Industries / Construction and property / Commercial and residential construction / Electrical contracting / Solar installation
Wiring and inverters
This work runs the cables from the panels to the inverter, the box that turns the panels' direct current (DC) into the alternating current (AC) a building uses, and fits the switches, fuses and fault detectors that keep the system safe.
6
Milestones achieved, from 2009 to 2020.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
2009
A fault the detector missed
In 2009 a 383 kW array on a store roof in Bakersfield, California, caught fire. A small 2.5 A leak to ground on one cable had gone unnoticed, because it was below what the ground-fault fuse could sense; when a second, 311 A fault came later, current ran back through the first one and set the cable's insulation alight. Bill Brooks called this gap the "blind spot" of US ground-fault protection, and in May 2011 he reported on a fire at a 1.1 MW plant on the National Gypsum building in Mount Holly, North Carolina. The 2011 National Electrical Code (NEC) then required arc-fault detection, which senses the sparking of a loose connection, on building-mounted systems above 80 V, and the 2014 edition extended it to ground-mounted systems. Sandia National Laboratories later reported a large installer whose commercial roofs had nine fires in six years with fuse-based protection and none in three years after it fitted finer current monitors.
LessonA protection device that cannot see small faults lets them grow until a second fault turns them into a fire.
Credit
Bill Brooks (Brooks Engineering); Sandia National Laboratories; NFPA's NEC committee
2013
Inverters cause most outages
A. Golnas published the service records of an operator running hundreds of solar plants worldwide with local crews. Outages of the parts a plant cannot run without made up 69% of the service issues and 75% of the energy lost, and most of them showed up at the inverter; failed panels were a small share.
LessonThe electronics and wiring that crews assemble on site fail more often than the factory-made panels.
Credit
A. Golnas, IEEE Journal of Photovoltaics
2014
Germany counts its PV fires
TÜV Rheinland, a testing company, and the Fraunhofer Institute for Solar Energy Systems (ISE) spent three and a half years collecting damage cases from Germany's more than 1.4 million solar systems. They found 210 fires caused by the systems: 12 buildings burned down entirely and 67 other systems were damaged. The most common causes were installation errors, then faulty products and design mistakes, and most fires came in the first year of operation. The researchers called for better installer training, acceptance checks by a third party and regular maintenance.
Credit
TÜV Rheinland; Fraunhofer ISE
2018
UK fires traced to installers
The Building Research Establishment (BRE) investigated 80 possible solar fires in the UK for the government, about 0.01% of the installations then in place. Of the fires the PV system caused, it put about 36% down to poor installation, 10% to design errors and 5% to faulty products; the rest had no known cause. DC isolators, switches that cut the panels' current, were the part found most often at the fire's origin, often because water got in.
LessonFire investigators often cannot name a cause, so counts depend on how carefully each fire is examined.
Credit
BRE National Solar Centre and BRE Global
2019
Walmart sues over roof fires
In August 2019 Walmart sued Tesla's energy unit, formerly SolarCity, which had put solar systems on the roofs of more than 240 of its stores. Fires had broken out on at least seven roofs since March 2018, and Walmart had all the systems switched off. The complaint described loose and hanging wires, defects visible to the naked eye, and inspectors who "lacked basic solar training and knowledge". Three days later the two companies issued a joint statement saying they would work to get the systems running safely.
LessonAn installer that inspects its own work needs someone checking the inspectors.
Credit
Walmart's own inspectors, through the complaint
2020
Whole arrays tested for firefighters
From 2019 the US code required cables inside a rooftop array to drop to 80 V or less within 30 seconds of shutdown, which in practice meant an electronic box under every panel. The 2020 NEC added another way to comply, a "PV hazard control system", and UL published its standard for one, UL 3741, in December 2020. A lab tests a named set of panels, racking, wiring and inverter together and certifies that, installed as listed, it keeps firefighters out of dangerous current paths without electronics at each panel. By IAEI Magazine's account only a few such combinations had been listed.
LessonA rule written around one kind of product can be met by testing the whole system instead.
Credit
UL (Underwriters Laboratories); NFPA's NEC committee
Watch for
Strings at 2,000 volts
Large solar plants wire their panels in strings of up to 1,500 V. Raising this to 2,000 V puts about a third more panels on each string, so fewer cables, combiner boxes and connectors are needed. ABB says its 2,000 V DC switch-disconnector, the first certified by UL to UL 98B, cuts the cost of a plant's electrical parts by up to 15% through faster installation and fewer components. China Datang has opened a 2,000 V test base in Wenchang, Hainan, beside a 100 MW plant, to compare 2,000 V with 1,500 V systems in hot, humid, salty air. A November 2026 search found no measured results published yet.
Achieved whenDatang's test base, a plant owner or an independent engineer publishes measured installed cost and energy yield for a 2,000 V plant against a 1,500 V plant built in the same period.
Claimed by
ABB, July 2025; China Datang, September 2025
Open
01
One form for fire reports
A shared case form and a public table of solar system fires from several countries, one row per fire, coded the same way: system size and age, how it was mounted, the part where the fire started, the root cause (design, product, installation or unknown) and how far the fire spread. Each code links to the sentence in the source report it came from.
The briefCredit · unclaimed02
Inverter fault cases with answers
A public test set of solved inverter faults from home and small business systems. Each case holds the error code, a day of monitoring data around the fault, the installer's notes and what actually fixed it. AI assistants are then scored on naming the cause and on saying whether a site visit is needed.
The briefCredit · unclaimed
Open 01 · the brief
One form for fire reports
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A shared case form and a public table of solar system fires from several countries, one row per fire, coded the same way: system size and age, how it was mounted, the part where the fire started, the root cause (design, product, installation or unknown) and how far the fire spread. Each code links to the sentence in the source report it came from.
Why it is still open
In November 2026 Nik Rus, Grunde Jomaas and four colleagues compared national statistics from the UK, Italy, Sweden and Slovenia in the Fire Safety Journal. They found about 4 to 11 fires per gigawatt (GW) a year in the first three countries and about 37 in Slovenia, said the gap probably comes from different ways of recording fires, and called for harmonised data collection. Case records exist but in different shapes: BRE's anonymised list of 80 UK incidents, Japan's accident reports and TÜV Rheinland's 210 German cases. No shared case table was found on GitHub, Hugging Face or Crossref.
Built with what exists
BRE's published case listing; Japan's consumer accident reports; the Swedish, Italian and Slovenian data described by Rus and colleagues; fire service reports; a document-reading model that drafts each code and cites the sentence behind it (see the Note "Findings that cite notes"); two people checking; Python to measure how often two coders agree.
First step
Code BRE's 80 listed UK incidents with a draft form and record which fields the listing cannot fill.
Done looks like
300 fires from at least four countries coded and published under an open license with the coding guide, and two independent coders agreeing on the starting part and the root cause in at least 80% of a 60-case sample.
Open 02 · the brief
Inverter fault cases with answers
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test set of solved inverter faults from home and small business systems. Each case holds the error code, a day of monitoring data around the fault, the installer's notes and what actually fixed it. AI assistants are then scored on naming the cause and on saying whether a site visit is needed.
Why it is still open
In Golnas's operator records most lost energy traced to outages at the inverter. Makers publish tables of alarm codes and steps, such as Solis's alarm code overview, and a 2026 conference paper describes an expert system for Huawei and Sungrow error codes in Romania, but no public set of solved cases was found on GitHub, Hugging Face, Crossref or the web; the solar fault sets on Hugging Face hold panel images, not inverter cases. Without one, claims for AI troubleshooting tools cannot be compared.
Built with what exists
Closed service tickets from installers, with owners' consent and personal data removed; monitoring exports from maker portals such as SolisCloud, SMA's Sunny Portal and Enphase's Enlighten; makers' alarm tables; an open document model; a fixed-answer scoring sheet (see the Note "A thousand questions, one pass").
First step
Ask three installers for 50 closed inverter tickets each and agree a redaction template with them.
Done looks like
500 solved cases from at least six inverter brands released openly, and three AI assistants scored on the right cause in their first and first three answers and on site-visit calls, against the judgement of two experienced service technicians.