310 includes a plurality of electrical and electronic components that provide power, operational control, and protection to the components and modules within the controller 310 and/or the AFCI receptacle 100. For example, the controller 310 includes, among other things, an electronic processor 615 (for example, a microprocessor or another suitable programmable device) and a memory 620. In some embodiments, the controller 310 further includes the arc fault detection circuit 530 and the zero cross detection circuit 602. That is, in some embodiments, the arc fault detection circuit 530 and the zero cross detection circuit 602 are integrated within the controller 310.

The memory 620 includes, for example, a program storage area and a data storage area. The program storage area and the data storage area can include combinations of different types of memory, such as read-only memory (ROM) and/or random-access memory (RAM). Various non-transitory computer readable media, for example, magnetic, optical, physical, or electronic memory may be used. The electronic processor 615 is communicatively coupled to the memory 620 and executes software instructions that are stored in the memory 620, or stored on another non-transitory computer readable medium such as another memory or a disc. Instructions may include instructions, which when executed by processor 615, cause the control system 600 to implement any of a variety of arc fault detection actions as described herein. The software may include one or more applications, program data, filters, rules, one or more program modules, and other executable instructions.

In some embodiments, the memory 620 stores a machine learning model that is to be implemented by the electronic processor 615. In some embodiments, the machine learning model is executed by the electronic processor 615 to cause the control system 600 to detect an arc fault within AFCI receptacle 100 and/or the circuit to which the AFCI receptacle 100 is connected. More particularly, the machine learning model may be executed by electronic processor 615 to cause the controller 310 and/or arc detection circuit 530 to detect an arc fault within AFCI receptacle 100 and/or the circuit to which the AFCI receptacle 100 is connected. The machine learning model may be implemented as, for example, a neural network, a fuzzy logic model, convolutional network, and/or other such model trained to detect arc faults as detailed herein.

The phase and neutral line contact arms, or terminals, 220, 225 are configured to receive a line power from source. The first and second coils 520, 525 are arranged to monitor current flowing through the phase and neutral line terminals 220, 225 respectively. As described above, the phase and neutral line terminals support contacts 240, 245 are selectively connected, via interrupting device 210, to the load contacts 250, 255 supported by phase and neutral load terminals 230, 235. The phase and neutral load terminals 230, 235 are configured to power an external load connected to an outlet 625 of AFCI receptacle 100. The zero cross detection circuit 602 is configured to measure the line voltage and frequency of the line terminals 220, 225 and/or the load terminals 230, 235.

In some embodiments, the first and second coils 520, 525 are arranged to monitor current flowing through the phase and neutral load terminals 230, 235. In some embodiments, the first and second coils 520, 525 are arranged to monitor current flowing through the phase and neutral line terminals respectively, and third and fourth coils (not shown) are arranged to monitor current flowing through the phase and neutral load terminals 230, 235, respectively. Current measurements taken by the first and second coils 520, 525 are provided to the arc fault detection circuit 530 and/or controller 310. In some embodiments, the arc fault detection circuit 530 further includes the zero cross detection circuit 602. That is, in some embodiments, the zero cross detection circuit 602 is integrated within the arc fault detection circuit 530.

The power supply circuit 605 is configured to convert line power to a nominal power for use by the controller 310. For example, the power supply circuit 605 may include a rectifier that is configured to rectify the line power to a nominal power for powering the controller 310. In some embodiments, the power supply circuit 605 rectifies alternating current (AC) power to a nominal direct current (DC) power. In some embodiments, the power supply circuit 605 includes one or more additional conversion circuits for converting line power to one or more additional power levels for use by control system 600.

The communication circuit 610 is configured to provide communication between the AFCI receptacle 100 and one or more external devices (for example, other receptacles, electrical devices, external computers, smart phones, tablets, etc.). For example, the communication circuit 610 is configured to provide communication between the AFCI receptacle 100 and an external device 635. In the illustrated embodiment, the external device 635 is shown as a laptop that includes an electronic processor and a memory. However, it should be understood that the external device 635 may be implemented as one or more of the above noted examples.

In such embodiments, the AFCI receptacle 100 communicates with the one or more external devices through a network using, for example, a transceiver 630. The network is, for example, a wide area network (WAN) (e.g., the Internet, a TCP/IP based network, a cellular network, such as, for example, a Global System for Mobile Communications [GSM] network, a General Packet Radio Service [GPRS] network, a Code Division Multiple Access [CDMA] network, an Evolution-Data Optimized [EV-DO] network, an Enhanced Data Rates for GSM Evolution [EDGE] network, a 3GSM network, a 4GSM network, a Digital Enhanced Cordless Telecommunications [DECT] network, a Digital AMPS [IS-136/TDMA] network, or an Integrated Digital Enhanced Network [iDEN] network, etc.). In other embodiments, the network is, for example, a local area network (LAN), a neighborhood area network (NAN), a home area network (HAN), or personal area network (PAN) employing any of a variety of communications protocols, such as Wi-Fi, Bluetooth, ZigBee, etc. In yet another embodiment, the network includes one or more of a wide area network (WAN), a local area network (LAN), a neighborhood area network (NAN), a home area network (HAN), or personal area network (PAN). In some embodiments, the communication circuit 610 communicates with the external device 635 using a wired connection.

In some embodiments, the transceiver 630 is configured to enable wireless communication between the AFCI receptacle 100 and an external device 635 using a wireless communication link 640. In other embodiments, rather than the transceiver 630, the AFCI receptacle 100 includes separate transmitting and receiving components, for example, a transmitter and a receiver. In operation, the controller 310 is configured to control the communication circuit 610 to transmit and receive data to and from the AFCI receptacle 100.

In some embodiments, the control system 600 further includes an oscilloscope 645. The oscilloscope 645 is configured to measure current, voltage, frequency, and/or other electrical characteristics of the AFCI receptacle 100. In the illustrated embodiment, the oscilloscope 645 is shown as measuring the line-side current and voltage; however, it should be understood that the oscilloscope 645 may also be configured to measure load-side characteristics of the AFCI receptacle 100. As shown, the oscilloscope 645 is configured to provide the current and voltage measurements to external device 635 via a communication link 650, which may be implemented as a wireless or wired connection.

During operation of the AFCI receptacle 100, the control system 600 may be in a standby mode or an operation mode. When in the standby mode, the interrupting device 210 electrically disconnects the line terminals 220, 225 from the load terminals 230, 235. Accordingly, in the standby mode, power is not provided to the outlet 625. When in the operation mode, the interrupting device 210 electrically connects the line terminals 220, 225 to the load terminals 230, 235. Accordingly, in the operation mode, power is provided to the outlet 625 and, thus, to an external load electrically connected to the outlet 625.

While power is provided to the outlet 625, the control system 600 is configured to monitor for the occurrence of an arc fault within the AFCI receptacle 100 and/or the circuit to which AFCI receptacle 100 is connected. In particular, the arc fault detection circuit 530 and/or the controller 310 are configured to perform one or more arc fault detection processing techniques to determine whether an arc fault is present. For example, when determining whether an arc fault is present, the arc fault detection circuit 530 and/or controller 310 may be configured to extract and analyze measurements indicative of volatility, power, and/or frequency content of an external load from the load terminals 230, 235. Furthermore, as another example, the arc fault detection circuit 530 and/or controller 310 are configured to employ statistical and spectral analysis to analyze line and/or load current measurements recorded by the first and second coils 520, 525. The arc fault detection circuit 530 and/or controller 310 analyzes the line and load terminal currents, voltages, and/or frequencies in the digital domain to reduce noise. As described below, the controller 310 may be further configured to determine whether an arcing fault is present by applying the analysis of the line and/or load electrical characteristics to a deep learning/machine learning model.

In some embodiments, when an electrical load is applied to the AFCI receptacle 100 (e.g., via an external load from the load terminals 230, 235), arcing characteristics indicative of an arc fault may be present in the line current measurements recorded by the first and second coils 520, 525. For example, FIG. 7 is a graph 700 illustrating an example alternating current (AC) signal 705 including turning points 710 for an electrical load as the AC signal 705 transitions from a non-arcing state 715 (e.g., a state in which no arc fault is present) to a series arcing state 720 (e.g., a state in which an arc fault is present). The AC signal 705 is a line current signal, indicative of the line current measurements, received by the arc fault detection circuit 530 and/or controller 310 from the first and second coils 520, 525. In particular, the AC signal 705 is representative of a 5 Ampere resistive load. As shown, the turning points 710 (illustrated as being within dashed circles) are local extremes within cycle peaks of the AC signal 705. The turning points 710 are indicative that an arc fault is present within the AC signal 705 as the AC signal 705 transitions from the non-arcing state 715 to the series arcing state 720 due to the sudden local changes in line current values within the cycle peaks of the AC signal 705. In addition, the position of the turning points 710 on the AC signal 705 relative to the zero cross points of the AC signal 705 is not constant.

FIG. 8 is a graph 800 illustrating an example AC signal 805 including turning points 810 for an electrical load as the AC signal 805 transitions from a non-arcing state 815 to a series arcing state 820. The AC signal 805 is a line current signal, indicative of the line current measurements, received by the arc fault detection circuit 530 and/or controller 310 from the first and second coils 520, 525. In particular, the AC signal 805 is representative of a 20 Ampere resistive load. As shown, the turning points 810 (illustrated as being within dashed circles) are local extremes within cycle peaks of the AC signal 805. The turning points 810 are indicative that an arc fault is present within the AC signal 805 as the AC signal 805 transitions from the non-arcing state 815 to the series arcing state 820 due to the sudden local changes in line current values within the cycle peaks of the AC signal 805. In addition, the position of the turning points 810 on the AC signal 805 relative to the zero cross points of the AC signal 805 is not constant.

FIG. 9 is a graph 900 illustrating an example AC signal 905 including turning points 910 for an electrical load as the AC signal 905 transitions from a non-arcing state 915 to a series arcing state 920. The AC signal 905 is a line current signal, indicative of the line current measurements, received by the arc fault detection circuit 530 and/or controller 310 from the first and second coils 520, 525. In particular, the AC signal 905 is representative of an electrical load applied from a home appliance such as, for example, a vacuum. As shown, the turning points 910 (illustrated as being within dashed circles) are local extremes within cycle peaks of the AC signal 905. The turning points 910 are indicative that an arc fault is present within the AC signal 905 as the AC signal 905 transitions from the non-arcing state 915 to the series arcing state 920 due to the sudden local changes in line current values within the cycle peaks of the AC signal 905. In addition, the position of the turning points 910 on the AC signal 905 relative to the zero cross points of the AC signal 905 is not constant.

FIG. 10 is a graph 1000 illustrating an example AC signal 1005 including turning points 1010 for an electrical load as the AC signal 1005 transitions from a non-arcing state 1015 to a series arcing state 1020. The AC signal 1005 is a line current signal, indicative of the line current measurements, received by the arc fault detection circuit 530 and/or controller 310 from the first and second coils 520, 525. In particular, the AC signal 1005 is representative of an electrical load applied from a home appliance such as, for example, an air compressor. As shown, the turning points 1010 (illustrated as being within dashed circles) are local extremes within cycle peaks of the AC signal 1005. The turning points 1010 are indicative that an arc fault is present within the AC signal 1005 as the AC signal 1005 transitions from the non-arcing state 1015 to the series arcing state 1020 due to the sudden local changes in line current values within the cycle peaks of the AC signal 1005. In addition, the position of the turning points 1010 on the AC signal 1005 relative to the zero cross points of the AC signal 1005 is not constant.

FIG. 11 is a graph 1100 illustrating an example AC signal 1105 including turning points 1110 for an electrical load as the AC signal 1105 transitions from a non-arcing state 1115 to a series arcing state 1120. The AC signal 1105 is a line current signal, indicative of the line current measurements, received by the arc fault detection circuit 530 and/or controller 310 from the first and second coils 520, 525. In particular, the AC signal 1105 is representative of a 1000 W dimmer switch load. As shown, the turning points 1110 (illustrated as being within dashed circles) are local extremes within cycle peaks of the AC signal 1105. The turning points 1110 are indicative that an arc fault is present within the AC signal 1105 as the AC signal 1105 transitions from the non-arcing state 1115 to the series arcing state 1120 due to the sudden local changes in line current values within the cycle peaks of the AC signal 1105. In addition, the position of the turning points 1110 on the AC signal 1105 relative to the zero cross points of the AC signal 1105 is not constant.

FIG. 12 is a flowchart of a method 1200 for detecting presence of an arc fault, according to some embodiments. For example, the method 1200 is implemented to detect whether an arc fault is present within the AFCI receptacle 100 and/or the circuit, including line terminals 220, 225, to which the AFCI receptacle 100 is connected. It should be understood that the order of the steps disclosed in method 1200 could vary. Although some steps are illustrated as occurring in parallel order, in other embodiments, the steps disclosed may be performed in serial order. Furthermore, additional steps may be added to the process and not all of the steps may be required.

In one embodiment, the method 1200 is performed by an electronic controller, such as the controller 310. However, in other embodiments, the method 1200 may be performed via other components within the control system 600, such as the arc fault detection circuit 530, or a combination of the controller 310 and the arc fault detection circuit 530.

At step 1205, a current sensor, such as but not limited to the first and second coils 520, 525, measures the current (e.g., a line current) flowing through line terminals 220, 225. As described above, in some embodiments, the first and second coils 520, 525 measure the current flowing through load terminals 230, 235 instead. In some embodiments, the first and second coils 520, 525 measure the current flowing through line terminals 220, 225 and third and further coils measure the current flowing through load terminals 230, 235. Current measurements are provided to the arc fault detection circuit 530 and/or controller 310 for digitization and further processing and analysis.

At step 1210, an analog-digital converter (ADC) (for example, an ADC of the arc fault detection circuit 530) converts the current measurements from the first and second coils 520, 525 to a digital current signal (e.g., a line current measurement signal). For example, the ADC receives an analog signal indicative of the current measurements from the first and second coils 520, 525 and converts the analog signal to the line current measurement signal to provide to the controller 310. In some embodiments, the ADC is a separate component included in control system 600.

At step 1215, the zero cross detection circuit 602 measures the voltage and frequency of line terminals 220, 225. In some embodiments, the zero cross detection circuit 602 measures the voltage and frequency of the load terminals 230, 235 instead. In some embodiments, the zero cross detection circuit 602 measures the voltage and frequency of the both the line terminals 220, 225 and the load terminals 230, 235. As described above, the zero cross detection circuit 602 may be implemented as a separate circuit, integrated within arc fault detection circuit 530, or integrated within controller 310. Voltage and frequency measurements (e.g., a line voltage measurement signal) are provided to the arc fault detection circuit 530 and/or controller 310 for further processing and analysis.

At step 1220, the controller 310 determines a plurality of zero cross positions of the line voltage measurement signal. In some embodiments, the line voltage measurement signal is converted to a digital line voltage measurement signal via the ADC. In some embodiments, the zero cross detection circuit 602 determines the plurality of zero cross positions. In such embodiments, the zero cross detection circuit 602 transmits the line voltage measurement signal including the plurality of zero cross positions to the controller 310.

At step 1225, the controller 310 samples 1 AC cycle each from the line current measurement signal and the line voltage measurement signal including the plurality of zero cross positions. For example, the controller 310 receives the line current measurement signal from the ADC and the line voltage measurement signal including the plurality of zero cross positions from the zero cross detection circuit 602. The controller 310 samples the 1 AC cycle such that the controller 310 only considers a portion (e.g., a window) of the line current measurement signal and the line voltage measurement signal including a positive peak and a negative peak of the line current measurement signal and the line voltage measurement signal. In some embodiments, the ADC is phase locked to the frequency of the line voltage measurement signal such that a frequency of the line current measurement signal corresponds to the frequency of the line voltage measurement signal. In addition, the controller 310 may dynamically determine which AC cycle to sample from the line current measurement signal and the line voltage measurement signal based on a power factor, a crest factor, and/or a symmetry of the line current measurement signal. In other embodiments, the controller 310 samples one or more 1/2 AC cycles each from the line current measurement signal and the line voltage measurement signal.

At step 1230, the controller 310 applies a digital filter to the line current measurement signal to differentiate turning points in the line current measurement signal (FIG. 16). That is, the digitized line current measurements are filtered. In some embodiments, the arc fault detection circuit 530 applies digital filters to the line current measurement signals. In some embodiments, the controller 310 is further configured to implement multi-rate analysis (e.g., down-sampling/decimation and up-sampling/interpolation) to optimize the digital filters applied to the line current measurement signals. The controller 310 estimates a derivative of the line current measurement signal in discrete time by applying the digital filter to the sampled and digitized form of the line current measurement signal.

In some embodiments, the filter differentiates between the 1 AC cycle sample (step 1225) and a previous sample of the line current measurement signal (e.g., a Two Tap Type IV FIR filter). In some embodiments, Type III or Type IV FIR (odd or even asymmetric filters) high pass or bandpass filters are applied via the controller 310. Type III or Type IV FIR filters may be preferable to Type I and Type II filters because, in some instances, Type I and Type II filters produce ringing artifacts in the filtered signal. Additionally, the controller 310 adjusts the number of taps in the applied filter to increase or decrease an amount of high frequency noise present in the estimated derivative of the line current measurement signal. In some embodiments, the controller 310 applies the filter to differentiate the line current measurement and determine the turning points in the line current measurement signal where the estimated derivative is zero and/or where the derivative changes from positive to negative or from negative to positive. In some embodiments, the following equations may be used to represent where the turning points are defined based on the differentiation of the line current measurement signal:

$\begin{matrix} \begin{matrix} Local Maximum \\ f^{'} (x) = 0 \\ f^{'} (x - r) > 0 \\ f^{'} (x + r) < 0 \end{matrix} & Equation 1 \end{matrix}$

$\begin{matrix} \begin{matrix} Local Minimum \\ f^{'} (x) = 0 \\ f^{'} (x - r) < 0 \\ f^{'} (x + r) > 0 \end{matrix} & Equation 2 \end{matrix}$

where f (x) is the line current measurement signal, f (x) is the estimated derivation of the line current measurement signal, x is a point in time on the line current measurement signal, and r is a small positive number.

At step 1235, the controller 310 calculates a power factor and a crest factor of the digitized and filtered line current measurement signal. In some embodiments, the power factor is a ratio of a real power absorbed by the load (e.g., the external load connected to the outlet 625) to the apparent power flowing through the control system 600. In some embodiments, the crest factor is a ratio of peak values of the line current measurement signal to an effective value (e.g., an average value or a root mean square (RMS) value) of the line current measurement signal.

At step 1240, the controller 310 creates a mask (e.g., a filter) based on the line voltage measurement signal (FIG. 17). The controller 310 may apply the mask to the digitized and filtered line current measurement signal to normalize the line current measurement signal. By normalizing the line current measurement signal, the mask zeroes out absolute peaks of the line current measurement signal (e.g., for each sampled AC cycle) such that only the turning points are present in the line current measurement signal.

At step 1245, the controller 310 applies the mask to the line current measurement signal (FIG. 18). In some embodiments, the controller 310 can dynamically adjust properties of the mask based on the determined power factor, the determined crest factor, and a symmetry of the line current measurement signal.

At step 1250, the controller 310 counts a plurality of turning points of the filtered line current measurement signal based on the masked output of step 1245. In some embodiments, the plurality of turning points correspond to points in the masked output of the filtered line current measurement signal where the masked output drops below zero (e.g., the masked derivative of the line current measurement signal is less than zero).

At step 1255, the controller 310 determines a position of each turning point of the plurality of turning points relative to a corresponding zero cross position of the plurality of zero cross positions. For example, the controller 310 may determine the position of each turning point of the masked output relative to the nearest zero cross position for each turning point.

At step 1260, the controller 310 determines a time between each turning point of the plurality of turning points. For example, the controller 310 determines a time between each turning point of the plurality of turning points and the subsequent turning point of the plurality of turning points. In some embodiments, the controller 310 determines the time between turning points based on a period between each turning point determined by the oscilloscope 645.

At step 1265, the controller 310 determines a local extreme derivative value (FIG. 18) around each turning point. In some embodiments, the local extreme derivative value is a local maximum value or a local minimum value at AC cycle peaks of the derivative signal, for example the filtered signal of step 1230. In some embodiments, the local extreme derivative value may include multiple values (e.g., a local minimum value and a local minimum value around each turning point.) For example, the controller 310 determines the local extreme derivative value (or multiple local extreme derivative values) around the determined position of each counted turning point, as described in steps 1250 and 1255 above.

At step 1270, the controller 310 applies turning point thresholds to the plurality of turning points, the positions the plurality of turning points relative to the corresponding zero cross positions, the time between each turning point of the plurality of turning points, and/or the local extreme derivative value around each turning point. For example, the turning point thresholds include one or more turning point thresholds such as, but not limited to, an upper turning point threshold relative to the counted turning points, a lower turning point threshold relative to the counted turning points, an upper position threshold, a lower position threshold, an upper time threshold, a lower time threshold, an upper extreme derivative value threshold, and/or a lower extreme derivative value threshold. In addition, the controller 310 compares the plurality of turning points to the upper turning point threshold and the lower turning point threshold. The controller 310 also compares the positions of the plurality of turnings points relative to the corresponding zero cross positions to the upper position threshold and the lower position threshold. Likewise, the controller 310 compares the time between each turning point of the plurality of turning points to the upper time threshold and the lower time threshold. Similarly, the controller 310 compares the local extreme derivative value around each turning point to the upper extreme derivative value threshold and the lower extreme derivative value threshold. In some embodiments, the controller 310 is configured to apply the turning point thresholds based on a correlation between the line current measurement signal and a reference waveform, such as the correlation-based arc fault detection method described in U.S. Pat. No. 10,243,343, issued Mar. 26, 2019, and entitled “SYSTEMS AND METHODS FOR DETECTING AND IDENTIFYING ARCING BASED ON NUMERICAL ANALYSIS,” the entire content of which is hereby incorporated by reference. In some embodiments, the controller 310 is configured to apply the turning point thresholds based on a spectral energy of the line current measurement signal, such as the arc fault detection methods described in U.S. patent application Ser. No. 18/106,854, filed Feb. 7, 2023, and entitled “ARC FAULT DETECTION THROUGH MIXED-SIGNAL MACHINE LEARNING AND NEURAL NETWORKS,” the entire content of which is hereby incorporated by reference. In some embodiments, the turning point thresholds are determined by a machine learning model stored in the memory 620 of the controller 310.

At step 1275, the controller 310 determines whether an arc fault is present within the AFCI receptacle 100 and/or the circuit, including line terminals 220, 225. For example, the controller 310 determines whether an arc fault is present based on the plurality of turning points, the position of each turning point relative to the corresponding zero cross position, the time between each turning point and the subsequent turning point and the local extreme derivative value around each of the plurality of turning points. In some embodiments, the controller 310 determines whether an arc fault is present based on the comparison of step 1270. As such, when the controller 310 determines that the counted plurality of turning points is greater than the upper turning point threshold or less than the lower turning point threshold, the controller 310 determines that an arc fault is present. When the controller 310 determines that the position of at least one of the plurality of turning points relative to the corresponding zero cross position is greater than the upper position threshold or less than the lower position threshold, the controller 310 determines that an arc fault is present. Likewise, when the controller 310 determines that the time between at least one of the plurality of turning points and the subsequent turning point is greater than the upper time threshold or less than the lower time threshold, the controller 310 determines that an arc fault is present. Similarly, when the controller 310 determines that the local extreme derivative value around each turning point is greater than the upper extreme derivative value threshold or less than the lower extreme derivative value threshold, the controller 310 determines that an arc fault is present. In some embodiments, when the controller 310 determines that an arc fault is present, the controller 310 activates the interrupting device 210. It should be understood that after the controller 310 determines whether an arc fault is present at step 1275, the method 1200 can be repeated for each AC cycle of the line current measurement signal. In some embodiments, after the controller 310 determines whether an arc fault is present at step 1275, the method 1200 returns to step 1205 to continuously monitor whether an arc fault is present.

FIG. 13 is a flowchart of a method 1300 for detecting presence of an arc fault, according to some embodiments. For example, the method 1300 is implemented to detect whether an arc fault is present within the AFCI receptacle 100 and/or the circuit, including line terminals 220, 225, to which the AFCI receptacle 100 is connected. It should be understood that the order of the steps disclosed in method 1300 could vary. Although some steps are illustrated as occurring in parallel order, in other embodiments, the steps disclosed may be performed in serial order. Furthermore, additional steps may be added to the process and not all of the steps may be required.

In one embodiment, the method 1300 is performed by an electronic controller, such as the controller 310. However, in other embodiments, the method 1300 may be performed via other components within the control system 600, such as the arc fault detection circuit 530, or a combination of the controller 310 and the arc fault detection circuit 530.

As shown in FIG. 13, the method 1300 includes many of the same steps as the method 1200. That is, steps 1305-1365 of the method 1300 are the same as steps 1205-1265 of the method 1200. However, at step 1370, the controller 310 applies a neural network to the plurality of turning points, the positions the plurality of turning points relative to the corresponding zero cross positions, the time between each turning point of the plurality of turning points, and/or the local extreme derivative value around each turning point based on a machine learning model stored in the memory 620. For example, the controller 310 executes the machine learning model to train the neural network. Training the neural network includes developing one or more thresholds, weights, and/or biases based on sample turning point and non-turning point data sets. For example, one or more of the thresholds described above with respect to the method 1200 (e.g., the upper turning point threshold, the lower turning point threshold, the upper position threshold, the lower position threshold, the upper time threshold, the lower time threshold, the upper extreme derivative value threshold, the lower extreme derivative value threshold, etc.) during training, testing, validation, or creation, of the neural network. In some embodiments, the controller 310 is configured to execute the machine learning model (and deploy the neural network) to detect arc faults, such as the arc fault detection methods described in U.S. patent application Ser. No. 18/106,854, filed Feb. 7, 2023, and entitled “ARC FAULT DETECTION THROUGH MIXED-SIGNAL MACHINE LEARNING AND NEURAL NETWORKS,” the entire content of which is hereby incorporated by reference.

At step 1375, the controller 310 determines whether an arc fault is present within the AFCI receptacle 100 and/or the circuit, including line terminals 220, 225. For example, the controller 310 determines whether an arc fault is present based on an output of applying the neural network of step 1370. As such, the neural network may be configured to label the turning points of the line current measurement signal. Based on the plurality of turning points, the positions the plurality of turning points relative to the corresponding zero cross positions, the time between each turning point of the plurality of turning points, and/or the local extreme derivative value around each turning point, the neural network calculates a probability of whether the arc fault is present, as described in more detail below with reference to FIGS. 14 and 15. In some embodiments, when the controller 310 determines that an arc fault is present, the controller 310 activates the interrupting device 210. It should be understood that after the controller 310 determines whether an arc fault is present at step 1375, the method 1300 can be repeated for each AC cycle of the line current measurement signal. In some embodiments, after the controller 310 determines whether an arc fault is present at step 1375, the method 1300 returns to step 1305 to continuously monitor whether an arc fault is present

In some embodiments, the controller 310 is configured to execute a machine learning model when predicting the presence of an arc fault based on determined arcing characteristics (e.g., the turning points) of the line current. For example, FIG. 14 illustrates a generic structure of a machine learning model 1400. The machine learning model 1400 is depicted as a neural network; however, the machine learning model may be implemented as any one or more of a deep learning algorithm, a neural network, a support-vector machine, and a long short-term memory.

As shown, N arcing features, or characteristics, of the line current measurement signal, such as the plurality of turning points, the positions the plurality of turning points relative to the corresponding zero cross positions, the time between each turning point of the plurality of turning points, and/or the local extreme derivative value around each turning point may be provided as inputs 1405 to the machine learning model 1400. The machine learning model 1400 further includes multiple hidden layers 1410, each hidden layer 1410 including an independent number of neurons. For example, the machine learning model 1400 may include 2-4 hidden layers 1410 with 2-256 nodes per layer. The hidden layers 1410 are configured to generate M outputs 1415 based on one or more weights, biases, and/or thresholds associated with the arcing characteristics provided as inputs 1405. As will be described in more detail below, the weights, biases, and/or thresholds that are used to calculate the outputs 1415 are generated using a supervised training algorithm that may executed by an external device, such as the external device 635. The outputs 1415 represent the probability of a binary, or multi-class classification, of the load current that includes arc fault, normal operation, and/or other classifications of the load. That is, the outputs 1415 indicate whether the load current is, or is likely to be, experiencing an arc fault condition or a normal operating condition. In some embodiments, the outputs 1415 include a probability of an arc fault being present within AFCI receptacle 100 and/or the circuit to which AFCI receptacle 100 is connected. In such embodiments, the controller 310 is configured to determine that an arc fault is present when the probability exceeds a probability threshold. In some embodiments, the probability threshold is a configurable value determined by a user. In other embodiments, the probability threshold is determined during creation of the machine learning model 1400.

In some embodiments, the controller 310 is configured to implement more than one arc fault detection method when determining whether an arc fault is present within the AFCI receptacle 100 and/or the circuit to which the AFCI receptacle 100 is connected, according to some embodiments. For example, the controller 310 may be configured to implement a machine learning, probability-based arc fault detection method, such as methods that are similar to the method described above with respect to method 1300. In some embodiments, the controller 310 is further configured to implement a correlation-based arc fault detection method, such as the arc fault detection methods described in U.S. Patent Application Publication No. 2020/0036183, published Jan. 30, 2020, and entitled “SYSTEM AND METHOD FOR DISCERNING ARCING IN ELECTRICAL WIRING,” the entire content of which is hereby incorporated by reference. In some embodiments, the controller 310 is configured to implement the correlation-based arc fault detection methods described in U.S. Patent Application Publication No. 2020/0264234, published Aug. 20, 2020, and entitled “APPARATUSES AND METHODS FOR PASSIVE FAULT MONITORING OF CURRENT SENSING DEVICES IN PROTECTIVE CIRCUIT INTERRUPTERS,” the entire content of which is hereby incorporated by reference.

FIG. 15 is a block diagram of a control system 1500 of the external device 635 used to create machine learning model 1400, in accordance with some embodiments. As described above, the external device 635 may be implemented as one or more of, a computer terminal, a desktop, a laptop, a smartphone, a tablet, a server, or any other electronic device that includes a memory and an electronic processor capable of training a machine learning model.

As shown, the control system 1500 of the external device 635 includes a controller 1505. The controller 1505 is electrically and/or communicatively connected to a variety of modules or electronic components of the external device 635. For example, the controller 1505 is connected to a power supply circuit 1510, a communication circuit 1515, and a user-interface 1520.

The controller 1505 includes a plurality of electrical and electronic components that provide power, operational control, and protection to the components and modules within the controller 1505 and/or the external device. For example, the controller includes, among other things, an electronic processor 1525 and a memory 1530.

The memory 1530 includes, for example, a program storage area and a data storage area. The program storage area and the data storage area can include combinations of different types of memory, such as read-only memory (ROM) and random access memory (RAM). Various non-transitory computer readable media, for example, magnetic, optical, physical, or electronic memory may be used. The electronic processor