From above curve, worst-case it is 0.03s at 200A PFC.
I guess the real question is what sort of fault will start a fire but only an AFDD will trip, and how common/likely are they?
In my limited experience of sorting out fixed wiring, the only two cases of a long-term serial arc were light bulb holders in the days of 60W/100W lights and you could sometimes here the buzzing and see the light dim slightly. Yes, those are circuits for which AFDD are not recommended...
Right, which at worse case would translate to 1 1/2 cycles on a 50Hz supply and about 2 cycles on a 60Hz supply if I have my math right. The maximum time which an arc fault can persist is 8 half cycles or 4 cycles on a 60Hz supply according to standard UL1699.
Testing done by UL in the United States has shown that the parallel resistance is only 0.03 ohms; page 337.
https://www.nfpa.org/assets/files/AboutTheCodes/70/70_A2022_NEC_P02_FD_PIReport_1of2_rev.pdf
When maximum Zs, 0.03 ohms arc impedance and the typical length of flex resistance are summed together the resulting current flow universally clears the fuse fast enough that the time (incident energy) does exceed those set by UL1699 and IEC-62606.
What folks in the UK have to realize is that before AFCIs were created in the US, the EU and British systems were extensively studied and determined to significantly limit short circuit energy (sputtering) which was theorized (key word) to be behind a sizable percentage of US residential electrical fires.
Where it all went horribly wrong was instead of mandating set earth fault loop impedance requirements in NFPA-70, research went out to determine the lowest reasonable short circuit current that may be found in a dwelling as is- which was discovered to be around 75 amps. And so the original concept was to lower the magnetic pickup on 15 and 20 amp breakers to 75 amps. When it was realized this would result in nuisance tripping on high inrush items like motors, tools and ballasts the electronic AFCI was created so tripping would be accomplished via AND logic looking for both ripple & magnitude instead of magnitude alone to achieve security. Although we know said security was not so- every time someone switched on a vaccuum cleaner with a brushed motor it was (and still is) another callback for the electrician.
Considering that Zs values almost always permit current flow over 125% of an MCB magnetic pickup there is absolutely no parallel event in the UK that will trip an AFDD but won't also trip an MCB.
Which leaves us with serial events. Serial events are the end stage of joule heating and rarely if ever occure on their own. The only way to stop a fire occurring from joule heating is by catching the event in its infancy- detecting local temperature rise before it reaches 1000*C. With AFDDs by then its to late, the only hope being the glowing connection will not have ignited any material long enough to produce a current signature.
