8-2 F60 Feeder Protection System GE Multilin8.1 HIGH-IMPEDANCE (HI-Z) FAULT DETECTION 8 THEORY OF OPERATION88.1.3 RANDOMNESS ALGORITHMThe Randomness algorithm monitors the same set of component energies as the Energy algorithm. However, rather thanchecking for a sudden, sustained increase in the value of the monitored component energy, it looks for a sudden increasein a component followed by highly erratic behavior. This type of highly erratic behavior is indicative of many arcing faults.Just as with the Energy algorithm, if the Randomness algorithm detects a suspicious event in one of its monitored compo-nents, it reports it to the Expert Arc Detector algorithm, resets itself, and continues to monitor for another suspicious event.8.1.4 EXPERT ARC DETECTOR ALGORITHMThe purpose of the Expert Arc Detector Algorithm is to assimilate the outputs of the basic arc detection algorithms into one"arcing confidence" level per phase. Note that there are actually 24 independent basic arc detection algorithms, since boththe Energy Algorithm and the Randomness Algorithm are run for the even harmonics, odd harmonics, and non-harmonicsfor each phase current and for the neutral. The assimilation performed by the Expert Arc Detector Algorithm, then, isaccomplished by counting the number of arcing indications determined by any one of the twenty-four algorithms over ashort period of time (e.g. the last 30 seconds). Also taken into account is the number of different basic algorithms that indi-cate arcing.In the Expert Arc Detector Algorithm, the arcing confidence level for each phase increases as the number of basic algo-rithms that indicate arcing (per phase) increases. It also increases with increasing numbers of indications from any onebasic algorithm. These increases in confidence levels occur because multiple, consecutive indications from a given algo-rithm and indications from multiple independent algorithms are more indicative of the presence of arcing than a single algo-rithm giving a single indication.8.1.5 SPECTRAL ANALYSIS ALGORITHMThe Spectral Analysis algorithm is the third and final confirmation algorithm performed only when a high impedance condi-tion is suspected.The Spectral Analysis algorithm receives five seconds of averaged non-harmonic residual current spectrum data and com-pares it to an ideal 1 / f curve. Depending on the result, three percent can be added to the arcing confidence level gener-ated by the Expert Arc Detector Algorithm.8.1.6 LOAD EVENT DETECTOR ALGORITHMThe Load Event Detector Algorithm examines, on a per-phase basis, one reading of RMS values per two-cycle interval foreach phase current and the neutral. It then sets flags for each phase current and for the neutral based on the followingevents:• an overcurrent condition• a precipitous loss of load• a high rate-of-change• a significant three-phase event• a breaker open condition.These flags are examined by the Load Analysis Algorithm. Their states contribute to that algorithm's differentiation betweenarcing downed conductors and arcing intact conductors, and inhibit the Expert Arc Detector Algorithm from indicating theneed for an arcing alarm for a limited time following an overcurrent or breaker open condition.Any of the above five flags will zero the Expert Arc Detector buffer, since the power system is in a state of change and thevalues being calculated for use by the Energy and Randomness algorithms are probably not valid.An extremely high rate of change is not characteristic of most high impedance faults and is more indicative of a breakerclosing, causing associated inrush. Since this type of inrush current causes substantial variations in the harmonics used bythe high impedance algorithms, these algorithms ignore all data for several seconds following a high rate-of-change eventthat exceeds the associated rate-of-change threshold, in order to give the power system a chance to stabilize.