Series arc detection firmware

Abstract

A faulty meter socket detection system is disclosed that is configured to detect a faulty socket as demonstrated by a series arc in an input current waveform or an elevated temperature of a meter. The system includes a metrology circuit configured to receive an input current waveform; and a controller including a power monitoring module. The meter may also include a temperature sensor. The power monitoring module is configured to detect a series arc current in the input current waveform data by: filtering the input current waveform data to produce filtered data; performing feature recognition on the filtered data; features of the filtered data are classified; performing time processing on the filtered data; optionally, scaling the temporal processing of the filtered data; and scoring the input current waveform based on temporal processing of the filtered data. The temperature sensor may detect an elevated temperature of the meter, which may indicate a faulty socket.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims the benefit of U.S. Provisional Patent Application No. 63 / 333,883, filed on April 22, 2022, the entire contents of which are incorporated herein by reference. Technical Field

[0003] Embodiments disclosed herein relate to methods and products for determining the integrity of an electrical outlet. Summary of the invention

[0004] Meters connected to a distribution network may have faulty sockets. The long-term effects of a faulty socket may be electrical socket failure, electrical faults, or even fire. Two causes of faulty sockets are arcing and high temperatures. If there is a connection fault to the connected load, an arc may occur across the faulty connection between the distribution network and the load. Arcs that occur in series may be referred to as series arcs. Series arc faults are difficult to detect because they typically do not cause current to flow through the grounded conductor or neutral conductor. However, detecting these series arcs can help identify faulty sockets. High temperatures can cause desoldering, the connection problems mentioned above, and other connection problems. As with series arcs, detecting high temperatures in the meter can also help identify faulty sockets.

[0005] In some embodiments, a series arc detection system is configured to detect a series arc in an input current waveform. The series arc detection system includes: a metering circuit configured to receive an input current waveform; and a controller including a power monitoring module configured to detect an input current waveform with a series arc current. The controller detects the series arc current by filtering the input current waveform data to produce a filtered waveform; performing feature recognition on the filtered waveform; classifying features of the filtered waveform; performing time processing on the filtered waveform; generating a score for the input current waveform based on an integration of the filtered waveform; and reporting the score to a user based on a comparison of the score with a programmable threshold.

[0006] In some embodiments, a method for determining socket integrity includes receiving an input current waveform via a metering circuit; and detecting a series arc current in the input current waveform via a controller including a power monitoring module by: filtering the input current waveform to produce a filtered waveform; performing feature recognition on the filtered waveform; classifying features of the filtered waveform; performing time processing on the filtered waveform; generating a score for the input current waveform based on an integration of the filtered waveform; and reporting the score to a user based on a comparison of the score with a programmable threshold.

[0007] In some embodiments, a meter in a power distribution network includes a metering circuit that communicates with at least one temperature sensor and is configured to receive an input current waveform and determine a temperature of the meter. The meter also communicates with a controller that includes a power monitoring module that is configured to detect an input current waveform with a series arc current by filtering the input current waveform data to produce a filtered waveform; applying a root mean square function to the input current waveform to produce a root mean square (RMS) value; performing feature recognition on the filtered waveform; classifying features of the filtered waveform; performing time processing on the filtered waveform; generating a score for the input current waveform based on an integration of the filtered waveform; determining a temperature value for the meter; using a fuzzy inference classifier to determine an alarm score based on the score for the input current waveform, the RMS value, and the temperature value for the meter; and reporting the alarm score to a user based on a comparison of the score with a programmable threshold.

[0008] Other aspects of the present technology will become apparent by consideration of the detailed description and accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 An electric distribution system is shown including an electric distribution network connected to meters of metered buildings.

[0010] Figure 2 A block diagram of a metering system is shown.

[0011] Figure 3 A graph showing the input current waveform of an arcing AC signal received at a meter from a power distribution network.

[0012] Figure 4 The correlation of the current waveform of an arc AC power signal received at a meter from a power distribution network with the high frequency component of the AC power signal is shown.

[0013] Figure 5 A schematic diagram of a system configured to detect series arcing occurring on a connection between a meter and an electrical distribution network is shown.

[0014] Figure 6 An input current waveform is shown including a series of shoulders at zero-crossings of the input current waveform.

[0015] Figure 7 A classification module in the form of a multilayer perceptron neural network is shown.

[0016] Figure 8 The output of a multilayer perceptron neural network is shown.

[0017] Fig. 9 A flow chart showing an embodiment in which the output of a classification module is temporally processed as input by a temporal processing module.

[0018] Fig.10 A graph showing the output of a temporal processing module as a result of the temporal processing module operating in response to the output of a classification module.

[0019] Fig.11 A graph of a sigmoid function configured to scale the output of a temporal processing module is shown.

[0020] Fig.12 A graph of the scaled temporal processing module output corresponding to the classification module output is shown.

[0021] Fig.13 A flow chart for processing an input current waveform from a power distribution network via a power monitoring module is shown.

[0022] Fig.14 A schematic diagram of a system configured to detect the occurrence of a series arc or high temperature in a meter on an electrical distribution network is shown.

[0023] Fig.15 A flow chart for processing an input current waveform from a power distribution network via a power monitoring module is shown.

[0024] Fig.16 A set of membership functions used by a fuzzy inference classifier to fuzzify a crisp input current waveform is shown.

[0025] Fig.17 A parametric model of the arc is shown.

[0026] Fig.18 A graph of the current of the arc over time is shown.

[0027] Fig.19 A parametric model of the arc fit is shown with a graph of the arc.

[0028] Fig. 20 A parametric model of the arc fit is shown with a graph of the arc.

[0029] Fig.21 A block diagram illustrating a pulse train detection algorithm approach for series arc detection using a folding algorithm is shown.

[0030] Fig. 22 A decibel versus hertz graph of the arc is shown.

[0031] Fig.23 A decibel versus hertz graph of the arc is shown.

[0032] Fig.24 The frequency response of the Butterworth high pass filter cutoff frequency is shown.

[0033] Fig.25 The signal filtered by a Butterworth high pass filter is shown.

[0034] Fig.26 A flow chart showing the approach of the power spectral density method for arc detection.

[0035] Fig. 27 A graph of a bispectrum for an input current waveform excluding arc current is shown.

[0036] Fig.28 A graph of a bispectrum for an input current waveform including an arc current is shown.

[0037] Fig.29 A graph showing the Cepstrum of the input current waveform is shown.

[0038] Fig.30 A graph of the power spectral density of an example input current waveform is shown.

[0039] Fig.31 The magnetic sensor output in response to sensing an arc in spherical coordinates is shown.

[0040] Fig.32 The magnetic field for an EMF plane wave is shown.

[0041] Fig.33 Another magnetic sensor output in response to sensing an arc in spherical coordinates is shown. DETAILED DESCRIPTION

[0042] Before explaining any embodiment of the present application in detail, it should be understood that the present application is not limited in its application to the details of the construction and arrangement of the components set forth in the following description or shown in the following drawings. The present application is capable of other embodiments and can be practiced or implemented in various ways.

[0043] In addition, it should be understood that the words and terms used herein are for descriptive purposes and should not be considered limiting. The use of "including," "comprising," or "having," and variations thereof, is meant to cover the items listed thereafter and their equivalents as well as additional items. Unless otherwise specified or limited, the terms "mounted," "connected," "supported," and "coupled," and variations thereof, are used broadly and cover direct and indirect mounting, connection, support, and coupling. As used within this document, the word "or" may mean inclusiveness. As a non-limiting example, if it is stated in this document that "item Z may include element A or B," this may be interpreted as disclosing item Z that includes only element A, item Z that includes only element B, and item Z that includes elements A and B.

[0044] A plurality of hardware- and software-based devices, as well as a plurality of different structural components may be used to implement various embodiments. In addition, embodiments may include hardware, software, and electronic components or modules, which, for the purposes of discussion, may be shown and described as most components being implemented only in hardware. However, a person of ordinary skill in the art will recognize, based on a reading of this detailed description, that in at least one embodiment, the electronic-based aspects of the present invention may be implemented in software executable by one or more electronic processors (e.g., stored on a non-transitory computer-readable medium). For example, the "control unit" and "controller" described in the specification may include one or more electronic processors, one or more memory modules including a non-transitory computer-readable medium, one or more input / output interfaces (I / O interfaces), one or more application specific integrated circuits (ASICs), and various connectors (e.g., a system bus) connecting various components.

[0045] Figure 1 A power distribution system 100 is shown including a distribution network 101 connected to a meter 112 of a metered building 104. The power distribution system 100 includes one or more distribution lines 106, a step-down transformer 108, and a feeder line 110. The feeder line 110 supplies alternating current (AC) power to the metered building 104 via the power meter 112. It is contemplated that the metered building 104 may be a residential, commercial, and / or industrial building. Figure 1 As shown, a meter 112 is coupled to the feeder line 110 and is configured to monitor parameters of the power supplied from the distribution line 106 (which in turn is provided by a utility such as a fossil fuel-based power plant, a nuclear power plant, a wind turbine, a solar collector, etc.) The parameters may include energy usage, power factor, voltage level, current level, etc.

[0046] In some embodiments, the meter 112 may be configured to communicate with one or more utility computer systems 114. The utility computer system 114 may be a collection device for communicating with multiple meters 112 in a region to collect data related to energy use of one or more metered buildings 104, and may communicate wired or wirelessly with a utility's substation 116. In some embodiments, the meter 112 may communicate with the utility computer system 114 via a wireless connection, such as RF (e.g., cellular or other RF communication types). However, other communication protocols are also contemplated and will be described in more detail below. In other examples, the meter 112 may communicate with the utility computer system 114 via a wired connection, such as, but not limited to, power line communication, fiber optic communication, or Internet communication. In some examples, the utility computer system 114 is a portable system operated by one or more utility personnel who move the utility computer system 114 within the range of one or more meters 112 in a given region, thereby allowing communication to be established between the meter 112 and the utility computer system 114. In other examples, utility computer system 114 may be a fixed system or may have multiple fixed receivers across the utility distribution area to allow communication between meter 112 and utility computer system 114 .

[0047] Now turn to Figure 2 , a block diagram of a metering system 200 according to some embodiments is shown. In one embodiment, the metering system 200 is similar to the above-mentioned Figure 1 The meter system 200 may include a meter 212 and a communication module interface 226. The meter 212 may include a processing circuit 222, a metering circuit 220, and a communication module interface 226. The processing circuit 222 may include one or more electronic processors 224 and one or more memory devices 228. The processing circuit 222 may be communicatively connected to one or more of the processing circuit 224 and the communication module interface 226. The electronic processor 224 may be implemented as a programmable microprocessor, an application specific integrated circuit (ASIC), one or more field programmable gate arrays (FPGA), a group of processing components, or implemented using other suitable electronic processing components.

[0048] Metering circuit 220 may be configured to meter the current drawn by metered building 204. Metering circuit 220 may also be configured to generate waveform data that models the input current. Such waveform data may be referred to herein as an "input current waveform" and may be directly associated with the input current waveform received from distribution network 201 at meter 212.

[0049] Memory 228 (e.g., non-transitory computer-readable media) includes one or more devices (e.g., RAM, ROM, flash memory, hard disk storage, etc.) for storing data and / or computer code for completing or facilitating the various process layers and modules described herein. Memory 228 may include database components, object code components, script components, or other types of code and information for supporting the various activities and information structures described herein. According to one example, memory 228 is communicatively connected to electronic processor 224 via processing circuit 222 and may include computer code for executing (e.g., by processing circuit 222 and / or electronic processor 224) one or more processes described herein.

[0050] The processing circuit 222 is configured to be coupled to at least a portion of the utility power associated with the metering system 200. The processing circuit 222 may be configured to execute modules stored in the memory 228. In one embodiment, the processing circuit 222 is configured to provide measured data to the power monitoring module 230 within the memory 228. The power monitoring module 230 is configured to determine one or more parameters associated with the utility power monitored by the processing circuit 222. For example, the power monitoring module 230 may determine power usage (e.g., kW, kWh, VAR, VA, etc.) based on the data measured by the processing circuit 222. In other examples, the power monitoring module 230 may be configured to determine other parameters, such as power factor, phasor data, phase balance, and other applicable parameters. As will be described in further detail below, the power monitoring module can also be configured to detect electrical outlet integrity problems (e.g., series arcing) caused by loose connections between the meter 212 and the distribution network 201, oxidized contacts, foreign objects interfering with the connection, differences in contact materials between the feeder 110 and the distribution network 201, and high meter temperatures that tend to indicate a faulty outlet.

[0051] The communication interface 226 is configured to provide communication between the meter 212 and an external device, such as the utility computer 214. In the illustrated embodiment, the communication interface 226 is disposed internal to the housing of the meter 212. In other embodiments, the communication interface 226 is external to the housing of the meter 212 and communicates with the meter 212 via a wired or wireless connection.

[0052] In some examples, the communication interface 226 can support various communication modes, such as RF, Bluetooth, cellular (e.g., 3G, 4G, 5G, CDMA, etc.), RF, Wi-Fi, NFC, power line communication (e.g., TWACS, PRIME, etc.), and the like. The communication interface 226 can be or include a wireless communication interface (e.g., antenna, transmitter, receiver, transceiver, etc.) for communicating data between the meter 212 and the utility computer system 214. In addition, the communication interface 226 can also use a communication protocol, such as cellular (3G, 4G, 5G, LTE, CDMA, etc.), Wi-Fi, LoRa, LoRaWAN, Z-wave, Thread, power line communication, short hop radio, and / or any other applicable wireless communication protocol.

[0053] Figure 3 Graph 300 depicts an input current waveform 302 of an arc AC signal received at meter 212 from distribution network 201. The power provided by distribution network 201 to meter 212 may be an AC signal having a frequency of 16 KHz or less. Figure 3 As shown, waveform 302 is an AC signal containing arc current. If the connection between meter 212 and distribution network 201 is faulty, arc current may occur, causing charge to accumulate across the insulator (e.g., air) disposed between the associated conductors of the meter and the associated conductors of the distribution network 201 until an arc is generated. In particular, the AC signal may begin to cause an arc on the faulty connection. The faulty connection may include a loose connection between meter 212 and distribution network 201, an oxidized contact, a foreign object interfering with the connection, a difference in contact material between feeder 110 and distribution network 201. These arcs may be identified in the input current waveform 302 of the AC signal in the form of a "shoulder" 304 at the zero crossing 306 every half cycle 308 (e.g., a zero current period before or after a rapid increase in current, and sometimes both).

[0054] The shape of the input current waveform 302 can depend on factors such as the arc gap size, the load impedance (e.g., the impedance of the meter), the fault connection, the degree of oxidation of the electrical contacts, and the temperature of the fault connection. These factors may particularly affect the input current waveform 302 in the area around the zero crossing. Oxidation of the connection surface can produce zero-current periods in the input current waveform 302, resulting in a flat input current waveform 302. In particular, the higher the degree of oxidation, the greater the voltage required to produce an arc. A large arc gap size also results in zero-current periods in the current received by the meter 212. An increase in the inductive load impedance can reduce the size of the zero-current shoulder. Changes in the gap temperature can affect the voltage required to produce an arc. In some cases, due to the increased temperature at the fault connection, molten copper can form a bridge across the fault connection, thereby creating a condition where the current flow is not interrupted and the arc discharge is therefore weakened. For example, in Figure 3 In the graph, the arc discharge ceased between 38 ms and 48 ms, which may indicate that the conductive bridge was temporarily formed by the molten metal.

[0055] Typically, during a series arc fault, the flat portion of the current waveform at the zero crossing 306 may not be consistent in size or appearance for all half cycles. In addition to the flat portion at the zero crossing 306, the input current waveform 302 may also exhibit steep current edges immediately following the zero current region of the input current waveform 302. The steepness of these edges may depend on the current gap length and the load impedance.

[0056] Analyzing the input current waveform 302 of the AC signal for these and other features can help identify arc currents in the input current waveform and thereby help prevent fires or other hazardous conditions that may be caused by arcs in the current signal. Therefore, in some embodiments, the power monitoring module 230 can first gate and filter the input current waveform 302, then perform feature recognition (including pulse train detection in some cases) on the filtered waveform, scale the identified features, classify the features of the filtered waveform, perform time integration on the classified features, consider additional factors, and generate an arc score by scoring the input current waveform 302 based on the integrated features and the additional factors.

[0057] Figure 4 Depicted is a correlation 400 of a current waveform 402 of an arc AC power signal received at meter 212 from distribution network 201 and a high frequency (HF) component 403 of the AC power signal. Figure 4As can be seen in FIG. 4 , each time a shoulder appears in the current waveform 402, a high frequency component of the current appears. This high frequency component of the current waveform 402 can be obtained by applying a filter to the current waveform 402. Other components or forms of the current waveform 402 can also be obtained by filtering the current waveform 402. The nature of the series arc fault current data indicates that the high frequency component of the signal is transient and only exists at a specific time during the signal. The filtering operation for highlighting such high frequency content can be configured to capture both the frequency and time content of the arc fault characteristics. A single filter with an ideal impulse response and frequency response can be used to highlight a target high frequency signal with a balance of time and frequency characteristics. The process of filtering the input current waveform 302 will be described below.

[0058] Figure 5 A schematic diagram 500 of a system configured to detect whether a socket of a meter 212 is faulty by identifying a series arcing that occurs on a connection between the meter 212 and the distribution network 201 is shown. In the illustrated embodiment, a root mean square (RMS) filter 502 is used for initial filtering of the input current waveform 302. The RMS filter 504 can be used to eliminate non-arcing current waveforms from further filtering analysis. For example, the RMS filter 504 can obtain a root mean square for the input current waveform 302 as an initial step to exclude the possibility that the input current waveform 302 contains evidence of a series current arc. The RMS filter 504 can also be configured to flag or disqualify the input current waveform 302 in situations where arcing is unlikely (e.g., zero current input) or in situations where the classification module 524 is likely to misclassify the input current waveform 302 after filtering to avoid the possibility of generating a false positive result (e.g., detecting the socket as faulty when in fact the socket is not faulty). For example, if the RMS value of the input current waveform 302 is below a certain threshold (e.g., below 1 amp to 6 amps, etc.), or if a first RMS current cycle of the input current waveform 302 is significantly different from a second RMS current cycle of the input current waveform 302, then the classification module 524 is likely to misclassify the input current waveform 302. The ΔRMS value can be compared to the RMS threshold to filter out the featureless two-cycle portion of the input current waveform 302 before feature extraction. This comparison can be expressed as:

[0059]

[0060] Among them, RMS c1 It is the first cycle of the two-cycle part, RMS c2 is the second cycle of the two-cycle part, and T ΔRMS is the RMS threshold.

[0061] In the schematic diagram shown, an alignment filter is also used in conjunction with the RMS filter 504. The alignment filter is configured to filter out features that do not occur at the zero crossings of the input current waveform 302.

[0062] After the RMS filter 504, an additional filtering module 508 applies an additional filter to the RMS filtered input current waveform 302. In some embodiments, a relatively simple 60 Hz filter 510 is used to remove noise from the input current waveform 302 in anticipation of feature extraction. In such embodiments, a high pass filter, a band pass filter, or a notch filter may be used to remove portions of the input current waveform near or below 60 Hz.

[0063] In some embodiments, the Gabor filter 512 is used to extract a target high frequency signal with a desired balance of time / frequency characteristics. The time-bandwidth product (TBW) is the product of the time variance and the frequency variance, and is a measure of localization in both the time domain and the frequency domain. The Gaussian function has a desired TBW. Further, the Fourier transform of the Gaussian function is also a Gaussian function, and the transform pair can be expressed as:

[0064]

[0065] The impulse response of the Gabor filter 512 is defined as a complex sine wave multiplied by a Gaussian function.

[0066]

[0067] Thus, the impulse response of Gabor filter 512 is effectively a Gaussian window modulated to frequency F0 - a bandpass filter with center frequency W0 and bandwidth determined by σ. This type of filter has a desirably compact representation in both time and frequency. The output of Gabor filter 512 can be windowed to a two-cycle window using the maximum consecutive zeros function, which can be expressed as:

[0068]

[0069] After using the 60 Hz filter 510 or the Gabor filter 512, feature extraction is performed on the gated filtered input current waveform 302. The feature extractor 514 uses filters and functions to extract (e.g., isolate) features from the input current waveform 302, and the extracted features are optionally scaled by the scaling module 522 so that the features can be classified by the classification module 524 as indicating a series arc based on a classification threshold. The output of the classification module 524 is then time processed (e.g., integrated using a leaky integrator) by the time processing module 526 to produce an arc score at the scoring module 528. In some embodiments, feature extraction is performed on two consecutive cycles of the current data. In some embodiments, feature extraction is performed on multiple current channels (e.g., Ia and Ic in some meters). At the output of the additional filtering module 508, the feature extraction module 514 extracts features indicative of an arc current.

[0070] In some embodiments, at the output of the 60 Hz filter 510, a Normalized Mean Squared Error (NMSE) function 516 measures the difference between the first cycle 312 and the second cycle 314 in a two-cycle window 316 of data. The NMSE function 516 can be used to detect signal randomness caused by series arcing. At the input of the NMSE function 516, a high pass filter can be used to suppress the fundamental frequency of the input current waveform 302 (e.g., 60 Hz in North America and 50 Hz in Europe). In some embodiments, a high pass infinite impulse response filter or a high pass finite impulse response filter can be used as the high pass filter mentioned above. In some embodiments, the cutoff frequency of the high pass filter can be between 60 Hz and 120 Hz or 50 Hz and 100 Hz, but is not limited to these ranges and can include other frequencies or ranges.

[0071] The output of the HP filter can be processed using the NMSE function 516. This function 516 can be expressed as:

[0072]

[0073] Among them, x j -x j+k is the previous cycle, x j+k is the current cycle. Typically, the correlation coefficient is normalized by the number of points in the window of interest. For the purpose of analyzing a two-cycle window of the input current waveform 302, the denominator may therefore be scaled by 0.5.

[0074] In some embodiments, at the output of the Gabor filter 512, the high frequency energy 520 detects arc characteristics of the high frequency (e.g., a frequency of approximately 2500 Hz) at the zero crossing of the input current waveform 302 by using a high pass filter centered at approximately 2500 Hz and a windowing process on the output of the Gabor filter 512.

[0075] Additionally, in some embodiments, neither the 60 Hz filter 510 nor the Gabor filter 512 is used, and instead, the maximum consecutive zero function 518 measures the number of points in the input current waveform 302 that are consecutive and below the dynamic threshold 318 (e.g., below zero), and can be used to detect the current shoulder at the zero crossing. If the absolute value of the point is below the threshold, and the absolute value of the previous point of the point is also below the threshold, the zero count is incremented. If the consecutive points rise above the dynamic threshold, the string of consecutive zeros is interrupted. If the consecutive zero condition is later restored in the two-cycle window 310 of data mentioned above, the zero count is created and incremented. The result of this process can be one or more zero counts. The maximum consecutive zero function can be expressed as:

[0076]

[0077] Zeros=MAX∑C_Zeros

[0078] The dynamic threshold is expressed as T z ,|x L | is the point currently being considered, |x L-1 | is the previous point. In some embodiments, T z =α*RMS, where α≈0.03 to 0.08. The maximum count of consecutive zeros within a two-cycle window is identified within one or more zero counts. In some embodiments, the maximum consecutive zero function 518 is used in conjunction with the high frequency energy function 520 to perform feature extraction prior to (optional) scaling by scaling module 522, classification by classification module 524, and temporal processing by temporal processing module 526.

[0079] Figure 6 An input current waveform 602 is depicted including a series of shoulders 604 at zero crossings 605 of the input current waveform 602, indicating a series of arcs between the meter 112 and the feeder line 110. A filtered scaled version of the input current waveform 602 is shown overlaid on an unscaled filtered version of the input current waveform 602 during the same time span. Figure 6As can be seen in FIG. 6 , the peak 610 in the unscaled filtered version of the input current waveform 602 corresponds to the shoulder 604 of the input current waveform 602. Both the unscaled filtered version of the input current waveform 602 and the scaled filtered version of the input current waveform 602 contain the extracted feature 609. Figure 6 , the scaled filtered version of the input current waveform 602 is relatively flat between the shoulders 604. This is due to good filter design. In particular, the filter used in the illustrated embodiment is designed to isolate the time instances of the high frequency content input current waveform 602. In the illustrated embodiment, the filter is designed to respond optimally to broadband events (1kHz to 4kHz) by ensuring that the peak filter response occurs at the zero crossing 605. However, if the filter design is not robust enough, any discontinuity in the waveform may result in an unsuppressed filter output. In the example shown, the filter produces a suppressed output when the input current waveform 602 is not at a zero crossing, and produces an unsuppressed filter output near the zero crossing 605 of the input current waveform 602. As described above, this is desirable for feature extraction, and is also desirable for subsequent (optional) feature scaling by the scaling module 522 prior to classification at the classification module 524.

[0080] In some embodiments, a normalization function is also used before classification. The normalization coefficients can be calculated using only the feature data extracted from the non-arcing waveform. The feature data can then be normalized to create a feature data set with unit variance and a standard deviation of 1.0. The normalization coefficients are then applied to the feature data set. This application of the normalization coefficients creates a non-arcing data set with a symmetrical shape in the feature space.

[0081] refer to Figure 7, shows an embodiment of a classification module 524 in the form of a multilayer perceptron neural network 700 that is configured to classify features extracted from an input current waveform 602. The multilayer perceptron shown includes a plurality of inputs provided by feature extraction functions F1, F2, and F3. The hidden layer 702 includes a node for each feature extraction function F1, F2, F3 (e.g., 3 nodes in this example) and 2 additional nodes 704 for acting as additional unique neurons. However, the number of feature extraction functions and the number of additional nodes may vary. For example, if both the normalized mean square error function 516 and the high frequency energy function 520 are used to extract features, the multilayer perceptron may include 6 nodes, 2 of which correspond to feature extraction inputs and 4 nodes act as additional neurons. In the illustrated embodiment, the multilayer perceptron neural network 700 also includes a single output node 706 and a back propagation training function 708 trained on the output and test data of the multilayer perceptron neural network 700. In some embodiments, the nodes of the hidden layer 702 act as multiple unique neurons because the hidden layer 702 can be configured to analyze the input according to a classification threshold (e.g., an activation threshold) that can directly affect how likely the hidden layer 702 is to classify a given feature as evidence of arc current. The multilayer perceptron neural network 700 can be trained on the output feedback produced by the training function 708. The training function 708 can be configured to compare the output of the multilayer perceptron neural network 700 with verified test data.

[0082] The test data provided for training the multilayer perceptron neural network 700 may be UL1699 test data, laboratory generated test data (e.g., hot socket simulator data or other test circuit data), or simulated data. Performance metrics such as F1 scores or Mathews correlation coefficients (MCC) may be calculated based on the classification output of the multilayer perceptron neural network 700 and the test data. In some embodiments, the strictness of the classification module 524 may be automatically modified based on the MCC or F1 score by adjusting the classification threshold mentioned above. For example, in training the multilayer perceptron neural network 700, the training function 708 may adjust the classification threshold within a certain value range to determine a threshold that maximizes the MCC or F1 score. In some embodiments, the classification threshold may be adjusted based on user input. The output of the classification module 524 (e.g., the perceptron neural network 700) may be time processed (e.g., integrated over time) and scaled to generate an arc score (e.g., a confidence score) at the scoring module 528, which indicates the confidence that the input current waveform data 602 contains evidence of a series arc current.

[0083] like Figure 8As shown, the output of the multilayer perceptron neural network 700 may be a classification in the form of a square wave waveform 800 having a value of 1 during a two-cycle period of the input current waveform 802 that is classified by the multilayer perceptron neural network 700 as containing an arc current; and the waveform 800 having a value of 0 during a two-cycle period that is classified by the multilayer perceptron neural network 700 as not containing an arc current. However, it is contemplated that in some embodiments, these classification values ​​may be reversed.

[0084] Fig. 9 Flowchart 900 shows an embodiment in which the output of the classification module 524 is temporally processed as input 902 by the temporal processing module 904. The temporal processing module 904 is configured to integrate the output of the classification module 524 according to the following expression:

[0085] y(k)=x(k)+Py(k-1)

[0086]

[0087] y(t)=ke -Pt

[0088] Wherein, P is the leaky integrator decay parameter. The integrator is configured to produce a decay in the output when the input from the classification module 524 is equal to 0 (e.g., when the classification module 524 is not outputting positive classification pulses). In the illustrated embodiment, the value of P can be adjusted based on input (e.g., user input). As will be discussed further below, in some embodiments, the time processing module output can be scaled by a scaling function 906 and output as an arc score 908.

[0089] Fig.10 A graph 1000 of a time processing module output 1002 is shown as a result of the time processing module 904 operating according to the above formula in response to the output of the classification module 1004. In the illustrated embodiment, the time processing module output 1002 value decreases by half exponentially before each new pulse 1006 is received, which means that for Fig.10 , e -Pt is chosen to be equal to 0.5. The decay parameter P may not be easily related to the decay per unit time of the time processing module output 1002, at least in part because the two-cycle sampling period may not be exactly 1 second. The decay parameter P can be expressed as:

[0090]

[0091] Among them, t secis the two-cycle sampling period. The attenuation parameter P may be adjustable by user input and enables the user to set the minimum acceptable arc current detection rate. However, it can be determined from the Z-transform of the leaky integrator function that for stability, e -Pt Should be less than 1:

[0092]

[0093] Once per second, a pulse 1006 indicating a positive classification of the input waveform 802 is output from the classification module 524, indicating that once per second a two-cycle window (i.e., in the illustrated embodiment, every 60 AC power cycles, or every 30 samples) 310 of the input current waveform 802 is classified as containing evidence of arc current. The time processing module output decays by half for each passing second, which means that the selected P is:

[0094]

[0095] It should be noted that the sampling rate can be adjusted by the user by changing the decay rate P according to the above equation.

[0096] Fig.11 A graph 1100 of a sigmoid function 1102 configured to scale the output of the time processing module 904 is shown. In the illustrated embodiment, the sigmoid function 1102 is configured to scale the time processing module output 1002 to a value between 0 and 1. The scaled time processing module output 1002 is a scaled arc score 908. The scaled arc score 908 indicates the confidence level that the input current waveform data 802 includes evidence of a series arc current. The sigmoid can be expressed as:

[0097]

[0098] Fig.11 The sigmoid shown can be expressed as:

[0099]

[0100] The slope of the sigmoid function 1102 can be expressed as:

[0101]

[0102] Where s is the sigmoid scaling factor, t0 is the sigmoid offset, and T h is the value of the time processing block output to be scaled.

[0103] Fig.12A graph 1200 of a scaled temporal processing module output 1202 corresponding to a classification module output 1204 is shown. Although the scaled temporal processing module output 1202 fluctuates during processing by the classification module 524, the temporal processing module 526, and the temporal processing module output scaling function 906, it can be seen from Fig.12 As can be seen in FIG. 9 , arc score 908 can be output at any time during processing in response to certain triggering events. In the illustrated embodiment, arc score 908 is output in response to a new positive classification being received by time processing module 526 .

[0104] Fig.13 13 is a flow chart 1300 for processing an input current waveform 802 from a distribution network 101 via a power monitoring module 230. The distribution network 101 may be configured to deliver power 112 to a meter 212 via a transformer 108 in communication with a substation 116. The meter 212 may be configured to analyze the input current waveform 802 via the power monitoring module 230 for evidence of arcing and generate a scaled arc score 908 indicating a likelihood that an arc is occurring on a connection between the meter 212 and the distribution network 101. Execution of the power monitoring module 230 may include passing the input current waveform 802 through a plurality of processing modules, such as the alignment and RMS filtering module 504, the additional filtering module 508, the feature extraction module 514, the feature scaling module 522, the classification module 524, and the temporal processing module 526, to generate an arc score at the scoring module 528 described above.

[0105] At block 1302, meter 212 may receive input current waveform 802 from power distribution network 101 (e.g., via transformer 108 or substation 116) and model input current waveform 802 via power metering circuit 220. Meter 212 may begin analyzing input current waveform 802 for evidence of an electrical arc by executing power monitoring module 230 using processing circuit 222 to first process input current waveform 802 using a gating module included in RMS filter module 504.

[0106] At block 1304, the processing circuit 222 may perform gating on the input current waveform 802 using the alignment and RMS filtering module 504. The alignment and RMS filtering module 504 may be configured to cancel the input current waveform 802 if aspects of the input current waveform 802 do not conform to a predefined range of values ​​that indicates a high likelihood (e.g., 95% to 100% likelihood) that the input current waveform 802 lacks evidence of arc current. For example, as mentioned above, the alignment and RMS filter 504 may calculate a root mean square (RMS) for the input current waveform 302. The alignment and RMS filter 504 may also be configured to flag or cancel the input current waveform 302 if arcing is unlikely to occur (e.g., the instrument line is disconnected or shorted) or if the classification module 524 is likely to misclassify the input current waveform 302 after filtering. After gating and filtering the input current waveform 802, the processing circuit 222 may further filter the input current waveform 802 via the additional filtering module 508.

[0107] At block 1306, the processing circuit 222 may use the additional filtering module 508 to apply a filter (e.g., a 60 Hz filter 510, a Gabor filter 512, an HP IIR filter) to the input current waveform 802 to highlight or isolate portions of the input current waveform 802, and output an unweighted filtered version 606 of the input current waveform 602 so that the highlighted or isolated features can be further analyzed. The additional filtering module 508 may also apply weighting to the unweighted filtered version 606 of the input current waveform 602 to produce a weighted filtered version 608 of the input current waveform 602. After filtering the input current waveform 802, the processing circuit 222 may classify the unweighted filtered version 606 of the input current waveform 602 or the weighted filtered version 608 of the input current waveform 602 via the classification module 524.

[0108] At block 1308, the processing circuit 222 may perform feature extraction on the input current waveform 602 using the NMSE function 516, the maximum consecutive zeros function 518, or the high frequency energy function 520 to identify arc currents in a two-cycle window of the filtered input current waveform 602. After extracting features from the filtered input current waveform 602, the extracted features may be scaled.

[0109] At block 1310, the processing circuit 222 may scale the features extracted by the processor during feature extraction. The extracted features (scaled or unscaled) may then be classified.

[0110] At block 1312, the processing circuit 222 may use the multilayer perceptron neural network 700 of the classification module 524 to classify the one or more binomial data windows 310 of features extracted from the input current waveform 602 as containing or not containing features that evidence arc current in the input current waveform 602. This classification may be performed by the classification module 524 based on a classification threshold. The classification module 524 may be configured to output the voltage pulse 1006 if the extracted features contain sufficient evidence of arc current. After classifying the features extracted from the input current waveform 802, the processing circuit 222 may temporally process the output of the classification module 524 via the temporal processing module 526 by integrating the pulse 1006 received at the temporal processing module over time.

[0111] At block 1314, the processing circuit 222 may integrate the input pulses 1006 from the classification module 524 over time using a leaky integration method of the time processing module 526. The decay parameter of the leaky integration method may be adjusted, for example, based on user input. The time processing module 526 is configured to generate decay in the output when the classification module 524 is not outputting positive classified pulses 1006. After the output of the classification module 524 is time processed via the time processing module 526, in some embodiments, the processing circuit 222 may scale the time processing module output 1002 via the scaling module 906.

[0112] At block 1316, the processing circuit 222 may process the time processing module output 1002 using a scaling function of the scaling module 906 to define the time processing module output 1002 for the purpose of assigning an arc score 908 to the input current waveform 802. The scaling function may be a sigmoid function configured to take the time processing module output 1002 and output a corresponding value within a predetermined range of values ​​(e.g., a value between 01 and 1, a value between 0 and 10, etc.). After (optionally) scaling the time processing module output 1002 via the scaling module 906, the processing circuit 222 may generate an arc score for the input current waveform 802 via the scoring module 528. The arc score may be based not only on the time processing of the filtered input current waveform (e.g., the time processing module output 1002), but also on one or more of the sensed temperature of the measurement circuit 220, the current time of day, the determined arc gap size, and the determined degree of oxidation of the connection between the measurement circuit 220 and the load.

[0113] At block 1318, the processing circuit 222 may output the arc score 908 via the scoring module 528. The scoring module 528 may be configured to output the arc score 908 only when the scaled temporal processing module output meets or exceeds a predetermined value. For example, the scoring module 528 may be configured to only output an arc score 908 of 1 when the scaled temporal processing module output is .95 or higher.

[0114] At block 1320, the meter 212 may be configured to take action in response to the arc score 908 indicating evidence that an arc current has been detected in the input current waveform 802. For example, the meter 212 may disconnect itself from the power distribution network, or generate a notification in response to the arc score 908 indicating evidence that an arc current has been detected in the input current waveform 802 and transmit it to a headend via wired or wireless communication.

[0115] Fig.14 A schematic diagram 1400 of a system configured to detect whether a socket (not shown) connected to a meter 212 of a power distribution network 201 is faulty is shown. In the illustrated embodiment, the RMS value 1406 and the scoring module 1428 are based on the information about Figure 5 The method described above is used to generate the temperature data 1430 of the meter 212, and the temperature data is also collected and processed to identify that the socket is faulty. The metering circuit 220 may include a temperature sensor (e.g., an internal contact or ambient temperature sensor, an infrared sensor, a sensor provided on the processing circuit 222). The metering circuit 220 may generate a temperature value 1430 via such a sensor. The metering circuit 220 may be configured to generate a temperature profile for the meter 212. The temperature profile may be affected by the location of the sensor within the meter (e.g., the delay, amplitude, and speed characteristics of the temperature rise). The temperature sensor may be configured to sample the temperature value 1430 of the meter 212 at a relatively slow rate. For example, the temperature sensor may be configured to obtain 3 to 10 temperature samples per minute.

[0116] In the schematic diagram shown, an alpha-beta filter 1432 processes the temperature value 1430 to produce a temperature velocity 1434. The alpha-beta filter can help smooth noisy measurements by combining the noisy measurements with the predicted output values. The temperature velocity 1434 provides an estimate of how quickly the temperature value 1430 is changing. To determine the temperature velocity 1434, the alpha-beta filter combines the temperature prediction term Combined with the temperature measurement result term T, where The temperature measurement is updated using the following equation:

[0117]

[0118]

[0119] Among them, 0 < α, β < 1. In the test, the α value is 0.2 and the β value is 0.05.

[0120] The fuzzy inference classifier 1436 uses fuzzy logic to generate an alarm score 1438 based on the scoring module 1428, the RMS value 1406, the temperature value 1430, and the temperature velocity 1434 as inputs. The scoring module 1428, the RMS value 1406, the temperature value 1430, and the temperature velocity 1434 are input to the fuzzy inference classifier 1436. The fuzzy inference classifier 1436 scores the inputs in a non-linear manner using fuzzy set theory and Gaussian membership functions. Sharp inputs (e.g., numeric values) are identified as having a specific type (e.g., temperature, temperature difference, RMS current, arc score) and are input to the fuzzifier, which associates the sharp input with a loosely defined value based on the input value (e.g., for a temperature input, the fuzzifier may associate the input with the value "cold") to generate a fuzzy input. The membership function then uses a Gaussian function to determine the degree of membership of the sharp input to the fuzzified association. For example, Fig.16 The Gaussian function 1600 shown can be used to determine the membership of a clear input to a fuzzy association. The reasoning engine applies rules to the fuzzy input to determine the criticality of the input (e.g., low, medium, high, critical). Example rules are provided below:

[0121] RI="IF(Arcing IS low)AND((Temp IS ideal)OR(Temp IS cold))THEN(ScoreIS low)."

[0122] R2="IF(Arcing IS medium)AND(Amps IS big)AND(Delta_Temp IS cooling)AND((Temp IS ideal)OR(Temp IS cold))THEN(Score IS low)."

[0123] R3="IF(Arcing IS medium)AND(Amps IS big)AND(Delta_Temp IS heating)AND((Temp IS ideal)OR(Temp IS cold))THEN(Score IS medium)."

[0124] R4="IF(Arcing IS medium)AND(Amps IS average)AND(Delta_Temp IScooling)AND((Temp IS ideal)OR(Temp IS cold))THEN(Score IS low)."

[0125] R5="IF(Arcing IS medium)AND(Amps IS avg)AND(Delta_Temp IS heating)AND(Temp IS ideal)THEN(Score IS high)."

[0126] R6="IF(Arcing IS medium)AND(Amps IS average)AND(Delta_Temp ISheating)AND(Temp IS cold)THEN(Score IS medium)."

[0127] R7="IF(Arcing IS medium)AND(Amps IS small)AND(Delta_Temp IS cooling)AND((Temp IS ideal)OR(Temp IS cold))THEN(Score IS low).”

[0128] R8="IF(Arcing IS medium)AND(Amps IS small)AND(Delta_Temp IS heating)AND((Temp IS ideal)OR(Temp IS cold))THEN(Score IS high)."

[0129] R9="IF(Arcing IS high)AND(Amps IS big)AND(Delta_Temp IS cooling)THEN(Score IS high)."

[0130] R10-'IF(Arcing IS high)AND(Amps IS big)AND(Delta_Temp IS heating)THEN(Score IS critical)."

[0131] R11="IF(Arcing IS high)AND(Amps IS average)AND(Delta_Temp IScooling)THEN(Score IS critical)."

[0132] R12="IF(Arcing IS high)AND(Amps IS average)AND(Delta_Temp ISheating)THEN(Score IS critical)."

[0133] R13="IF(Arcing IS high)AND(Amps IS small)AND(Delta_Temp IS cooling)THEN(Score IS critical)."

[0134] R14="IF(Arcing IS high)AND(Amps IS small)AND(Delta_Temp IS heating)THEN(Score IS critical)."

[0135] R15="IF(Temp IS hot)THEN(Score IS critical)."

[0136] Based on the determined criticality of the input, the output of the Gaussian membership function is weighted according to a weighting function:

[0137]

[0138] Fig.15 15 is a flow chart 1500 for processing an input current waveform 802 from a distribution network 101 via a power monitoring module 230. The distribution network 101 may be configured to deliver power 112 to a meter 212 via a transformer 108 in communication with a substation 116. The meter 212 may be configured to analyze the input current waveform 802 via the power monitoring module 230 for evidence of arcing and generate a scaled arc score 908 indicating a likelihood that an arc is occurring on a connection between the meter 212 and the distribution network 101. Execution of the power monitoring module 230 may include passing the input current waveform 302 through a plurality of processing modules, such as the alignment and RMS filtering module 1402, the additional filtering module 1408, the feature extraction module 1414, the feature scaling module 1422, the classification module 1424, and the temporal processing module 1426, to generate an arc score at the scoring module 1428 described above.

[0139] At block 1502, the meter 212 may receive an input current waveform from the power distribution network 101 (e.g., via the transformer 108 or the substation 116) and model the input current waveform 302 via the power metering circuit. The meter 212 may begin analyzing the input current waveform 302 for evidence of an electrical arc by executing the power monitoring module 230 using the processing circuit 222 to first process the input current waveform 802 using a gating module included in the RMS filtering module 1404.

[0140] At block 1504, the processing circuit 222 may perform gating on the input current waveform 802 using the alignment and RMS filtering module 1402. The alignment and RMS filtering module 1402 may be configured to cancel the input current waveform 802 if aspects of the input current waveform 802 do not conform to a predefined range of values ​​that indicates a high likelihood (e.g., 95% to 100% likelihood) that the input current waveform 802 lacks evidence of arc current. For example, as mentioned above, the alignment and RMS filter 1404 may calculate a root mean square (RMS) for the input current waveform 302. The alignment and RMS filter 1404 may also be configured to flag or cancel the input current waveform 302 if arcing is unlikely (e.g., the instrument line is disconnected or shorted) or if the classification module 1424 is likely to misclassify the input current waveform 302 after filtering. After gating and filtering the input current waveform 802, the processing circuit 222 may further filter the input current waveform 802 via the additional filtering module 1408.

[0141] At block 1506, the processing circuit 222 may use the additional filtering module 1408 to apply a filter (e.g., a 60 Hz filter 1410, a Gabor filter 1412, an HP IIR filter) to the input current waveform 802 to highlight or isolate portions of the input current waveform 802, and output an unweighted filtered version 606 of the input current waveform 602 so that the highlighted or isolated features can be further analyzed. The additional filtering module 1408 may also apply weighting to the unweighted filtered version 606 of the input current waveform 602 to produce a weighted filtered version 608 of the input current waveform 602. After filtering the input current waveform 802, the processing circuit 222 may classify the unweighted filtered version 606 of the input current waveform 602 or the weighted filtered version 608 of the input current waveform 602 via the classification module 1424.

[0142] At block 1508, the processing circuit 222 may perform feature extraction on the input current waveform 602 using the NMSE function 1416, the maximum consecutive zeros function 1418, or the high frequency energy function 1420 to identify arc currents in a two-cycle window of the filtered input current waveform 602. After extracting features from the filtered input current waveform 602, the extracted features may be scaled.

[0143] At block 1510, the processing circuit 222 may scale the features extracted by the processor during feature extraction. The extracted features (scaled or unscaled) are then classified.

[0144] At block 1512, the processing circuit 222 may use the multilayer perceptron neural network 700 of the classification module 1424 to classify one or more binomial data windows 310 of features extracted from the input current waveform 602 as containing or not containing features that evidence arc current in the input current waveform 602. This classification may be performed by the classification module 1424 based on a classification threshold. The classification module 1424 may be configured to output the voltage pulse 1006 if the extracted features contain sufficient evidence of arc current. After classifying the features extracted from the input current waveform 802, the processing circuit 222 may temporally process the output of the classification module 1424 via the temporal processing module 1426 by integrating the pulse 1006 received at the temporal processing module over time.

[0145] At block 1514, the processing circuit 222 may integrate the input pulses 1006 from the classification module 1424 over time using a leaky integration method of the time processing module 1426. The decay parameter of the leaky integration method may be adjusted, for example, based on user input. The time processing module 1426 is configured to generate decay in the output when the classification module 1424 is not outputting positive classified pulses 1006. After the output of the classification module 1424 is time processed via the time processing module 1426, in some embodiments, the processing circuit 222 may scale the time processing module output 1002 via the scaling module 906.

[0146] At block 1516, the processing circuit 222 may process the time processing module output 1002 using a scaling function of the scaling module 906 to define the time processing module output 1002 for the purpose of assigning an arc score 908 to the input current waveform 802. The scaling function may be a sigmoid function configured to take the time processing module output 1002 and output a corresponding value within a predetermined range of values ​​(e.g., a value between 01 and 1, a value between 0 and 10, etc.). After (optionally) scaling the time processing module output 1002 via the scaling module 906, the processing circuit 222 may generate an arc score 908 for the input current waveform 802 via the scoring module 1428. The arc score 908 may be based not only on the time processing of the filtered input current waveform (e.g., the time processing module output 1002), but also on one or more of a sensed temperature of the measurement circuit 220, a current time of day, a determined arc gap size, and a determined degree of oxidation of the connection between the measurement circuit 220 and the load.

[0147] At block 1518, the processing circuit 222 may output the arc score 908 via the scoring module 1428. The scoring module 1428 may be configured to output the arc score 908 only when the scaled temporal processing module output meets or exceeds a predetermined value. For example, the scoring module 1428 may be configured to only output an arc score 908 of 1 when the scaled temporal processing module output is .95 or higher.

[0148] At block 1520 , a temperature sensor coupled to processing circuit 222 generates temperature value 1430 by measuring the temperature of meter 112 .

[0149] At block 1522 , processing circuitry 222 applies alpha-beta filter 1432 to temperature value 1430 to produce temperature velocity 1434 , which indicates the rate at which temperature value 1430 is changing.

[0150] At block 1524 , the processing circuit 222 inputs the scoring module 1428 , the RMS value 1406 , the temperature 1430 , and the temperature velocity 1434 to the fuzzy inference classifier 1436 to generate an alarm score 1438 .

[0151] At block 1526, the meter 212 may be configured to take action in response to the arc score 908 indicating evidence that an arc current has been detected in the input current waveform 802. For example, the meter 212 may disconnect itself from the distribution network, or generate a notification in response to the arc score 908 indicating evidence that an arc current has been detected in the input current waveform 802 and transmit it to the headend via wired or wireless communication.

[0152] Other methods of determining socket integrity are contemplated (eg, detecting a series arc). These other methods are discussed below.

[0153] In some embodiments, evidence of arc current can be detected in the input current waveform 802 using a pulse train detection algorithm. In particular, by using the pulse train detection algorithm on the output of the filtering module 504, the arc can be determined to be part of a series of arcs. Typically, energy spikes can be identified in the unweighted filtered version 606 of the input current waveform 602 and the weighted filtered version 608 of the input current waveform 608 near the half-cycle point (i.e., at the zero crossing 605). The arc typically produces high-frequency pulse data at the zero crossing 605, but the exact location, amplitude, and time length of each pulse can vary. The pulse train detection algorithm can be used to detect evidence of series arcs in the weighted filtered input current waveform 602 over time to determine whether a series arc exists. A fast folding algorithm (FFA) can be used as such a pulse detection algorithm. The FFA can take a segment of noisy frequency data obtained from the input current waveform 802 and perform a series of additions and shifts to extract the periodic signal. If a pulse train of period P is embedded in the unweighted filtered version 606 of the input current waveform 602 of length N or the weighted filtered version 608 of the input current waveform 608 of length N, it can be detected by additively folding the data onto itself with period P.

[0154]

[0155] The folding operation may form a matrix of size (N / P, P) and perform a summation on each column. If multiple P values ​​are used, many additions may be redundant and this algorithm helps remove redundancy. In addition to providing computational savings, this FFA may also provide sub-sampling accuracy with respect to the periodic calculation P according to the following formula:

[0156]

[0157] Where N is the length of the data segment and P0 is the base period (expressed in number of samples). The FFA algorithm can be used to calculate the output matrix for a variety of periods (e.g., P=3, 3.25, 3.5, 3.75, and 4.0). In many embodiments, the size of the output matrix is ​​the same as the input matrix (N / P, P). The rows of the output are the sum of the data for each period P. Then, the maximum value in each of the output matrices (rows) can be taken as a single point in the final output at that period P. In order to fully analyze the waveform, multiple values ​​of P0 can be run within the design range. For example, if the sample rate is 16.080kHz, the possible period P0 is 134 samples, but it may be useful to use multiple values ​​of P0. The FFA can be used to calculate partial sample periods with a reduced number of additions. In addition, calculating fractional periods may involve shifting each row and then adding, where the shift does involve wrapping around so that all data is used in each calculation. In general, the FFA algorithm may involve mainly additions and possibly a single division for normalization and is therefore very efficient. Further computational savings may be obtained by limiting the number of folding cycles computed.

[0158] The output of the fast folding algorithm can be further processed to extract features that can be used to determine whether a series arc fault exists. Certain features can be targeted during the extraction process so that once extracted, the extracted features provide a good basis for distinguishing between fault data and non-fault data. One feature extraction method can be to define an example arc 1800 (e.g., Fig.18 The parameter model 1700 (as shown) Fig.17 1700) as it may appear in the unweighted filtered version 606 of the input current waveform 602 and the weighted filtered version 608 of the input current waveform 608. The parameter model 1700 may be defined by:

[0159]

[0160] Where α is an amplitude parameter, μ is a location parameter, and β is a scaling parameter. Fitting the parameter model 1700 to the example arc 1800 may include using a curve fitting function on the graph of the example arc 1500 by adjusting the amplitude parameter α, the location parameter μ, and the scaling parameter β. The curve fitting function may be configured to select parameters with the goal of minimizing the residual sum of squares. Fig.19 and Fig. 20 Some example curve fits are shown. Fig.21 A process block diagram 2100 illustrating a fast collapse algorithm method for socket integrity detection is shown.

[0161] Further computational savings may be obtained by limiting the number of folding cycles that are computed.The output of the fast folding algorithm may be further processed to extract features that may be used to determine if a series arc fault is present.

[0162] In the test, Kellems example arc_5a_arc2 was divided into 12 10-cycle segments to analyze and test the pulse train detection algorithm. The first four segments contain data without series arc events, the next seven segments contain series arc events, and the fifth segment is a combination of both types of events. The parameter data is listed in Table 1 below.

[0163]

[0164]

[0165] Table 1

[0166] In some embodiments, the difference in high frequency spectral content of waveforms containing evidence of arcing and waveforms not containing evidence of arcing can be used to distinguish the waveforms based on power spectral density analysis methods. Evidence of arcing may appear in the form of a more "peaky" spectrum than the relatively flat spectrum of the non-arcing waveform. Harmonics generated by the load can be seen in a zoomed-in view of the spectrum. For example, in Fig. 22 In the example, the 15th harmonic (900Hz) is 60dB lower than the fundamental peak of signal 2200 and is Fig.23 , the 15th harmonic is buried in the noise contained in signal 2300. It may be possible to use the high frequency noise component of the signal to distinguish between signals that contain evidence of arc current and signals that do not contain evidence of arc current. A 12th order Butterworth high pass filter may be used to extract the high frequency "noise" component from the input current waveform 802 for analysis. In testing, the cutoff frequency of the Butterworth high pass filter was designed to be 889 Hz. Fig.24 The filter frequency response 2400 is shown. The signal ( Fig.25 506 ) looks like random noise, therefore, a power spectral density (PSD) that estimates the power of a random signal across a frequency spectrum is a suitable analysis tool. The PSD can be integrated over a wider frequency band (~500 Hz) and analyzed to determine the total energy of the frequency band extracted by the Butterworth high pass filter as a feature for decomposition into the classification performed by the classification module 506. For example, a plurality of such frequency band power values ​​can be calculated and used to train the multilayer perceptron neural network 700. Fig.26 A flow chart 2600 illustrating an approach to the power spectral density method for socket integrity detection.

[0167] In some embodiments, high-order statistics can be used in many different applications from EEG signal analysis to vibrating machinery diagnostics. In some embodiments, these techniques can be applied to series arc fault detection. The bispectrum is the first in a series of high-order statistics. It is similar to the power spectral density in that it is the Fourier transform of a third-order cumulant and generally defines the skewness or asymmetry of the distribution.

[0168]

[0169] Among them, C3 is defined as:

[0170] c3(λ1,λ2)=E(x(k),x(k+λ1),x(k+λ2))

[0171] A more intuitive way to express the bispectrum is to use the target signal X(w t )

[0172] B x (w1,w2)=X(w1)X(w2)X * (w1+w2)

[0173] Fig. 27 and Fig.28 Example graphs 2700 , 2800 are shown for bi-spectra of an input current waveform excluding arc current and an input current waveform including arc current.

[0174] In some embodiments, the cepstrum 2900 is used to detect series arcs in the input current waveform 802. Fig.29 The cepstrum is defined as the power spectral density 3000 (as shown in Fig.30 In some embodiments, the cepstrum 2900 can be used to find the periodic components of the power spectral density. The cepstrum 2900 can be defined as:

[0175] C p =|F[log(|F[f(t)] 2 )]

[0176] Where F is the Fourier transform operation. The natural logarithm function acts as an element of the compressed PSD so that the cepstrum 2900 can capture a large dynamic range and enable low amplitude components to become visible. The natural logarithm function also converts multiplication in the frequency domain (convolution in the time domain) to addition in the cepstrum domain. Therefore:

[0177] y(t)=h(t)*x(t),Y(f)=H(f)X(f),yc (λ) = h c (λ)+x c (λ)

[0178] The cepstrum 2900 is a time domain function, however, the x-axis is not the same as the time axis. The x-axis is called the quefrency-axis and is in units of time. The quefrency-axis can be interpreted as 1 / F, where F is the frequency from the PSD. Therefore, the time axis represents a periodic signal in the PSD, with the frequency decreasing towards the left side of the axis. As the time value approaches zero, the PSD frequency approaches infinity. The cepstrum 2900 can be used to capture the higher energy of the harmonics present in the input current waveform 802. From Fig.29 and Fig.30 As can be seen in the figure, there are clear visible differences in the amplitudes of the odd harmonics at high frequencies.

[0179] In some embodiments, a magnetic 3D sensor can be used to detect a series arc fault located at a meter socket. A series arc fault can be considered a spark gap transmitter. The spark produced by the arc can initiate a strong EMF pulse that moves radially outward from the arc location. This EMF pulse should change the local magnetic field in such a way that it can be detected by the 3D magnetic sensor.

[0180] During normal operation, the magnetic field generated by a current-carrying conductor can surround the conductor in concentric rings perpendicular to the direction of the current. Since this is an AC current, the magnetic field can switch direction over time. During testing, the magnetic 3D sensor was placed directly below the wire to test the operation of the sensor. The test results showed that the sensor did detect the magnetic field near the conductor. Fig.31 Magnetic sensor output 3100 in spherical coordinates is shown.

[0181] Fig.32 Magnetic field 3200 is shown for an EMF plane wave. The E field and H (magnetic) field are perpendicular to each other and to the direction of propagation. The dotted line graph is the normalized magnetic field magnitude, the dashed line graph is the angle in the XY plane, and the solid line graph is the angle in the Z plane. The vector direction is either the positive X direction or the negative X direction. The orientation in the XY plane is usually random and is specified by the term polarization. The XY orientation can be determined primarily by the transmit antenna orientation. For spark gap transmission, this is a random XY orientation.

[0182] Preliminary results of the test 330 Fig.33 It is shown in FIG. 1 and demonstrates that some distortion in the sensed magnetic field can help determine the occurrence of an arc fault.

[0183] In some embodiments, the Pearson correlation coefficient is calculated cycle by cycle for the input current waveform 802. The Pearson correlation coefficient provides a measure of the correlation between adjacent 60 Hz cycles. The Pearson correlation coefficient can be defined as:

[0184]

[0185] Among them, r is the correlation coefficient, x i is the value of the x variable in the sample, is the mean of the values ​​of the x variable, y i is the value of the y variable in the sample, is the mean of the values ​​of the y variable. The variable n is the sample x k-j With x k-n-j The time offset in samples between . AC signals have zero mean, so the equation can be simplified to analyze AC signals:

[0186]

[0187] The calculation can be done by only considering n = T0 = f s / f0r k (n) or the number of samples per cycle for simplification. For the case of a 16 kHz sample frequency, n = n0 = 266.66 sample points. For a 60 Hz sine wave, the correlation coefficient at n = 1 cycle can be exactly 1.0. The basic assumption for using the inter-period correlation coefficient for socket integrity diagnostics is that the statistics of the line current through a single cycle are consistent in the absence of arcing even if the load is nonlinear, and the arcing process can be represented by a non-stationary random process. For non-arcing current signals, the correlation coefficient r k (T0) is likely to be close to 1.0 and consistent across multiple cycles (low variance). In the presence of an arcing event, r k (T0) is likely to be significantly less than 1.0, with a larger variance. When a series arc fault occurs, the arc signal level is much smaller than the fundamental component and is therefore completely masked when the correlation is calculated on the original signal. The deterministic part of the signal can be attenuated so that the resulting signal r k The (T0) value is affected by the arcing event. However, if the fundamental is completely removed, the r k The (T0) value may not be maintained at a value close to 1.0. Basic candidate IIR filters can be suitable for:

[0188]

[0189] Various features and advantages of the invention are set forth in the following claims.

Claims

1. A series arc detection system, configured to detect a series arc in an input current waveform, the series arc detection system comprising: a metering circuit configured to receive an input current waveform; and A controller includes a power monitoring module configured to detect a series arc current in an input current waveform by: filtering the input current waveform to generate a filtered waveform; performing feature recognition on the filtered waveform; classifying one or more features of the filtered waveform identified during feature recognition; performing time processing on the filtered waveform; generating a score for the input current waveform based on time processing of the filtered waveform; as well as The score is reported to a user based on a comparison of the score to a programmable threshold.

2. The series arc detection system according to claim 1, wherein: Performing feature recognition on the filtered waveform includes using a neural network to detect one or more features that can be indicative of arcing.

3. The series arc detection system according to claim 2, wherein: The neural network is a multi-layer perceptron.

4. The series arc detection system according to claim 1, wherein: Performing feature recognition on the input current waveform includes applying a maximum consecutive zeros function to the filtered waveform.

5. The series arc detection system according to claim 1, wherein: Filtering the input current waveform includes: inputting the input current waveform into a Gabor filter.

6. The series arc detection system according to claim 1, wherein: The time processing of the filtered waveform includes integrating the filtered waveform and inputting the filtered waveform into a sigmoid function.

7. The series arc detection system according to claim 1, wherein: Scoring the input current waveform includes scoring the input current waveform based on time processing of the filtered waveform and one of a temperature of the metering circuit, a current time of day, an arc gap size, and a degree of oxidation of a connection between the metering circuit and a load.

8. A method of determining socket integrity, the method comprising: receiving an input current waveform via a metering circuit; and The series arc current in the input current waveform is detected by a controller including a power monitoring module in the following manner: filtering the input current waveform to generate a filtered waveform; performing feature recognition on the filtered waveform; classifying one or more features of the filtered waveform identified during feature recognition; performing time processing on the filtered waveform; generating a score for the input current waveform based on time processing of the filtered waveform; as well as The score is reported to a user based on a comparison of the score to a programmable threshold.

9. The method according to claim 8, wherein: Performing feature recognition on the filtered waveform includes using a neural network to detect one or more features that can be indicative of arcing.

10. The method according to claim 9, wherein: The neural network is a multi-layer perceptron.

11. The method according to claim 8, wherein: Performing feature recognition on the input current waveform includes applying a maximum consecutive zeros function to the filtered waveform.

12. The method according to claim 8, wherein: Filtering the input current waveform includes: inputting the input current waveform into a Gabor filter.

13. The method according to claim 8, wherein: The time processing of the filtered waveform includes integrating the filtered waveform and inputting the filtered waveform into a sigmoid function.

14. The method according to claim 8, wherein: Scoring the input current waveform includes scoring the input current waveform based on time processing of the filtered waveform and one of a temperature of the metering circuit, a current time of day, an arc gap size, and a degree of oxidation of a connection between the metering circuit and a load.

15. A meter in a power distribution network, the meter comprising: a metering circuit in communication with at least one temperature sensor and configured to receive an input current waveform and determine a temperature value for the meter; and A controller includes a power monitoring module configured to detect a series arc current in an input current waveform by: filtering the input current waveform to generate a filtered waveform; applying a root mean square function to the input current waveform to generate an RMS value; performing feature recognition on the filtered waveform; classifying one or more features of the filtered waveform identified during feature recognition; performing time processing on the filtered waveform; generating a score for the input current waveform based on time processing of the filtered waveform; determining a temperature value for the instrument; determining an alarm score based on the score for the input current waveform, the RMS value, and a temperature value for the meter using a fuzzy inference classifier; as well as The alert score is reported to a user based on a comparison of the score to a programmable threshold.

16. The method according to claim 15, wherein: Performing feature recognition on the filtered waveform includes using a neural network to detect one or more features that can be indicative of arcing.

17. The method according to claim 16, wherein: The neural network is a multi-layer perceptron.

18. The method according to claim 15, wherein: Performing feature recognition on the input current waveform includes applying a maximum consecutive zeros function to the filtered waveform.

19. The method according to claim 15, further comprising: Applies an alpha-beta filter to the temperature values ​​to produce the temperature velocity.

20. The method according to claim 19, wherein: The alarm score is also based on the temperature velocity.

Images