Recently, I needed to dig up a paper I wrote more than a decade ago for a conference in the United Kingdom. Part of the paper was devoted to expert systems that were beginning to gain momentum as a possible way to solve power quality problems, rather than just provide more data or information. In the near future, such promising systems would provide answers to questions, such as, “Is the waveform in the figure below the result of a single line-to-ground fault on the distribution system?” Responding to the need to remove the human from having to analyze megabytes of trend and event data (now gigabytes in today’s power quality monitoring systems), a number of projects at universities and a few commercial products had incorporated advanced mathematical techniques, such as neural nets, fuzzy logic and rules-based expert systems.
By developing connections within the network, based on the data on which it is trained, and modifying the network parameters (weighting factors) to minimize error, neural networks emulate the manner by which the human brain learns. However, a specific neural network architecture is required to detect a particular type of (power quality) disturbance and, in general, is not appropriate for detecting and identifying other types of disturbances. Software programs using fuzzy logic combined with rules-based expert systems have seen some commercial success, using a set of rules and weighting factors to make decisions, rather than the traditional binary if/then logic. The accuracy of the system is based on the knowledge of human experts who determine the rules and the weighting factors; hence, it is only as good as the validity of the rules.
Generally, expert system accuracy needs to be 95 percent or better for people to feel confident in using it. Too many inaccurate answers plagued some of the earlier power quality expert systems with confidence factors of 75 percent or less, and the low confidence factors affected market acceptance. Many of the projects have remained in the university research realm. However, there have been intelligent systems algorithms implemented in products, such as harmonic source determination, capacitor-bank switching transient waveform recognition and direction, voltage-sag waveform recognition and direction, and fault location on radial and mesh networks. But it looks like it is still up to us humans to be the experts in analyzing the squiggly lines and lists of minimum/maximum/average values of numerous parameters.
So why not become an expert? It really isn’t that difficult if one remembers the two golden rules that I have covered in previous articles (Ohm’s and Kirchoff’s Laws), and if you break the data down into small bits and consider each one separately, then look at how the system transgressed from one to the other, it’s pretty understandable. Here, I have broken it down into six segments, A to F.
Background: The voltage waveform shows a voltage sag at the service entrance in a facility on a wye circuit where the monitor was connected to line-to-line voltages.
Stage A—prefault: the voltage waveform peak is 680 volts (V), which covers to 482V (multiple peak 0.707 for sinusoidal waveforms), which is consistent with it being a 277/480V service. The waveform looks very sinusoidal, indicating little harmonic content, if any.
Stage B—fault begins: the sag begins suddenly at approximately 270 degrees on the sine wave, not at a peak. This probably rules out a voltage breakdown condition from tree branch contact or the like, which usually would occur near the peak of the waveform. There is some arcing apparent that lasts almost half a cycle before decreasing to a more sinusoidal signal but at a reduced amplitude.
Stage C—mid-fault: the peak value is approximately 390V, which is close to 277V rms, the line-to-ground value.
Stage D—fault ending: the fault lasts less than one cycle before a large arcing signal appears that decays in less than a quarter cycle. This arcing signature is characteristic of what happens inside a fuse when it blows, as the molten metal provides a discontinuous path as the fuse vaporizes, until it is all deposited on the interior walls of the fuse and current flow ceases.
Stage E—fault cleared: as abruptly as it started, the fault is cleared, and the voltage returns to the prefault sinusoidal pattern.
Stage F—return to normal: a small oscillating transient occurs upon recovery, not like one would see from a power-factor capacitor switch transient. After that, things look like they did before the fault.
Conclusions: the line-to-line voltage, ~480V, sagged to nearly line-to-ground potential, ~277V, indicating the cause of the sag would be a single-line-to-ground fault. Based on the waveforms showing that the voltage returned to normal after one cycle, the fault was most likely cleared by the operation of a fuse (since protective relays generally do not operate that quickly and by evidence of arcing waveforms characteristic of a fuse blowing). These are all situations found on the utility distribution system. The location of the facility was not on the feeder where the fuse was, or the voltage would have dropped to zero when the fuse blew.
It only takes a little practice while remembering the rules to become an expert in power quality analysis.
BINGHAM, a contributing editor for power quality, can be reached at 732.287.3680.