Arc-fault interrupters (AFIs) are starting to show up in the marketplace and are under consideration within NEC (National Electrical Code) regulations. They are the next big thing in circuit protection following the introduction of ground fault interrupters (GFIs). The AFI is intended to detect arcing in a circuit when a connection node deteriorates. Current flowing to the load arcs over microscopic gaps in the conduction path. This is obviously undesirable.
A bit of background: A GFI monitors the AC current flowing into and out of a load. The load can be anything in your house, a commercial setting, or an industrial setting. This includes (but is not limited to) incandescent or fluorescent lights, refrigerators, hairdryers, power tools, air compressors, etc. Current is monitored in various modes depending on how power is supplied. In a simple residential environment, current is simultaneously monitored in the “hot” line and the neutral line. As long as they match in amplitude (but opposite in polarity), the assumption is that no current is sneaking away through an earth ground path.
This same logic can be applied to loads connected across a 240V source (e.g., your electric range; both lines are “hot” with respect to ground) or to loads connected across a 3-phase delta connection. In all cases, the detection circuitry simply looks for an imbalance in current flows, which indicates a ground fault.
While the method for detecting a ground fault is simple and readily lends itself to analog techniques, arc-fault detection is the opposite. To properly detect an arc that is a fault, it's necessary to characterize the signature of the arc. Current is still flowing, but it is probably no longer sinusoidal or even vaguely cyclical. We could assume there is more high-frequency content present. One method of purely analog detection would be to use a small current transformer to monitor the hot line's current. Pass that signal through a high-pass filter; then rectify and filter the result and monitor with a comparator.
This would probably detect arc faults. Unfortunately, it would also detect the opening of switches and relay contacts that are controlling inductive loads. Strictly speaking, this arcing fits the general definition of an arc fault, but it's very short term and can be ignored. So we could add a time delay to the circuitry described above. But what about motors that use brushes on a commutator (think vacuum cleaner and blender motors)? They produce continuous arcing, so the time delay won't help.
I've read a little of the available technical literature on AFIs. I've also had a few discussions with design engineers who are working on AFI products. The trick here seems to be to characterize the arc phenomena. The means to do this, though it pains me to admit it, go beyond what can be done using purely analog techniques — at least given cost and PC board size limitations.
To characterize the arc, start again with the current transformer as above. That's where the analog portion ends, because the output of the current transformer goes to a DSP IC where the magic occurs. It's magic because designers of AFIs are loath to disclose the techniques used for detection. Like most software or firmware, the intellectual property represents lots of work, and manufactures don't want to share it.
In spite of that limitation, I will try to get more information on the techniques used and share them. I welcome comments from anyone who can share some of their knowledge.