EPRI Attributes 11 Percent of Battery Failures to Cells as 2026 Fire Codes Tighten Cell Testing

Roughly one in nine battery storage failure incidents in the Electric Power Research Institute’s Failure Incident Database is attributed to the cells. The rest sits elsewhere in the system.

EPRI’s root-cause attribution runs approximately 46 percent to controls, 43 percent to balance of system, and 11 percent to cells. Combined, controls and balance of system account for about 89 percent of attributed causes. Balance of system is the category that covers the equipment around the battery: wiring, enclosures, power conversion, thermal management, fire suppression, and the rest of the installed assembly.

The decline. EPRI released Risk Factors and Failure Statistics of Lithium-Ion BESS on October 5. Energy-Storage.News reported the headline finding: annual failures per deployed gigawatt-hour fell roughly 99 percent globally between 2018 and 2025. EPRI benchmarked the current incident rate against large oil-filled transformers, another utility asset with infrequent fire events.

That comparator is the part of the report most likely to travel. A transformer is equipment that facility engineers, insurers and authorities having jurisdiction already accept in and around occupied buildings, and EPRI has now placed the current battery failure rate in the same reference frame.

The first two years. Seventy-two percent of failures in the database surfaced within the first two years of operation. Taken alongside the root-cause split, the distribution describes a failure profile concentrated at the front of the asset life and concentrated outside the electrochemistry.

Both findings point toward the same set of activities: how a system is assembled on site, how components from different suppliers are made to work together, how commissioning is documented, and whether a monitoring alarm reaches someone who acts on it. None of those are properties of a cell.

Washington. The Seattle Fire Marshal’s office has pressed the Washington State Building Code Council to move the state energy storage reference from NFPA 855’s 2023 edition to the 2026 edition, and to fix multiple correlation errors between the battery provisions of the International Building Code and the International Fire Code.

Washington currently operates under an extended emergency rule that directs fire code officials to the 2023 edition in IFC Chapter 12. The same Seattle office drove that emergency rule, and previously pushed to bring NFPA 855 and UL 9540 into the state residential code.

The 2026 edition changes what an indoor installation must demonstrate. It makes hazard mitigation analysis the default, replaces standalone deflagration venting with an NFPA 69 gas-concentration requirement, and sets a dedicated-room threshold. A state that pins its code to that edition converts those provisions into enforceable law and creates a permitting path that diverges from states still referencing 2023.

The correlation errors matter independently of the edition question. Ambiguity between building code and fire code battery provisions is the kind of conflict that stalls an indoor plan review, and an adoption that updates the reference while leaving the conflicts in place would produce a worse permitting environment than the one it replaced.

The code cycle. Washington is one jurisdiction inside a year that has moved consistently in one direction. UL 9540A reached Edition 6 in March. The first indoor large-scale fire test under the updated standard was witnessed on September 18. NFPA 855’s 2026 edition tightened thermal-runaway propagation requirements.

Each of those tightens the evidence a manufacturer must produce about how a cell behaves in thermal runaway and whether that runaway propagates. The direction of travel is toward more cell-level and unit-level testing as the condition of entry to an occupied building.

The gap. Product certification tests the category that EPRI attributes 11 percent of root causes to. The 89 percent attributed to controls and balance of system is governed by installation workmanship, interface compatibility between components from different suppliers, commissioning procedure, and alarm escalation.

A listing mark does not cover those. A test report describes a unit that passed under laboratory conditions. It does not describe how that unit was installed on a particular site, or whether a fault alarm on that site routes to a person who responds.

The documents that correspond to EPRI’s failure distribution are commissioning records, interface test results, alarm-escalation procedures, and the first two years of monitoring data. On the evidence available here, those sit in commercial contracts and operations-and-maintenance scopes rather than in a certification file, which is an observation about where the paperwork lives, not a measured finding about how closely it is scrutinized.

The practical reading is that the certification cycle and the failure data are tightening different things at the same time. Both are moving, and they are not moving on the same surface.

The database limits. The Failure Incident Database covers commercial and industrial systems as well as utility-scale systems. The reported root-cause percentages and the two-year concentration apply to that combined population, and EPRI’s published figures do not separate the two segments.

A building owner reading the 99 percent headline is therefore reading a global rate across a population dominated by grid-scale deployment, with a root-cause attribution that spans both scales. That is a caveat about granularity rather than about accuracy. The 99 percent decline is EPRI’s own measurement against its own database, and the transformer benchmark is EPRI’s own framing.

What the data supports is narrower than the headline implies and more useful than it first appears. The failure rate has fallen by two orders of magnitude. Where failures still occur, the attributed cause is far more often the system around the cell than the cell itself, and the exposure is concentrated in the first two years after a system is energized. The code cycle now closing in on 2026 addresses the first of those findings in detail. The second and third are addressed, if at all, in commissioning scopes and service contracts that no listing mark reaches.

Sources