External disturbance, e.g. around the 23 MHz range, shouldn't be such a trouble as it might sound. Descriptions I've seen of the algorithms suggest they look at e.g. the correlation of signal in this band to the ac cycle (the noise should have big changes in strength during the start and stop of arcing in each half-cycle), or at least that there should be abrupt changes. It's not just a matter of there being RF noise in that band. For more background it would be worth looking at the manufacturer's brochure that I linked to earlier (in thread  AFDD AMD2: it was this file 'Primer' - see Chapter 7).


Broadgage's case is interesting. I wouldn't quite say that a single case proves nothing - unless it's not reliably reported. If some existing AFDDs really have tripped for an upstream fault - even in the tiny currently-installed population - it helps support the very plausible reasoning that has gone on already in this thread. We can't be sure exactly how different models will behave, but it's worth noting even if one model does this.

The link above (p31) shows no use of voltage signal to assess 'fault direction' (and if it did, perhaps we'd complain that the AFDD wouldn't work well on a series fault on a subcircuit to a PV inverter). 

However, the AFDD is electronic, powered from the supply voltage. If the upstream fault causes big enough breaks in supply to cause the AFDD electronics to reset its thought-process before tripping, that would make it less sensitive than for a downstream fault. 

It's not altogether a bad feature if AFDDs trip on upstream arcing, as this could provide a clear indication of a meter/service problem worth fixing, when multiple AFDDs trip. On the down-side, dependence on the AFDD make/model and the connected load could result in just one downstream AFDD tripping, resulting in fault-finding being done at the wrong side of the AFDD; and a householder wouldn't be delighted to return from holiday to find that an earlier fault in the street had caused their boiler/freezer to be off for a week. 


I tried an AFDD just now on the table. Load: a 2 kW heater.  Arc: pencil-lead to copper, having given up the copper-copper (others who've tried this a lot can get 'loose contact' arcs with just copper, tripping some of the AFDDs consistently; but I'm not so practised).  It tripped several times with the arc on the load side. But not with it on the source side, so far. That could well be just because it's hard to avoid having some times when the AFDD loses its power. The brief tests were stopped due to time constraint and worry about UV exposure. 


Another case of interest is faults in parallel circuits. The IEC standard requires the AFDD not to trip when arcing happens in a parallel circuit (connected upstream of the AFDD). This is tested with the AFDD feeding a resistive load, yet loads with capacitance are common (power-supply filters or power-factor compensation) and should be better at attracting some high frequencies from the arcing in another circuit to pass through their AFDD. 


A question came up a few days ago about arcing on the secondary side of a transformer, when its primary side is supplied from an AFDD. This was something we tried recently, using a ~1:1 transformer without any shielding between the windings, with the secondary supplying a 2 kW resistive heater in series with a copper-copper loose contact (manipulated to get the arc). The AFDD on the primary tripped for this, about as readily as when the transformer wasn't in the way. This is not surprising, although a very different design of transformer could reduce how much of the high frequency of the current gets transferred to the primary.

External disturbance, e.g. around the 23 MHz range, shouldn't be such a trouble as it might sound. Descriptions I've seen of the algorithms suggest they look at e.g. the correlation of signal in this band to the ac cycle (the noise should have big changes in strength during the start and stop of arcing in each half-cycle), or at least that there should be abrupt changes. It's not just a matter of there being RF noise in that band. For more background it would be worth looking at the manufacturer's brochure that I linked to earlier (in thread  AFDD AMD2: it was this file 'Primer' - see Chapter 7).


Broadgage's case is interesting. I wouldn't quite say that a single case proves nothing - unless it's not reliably reported. If some existing AFDDs really have tripped for an upstream fault - even in the tiny currently-installed population - it helps support the very plausible reasoning that has gone on already in this thread. We can't be sure exactly how different models will behave, but it's worth noting even if one model does this.

The link above (p31) shows no use of voltage signal to assess 'fault direction' (and if it did, perhaps we'd complain that the AFDD wouldn't work well on a series fault on a subcircuit to a PV inverter). 

However, the AFDD is electronic, powered from the supply voltage. If the upstream fault causes big enough breaks in supply to cause the AFDD electronics to reset its thought-process before tripping, that would make it less sensitive than for a downstream fault. 

It's not altogether a bad feature if AFDDs trip on upstream arcing, as this could provide a clear indication of a meter/service problem worth fixing, when multiple AFDDs trip. On the down-side, dependence on the AFDD make/model and the connected load could result in just one downstream AFDD tripping, resulting in fault-finding being done at the wrong side of the AFDD; and a householder wouldn't be delighted to return from holiday to find that an earlier fault in the street had caused their boiler/freezer to be off for a week. 


I tried an AFDD just now on the table. Load: a 2 kW heater.  Arc: pencil-lead to copper, having given up the copper-copper (others who've tried this a lot can get 'loose contact' arcs with just copper, tripping some of the AFDDs consistently; but I'm not so practised).  It tripped several times with the arc on the load side. But not with it on the source side, so far. That could well be just because it's hard to avoid having some times when the AFDD loses its power. The brief tests were stopped due to time constraint and worry about UV exposure. 


Another case of interest is faults in parallel circuits. The IEC standard requires the AFDD not to trip when arcing happens in a parallel circuit (connected upstream of the AFDD). This is tested with the AFDD feeding a resistive load, yet loads with capacitance are common (power-supply filters or power-factor compensation) and should be better at attracting some high frequencies from the arcing in another circuit to pass through their AFDD. 


A question came up a few days ago about arcing on the secondary side of a transformer, when its primary side is supplied from an AFDD. This was something we tried recently, using a ~1:1 transformer without any shielding between the windings, with the secondary supplying a 2 kW resistive heater in series with a copper-copper loose contact (manipulated to get the arc). The AFDD on the primary tripped for this, about as readily as when the transformer wasn't in the way. This is not surprising, although a very different design of transformer could reduce how much of the high frequency of the current gets transferred to the primary.