors

A smaller category: sensors embedded in cable connectors or terminations. An example: some companies offer wireless temperature sensors that attach to cable lugs or elbows to monitor connection temperature (E.g., Eaton or ABB have devices to monitor bolted connections in switchgear). These are useful because a loose connection in a pad-mounted transformer will get hot and eventually fail.

Data Provided: Point temperature readings of critical connection points. These often communicate via low-power radio to a local gateway. A spike in temperature at a particular elbow might indicate an incipient failure (or just an overload).

8.6 Advanced Fault Indicators (Stand-alone)

Companies like SEL and Horstmann (in Europe) provide underground fault indicators that can detect both short-circuit and earth-fault currents continuously84. Some models mount inside transformer cabinets and use current transformers around cables. Modern ones can communicate via cellular or radio.

Data Provided: Indication of a fault pass and sometimes load values. Not as rich as Sentient's full sensor, but if networked, they feed into the outage analysis.

8.7 The Economics of Instrumenting a Territory

Every sensor class above works. The constraint is arithmetic. U.S. utilities reporting to EIA operate more than 200,000 distribution circuits, and the largest individual utilities run between two and six thousand each135 (Figure 8.1). Each circuit divides further into many cable sections between splices, switches and transformers.

02,0004,0006,000ComEd: 5,639 distribution circuits (EIA-861, 2024)ComEd5,639SCE: 4,323 circuitsSouthern California Edison4,323FPL: 3,909 circuitsFlorida Power & Light3,909Oncor: 3,757 circuitsOncor3,757DTE: 3,518 circuitsDTE Electric3,518PG&E: 3,422 circuitsPG&E3,422PSE&G: 2,909 circuitsPSE&G2,909Georgia Power: 2,796 circuitsGeorgia Power2,796Con Edison: 2,218 circuitsCon Edison2,218
Figure 8.1 · Distribution circuits operated by the largest U.S. utilities All distribution circuits, overhead and underground, as reported for 2024. Source: EIA Form 861135.

Permanent monitoring scales with that count. Distributed temperature sensing needs optical fibre running alongside the cable. Where fibre went in with the cable, it is an option; for the direct-buried cable that makes up most residential underground plant, adding fibre means reopening the trench — at which point the work approaches the cost of replacing the cable. Permanent partial-discharge monitors sit at terminations and switchgear and see only the cable within their reach; line sensors and fault indicators sit at sectionalizing points. Each is a per-location purchase, installation and communications cost, multiplied across thousands of circuits — which is why, outside transmission cable and a few critical urban networks, permanent monitoring remains confined to pilots.

Even periodic testing adds up. CDFI’s consensus cost for offline diagnostic testing is $850–1,600 per segment from a service provider, or $400–650 when the utility tests in-house — and the price is per segment regardless of length117. Table 8.1 scales that to a territory.

Segments tested (one round)Utility in-house
($400–650 / segment)
Service provider
($850–1,600 / segment)
1,000$0.4–0.65M$0.85–1.6M
10,000$4.0–6.5M$8.5–16M
50,000$20–32.5M$42.5–80M

Table 8.1 · Cost of one round of offline diagnostic testing at territory scale. Arithmetic on the CDFI per-segment ranges117; segment counts are illustrative, and outage coordination and follow-up work are excluded.

None of this argues against sensors or testing. It argues for putting them where they change a decision. The CDFI itself concluded that diagnostics do not make sense in every situation117, and the same holds for permanent monitoring. A condition ranking built from data the utility already holds is what tells it which few hundred segments justify a test or a sensor — and which tens of thousands do not.

Sensor-Agnostic by Design Sensor hardware ranges from simple fault flags to multi-sensor IoT units. A comprehensive analytics solution should be sensor-agnostic, taking in data from any of these devices. And because no utility can instrument every segment, sensors earn their cost where they change decisions — on the highest-risk, highest-consequence segments — rather than everywhere at once.

Part III turns from hardware to the analytics layer: what it needs to ingest, and how it can be deployed quickly on data the utility already holds.

Part III
The AI/ML Solution

9 · Data, Sensors, and Analytics for an AI/ML-Driven Underground Cable Solution

Building a robust AI/ML-driven analytics solution for underground cable health requires integrating many data sources and sensor inputs, and presenting the results on a map that engineers and crews can act on. This chapter sets out the data types, sensors and analytics capabilities involved.

9.1 Asset Data (Inventory and Attributes)

The foundation is the utility's existing asset registry and GIS data for the underground system. This includes for each cable segment: installation date (age), length, route geometry (for mapping), voltage level, conductor size and material, insulation type (XLPE / PILC / etc.), manufacturer if known, and any construction notes (duct vs direct-buried, etc.). It also includes connectivity – which cable connects which equipment (from substation breaker down to loads). Using the GIS, the solution can map each cable segment to specific coordinates. Off-the-shelf, most utilities have this data in their GIS or asset database. A solution should readily ingest this "already available data"63 – as Siemens notes in its Cable Analytics offering, analyzing underground cables based on existing utility data (age, load, etc.) can reduce failures dramatically63. In practice, the solution must parse GIS line data — typically from an Esri geodatabase or similar — and assign a stable ID to each cable section.

9.2 Historical Performance Data

This includes failure/outage history for the cables. An ideal solution will interface with the utility's Outage Management System (OMS) or reliability database to pull records of past underground cable faults, including date, location (which cable or approximate), cause (if determined), and outage duration/customers affected. This is what lets a model learn which factors actually correlate with failure on this utility's network — a particular vintage in wet ground, say. Work-order history (repairs, splice installations) matters too: splices are common weak points, so the number of joints on a segment is worth carrying as a field wherever it is recorded.

9.3 Load and Electrical Data

Integration with SCADA or distribution management systems is needed to get electrical measurements on the cables/circuits. Minimum data would be feeder-level loading profiles (how much current the cable sees over time). If SCADA gives only substation feeder current, the solution can assume that load is distributed, or better, use power flow in the network model to estimate each segment's loading. Ideally, AMI (smart meter) data could also be used to estimate loading on each segment by summing the meters beyond it. This gives a more granular view of how heavily each cable is loaded at peak times. For circuits with switching, the system should know the current network configuration to assign loads correctly. Additionally, any voltage or power quality data from sensors could be input. Repeated voltage sags recorded by power-quality monitors, for instance, can point to an incipient fault. At a minimum, feeder peak currents by season and any overload events should be fed in.

9.4 Sensor and Condition Data

Where sensors exist, their data sharpens the picture considerably:

9.5 External Data (for Context)

External factors like weather and events can be relevant. For instance, the system might ingest severe weather data – if an area was flooded or experienced a lightning surge, that could stress cables. Soil maps or corrosion indices can also be input. At minimum, noting environment type (e.g. coastal salty area vs dry inland, or direct buried vs conduit) helps the ML model cluster cables by similar environment.

9.6 Analytics and AI/ML Needed

With these inputs, the solution performs six analytic functions:

9.7 Example of Solution Implementation

Siemens offers a Cable Analytics solution via its Advanta services, which emphasizes using advanced data analytics on existing utility data to target cable failures. They claim it can significantly reduce failures (by up to 10×) and defer capital by 10+ years63. The approach uses the kinds of data above to separate cables that can be treated to extend their life from those that need replacing. Hitachi's Lumada APM likewise ingests these data streams into its health and reliability models, providing a "single pane of glass" for asset condition7371. The combination of physics-based models (e.g. thermal aging calculations) and AI models (pattern recognition from big data) yields a comprehensive view.

The harder practical question is how quickly such a solution can be stood up. Chapter 10 shows that most of it can run on data the utility already holds.

10 · Rapid Implementation Using Off-the-Shelf Utility Data

Where to point the first pilot — the primary-main “getaway”. Value density is highest in the first ~300–1,000 ft of the three-phase main leaving the distribution substation. That run is simultaneously the highest-consequence segment on the feeder — it carries the summed load of everything downstream, so its failure is the largest single outage — and the most thermally stressed, carrying the highest current, frequently through a congested duct bank at the station exit. It is therefore the one location that demonstrates a Capacity result and a Condition result on the same asset. A workable first scope is 15 kV class (12.47 / 13.8 kV), including both direct-buried and duct/conduit runs where available so the derating difference can be quantified, on suburban or non-urban radial or open-loop topology, across two to three substations with two to three feeders each — a handful of circuit-miles. Large enough to back-test a ranking; small enough to assemble quickly.

An underground cable analytics solution does not have to wait for sensors. It can be deployed on the utility's existing data — cable age, material, voltage, outage history, load — and enriched as new sources arrive. The sequence below is how that works in practice.

  1. Leverage Existing GIS and Asset Databases: As a first step, ingest all readily available asset data on underground cables. Most utilities have GIS mapping of their network and attribute tables for equipment. By importing this, the solution immediately knows the inventory of cables and their key attributes (length, install date, type). This alone enables basic risk modeling – e.g., identifying all cables older than 40 years or all cables of a certain type that has known issues. Without any new sensor, industry failure curves for each cable type give a baseline failure probability for every segment from its age and type. Many utilities have done paper analyses of this; the software makes it systematic and map-based.
  2. Integrate Historical Outage Records: Pull in the utility's last five to ten years of outage records for underground equipment. This will show which segments failed and how often. From this, a model finds the utility's own patterns — a vintage or a district failing well above the fleet average. Off-the-shelf data here is just the utility's own outage logs. This can typically be exported from OMS or reliability reporting spreadsheets. The AI can be initially trained on this supervised dataset (features = cable attributes, label = failed or not within period). Even a simple model will often surface a high-risk cluster such as early-generation XLPE past thirty years of service.
  3. Use Existing Load Data: Without adding sensors, use substation SCADA and AMI meter data that the utility already collects. Many utilities have hourly load data per feeder and interval meter data per customer. By correlating those, the solution can estimate loading on each cable section (especially for radial systems). Where detailed allocation is not yet possible, feeder peak load as a percentage of feeder capacity is a workable proxy — cables on heavily loaded feeders get a higher stress score. Off-the-shelf, utilities have this from their PI historian or similar. Incorporating it requires some data mapping but no new hardware.
  4. Apply Industry Models and Benchmarks: In absence of utility-specific condition tests, the solution can incorporate industry failure models (like IEEE or CIGRÉ guides on cable failure rates by age) to predict remaining life. It can also use known problem lists – e.g., if the utility can list "we have 50 miles of unjacketed PILC cable from 1950s" – industry experience flags them for lead-sheath corrosion, and they can be treated as high-risk until evidence says otherwise.
  5. Quick GIS-Based Visualization: Using the imported data, the solution can very quickly produce a map of the service territory highlighting the oldest or most failure-prone cable segments. This addresses one of the biggest challenges: just knowing where all the old cables are. Many utilities have this info but not visualized intuitively. A map-based risk heat map can be generated in weeks by combining age and past outages. For example, circuits could be color-coded by "cable health index" derived from age+failures. This immediate visualization often provides insights (e.g., one neighborhood might stand out as all red = needs attention).
  6. Prioritization & "Low-Hanging Fruit": With these analyses, the utility can take initial actions even before deploying new sensors. Proactive replacement of the worst few segments — often ones that have already failed more than once — is the obvious first step. Where capital is constrained, cable rejuvenation (silicone fluid injection) on eligible aged cable is the main alternative; Chapter 14 sets out how the inject-or-replace decision is made.
  7. Iterative Improvement with New Data: Once the system is live with basic data, the utility can start feeding new sources as they become available. For example:
    • After a season of smart meter outage data, the system might get more granular fault location info.
    • If the utility does a round of offline tests on sample cables (say tan-delta tests on 20 cables this year), those results can be input and the model refined (and similar cables without tests inferred about).
    • Over time, deploy a few pilot sensors on the worst feeders. The data from those pilots can be used to calibrate the model. If the sensors catch a run of momentary faults ahead of a failure, that signature can then be sought in feeder SCADA records elsewhere.
    • In essence, you start with data at hand, get actionable output in a matter of months, and then continuously enrich the model with better data (sensors, test results, etc.) to improve accuracy.

10.1 Rapid Deployment Feasibility

Most utilities already hold the core data needed to start:

All these are "off-the-shelf" internal data. The implementation mainly involves data extraction, cleansing, and loading into the analytics platform. Modern utility analytics tools often come with adapters for common systems (GIS, OMS, etc.), making integration faster. With a cloud platform, that can be as simple as CSV exports or an API connection. Within a few weeks the data model can be populated and initial risk scores generated.

Siemens explicitly notes their approach uses "already available data – using advanced analytics and ML"63 to reduce failures by up to 10x. The point is broader than any one vendor: significant insight is available from existing data before a single sensor is added.

Rapid deployment means existing data plus proven methods. Sensor integration runs in parallel as a longer-term enhancement. Phasing it this way delivers value early and builds the evidence that justifies sensors where they are needed.

Day 1 — Data Drop
Day 1
Utility drops GIS + OMS data into the analytics platform.
Day 2-3 — Schema Mapping
Day 2 - 3
Auto-mapper aligns utility schema; analyst validates.
Day 4-5 — Models Run
Day 4 - 5
The Physics AI modules run; Health Index, remaining-life band, soil and flood scored.
Day 6-7 — Live 3D Twin
Day 6 - 7
Asset Management, T&D and CIO view per-segment Health Index in 3D.

Figure 10.1 · The onboarding pattern from data drop to live 3D twin. The program target tracked through a pilot is under 14 days; the day-by-day sequence above is the pattern, not a guarantee. Reference implementation: Orbyfy Earthflow Underground. See it live: earthflow.orbyfy.com/earthflow-underground

Chapters 11 and 12 describe one concrete implementation of these principles; Part IV then turns to the business case.

11 · Earthflow Underground — Reference Architecture

Chapters 1–9 describe what an underground distribution cable analytics solution should look like in principle. This chapter describes one such design: Orbyfy Earthflow Underground, a virtual-sensor physics engine for buried medium-voltage cable. It is presented as a reference architecture — one concrete instantiation of the abstract principles in this paper — not as the only possible design.

11.1 Two-Layer Module Stack: 11 + 19

Earthflow Underground combines 11 cable-specific Physics AI modules with a 19-module platform engine — 30 modules in total. The two layers run together: the Underground modules consume the platform's environmental outputs, so cable scoring is grounded in the same authoritative environmental intelligence used across Orbyfy's broader product line.

The table below states what each module answers and what it returns. What matters to a utility evaluating the approach is the question each module settles, the form of the answer, and where its limits lie.

#Underground ModuleWhat it answersWhat it returns
1AHI CompositeHow degraded is this segment relative to the rest of the fleet?1–5 Health Index per segment
2RUL SurvivalHow much service life is likely left?Remaining-life band, never a single year
3PD Analytics (field sensing)Is partial discharge present, where, and of what class?Presence, location and class — magnitude is deliberately not trended
4Fault Triangulation (field sensing)Where along the run did the fault occur?Narrowed search length along the segment
5Soil CorrosionHow aggressive is the ground around this cable, relative to others?Relative corrosion-driver ranking — never predicted metal loss
6Flood HydrologyHow exposed is this segment to inundation and standing water?Flood-exposure score per segment, splice and vault
7Manhole Risk (field sensing)Which structures carry elevated gas or thermal risk?Structure-level risk ranking
8Capex DeferralWhich deferrals buy the most reliability per dollar?Ranked deferral candidates with reliability trade-off
9AMI Dark-SectorWhich part of the network went dark, and where does the evidence point?Narrowed candidate section
10Dual-Basis Thermal RatingHow much current can this segment actually carry, in the ground it sits in?Two ratings — a planning basis comparable to nameplate, and an environmentally-adjusted in-situ basis
11Loading & HeadroomHow much of that capacity is already used?% capacity used · % headroom remaining

Table 11.1 · The 11 Underground-specific Physics AI modules, organized by what they serve — Condition (1, 2, 3, 5, 6, 7), Capacity (10, 11) and Operations & Planning (4, 8, 9). Eight of the eleven run in the first week with no new hardware; the three marked field sensing genuinely require instrumentation and arrive when that data does. Every output is per-cable-segment and carries its uncertainty and provenance.

11.2 The Platform Engine Underneath

Beneath the cable-specific layer sits a 19-module platform engine that turns public environmental science into per-location physics. It covers terrain and hydrology, soil state and chemistry, thermal and moisture regimes, weather and storm climatology, and hazard physics — resolved to the individual cable segment rather than to the service territory.

This paper describes those inputs by the property they measure rather than by dataset name, in the same way the FAQ table in Chapter 18 does. What the utility needs to know is that every input is an authoritative public source, that roughly 1,100+ environmental fields resolve to each location, and that every one of them is traceable in the provenance record attached to any score.

Beneath the eleven modules sits a physics layer that models the five ways buried cable degrades — heat, aggressive soil, water, electrical stress and ground movement — on top of an age baseline. Each driver is computed from published cable-engineering standards, expressed as a relative exposure for every segment, and combined into the Health Index, remaining-life band and Capacity Index.

Your data · per utility
GIS · OMS · SCADA · AMI · test records
Single-tenant. Never shared.
Earth's data · per geography
Soil · moisture · flood · climate · lightning
Pre-built national layer, sampled at each cable's coordinates.
↓   ↓
The Earthflow Underground Engine
11 Physics AI modules
Physics & standards
Published cable-engineering standards, the same everywhere.
Calibrated to you
Tuned to your own failure record.
Uncertainty & provenance
Every number leaves with a confidence band and its source record.
↓
Health Index 1–5+ remaining-life band
Capacity Index% used · % headroom
Ranked action queuereplace · test · defer
Re-scored monthly

Figure 11.0 · Architecture at a glance The utility's own records and the pre-built environmental layer meet in one engine, which returns two indices and a ranked action queue for every cable segment.

11.3 Onboarding and the Schema Auto-Mapper

The most important architectural choice in Earthflow Underground is the Schema Auto-Mapper. Most utility analytics deployments require multi-month integration projects to align the vendor's data model with the utility's GIS, OMS, SCADA, and AMI schemas. The Schema Auto-Mapper detects field-name patterns automatically across common utility geodatabase conventions and produces a candidate mapping in hours rather than months. An analyst validates that mapping early in onboarding, after which the Physics AI modules can run.

The program target tracked through a pilot is under 14 days from data drop to a live twin. It is a target the pilot measures and reports.

11.4 Update Cadence — Monthly, By Design

Once the twin is live it stays current, at the pace buried cable actually changes. Scores and the ranked queue are re-published monthly. That cadence is set by two things: the physics, because soil filters anything faster than weekly into noise at burial depth, and the utility, because crews, capital and outage windows are monthly-to-quarterly decisions. A faster refresh would add noise, not information.

11.5 Cirra — The Agentic Layer

Cirra is the natural-language orchestration layer that sits on top of the full module stack. A T&D Asset Management VP can ask Cirra: "Which cables are most exposed going into storm season?" — and Cirra composes a query across the relevant condition and environmental modules, returns a ranked list of segments with citations back to the underlying source records, and writes a model card explaining the reasoning. This converts a multi-week analyst engagement into a short answer that can be checked.

11.6 Citeable Provenance — Defensible Before FERC and the PUC

Every numeric output Earthflow produces carries a provenance record citing the authoritative source, the method used, and the data vintage, alongside the value itself and its uncertainty. When a utility's Regulatory Affairs team takes a cable-replacement recommendation into a rate-case prudency review, the model card is auditable back to the input record. This is the foundational design choice of the approach: a regulator cannot audit a proprietary score, but can audit a citeable physics result.

11.7 Living 3D Twin

Earthflow3D — living 3D twin of underground network
Figure 11.1 · Earthflow3D · Living Twin Per-segment Health Index is rendered in 3D over the network. Asset Management, T&D Engineering and Operations each see a tailored view derived from the same underlying model.  ·  See it live →

The portfolio resolves to two headline indices per cable segment. The Condition modules roll into a Health Index — a 1–5 score answering how degraded the segment is relative to its fleet. The Capacity modules produce a Capacity Index — measured loading against the segment's environmentally-adjusted in-situ rating, expressed as percent used and percent headroom remaining. The Operations and Planning modules consume both to inform restoration and capital plans. Every value carries its uncertainty and a provenance record identifying which inputs were measured and which were defaulted.

The visualization layer is a 3D twin tied to the GIS network. Per-segment Health Index is colour-graded, structure-level status is surfaced where sensing exists, and clicking a segment opens the full record including provenance. The same twin produces tailored views for Asset Management, T&D Engineering, Operations, Finance and Reliability — one network, many decision-ready surfaces.

11.8 Zero New Sensors for v1

Earthflow Underground requires no new sensors for a v1 deployment. The twin ships from the utility's existing GIS, OMS, SCADA and AMI fused with the authoritative environmental layer. Field sensing — line monitors, structure monitors, partial-discharge systems — is a Phase 2 enhancement the platform is designed to ingest as those sources come online, and it is what unlocks the three modules marked field sensing above.

Earthflow Underground is read-only decision support. It takes no live SCADA connection, no customer personal data, no write access to any utility system, and holds no control function.

Architecture Summary 11 Underground modules + 19 platform modules = 30, plus the Cirra agentic layer, the Schema Auto-Mapper and citeable provenance on every output — running on existing utility data with zero new sensors for v1, re-scored monthly, with eight of the eleven Underground modules live in the first week. Field sensing integrates as Phase 2 when the utility chooses to add it. See it live →

12 · Reference Deployment Pattern

This chapter outlines the deployment pattern for an Earthflow Underground engagement, illustrated on a synthetic demonstration feeder modeled on a coastal Florida service area. It is a worked illustration of the pattern — representative geography, representative soil and flood profile, modeled network — and not a record of a utility deployment. No utility's own operational records have yet been scored through this engine.

12.1 Demonstration Profile

12.2 Engagement Timeline

Week 1-4 — Onboard and score
Week 1 - 4
Onboard and score every segment for Health Index and Capacity Index.
Week 5-8 — Back-test
Week 5 - 8
Back-test the ranking against the utility's own multi-year outage history.
Week 9-12 — Validate
Week 9 - 12
Optional: validate against a partial-discharge or structure-monitoring pilot.
Week 13-16 — Review and executive briefing
Week 13 - 16
Results review, written findings, executive briefing.

Figure 12.1 · The sixteen-week engagement pattern. The twin goes live early; the remaining weeks are spent grading it against the utility's own record.

12.3 What a Pilot Sets Out to Measure

These are the criteria a pilot is designed to test. They are program targets, stated in advance so the result can be judged against them:

A null result is a valid outcome If the ranking does not beat the utility's own outage-history baseline, the pilot reports that as plainly as a positive result. A method that cannot fail its own validation is not a method. The right baseline is not random and not age — it is selection by outage history, and a well-run dielectric testing program achieves roughly 2–5× failure-rate discrimination. Matching the floor of that range is table stakes, not success.

12.4 Fleet Health Dashboard — the Prudency View

Fleet Health dashboard — portfolio Health Index rollup, material mix, age distribution
Figure 12.2 · Fleet Health · Rate-Case View Portfolio Health Index rollup, paper-insulated-cable exposure, material mix, age distribution and program-aligned benchmarks. The view a Regulatory Affairs lead takes to a state commission prudency review.

Each cell of the dashboard ties back to a provenance record pointing at the authoritative source used. A state commission asking "how did you decide to defer this cable-replacement project?" gets an auditable answer rather than a proprietary score.

Part IV
Business Case

13 · Value Proposition and Benefits

The value of underground cable analytics comes from one thing: acting on the right segment before it fails. Every benefit below follows from that — fewer outages, faster restoration, capital spent where it buys the most reliability, and a record that stands up in front of a regulator.

0%10%20%30%40%50%60%70%Fault-patrol time reduced up to 65% with intelligent line monitoringFault-patrol time−65%CMI reduced 20%+ on instrumented feedersCustomer minutes interrupted−20%+Reactive maintenance reduced 20% (GE Vernova APM customer)Reactive maintenance−20%Distribution O&M cost reduced 11% after line sensing + analyticsDistribution O&M cost−11%
Figure 13.1 · Published reductions from sensing and analytics programs Reported by utilities and vendors for their own programs; these are industry results, not Earthflow results. Sources: patrol time and O&M58, CMI79, reactive maintenance102.

13.1 Improve Reliability and Reduce Outages

A cable replaced or treated before it fails is an outage that never happens. Predictive analytics on line-sensor data have been shown to cut feeder outage minutes by more than 20%79. Because underground failures are rare but long, each one avoided moves SAIDI and SAIFI disproportionately — and removes an emergency repair from the crew schedule.

13.2 Faster Restoration When Failures Do Occur

Not every failure can be prevented. When one occurs, knowing where to dig is most of the restoration time. Utilities using intelligent line monitoring report 50–65% reductions in fault-location patrol time58, which shortens outages (lower CAIDI) and replaces exploratory excavation with a targeted repair.

13.3 Cost Savings in Operations and Maintenance

Knowing the condition of each cable lets a utility move from time-based to condition-based maintenance, spending crew hours where the risk is. One large utility reported an 11% O&M cost reduction after deploying line sensors and analytics, through more efficient patrols and fewer reactive repairs58. Underground lines already avoid most routine maintenance — 75–80% less than overhead28 — so the remaining savings sit almost entirely in the emergency repairs that analytics are designed to prevent.

13.4 Extend Asset Life and Defer Capital Expenditures

Age-based replacement retires good cable early and bad cable late. Condition data separates the two: cable in sound condition stays in service, and capital goes to the segments genuinely near end of life. Siemens reports deferrals of capital replacement of up to 10 years for cables managed this way63. For a replacement program measured in hundreds of millions of dollars, moving even a fraction of that spend to the right segments is the largest financial lever in this paper.

13.5 Enhanced Safety and Risk Mitigation

Underground failures carry a safety risk overhead failures do not: confined spaces, energized structures and manhole fires. Monitoring gas, temperature and stray voltage gives warning before conditions turn dangerous — Con Edison reports that its manhole sensors identify cables in need of attention or repair before they fail55. Fewer run-to-failure events also means fewer crews working in hazardous conditions under time pressure.

13.6 Wildfire and Storm Resilience

Much of the cable now going into the ground is there for resilience — to remove ignition sources in wildfire country and to keep laterals energized through hurricanes. That investment only pays off if the cable holds. Knowing which buried laterals are most exposed to flooding, surge and heat before storm season lets a utility reinforce them while the overhead system is still standing.

13.7 Data-Driven Planning and Investment

Regulators increasingly ask utilities to show why a given cable, and not another, is being replaced. A per-segment condition and risk record answers that question directly, and it does the same for undergrounding proposals by quantifying the reliability improvement a project buys. This is the same shift toward asset performance management already under way across the industry7371; Chapter 3 sets out what regulators expect to see.

13.8 Customer Satisfaction and Regulatory Incentives

Many utilities now operate under performance-based rates or scorecards in which reliability indices and customer satisfaction affect earnings directly. Fewer and shorter underground outages improve those metrics, and with them the utility's position under incentive and penalty mechanisms.

13.9 Ease of Integration and Rapid Benefits

Because the analytics run on data the utility already holds, there is no infrastructure program to fund before results appear. The work can start on a few circuits, prove itself against the utility's own outage history, and scale — a modular analytics layer rather than an IT overhaul, with actionable output in months rather than years.

Taken together, the case is simple: a more reliable, safer underground network for the same or less capital, with every decision traceable to evidence. Chapter 14 shows how a utility puts numbers on that case from its own books.

14 · Estimated ROI of an Underground Cable Analytics Solution

The return on underground cable analytics arrives through three channels — avoided outages, lower reactive O&M, and better-targeted capital — and a fourth that is specific to cable: choosing correctly between rejuvenating a cable and replacing it. This chapter sets out what the published evidence says about each, then gives the framework a utility uses to compute its own figure.

14.1 Avoided Outage Costs (Value of Reliability Improvement)

Every failure prevented avoids both a repair and an interruption, and the interruption is usually the larger cost. Published estimates of the economic cost of outages run as high as $110,000 per outage for a medium-to-large commercial customer28. Regulators have put figures on the same thing: the justification for Dominion's Strategic Underground Program valued the loss to Virginia's GDP at $35,458 per customer outage hour, and found $1.76 in avoided GDP loss per $1 spent17. On instrumented feeders, analytics-driven programs have reduced customer minutes interrupted by more than 20%79. Analytics does not move reliability the way physically undergrounding a line does, but its cost is a small fraction of a construction program’s.

$0.00$0.50$1.00$1.50$2.00Dominion SUP: $1 spentProgram cost$1.00Dominion SUP: $1.76 avoided GDP loss per $1 spentAvoided GDP loss$1.76
Figure 14.1 · Return per dollar in a filed undergrounding justification Dominion’s Strategic Underground Program justification: avoided loss to Virginia’s GDP per dollar spent. Source17.

14.2 Reduced Reactive Repair and O&M Costs

An unplanned underground failure is the most expensive way to replace a piece of cable. It brings emergency crew labour, often on overtime; excavation and surface restoration where the cable is not in accessible conduit; materials; and, occasionally, collateral damage to neighbouring equipment. Each failure prevented removes all of these at once.

Analytics also reduces the cost of the failures that still occur. One utility reported a 65% reduction in fault-patrol time with line sensing, and an 11% reduction in distribution O&M cost overall58. Continuous monitoring of underground structures can stand in for a share of routine physical inspections — Con Edison describes its manhole sensors as a "virtual inspection team"83.

14.3 Capital Deferral and Life Extension

Capital is where the largest sums sit. Published figures for underground cable work span roughly $200,000 to $1 million per mile depending on context8586, and Siemens reports deferrals of replacement of up to ten years for cables managed on condition63. Deferral is only half the value. The other half is redirection: capital that would have replaced healthy cable on an age schedule goes instead to the failing segment the schedule would have missed.

14.4 Inject or Replace — The Rejuvenation Decision

For aged polyethylene-insulated cable, replacement is not the only capital option. Rejuvenation — injecting a silicone-based fluid into the stranded conductor to restore insulation degraded by water treeing — applies to polyethylene-based insulation such as XLPE117. The economic case is the price gap. CDFI’s consensus range is $4–13 per foot to rejuvenate a single-phase run against $15–25 per foot to replace it117 (Figure 14.2).

$0$5$10$15$20$25$30Rejuvenate a single-phase cable run: $4–13 per ft (CDFI consensus)Rejuvenate (inject)$4–13 / ftInstall a new single-phase run of cable: $15–25 per ft (CDFI consensus)Replace (new run)$15–25 / ft
Figure 14.2 · Cost per foot: rejuvenate versus replace (single-phase run) Consensus ranges from the DOE-funded Cable Diagnostic Focused Initiative. Source117.

Utilities report similar ratios. Mountain View Electric Association in Colorado found injection "about 40% less than cable replacement"133; Veridian Connections in Ontario put it at "one-third the cost" and reported that failure rates on treated cable "plummeted from multiple failures every year to near zero"134. Suppliers offer warranties of up to 40 years on current fluids; those terms are vendor-reported133134.

The decision is made segment by segment, and it turns on condition evidence most utilities hold only in part:

EvidencePoints toward rejuvenationPoints toward replacement
InsulationPolyethylene-based (XLPE, HMWPE) with water-tree ageing117Insulation types the fluid does not treat, such as paper-insulated cable
Concentric neutralNeutral largely intact50% or fewer strands intact — the typical replacement threshold used by utilities and rejuvenation providers117
SplicesFew splices, passable to fluidToo many blocked splices for injection to be cost-effective133
Failure historyFailures concentrated in an otherwise eligible area — Veridian targeted its highest-failure subdivision first134Repeated failures on cable that also fails the neutral or splice checks
CapacityLoad stays within the cable’s rating over the planning horizonLoad growth will exceed the rating — replacement can upsize the conductor; rejuvenation cannot

Table 14.1 · The inject-or-replace decision, by evidence type.

The field checks already run at scale: IEEE Std 1617-2022 reports time-domain reflectometry tests on nearly 300,000 cable segments — more than 30 million metres — carried out as a pre-qualification check for rejuvenation programs132. What utilities lack is a way to decide where to look first. A condition ranking narrows the fleet to the segments where the inject-or-replace question is live; the capacity view settles the cases where load growth makes upsizing the better answer; and the neutral and splice checks make the final call.

14.5 How to Run the ROI Calculation

This paper does not publish a worked ROI example, because the inputs that decide the answer are specific to each utility. What follows is the framework. Every term should be filled from the utility's own books.

TermWhere the utility gets it
Cost of the analytics programmeVendor quote plus internal integration effort, amortised over the contract term
Avoided failure costUnderground failures expected per year × the utility's own average emergency repair cost, including excavation, restoration and overtime
Avoided customer-interruption costCustomer-minutes avoided × the utility's own value of customer reliability. Published figures differ widely; one filed analysis valued the GDP loss at $35,458 per customer outage hour17
O&M labour savingsPatrol and inspection hours redirected by a ranked queue, at loaded labour cost
Deferred and redirected capitalCable miles whose replacement is deferred, or rejuvenated instead of replaced, × the utility's own cost per mile, amortised over the deferral period
Risk-adjusted upsideThe tail events — a downtown network failure, a hospital outage — which dominate the distribution of outcomes and are usually left out of a base case

Table 14.2 · The ROI framework. A business case built from these six rows, filled from the utility's own records, is one it can defend.

Two structural observations hold regardless of the inputs. First, the benefit is concentrated: a small share of segments carries a disproportionate share of failures, so the value of the programme rests on whether the ranking is right, not on how many segments are scored. Second, the asymmetry favours action — the cost of the analytics layer is bounded and known up front, while the cost of the failures it is meant to prevent is unbounded and arrives at the worst possible moment.

14.6 Real-world Competitive Context

The major platform vendors make the same economic argument for their own products. GE Vernova cites 10–40% reductions in reactive maintenance from its APM software87; Hitachi Energy positions Lumada APM around optimising maintenance spend and extending asset life8889. The convergence matters more than any single figure: every major grid-technology vendor now treats condition-based asset analytics as a source of measurable return.

14.7 ROI Beyond Dollars — Strategic ROI

The return depends on the utility's own numbers, but its shape does not: the costs of the analytics are small and known, and the costs it targets — failures, emergency repairs, misdirected capital — are large and recurring. Chapter 15 looks at how the major vendors approach the problem.

Part V
Competitive Landscape

15 · Where the Incumbents Stop — Siemens, Hitachi Energy, GE Vernova, Schneider Electric

Four major vendors address grid asset analytics, each from a different starting point. This chapter sets out what Siemens, Hitachi Energy, GE Vernova and Schneider Electric offer for underground distribution cable, based on their own published material — and where each stops short of a solution that is cable-specific, rapidly deployable, multi-hazard and defensible before a regulator.

15.1 Siemens

Siemens is the only one of the four to market a product aimed specifically at underground cable. Its Cable Analytics solution, delivered through Siemens Advanta, applies data analytics and machine learning to readily available utility data — cable attributes, loading and failure history — to identify at-risk cables and prioritise maintenance63. Siemens states that the approach can "reduce cable failures by up to a factor of 10" and defer replacement by up to 10 years63.

Around the analytics, Siemens supplies field hardware and automation: SICAM fault indicators that detect earth faults in MV cable networks90, and distribution automation and FLISR systems that isolate faults. Its delivery model pairs software with advisory services, and it has argued publicly for the 50%-underground-by-2040 goal, calling for "underground-line-sensing technology married with artificial intelligence"91.

Where Siemens Stops Cable Analytics is a narrow product within a broad portfolio, delivered with consulting hours to interpret its outputs. Environmental drivers — soil, flood, lightning — are not fused into the scoring at segment level. There is no natural-language query layer and no rapid onboarding path from raw utility GIS to a live twin.

15.2 Hitachi Energy (formerly ABB)

Hitachi Energy approaches the problem through Lumada Asset Performance Management (APM), a platform giving a unified view of asset health, reliability and investment planning9273. Lumada APM uses AI-driven prognostics and statistical models to predict failures and recommend proactive maintenance, and supports scenario analysis of when to intervene7193.

Its distinguishing strength is integration with work management: Hitachi offers APM alongside enterprise asset management (EAM) and field service management96, so a flagged asset can flow directly into an inspection work order. It also carries a hardware and monitoring pedigree from ABB, including digital-substation and fibre-sensing work9495. Hitachi does not, however, market a product specific to underground cable; cables are one asset class among many in a general APM.

Where Hitachi Energy Stops Lumada APM is not cable-specific. To run an underground program, a utility must shape a generic APM around cable data itself. There is no cable module library, no automated schema mapping and no rapid deployment path. Strong on EAM and work-management integration; weak on physics-grounded, multi-hazard scoring at the cable-segment level.

15.3 GE Vernova (General Electric)

GE Vernova's offering is its Asset Performance Management software — the former Meridium platform — recognised by industry analysts99 and positioned to deliver "increased reliability, decreased maintenance costs, [and] lowered risk" across energy assets100. Utilities including Ameren have used it to bring condition data from multiple sources together for T&D assets101, and GE cites outcomes such as a 20% reduction in reactive maintenance102.

GE's advantage is breadth: APM sits within the GridOS portfolio alongside ADMS and DER management, so a utility already running GE's operational software can add asset analytics within a familiar stack. APM's roots, though, are in generation and transmission. For underground distribution cable it is a configurable framework rather than a dedicated product.

Where GE Vernova Stops An integrated platform, but cable analytics must be configured into a generic APM — not a turnkey underground product. There is no natural-language layer that turns an asset manager's question into an auditable answer, and provenance sits at the asset level rather than at the level of each calculation.

15.4 Schneider Electric

Schneider Electric starts from the data layer. It owns ArcFM, one of the most widely used utility GIS platforms, and Conduit Manager, which records which cable sits in which duct and where manholes and splices are located3810346104. No underground analytics program can work without that record, which makes Schneider's position unusually strong at the foundation.

On top of it, Schneider's EcoStruxure Asset Advisor services apply AI and machine learning, remote monitoring and predictive analytics to electrical assets105106, using an Asset Health Index and Maintenance Index to track condition107108, and include partial-discharge monitoring for medium-voltage switchgear109. Asset Advisor began with critical facility assets such as UPS systems110 and has since extended to transformers111, supported by a 24/7 remote-monitoring service112113. Schneider does not market an analytics product specific to underground cable.

Where Schneider Electric Stops Strongest on the data-management layer (ArcFM, Conduit Manager). The analytics are general-purpose EcoStruxure services, not cable-specific physics-based scoring. There is no rapid deployment path, no natural-language layer and no calculation-level provenance a regulator can audit.

15.5 The Shared Gap — What None of the Incumbents Deliver

Each vendor comes at the problem from its own strength: Siemens from cable-specific analytics and field devices, Hitachi Energy from enterprise APM and work management, GE Vernova from an integrated operations software suite, Schneider from the GIS record of the network itself. A utility evaluating them will weigh fit with its existing systems, evidence behind each vendor's claims, and the commercial model — software subscription or consulting engagement.

All four recognise underground cable analytics as a real need, and that competition is good for utilities.

But the shared gap remains. Across all four incumbent platforms, none deliver all six of the following in a single product:

CapabilitySiemensHitachi EnergyGE VernovaSchneiderEarthflow Underground
Rapid deploy from utility GIS / OMS / SCADA / AMI ✗ ✗ ✗ ✗ ✓
Zero new sensors required for v1 ~ ~ ~ ~ ✓
Multi-peril Physics AI fusion, cable-specific ✗ ✗ ✗ ✗ ✓
FERC / PUC defensible citeable provenance per output ✗ ✗ ✗ ✗ ✓
Natural-language query layer with cited answers ✗ ✗ ✗ ✗ ✓
70+ authoritative environmental datasets fused with utility data ✗ ✗ ✗ ✗ ✓

Table 15.1 · Capability matrix · The combination only Earthflow Underground offers in a single product. ✓ = full capability · ~ = partial / requires extensive customization · ✗ = not in product.

This is the gap Earthflow Underground was built to close. Chapter 16 describes how.

16 · Where Physics AI + Agentic AI Begin — Earthflow's Differentiator

The incumbents' tools are sound; they were built for other jobs. Producing a defensible, rapidly deployable, multi-hazard picture of underground cable is a specific job, and it calls for specific design choices. This chapter describes the ones that distinguish Earthflow Underground.

16.1 What Takes Years to Build

The language model is the easy part Any capable AI model can summarise a document. What it cannot do on its own is say how hot a particular buried cable runs, or how aggressive the soil around it is. That requires a citeable environmental data foundation — 70+ authoritative datasets resolved to each location and joined to cable physics — and the utility-specific engineering judgement to apply it, drawn from the founding team's combined 40+ years at Siemens, Alstom, GE, Hatch, Bloomberg, BMO and SAS Institute. The conversational layer (Cirra) sits on top of that foundation as the way to query it, not as the product itself.

16.2 Citeable Provenance — Built for the Prudency Review

The deepest design choice in Earthflow Underground is that every numeric output emitted by the platform traces to a citeable authoritative dataset. The provenance JSON for a Health Index of 4.2 on cable segment XYZ breaks the score into its sub-scores, and each sub-score cites the specific dataset, record and published standard it was computed from — including the utility's own outage history where that is the source. No part of the score rests on an undisclosed weight.

Why this matters: when a utility's Regulatory Affairs lead walks into a state Public Utility Commission rate-case prudency review and says "we deferred replacing these 47 miles of cable and accelerated replacement on these other 31 miles," the PUC's first question is "on what basis?" — and the auditable answer must trace back to a public dataset and a published equation, not to a vendor's proprietary score. Earthflow's provenance record makes that the default. None of the incumbent solutions surveyed in Chapter 15 carry provenance down to the individual calculation.

16.3 Physics-Rooted Modules — Not Statistical Regression

Each of the 11 Underground modules described in Chapter 11 is grounded in published cable-engineering standards and peer-reviewed methods. Where the science supports an absolute quantity — a thermal rating, for example — the module computes one. Where it does not, the module says so: soil corrosion is reported as a relative ranking of corrosion drivers, not a predicted metal-loss rate, and remaining life is reported as a band, never a single year.

A deep-learning model trained on ten years of outage records will find statistical patterns, but it struggles to give an auditable reason for any single prediction. A physics-rooted module answers both what and why — and the why is what an engineer, a planner and a regulator each need before acting.

16.4 Cirra — Asking the Network a Question

Most AI assistants added to enterprise software search documentation. Cirra instead works over the full 30-module Physics AI stack. When a T&D asset management lead asks Cirra "which cables are most likely to fail in storm season?", Cirra:

  1. Decomposes the question into the relevant module queries (AHI Composite · Flood Hydrology · Soil Corrosion · Manhole Risk)
  2. Executes those queries against the live data layer for the utility's network
  3. Ranks the resulting segments by combined risk score
  4. Composes a written response with inline citations back to each contributing module and source dataset
  5. Saves the conversation as a defensible model card for the utility's records

A question that would otherwise become an analyst engagement is answered directly, with the audit trail attached.

16.5 The Reinsurance Adjacency — Same Engine, Different Buyer

Earthflow Reinsurance Underwriter — adjacent vertical
Figure 16.1 · Earthflow Underwriter · Reinsurance vertical The same Physics AI engine powers underground cable risk for utilities and treaty-pricing for reinsurance underwriters. The physics that says "this cable is at risk" is the same physics that says "this exposure has this expected annual loss." One platform, two adjacent markets.

The architecture is not specific to one market: the 70+-dataset provenance foundation and the multi-peril fusion engine that power Underground also power Earthflow Underwriter (reinsurance) and Earthflow Detect (EPC bid evaluation). The same physics that flags a cable segment as high-risk drives a treaty premium calculation on the reinsurance side. One engine, multiple adjacent verticals.

16.6 What Remains Hard

Three parts of the approach take real work in every deployment:

Earthflow Underground is not the first analytics platform for underground distribution cable, and the incumbents solve parts of the problem well. What it adds is the whole: rapidly deployable, multi-hazard, physics-rooted and auditable before FERC and the PUC, with a natural-language layer on top. The advantage lies in the data foundation and the engineering behind it, not in the AI model. See it live →

17 · Conclusion — A Citeable, Physics-Rooted Path Forward

Utilities are burying their networks for resilience faster than ever, and every mile buried is a mile that can no longer be inspected. The binding constraint is no longer the cost of putting cable in the ground. It is knowing the condition and capacity of what is already there — and being able to show a regulator why each dollar goes where it does.

The technology discussed in this report is no longer theoretical. Published industry programmes report measurable gains — reliability improvement on instrumented feeders79 and vendor-reported failure reduction in trials63 — and those results, not Earthflow's, are what establish that the category works. The data to act on is largely already in the utility's hands. Combined with an environmental picture of the ground each cable sits in, it is enough to rank every segment by condition and capacity, decide where to replace, where to rejuvenate and where to wait, and defend those choices in a rate case.

17.1 What This Paper Asked, and What It Found

This paper began with a comprehensive question — what does it take to digitally manage North America's 1.1+ million miles of underground distribution cable? It surveyed:

The evidence cited makes a clear case that the need is pressing and the economics favour action, especially as electrification and resilience goals push utilities toward more undergrounding. The incumbent vendors — Siemens, Hitachi Energy, GE Vernova, Schneider Electric — have produced credible point tools, but none of them ship a rapidly deployable, multi-peril-fusion, FERC/PUC-defensible solution as a single product. The market still has a real gap.

17.2 The Path Forward

Digital analytics for underground distribution is fast becoming standard practice among utilities aiming for top-tier reliability with ageing assets and a harsher climate. The utilities that build the intelligence to manage buried cable alongside the cable itself will spend less, fail less, and carry stronger cases to their regulators.

Orbyfy Earthflow Underground is built to close that gap. Citeable provenance instead of black-box scoring. Weeks instead of months. 11 cable-specific Physics AI modules on top of a 19-module platform engine. FERC and PUC defensible by design. One living twin, six decision-ready surfaces — for the T&D Asset Management VP, the OMS Operator, the CFO, the CIO, the Reliability Engineer, and the field crews who actually go fix the cables.

17.3 Call to Action

18 · FAQ for Utilities

18.1 What data is already in place before we ever talk to the utility?

Earthflow Underground starts with a data foundation we have already gathered, cleaned, and kept current — the Orbyfy Environmental Intelligence Layer. It is national and pre-built at 100% spatial coverage, carries 1,100+ data points at every lat/long, and is resolved to the individual cable segment. Nothing on this list requires anything from the utility — it is all in place on the day the conversation begins, and the only input needed to unlock it is accurate route geometry. Each dataset feeds one or more of the eleven Physics AI models, listed here alongside the answer it produces. (Chapter 11 sets out the layer in technical detail.)

Data SourceWhat it tells usPhysics AI answer it produces
National soil-property database How acidic, salty, wet, or rich in organic matter the ground is around every cable trench in the country Soil Corrosion model — ranks how aggressive the soil is toward each cable's metal sheath
Flood and water history Which patches of ground sit inside a 100-year floodplain, and which have actually been under water in the last 35 years Flood Risk model — flags which manholes, splices, and segments are exposed to flood damage
Lightning-strike history How often lightning has actually struck the ground above each feeder over the past 35 years Lightning Surge model — estimates surge-damage exposure for each underground feeder
Ground-motion maps Which neighborhoods sit on soft soil that amplifies earthquake shaking Seismic Risk model — flags segments crossing ground-motion amplification zones
Storm and climate history The long-run climate exposure of every neighborhood the utility serves Storm Hardening model — scores neighborhood-by-neighborhood storm risk and informs undergrounding priorities
Soil moisture and vegetation maps Whether the soil around each cable sheds heat or traps it Thermal Loading model — flags cables in heat-trapping soil, where a hot summer day can age a cable like a hot summer year
International cable-aging reference curves (IEEE and CIGRÉ standards) Industry-baseline failure rates by cable age, insulation type, and operating environment Used as the starting point for the Health Index — replaced by the utility's own outage history once that is available

Every input carries provenance metadata. For any score we report, we can show the utility exactly which dataset contributed and where we used an industry default instead of the utility's own history.

18.2 What does the utility need to provide — and how do we handle it?

We work with whatever the utility can provide, in whatever shape it is in. Underground distribution data is rarely clean — that is a structural feature of the asset class (see Chapter 6), not a utility-specific failing — and we never ask a utility to scrub its records before sending them to us. The table below lists the inputs that compound value fastest, the ones that sharpen results when they happen to exist, and the ones we explicitly do not want the utility spending effort on.

TierWhat we ask forPhysics AI answer it powers
Most valuable to start with Cable-map system — in whatever shape it currently exists, including segments with incomplete attributes or legacy paper-only records Health Index (1 = best, 5 = worst), Remaining Useful Life forecast, and every spatial overlay
Most valuable to start with Outage log — last 5 to 7 years; we work with “cause = unknown” entries the same way we work with detailed cause codes Calibrates the Health Index and Remaining Useful Life forecast to the utility's own fleet
Most valuable to start with Daily-load readings — from the historian, at whatever cadence is available (5-minute through 1-hour) Thermal Loading model and capacity-stress indicators per feeder
Most valuable to start with Smart-meter data — meter-to-feeder topology and last-gasp signals; we handle the reliability gaps across different meter vintages Dark-Sector Fault Triangulation — narrows an outage to about a 200-foot stretch of cable, removing roughly 65% of patrol time
Sharpens results if available Partial-discharge test reports — PDFs, scanned paper, or maintenance-system attachments; we do the extraction Sharpens the Partial Discharge Progression forecast
Sharpens results if available Splice locations, dielectric-loss test history, manhole inspection records — in whatever form they exist Tightens the Remaining Useful Life forecast
Sharpens results if available Manhole sensors — temperature, humidity, combustible gas (works fine without them) Manhole Explosion Risk model
We do not need this Live control-room data streams Daily summaries from the historian are sufficient
We do not need this Customer billing data Outside the scope of the analysis
We do not need this Any operational control authority Decision-support only — we never write back to the control system
We do not need this Staff time digitizing dirty records The Schema Auto-Mapper handles the cleanup; we do the unstructured extraction
We do not need this Whole-system data — just the pilot feeders A bounded scope is what makes the pilot quick to assemble and quick to prove
We do not need this Customer names, addresses or account identifiers De-identified interval load is sufficient — we need the load shape, not the customer

Bottom line: six of the eleven Physics AI models go live from the cable map and outage log alone — even when those records carry the typical underground-data gaps. The daily-load readings and smart-meter data light up two more — the Capacity pair, which turn measured loading into an environmentally-adjusted in-situ line rating and a headroom figure — leaving three that genuinely require field sensing.

The minimum viable dataset. If only a subset can be supplied to begin with, the floor is cable-map geometry and attributes, five years of outage history, and one year of feeder loading. With those three, plus the Environmental Intelligence Layer described in §18.1, both headline indices can be produced — a per-segment Health Index and a first Capacity screen. Everything else in the table above refines the confidence bands rather than unlocking new outputs.

18.2b How the data is handled — delivery, security and governance

Underground asset data is sensitive, and utility security teams are right to ask how it moves before they ask what it produces. The arrangement is deliberately narrow:

The practical consequence is that a pilot is a data-transfer exercise rather than a systems-integration project. Nothing is installed on utility infrastructure, no agent runs inside the utility network, and the utility's systems of record remain the systems of record.

How we handle the data itself

Underground distribution data is heterogeneous, partial, and often decades stale. Our architecture assumes that on day one, not as an exception to be managed:

Seven days from data drop to a live risk twin. That is the engineering choice that makes this possible.

18.3 Do you need historical data to identify alarms or similar events?

To start, no. Six of the eleven Physics AI models run on day one using the data foundation above plus the utility's cable map, and the two Capacity models come online as soon as any feeder-loading history exists. Where the utility's own failure history is missing, we fall back to internationally cited industry failure curves (IEEE and CIGRÉ standards). The utility receives a ranked, citeable list of its worst-5%-of-segments within seven days of the data drop — even with zero labeled failures on file.

For a defensible back-test and tighter accuracy, yes. Our primary success measurement — getting at least 70 of every 100 “worst-cable” predictions correct against the utility's own historical fault locations — only becomes possible once five years of outage history is in hand. With that history, the Remaining Useful Life model calibrates to the utility's own vintages and failure patterns. Without it, results are still produced; they simply carry wider confidence intervals, surfaced explicitly in every output rather than hidden behind a precise-looking number.

In short: historical data is what proves the worst-5% list is right. It is not what makes the list possible.

For the deployment timeline behind these answers, see Chapter 11 — Earthflow Underground Reference Architecture and Chapter 12 — Case Study: Reference Deployment Pattern.

About Orbyfy Earthflow · About the Authors

About Orbyfy Earthflow

Orbyfy is a Boston-based Climatetech company building Earthflow — a Physics AI + Agentic AI living digital twin for critical infrastructure. The platform fuses authoritative environmental intelligence with utility, EPC, and reinsurance data through a 19-module Physics AI engine and an agentic AI orchestration layer (Cirra). Two product verticals are in active development: Earthflow Underground (underground electric distribution cable risk) and Earthflow Underwriter (reinsurance treaty pricing). Orbyfy was founded in 2025, is a member of Greentown Labs in Somerville, MA, won Edison Awards Gold 2026, and is part of NVIDIA Inception and the Black & Veatch IgniteX accelerator cohort. Production runs on Google Cloud Platform with NVIDIA NIM-compatible inference paths. The company is partnered with EPRI (AI for Power Challenge 2026 submission, Category 1 · Grid Asset Management) and is assembling a multi-utility Demonstration Partner Consortium across the Northeast.

About the Authors

Hara Brar — Founder, CEO, CTO of Orbyfy
Hara Brar
Founder · CEO · CTO · Orbyfy

20+ years at the intersection of power grid and AI — Siemens, Alstom Power, GE, Hatch. IDC-ranked digital solutions architect for utility infrastructure. Hands-on developer behind the Earthflow geospatial core and the Underground Physics AI engine. Leads product, engineering, and customer engagements. Boston, MA.

Greg Elliott — Founder, Chief Physics AI Officer of Orbyfy
Greg Elliott
Founder · Chief Physics AI Officer · Orbyfy

20+ years across fin-tech, energy & utilities, and applied Physics AI research — Bloomberg, Thomson Reuters, Siemens, Cinchy, BMO, SAS Institute. Architects the Physics AI stack — 19 physics-informed modules fusing authoritative environmental datasets with utility telemetry into Health Index, RUL, and risk forecasts.

Glossary of Acronyms

ADMSAdvanced Distribution Management System — integrated software suite for managing the operations of a distribution grid (SCADA + outage management + network analysis).
Health IndexAsset Health Index — a composite scoring (typically 1–5) indicating overall asset condition combining age, type, load, environment, and test results.
AMIAdvanced Metering Infrastructure — smart meter network capable of two-way communication with utility systems.
APMAsset Performance Management — software platform that combines asset condition data with predictive analytics for maintenance and risk planning.
CAIDICustomer Average Interruption Duration Index — the average time required to restore service to interrupted customers.
CMICustomer Minutes Interrupted — total minutes of interruption summed across all affected customers.
DPUDepartment of Public Utilities — state-level regulator (e.g., Massachusetts DPU oversees the Eversource ESMP filings).
DTSDistributed Temperature Sensing — optical-fiber technique that measures temperature along the entire length of a cable run.
EAMEnterprise Asset Management — software for tracking work orders, inspections, and lifecycle data of physical assets.
ESMPElectric Sector Modernization Plan — utility filings with state PUCs outlining grid modernization and resilience investments.
FCIFaulted Circuit Indicator — a device that flags the passage of a fault current through a cable, used to localize faults.
FERCFederal Energy Regulatory Commission — US federal regulator of interstate electric and gas markets.
FLISRFault Location, Isolation, and Service Restoration — distribution-automation function that automatically isolates faults and restores power.
GISGeographic Information System — spatial database and mapping software (e.g., Esri ArcGIS / ArcFM).
IOUInvestor-Owned Utility — privately-owned, regulator-overseen electric utility (e.g., Eversource, FPL, PG&E).
IRInfrared — thermal-imaging technique used to detect hot spots on cable terminations and connections.
LLMLarge Language Model — the class of foundation models (e.g., GPT, Claude) used to power natural-language interfaces.
OMSOutage Management System — utility software that tracks outage events, customer-impact, and crew dispatch.
PDPartial Discharge — localized electrical breakdown in cable insulation that precedes full failure; measurable with online or offline sensors.
PILCPaper-Insulated, Lead-Covered — legacy distribution-cable construction; long-lived but susceptible to moisture if lead sheath degrades.
PUCPublic Utility Commission — state-level utility regulator (also DPU in some states); oversees rate cases and prudency reviews.
RULRemaining Useful Life — estimated time until an asset reaches end-of-life condition, typically expressed in years with a confidence interval.
SAIDISystem Average Interruption Duration Index — average outage duration per customer per year.
SAIFISystem Average Interruption Frequency Index — average number of outages per customer per year.
SCADASupervisory Control and Data Acquisition — real-time monitoring and control system for electric grid operations.
SSUPStorm Secure Underground Program — FPL's program for converting overhead distribution laterals to underground.
XLPECross-Linked Polyethylene — the dominant modern insulation material for distribution cables; susceptible to water-tree degradation in older formulations.

References & Data Sources

Supporting data and examples have been drawn from industry reports, utility case studies, vendor literature, DOE and regulatory reports, and product information from Siemens, Hitachi Energy, GE Vernova, and Schneider Electric, among others. These illustrate the current state and benefits of underground cable analytics in practice.

[1]U.S. Department of Energy Announces $34 Million to Improve the Reliability, Resiliency and Security of America's Power Grid. arpa-e.energy.gov
[2, 5, 62]Undergrounding Transmission and Distribution Lines Resilience Investment Guide. DOE Grid Deployment Office, Nov 2024. energy.gov
[3]Distribution of high voltage transmission lines across Canada. researchgate.net
[4, 22, 23]Underground Distribution Lines · Hydro-Québec. hydroquebec.com
[6, 8, 9, 10, 12, 13, 14, 18, 19, 20, 21, 91]Distribution Goal: 50% Underground by 2040 · T&D World. tdworld.com
[7, 11, 15, 16, 17, 40]The Changing Economics of Utility Investment in Undergrounding. Technical Paper 100-T144. sandc.com
[24]Why not avoid falling trees by putting lines underground? · BC Hydro. bchydro.com
[25, 26, 33, 34, 35]CPUC Undergrounding Programs Description. cpuc.ca.gov
[27]Transmission and Distribution in Canada · POWER Magazine. powermag.com
[28, 29, 30, 85, 86]MergedFile (ClearSkiesAhead). scenic.org
[31, 32, 37, 49, 50, 51, 60, 61, 94]Predictive Maintenance and Remaining Useful Life for Underground Cable Systems. powersystems.technology
[36, 41, 45, 52, 53, 54, 55, 80, 81, 82, 83]IoT technology prevents exploding manhole covers · Energy Magazine. energydigital.com
[38, 39, 46, 103, 104]Conduit Manager by Schneider Electric · Esri Partner Solution. esri.com
[42]Wired for Growth: Unveiling Opportunities Amid Challenges in the US · T&D World. tdworld.com
[43]The U.S. is facing an "unprecedented" shortage of electric transformers, NREL says. euci.com
[44]A Mid-2024 Assessment of the U.S. Distribution Transformer Industry. newton-evans.com
[47, 48, 56, 64, 65, 74, 75, 76, 77, 78]Fault Indicator Underground · UM3+ by Sentient Energy. sentientenergy.com
[57, 58, 66, 67, 68, 69, 70, 79]Determining the ROI of Your Intelligent Line Sensor Projects. electricenergyonline.com
[59]Online Partial Discharge Monitoring of Power Cables · OMICRON. omicronenergy.com
[63]Cable Analytics · Siemens Xcelerator Global. xcelerator.siemens.com
[71, 72, 73, 88, 89, 92, 93, 97, 98]Hitachi Energy Launches the Next Generation of its Asset Performance Management Solution, Lumada APM. hitachienergy.com
[84]Evolution of Circuit Indicators & Line Sensors · Sentient Energy. sentientenergy.com
[87, 99, 100, 102]Asset Performance Management System — APM Software · GE Vernova. gevernova.com
[90]Earth fault indicator for cables · SICAM EFI · Siemens Global. siemens.com
[95]Monitoring and Control Systems for the IoT in the Water Supply. hitachihyoron.com
[96]Asset Management · Hitachi Energy. hitachienergy.com
[101]Digitalizing Energy Asset Performance Management for Smarter Grids. social-innovation.hitachi
[105, 106, 107, 108, 109, 111, 112, 113]Our digital services · Schneider Electric. se.com
[110]Schneider Electric Launches New Monitoring & Dispatch Services to Manage Distributed IT. prnewswire.com
[114]IEEE Std 835-1994 (R2006) and 835a-2012, IEEE Standard Power Cable Ampacity Tables. IEEE.
[115]IEC TR 60287-3-1, Electric cables — Calculation of the current rating — Part 3-1: Operating conditions — Site reference conditions, Annex A (national practices). IEC.
[116]IEEE Std 442-2017, IEEE Guide for Thermal Resistivity Measurements of Soils and Backfill Materials. IEEE.
[117]NEETRAC / Georgia Tech Research Corp., Cable Diagnostic Focused Initiative (CDFI) Phase II, DOE Award DE-OE0000188, 2016. osti.gov
[118]LBNL for U.S. DOE Grid Deployment Office, Undergrounding Transmission and Distribution Lines: Resilience Investment Guide, September 2024. energy.gov
[119]Alectra Utilities, interrogatory response on cable injection and replacement capital, OEB EB-2022-0013. rds.oeb.ca
[120]Ontario Energy Board, backgrounder on Alectra Utilities Decision EB-2023-0004, February 2024. oeb.ca
[121]Ontario Energy Board, Filing Requirements for Electricity Distribution Rate Applications, Chapter 5 (Distribution System Plan), December 2025. oeb.ca
[122]Florida Statutes s. 366.96, Storm protection plan cost recovery. flsenate.gov
[123]Florida Administrative Code Rule 25-6.030, Storm Protection Plan. law.cornell.edu
[124]Florida Power & Light, 2023–2032 Storm Protection Plan, Docket 20220051-EI, Exhibit MJ-1. floridapsc.com
[125]Code of Virginia § 56-585.1(A)(6). law.lis.virginia.gov
[126]Virginia SCC Division of Utility Accounting & Finance, letter on continuation of the strategic underground program, February 24, 2026. baconsrebellion.com
[127]PG&E files 10-year plan for electrical undergrounding in high wildfire-risk areas · T&D World, October 2026. tdworld.com
[128]PG&E press release on 1,000 miles undergrounded and cost per mile, October 2025. stocktitan.net
[129]Florida Public Service Commission, Annual Status Report on Storm Protection Plan Activities of Florida IOUs, November 2025. floridapsc.com
[130]Kihm, Beecher, Lehr, Regulatory Incentives and Disincentives for Utility Investments in Grid Modernization, Future Electric Utility Regulation Report No. 8, LBNL, 2017. eta-publications.lbl.gov
[131]Ofgem, Handbook for implementing the RIIO model, October 2010. ofgem.gov.uk
[132]IEEE Std 1617-2022, IEEE Guide for Assessment, Mitigation, and Control of Corrosion of Metallic Shields for Extruded Dielectric Cables Rated 5 kV to 46 kV, Annex B. IEEE.
[133]MVEA implements an alternative to cable replacement · T&D World, October 2017. tdworld.com
[134]Ontario utility benefits from rejuvenating cable · T&D World, June 2018. tdworld.com
[135]U.S. Energy Information Administration, Form EIA-861, Distribution Systems, 2024 data. eia.gov
[136]California Assembly Committee on Utilities and Energy, analysis of SB 884 (McGuire), June 2022. autl.assembly.ca.gov