Fire safety and DC arc risk on a factory rooftop array

A rooftop plant puts hundreds of volts of direct current across a factory roof, live in daylight, that no switch in the building can turn off. Why DC arcs behave differently, what rapid shutdown really achieves, and ten questions a safety officer should ask.

The electrical safety file in a Bangladeshi factory is written around alternating current. Fuses, MCCBs, earthing, the Fire Service inspection for the fire licence, the electrical checklist an RMG building goes through for the RSC. A rooftop solar plant adds several hundred volts of direct current running the length of the roof, live whenever there is light on the glass, that no switch in the building can turn off. It is a different hazard and it needs its own set of questions.

A DC arc does not put itself out

An alternating current at 50 Hz passes through zero a hundred times every second, and each crossing gives an arc a chance to extinguish. That is why AC switchgear works the way it does. A direct current circuit has no zero crossing. Once an arc is struck it is stable, and it keeps burning for as long as the source can feed it. A PV string behaves close to a constant current source, which is precisely the sort of source that sustains an arc rather than collapsing it.

The temperature at the arc root is high enough to ignite the module backsheet, cable insulation, a bitumen roof membrane, or whatever is stored on the floor below a light steel roof. And the second difference matters as much as the first: you cannot switch it off. Opening the DC isolator at the inverter de-energises the run between isolator and inverter and does nothing to the string, which sits at open-circuit voltage all day. Modules have no off switch.

Three kinds of arc, three different causes

• Series arc. A break in a current-carrying conductor. A badly crimped connector, a connector never fully clicked home, a fractured cell interconnect ribbon, a loose terminal in a combiner box. It is the most common and the easiest to miss, because the string keeps working at slightly reduced output while the joint heats.
• Parallel arc. Insulation failure between positive and negative. Rodent damage, cable chafed on the sharp edge of a purlin or tray, ordinary PVC building cable used outdoors and cooked by two seasons of sun.
• Earth arc. Conductor to earthed metal. Water into a junction box, damaged frame bonding, a self-drilling screw put through a cable during later roof works.

Cross-mated connectors deserve a sentence of their own. Two connectors from different manufacturers that physically click together are not a rated connection. The contact pressure, the seal geometry and the ageing behaviour were never qualified as a pair. It is the most common cause of the hot connector we find on a thermal survey, and it is almost always the result of an installer using whatever was in the van that week.

Arc fault detection, and what it can and cannot see

Most current string inverters offer arc fault detection. It listens for the broadband electrical noise signature of an arc on the DC side and shuts the inverter down. IEC 63027 sets out how that detection is tested. It is worth having, and it is worth asking two questions before relying on it: how many strings share one detector, and what the false trip history is on similar installations.

Detection across a long, electrically noisy factory roof is harder than in a test laboratory. A detector covering many strings will struggle to hear a small series arc at the far end of one of them, and a detector that trips spuriously gets switched off by the maintenance team within a month, which is worse than not having it. What none of them can do is de-energise the array. The inverter stops. The strings stay live.

Rapid shutdown, and what it actually achieves

Module-level or string-level shutdown devices sit up at the array and, on loss of a keep-alive signal, drop the voltage on conductors outside the array boundary to a low value. The US National Electrical Code rule, NEC 690.12, is the reference most manufacturers design their devices against. Bangladesh does not mandate it, and most rooftops here do not have it.

The honest position is that it adds cost and adds components that can themselves fail, and that there are sites where it earns that cost. Specify it on a building where people sleep or are treated, on a factory with a large workforce and a single stair to the roof, and on any building where the fire service would realistically be asked to work on the roof rather than from the ground. On a low warehouse roof with good access, the same money often does more good spent on connector quality and a funded thermal survey programme.

What every plant should have, mandate or not, is a clearly labelled DC isolator at the array as well as at the inverter, and a written instruction saying which one is opened first and by whom.

Cable routing and fixings are a fire control

• Cable rated for the duty. Double-insulated, UV-stable solar cable to the H1Z2Z2-K description. Not ordinary PVC building cable that a trader has described as solar cable.
• Nothing lies on the roof. Cable sitting in a puddle on an RCC roof through a monsoon will fail. Route in trays or along the mounting rail, clear of standing water, with a drip loop before every entry.
• No ordinary cable ties as the primary fixing. Nylon ties fail within a couple of seasons under this sun and the cable ends up hanging. Use UV-stabilised ties or stainless clips, and inspect them.
• Keep the positive and negative of a string together. It reduces the induced loop area during lightning activity, and it means an insulation fault is less likely to become a parallel arc across a wide loop.
• Respect the bend radius, and gland every entry. An unglanded entry into a combiner box is a water path and eventually an earth fault.
• Look at every edge the cable crosses. Purlin lips, tray ends, flashings. That is where abrasion happens, over years of thermal movement, long after anyone has looked.

Getting a firefighter onto the roof

An array laid out purely for maximum installed capacity, with modules to the parapet and no walkways, is a roof the fire service cannot work on. The design decisions that decide this are made in an afternoon and cannot be undone later.

• A clear perimeter setback wide enough to walk the roof edge without standing on glass.
• Access routes through the array so the roof can be crossed and a ladder can be pitched.
• Smoke vents, roof hatches and existing firefighting penetrations left clear and marked on the layout drawing.
• A laminated single-line diagram and array layout at the main LT panel and at the plant entrance, showing every DC isolator and the array boundary.
• Labels in Bangla as well as English, including a warning that the array is live in daylight.
• An agreed instruction on what happens at night, when a shutdown device or an isolator behaves differently because the array is dark.

The step that actually helps is to walk the roof with the local Fire Service and Civil Defence station before the fire licence inspection rather than after it, and to leave them a copy of the layout drawing. A station that has already seen the roof will make better decisions at two in the morning than one reading a label for the first time.

Ten questions for a factory safety officer to ask

1. Show me the DC isolators. Where are they, are they labelled, and which one do we open first?
2. Are all the connectors on this roof from one manufacturer, and can you show me the crimp tool that was used?
3. What is the printed marking on the DC cable, and does it match what the specification asked for?
4. Do the inverters have arc fault detection, how many strings share a detector, and has it ever tripped?
5. When was the last thermal survey of connectors, combiner boxes and terminations, and can I see the images?
6. Where does the array earthing and equipotential bonding terminate, and when was continuity last measured?
7. Is there a walkway to the roof edge and a route across the array?
8. Where is the array layout drawing kept, and does the fire service have a copy?
9. What is the insulation resistance of each string today, and how does it compare with the commissioning value?
10. If a fire starts under the array at two in the afternoon, what is the written instruction for the shift electrician?

The last one is the real test. If the answer exists only in somebody's head, that is the gap, and it is the cheapest of all of these to close.

Back to all Insights