91TV

Facilities that track what they intercept, and where it came from, can start to characterize their inbound risk, hold problem sources accountable, and feed real numbers back into public education and producer-responsibility programs. Detection gives operators the information to reduce risk year-over-year, rather than just survive it.
By Lia Kiam

Last year was the worst on record for fires at North American waste and recycling facilities. Fire Rover’s annual report counted 448 publicly reported incidents across the U.S. and Canada, up from 430 in 2024 and 373 in 2023, and nearly 25 percent above the long-run average of 360. July alone produced 56 reported fires, the highest single month ever tracked. A separate analysis put direct property damage near $2.5 billion, including roughly 100 catastrophic losses running from half of a million dollars into the tens of millions. And those are just the fires that made the news. The National 91TV and Recycling Association estimates more than 5,000 fires occur at recycling facilities every year once you count the small ones that crews put out before anyone outside the building notices.

The June 2026 fire report put it bluntly: the industry is now operating in a permanently elevated fire risk environment. It is a useful phrase because it forecloses the comforting explanation. This is not a bad stretch that better suppression will eventually correct. Something has changed structurally in what arrives at the tip floor, and mitigation must change with it.

Two Hazards that Punish How We Process Material
The main driver is lithium-ion batteries, specifically the ones buried inside everyday products never designed to be recovered. A punctured, crushed, or shorted lithium cell can enter thermal runaway, a self-sustaining reaction that generates its own heat and oxygen and resists cooling. The trouble is that a MRF is, functionally, a machine for puncturing, crushing, and compacting. A battery that sits inert in a kitchen drawer becomes an ignition source the instant it meets a baler ram, a shredder, or the weight of a loaded conveyor.

Disposable vapes have made this considerably worse. Roughly 1.2 billion of them enter the waste stream every year, each one a small lithium cell in a casing that offers almost no protection against mechanical insult. Consumer habits do not help: a recent UL Standards & Engagement survey found 36 percent of people admit to throwing old lithium-ion batteries straight in the trash, and another 30 percent mix them into their recyclables. Most facility fires that trace back to a battery, trace back to one that was never identified, segregated, or deactivated before it came through the door.

The second hazard gets less attention, but is just as capable of putting a worker in the hospital: pressurized and flammable containers. Propane cylinders, camping canisters, aerosol cans, and partially full fuel containers all store energy that releases violently when the vessel is breached. A propane tank that ends up in a shredder with a load of scrap does not smolder—it detonates. Aerosol cans, which typically use flammable propane or butane as a propellant, behave the same way when crushed in volume. What makes them dangerous is also what makes them hard to catch: a worker on a sort line has no way to know whether a dented cylinder is empty or still holds fuel, and these containers routinely arrive hidden in bags, mislabeled, or buried in mixed material.

The Case for Moving Detection Upstream
The default mitigation posture for years has been suppression-centric: install sprinklers, keep extinguishers and hose lines close, train crews to respond, and treat some fires as the cost of doing business. Suppression is necessary (no facility should run without it), but it is the last line of defense, and by the time it engages, the hazard has already reached the equipment, and the damage clock is running.

The better organizing principle is defense in depth, weighted upstream. Every hazard has a window between arriving at the facility and reaching a piece of equipment that can set it off. The job of a modern mitigation program is to widen that window and exploit it: to find the battery or the cylinder while it is still on the tip floor or the pre-sort line, where the response is a gloved hand and a segregation bin instead of a fire crew and an insurance claim. Detection, not suppression, is where the leverage sits.

That reframing changes what gets measured. A suppression-first operation tracks how fast it extinguishes a fire. A detection-first operation tracks how often a hazard is caught before it becomes one, and counts every battery pulled from the inbound stream as a save rather than a near-miss nobody logs.

Layers of Detection: From the Tip Floor to the Burden Depth
Detection works best as a series of overlapping layers, each one catching what the last one missed. Two technologies are doing most of the real work right now, and they catch different things for a reason: one sees what is on the surface, the other sees what is buried inside it.

The first layer is AI-driven visual detection on the tip floor and sort line. Cameras trained on a facility’s actual material stream can recognize a battery, a propane cylinder, or an aerosol can as it moves past, and flag it for removal in real time. These models improve with exposure: the more footage they see of a given facility’s material mix, the better they get at telling a genuine hazard from a look-alike, which is what eventually brings false alarms down to a level crews actually trust and act on. This layer is powerful for exactly the hazards it can see, meaning anything sitting on top of the pile or visible on a conveyor. Its limitation is the same as a human sorter’s: it cannot see through the material in front of it.

That is the gap the second layer closes. A lot of the most dangerous batteries never sit on top of anything. They are inside a phone, a power tool, or a set of earbuds that is buried under other material, and a camera pointed at the surface of the stream will never register them. AI-powered X-ray systems solve for this by imaging straight through burden depth, reading density and material composition rather than surface appearance, which means they can pick up a small, embedded lithium cell even when it is wrapped in plastic, packed inside a bag, or sitting under a layer of other debris. For a hazard class defined by hiding inside things that were never meant to be taken apart, seeing through the pile rather than just at it is the difference between catching the battery and finding out about it later—from the fire.

Used together, the two systems cover the stream in a way neither can alone: visual detection handles what is exposed and moving fast, while X-ray handles what is hidden and would otherwise slip through every downstream stage undetected.

What is Different About C&D Streams
Construction and demolition facilities face the same two hazards through a different door. C&D loads are heavier, dirtier, and processed with more aggressive equipment, which raises the stakes on anything that slips through. The material is also less predictable: a demolition load can carry cordless tool batteries, partially full propane and fuel cylinders from a job site, solvent and adhesive aerosols, and the occasional abandoned tank, all mixed into debris that gives a sorter very little time to react. Because the stream looks nothing like clean single-stream recycling, detection systems and staff training must be tuned to what actually shows up in demolition debris, not to a generic battery-in-the-bin model. The principle holds—the hazard catalog and the points where it is most exposed are specific to the operation.

Detection as the Foundation, not the Finish
Regulators and industry groups are finally catching up to what operators have been living with. In early 2026 the National 91TV and Recycling Association, the Recycled Materials Association, and the Solid 91TV Association of North America jointly published a guide for managing lithium-ion batteries at materials recovery facilities—the first cross-industry framework of its kind. The EPA, meanwhile, is working toward a new universal-waste category written specifically for lithium batteries, aimed at keeping them out of the trash stream altogether. Both efforts point in the same direction: identify and segregate the hazard as early as possible.

There is a longer-term payoff to detection that suppression cannot offer. Every battery and cylinder a system catches is also a data point. Facilities that track what they intercept, and where it came from, can start to characterize their inbound risk, hold problem sources accountable, and feed real numbers back into public education and producer-responsibility programs. Suppression tells you how a fire ended. Detection tells you how the hazard arrived, and that is the information that lets an operator reduce risk year-over-year, rather than just survive it.

The fires are not going to stop arriving. Batteries are in everything now, the volume entering the waste stream keeps climbing, and the equipment that makes recycling economical is the same equipment that turns a hidden hazard into a loss. What an operator can control is how early the hazard is found. The facilities that come through this elevated-risk era in the best shape will be the ones that stopped treating detection as an add-on to suppression and started treating it as the foundation of the whole program | WA.

Lia Kiam is Head of Strategy and Operations at Visia. She can be reached at [email protected].

Resources