Fire risk detection method using electrical signatures for an electrical system in a building

Abstract

A method and system detect fire risk in a building electrical system. The method monitors loads connected to a circuit in real time, distinguishes loads by category (resistive, inductive / reactive, electronic), and utilizes features defined for each load category to recognize when a specific load category is turned ON in the circuit. The method is agnostic of appliance combinations connected to the circuit, agnostic of inherent load behaviors, and distinguishes arc faults from overcurrent faults. The method informs a user when conditions in the circuit will cause a fault, by distinguishing between load categories and between normal and abnormal behavior for each load category. The user can thus proactively prevent the fault from occurring without disrupting the circuit, in contrast with existing approaches where the user only learns of a fault after the fault has occurred and the user has to manually troubleshoot the fault.

Description

FIRE RISK DETECTION METHOD USING ELECTRICAL SIGNATURES FOR AN ELECTRICAL SYSTEM IN A BUILDINGCROSS REFERENCE TO RELATED APPLICATIONS

[0001] This patent application claims the priority benefit of Indian Provisional Patent Application Serial No. 202411068150, filed September 10, 2024 entitled, “Fire Risk Detection Method Using Electrical Signatures For An Electrical System In A Building”, the contents of which are incorporated by reference.FIELD OF THE INVENTION

[0002] The disclosed concept relates generally to electrical systems in buildings, and in particular, to methods and systems for detecting fire risk in electrical systems in buildings.BACKGROUND OF THE INVENTION

[0003] According to a report issued by the National Fire Protection Association (NFPA) for the U.S. and a report issued by ElectricalDirect for the U.K., electrical fires are among the most common causes of fire. Per the NFPA report, local fire departments responded to an estimated average of 46,700 home fires involving electrical failure or malfunction each year for the years 2015-2019. Home fires involving electrical failure or malfunction caused an estimated average of 390 civilian deaths and 1,330 civilian injuries each year in 2015-2019, as well as an estimated $1.5 billion in direct property damage per year. Per firefighter data, in the U.S. and Canada, electrical fires are the most common household fires in Canada and the U.S.Firefighters in the U.S. respond to 44,880 home fires caused by electrical failure every year.

[0004] The most common causes of electrical fires in the U.S. are due to failures / malfunctions of the following categories of equipment in the following distributions: approximately 50% are due to household wiring, lighting, and power cord failures / malfunctions; approximately 15% are due to cooking equipment; approximately 9% are due to heating equipment; approximately 6% are due to fans / ventilators; approximately 3% are due to air conditioners; and approximately 3% were due to clothes dryers. According to ElectricalDirect report for the U.K., 34% of electrical fires are caused by misuse of appliances and 15% of electrical fires are caused by faulty appliances. Major causes of electrical fires include faultyoutlets and old / outdated outlets; outdated wiring in a home that is very old and may not have the wiring capacity to handle the increased amounts of electrical appliances in today’s average home; installing a bulb with a wattage that is too high for the lamp / light fixture in which the bulb is installed; and appliance misuse wherein an appliance draws too much current, and thus gets overheated, leading to the melting of wire insulation.

[0005] Two types of arc faults can occur in electrical appliances, a parallel arc fault and a series arc fault. A parallel arc occurs when electricity intermittently jumps a gap between wires of different voltages, such as line to line, line to neutral, or line to ground. Because a parallel arc occurs between conductors at different voltages, the amount of current can vary widely, from very little to a very high amount. A series arc occurs when electricity intermittently jumps a gap between two or more points within the same phase. Because a series arc occurs within the same phase, the amount of current is no higher than the load current itself. In contrast with a parallel arc, the amplitude of a series arc fault current tends to be close to that of the normal load current, so it is difficult to detect.

[0006] Although there are products available that can protect circuits in case of excessive current or dangerous arc, detecting fires caused by currents that are lower than set thresholds, as in the case of a series arc, can prove challenging. Certain devices can monitor an electrical waveform and interrupt the circuit when they detect higher magnitude arcs. However, these devices are not present in all homes and some homeowners may find such devices costly. And, while such devices can detect arcs and interrupt circuits, they are unable to give recommendations or insights into the causes of electrical fires.

[0007] The present invention thus provides a fire risk detection solution.SUMMARY OF THE INVENTION

[0008] These needs, and others, are met by embodiments of a system and method for detecting fire risk in a building electrical system. The system monitors the loads connected to a given circuit in the electrical system in real time. The method distinguishes loads by category (resistive, inductive / reactive, electronic) and utilizes computationally efficient features that are defined for each load category to recognize when a specific load type is turned ON in the circuit. The method is agnostic of appliance combinations that can be connected to the circuit and agnostic of inherent load behaviors. The method is further designed to distinguish arc faultsfrom overcurrent faults. By distinguishing between load types and between normal and abnormal behavior for each load type, the system is able to notify a user when present conditions could lead to a fire, so that the user can take preventative action.

[0009] In accordance with one aspect of the disclosed concept, a method for detecting fire risk for a circuit in an electrical system of a building is provided, the circuit being structured to be connected to a number of loads, and the method being executable by a controller. The method comprises: at a Stage 1 of the method, collecting raw electrical data for one voltage cycle of the circuit; at a Stage 2 of the method that immediately follows Stage 1 once one voltage cycle of raw electrical data has been collected, detecting an operating mode for each load in the number of loads using data derived from the raw electrical data; at a Stage 3 of the method that immediately follows Stage 2 once at least one load in the number of loads is determined to be powered ON based on the operating mode detected for each load in the number of loads, evaluating whether the circuit is in a steady state and performing event detection for the circuit; at a Stage 4 of the method that immediately follows Stage 3 when an event is detected at Stage 3, extracting an event current for the event; at a Stage 5 of the method that immediately follows Stage 4, performing a load category evaluation of the event current to determine if the event current corresponds to a resistive load category, a reactive load category, or an electronic load category; at a Stage 6 of the method that immediately follows Stage 5, determining if excessive current is being drawn from the circuit; at a Stage 7 of the method that immediately follows Stage 6, performing an overload condition evaluation for the circuit; at a Stage 8 of the method that immediately follows Stage 7, performing an overcurrent check for the circuit; at a Stage 9 of the method that immediately follows Stage 8, making a decision to perform a load categoryspecific evaluation for each load in the number of loads whose load category was determined at Stage 5, and making a decision not to perform the load category-specific evaluation for any load in the number of loads whose load category was not determined at Stage 5; at a Stage 10 of the method that immediately follows Stage 9, performing a reactive load-specific evaluation when the decision to perform a load category-specific evaluation was made at Stage 9 and at least one load in the number of loads was determined to be in the reactive load category at Stage 5; at a Stage 11 of the method that immediately follows Stage 10, performing an electronic loadspecific evaluation when the decision to perform a load category-specific evaluation was made at Stage 9 and at least one load in the number of loads was determined to be in the electronic loadcategory at Stage 5; at a Stage 12 of the method that immediately follows Stage 11, performing a resistive load-specific evaluation when the decision to perform a load category-specific evaluation was made at Stage 9 and at least one load in the number of loads was determined to be in the resistive load category at Stage 5; at a Stage 13 of the method that immediately follows Stage 3 when no event is detected at Stage 3 and otherwise immediately follows Stage 12, determining if any fault flags are set in the controller; and at a Stage 14 of the method that immediately follows Stage 13, determining if the controller has received any reset request for the fault flags

[0010] In accordance with another aspect of the disclosed concept, a system for detecting fire risk in a building electrical system is provided, the building electrical system including a main panel and a number of branch circuits. The system includes: a load evaluation database, the load evaluation database including a load features hierarchy having a number of load feature layers; an analog to digital converter (ADC), the ADC being structured and configured to collect raw electrical data from a monitored circuit of the number of branch circuits, the raw electrical data including raw voltage and current samples; and a controller, the controller being configured to access the load evaluation database and to receive data from the ADC. The monitored circuit is structured to be connected to a number of loads. The controller is structured and further configured to execute a method for detecting fire risk for the monitored circuit, the method comprising: at a Stage 1 of the method, collecting raw electrical data for one voltage cycle of the monitored circuit; at a Stage 2 of the method that immediately follows Stage 1 once one voltage cycle of raw electrical data has been collected, detecting an operating mode for each load in the number of loads using data derived from the raw electrical data; at a Stage 3 of the method that immediately follows Stage 2 once at least one load in the number of loads is determined to be powered ON based on the operating mode detected for each load in the number of loads, evaluating whether the monitored circuit is in a steady state and performing event detection for the monitored circuit; at a Stage 4 of the method that immediately follows Stage 3 when an event is detected at Stage 3, extracting an event current for the event; at a Stage 5 of the method that immediately follows Stage 4, performing a load category evaluation of the event current to determine if the event current corresponds to a resistive load category, a reactive load category, or an electronic load category; at a Stage 6 of the method that immediately follows Stage 5, determining if excessive current is being drawn from the circuit; at a Stage 7 of themethod that immediately follows Stage 6, performing an overload condition evaluation for the monitored circuit; at a Stage 8 of the method that immediately follows Stage 7, performing an overcurrent check for the monitored circuit; at a Stage 9 of the method that immediately follows Stage 8, making a decision to perform a load category-specific evaluation for each load in the number of loads whose load category was determined at Stage 5, and making a decision not to perform the load category-specific evaluation for any load in the number of loads whose load category was not determined at Stage 5; at a Stage 10 of the method that immediately follows Stage 9, performing a reactive load-specific evaluation when the decision to perform a load category-specific evaluation was made at Stage 9 and at least one load in the number of loads was determined to be in the reactive load category at Stage 5; at a Stage 11 of the method that immediately follows Stage 10, performing an electronic load- specific evaluation when the decision to perform a load category-specific evaluation was made at Stage 9 and at least one load in the number of loads was determined to be in the electronic load category at Stage 5; at a Stage 12 of the method that immediately follows Stage 11, performing a resistive load- specific evaluation when the decision to perform a load category-specific evaluation was made at Stage 9 and at least one load in the number of loads was determined to be in the resistive load category at Stage 5; at a Stage 13 of the method that immediately follows Stage 3 when no event is detected at Stage 3 and otherwise immediately follows Stage 12, determining if any fault flags are set in the controller; and at a Stage 14 of the method that immediately follows Stage 13, determining if the controller has received any reset request for the fault flags.BRIEF DESCRIPTION OF THE DRAWINGS

[0011] A full understanding of the invention can be gained from the following description of the preferred embodiments when read in conjunction with the accompanying drawings in which:

[0012] FIG. 1 is a waveform graph showing multiple random power changes in an electrical circuit powering a furnace due to inherent load behavior of the furnace;

[0013] FIG. 2A is a waveform showing how a single non-dedicated circuit having one or more loads connected is affected by the load(s) being turned OFF and ON in any random sequence;

[0014] FIG. 2B shows a pair of waveforms showing the electrical activity for a blenderconnected to the non-dedicated circuit, as indicated in FIG. 2A;

[0015] FIG. 2C shows a pair of waveforms showing the electrical activity for an air conditioner (AC) connected to the non-dedicated circuit, as indicated in FIG. 2A;

[0016] FIG. 2D shows a pair of waveforms showing the electrical activity for a fan connected to the non-dedicated circuit, as indicated in FIG. 2A;

[0017] FIG. 2E shows a pair of waveforms showing the electrical activity for a vacuum cleaner connected to the non-dedicated circuit, as indicated in FIG. 2A;

[0018] FIG. 3 is a group of current and voltage waveforms that show how the voltage and current signatures for inductive loads, electronic loads, and resistive loads differ from one another;

[0019] FIGS. 4A, 4B, and 4C each show a pair of electrical waveforms for one type of load under both arcing and non-arcing conditions, with FIG. 4A showing graphs for an electrical circuit powering a resistive load (an electric kettle), with FIG. 4B showing graphs for an electrical circuit powering an inductive / reactive load (an air conditioner), and with FIG. 4C showing graphs for an electrical circuit powering an electronic load (a laptop);

[0020] FIG. 5 is a block diagram representation of a fire risk detection system, in accordance with an example embodiment of the disclosed concept, configured to monitor the electrical system of a building;

[0021] FIG. 6 is a flow chart of a method for detecting fire risk in an electrical system, in accordance with an example embodiment of the disclosed concept;

[0022] FIG. 7 shows a waveform of the electrical signal detected at a single nondedicated circuit having one or more loads connected in random order, annotated to provide an example of how an event detection current is obtained during an event current extraction step of the method shown in FIG. 6;

[0023] FIGS. 8 A, 8B, and 8C are waveforms for a resistive load, with FIG. 8 A showing a voltage waveform, FIG. 8B showing a current waveform under a normal operating condition, and FIG. 8C showing a current waveform under an arc fault condition;

[0024] FIG. 9 is a hierarchical load feature database used during a load category evaluation performed during the fire risk detection method;

[0025] FIG. 10 is a block diagram of data acquisition functions used during the load category evaluation performed during the fire risk detection method; and

[0026] FIG. 11 is a block diagram of a hierarchical load identification system architecture used during the load category evaluation performed during the fire risk detection method.DETAILED DESCRIPTION OF THE INVENTION

[0027] Directional phrases used herein, such as, for example, left, right, front, back, top, bottom and derivatives thereof, relate to the orientation of the elements shown in the drawings and are not limiting upon the claims unless expressly recited therein.

[0028] As used herein, the singular form of “a”, “an”, and “the” include plural references unless the context clearly dictates otherwise.

[0029] As employed herein, employed herein, when ordinal terms such as “first” and “second” are used to modify a noun, such use is simply intended to distinguish one item from another, and is not intended to require a sequential order unless specifically stated.

[0030] As employed herein, the term “controller” shall mean a programmable analog and / or digital device that can store, retrieve and process data; a processor; a control circuit; a computer; a workstation; a personal computer; a microprocessor; a microcontroller; a microcomputer; a central processing unit; a mainframe computer; a mini-computer; a server; a networked processor; or any suitable processing device or apparatus.

[0031] As employed herein, the term “number” shall mean one or an integer greater than one (i.e., a plurality).

[0032] Most fires start with a spark. The innovative fire risk detection system 10 and fire risk detection method 100 disclosed herein, which are detailed further in connection with FIG. 5- 6, enable monitoring of voltage and current levels of electrical appliances in order to detect conditions that are likely to lead to a spark so that corrective action can be taken to prevent a fire from starting. In order to provide context for the needs addressed by the disclosed fire risk detection system 10 and method 100, some additional information on the complexities of appliance usage that can lead to electrical fires and that are accounted for by the disclosed system 10 and method 100 will be discussed first in conjunction with FIGS. 1-3.

[0033] In a home, there can be dedicated and non-dedicated circuits having a hardwired connection to the bus of the building’s main power supply. A dedicated circuit is a circuit having a fixed load connected, e.g., refrigerator, air conditioner (AC), washing machine,dishwasher, etc. A non-dedicated circuit is a circuit that can have movable loads connected, e.g., laptop charger, table lamp, coffee maker, vacuum cleaner, iron, etc. Loads connected to nondedicated circuits may change in real time, and as such, observing usage of appliances connected to non-dedicated circuits over some period of time and trying to use that historic data to prevent conditions leading to fire may lead to nuisance trips, because historic usage of non-dedicated circuits tends not to be predictive of future usage.

[0034] With some appliances such as furnaces, multiple power changes are observed due to the inherent load behavior itself. For example, FIG. 1 shows multiple random power changes in an electrical circuit powering a furnace due to inherent load behavior of the furnace. Although the highest peak in the electrical waveform shown in FIG. 1 is significant, the duration of the high magnitude wattage is momentary and would not necessitate tripping of the connected circuit breaker. However, if the circuit breaker connected to the circuit is only configured to monitor a rise in current and compare the monitored current to a prestored threshold without considering additional factors (e.g. such as duration of the over-threshold current value), then the circuit breaker would trip the circuit unnecessarily, which would constitute a nuisance trip.

[0035] FIG. 2A shows how a single non-dedicated circuit having a single load or multiple loads connected is affected by the load(s) being turned OFF and ON in any random sequence. In such a scenario, loads consuming higher power may dominate loads consuming lower power. Specifically, in FIG. 2A, a blender, an air conditioner (AC), a fan, and a vacuum cleaner are each turned ON at different times and turned OFF at different times. It can be seen in FIG. 2A that when the fan turns ON after the AC is turned ON, the contribution of the fan to the electrical waveform is not easily discernible. The current vs. time and current vs. voltage waveforms for each appliance in FIG. 2 A is enlarged in one of FIGS. 2B-2E.

[0036] Still referring to FIG. 2A, for a single non-dedicated circuit having a single load or multiple loads connected, a series arc fault caused by low power loads may not be detectable. Another scenario that can prevent detection of pre-fire conditions is that in which a power strip is connected to the circuit, which cannot be detected by monitoring software. The power strip can be overloaded unknowingly, and an overloaded power strip will cause voltage to drop in the circuit. However, it is difficult to differentiate between a voltage drop due to overload and a voltage drop due to utility undervoltage.

[0037] FIG. 3 provides a side-by-side comparison of how the voltage and currentsignatures for inductive loads, electronic loads, and resistive loads differ from one another. In FIG. 3, a fluorescent lamp (inductive load), an LED lamp (electronic load), and an incandescent lamp (resistive load) are compared. While all three have a startup time of 0.15 seconds or less, the startup current profiles and V-I trajectory vary widely among the three types of loads. Because the current and voltage profiles of the different types of loads vary so widely, using only transient duration to classify load type can lead to the load category being misclassified. Noise may not be seen in resistive loads when arcing occurs, so there is a need for special handling.

[0038] The foregoing discussion of FIGS. 1-3 is provided to make apparent the complexities that arise in monitoring loads and in attempting to predict future behavior based on observed appliance usage. Due to these complexities, existing approaches are directed toward analyzing the harmonics in high frequency bands (up to MHz). Such analysis requires the use of additional computation resources and specific hardware that increase the overall system cost. Also, there is a need for specific communication between a main breaker and branch breakers. For the specific case of a series fault, the fire risk detection features claimed to be included in a fire risk management system may not work for all resistive and inductive / reactive loads. In addition, a real time implementation approach wherein the fire risk prevention strategy can be adapted to changing conditions in the monitored circuit does not appear to exist. Furthermore, in existing approaches, the user only becomes aware of a fault after the fault has already occurred, i.e. due to one of the circuits having tripped, and identifying the fault requires manual troubleshooting.

[0039] The improved method 100 and system 10 disclosed herein address all of the challenges of existing approaches noted above, as the method 100 and system 10 use minimal memory and computing resources to monitor electrical conditions in real time, and notify a user of fault-imminent conditions in advance of the fault actually occurring so that the user can act to prevent the fault from occurring, thus eliminating the need to perform manual troubleshooting.

[0040] FIGS. 4A, 4B, and 4C respectively show electrical waveforms under arcing and non-arcing conditions for an electrical circuit powering a resistive load (electric kettle), an electrical circuit powering an inductive / reactive load (air conditioner), and an electrical circuit powering an electronic load (laptop), provided by the reference paper Jiang et al. (Jiang, W., Eiu, B., Yang, Z., Cai, H., Fin, X., & Xu, D. (2023). Non-Intrusive Arc Fault Detection and Eocalization Method Based on the Mann-Kendall Test and Current Decomposition. Energies,16(10), 3988. Available at https: / / doi.org / 10.3390 / enl6103988.). It is apparent from FIGS. 4A- 4C that each type of load (i.e. either resistive, inductive, or electronic) has a current signature distinct from the other two types of loads during normal use, and that arcing causes variations in the current signature for each type of load. The disclosed improved fire risk detection system 10 and method 100 for detecting fire risk in an electrical system utilize the unique features of the current signature under non-arc conditions for each type of load that are shown in FIGS. 4A-4C.

[0041] Reference is now made to FIG. 5 and FIG. 6, which respectively show the innovative fire risk detection system 10 and a flow chart of the innovative method 100 for detecting fire risk in an electrical system in a building, in accordance with an exemplary embodiment of the disclosed concept. The method of FIG. 6 may be employed, for example and without limitation, with the fire risk detection system 10 shown in FIG. 5 and is described in conjunction with the system fire risk detection system 10 shown in FIG. 5. It is expected that the system fire risk detection system 10 and method 100 would be especially useful in a residential building, but it will be apparent from the following description of the method 100 that the method 100 can be readily implemented in other types of buildings as well.

[0042] The fire risk detection system 10 is a controller configured to execute a software routine that implements the method 100. The controller of the fire risk detection system 10 is thus also referred to herein as the “controller 10”. The fire risk detection system 10 includes a load evaluation database 11 that is used to store and provide data during the execution of the method 100. Among other things, the load evaluation database 11 includes fault flags for overload and arc faults. As is customary, each fault flag has a default value of 0 to indicate that no fault has been detected, and when a fault is detected during the method 100, the appropriate fault flag is set to a value of 1. When a fault flag is described as being “set” herein, that signifies that the fault flag has been set to a value of 1. When a fault flag is described as being “reset” herein, that signifies that the fault flag has been reset to a value of 0 after previously having been set to a value of 1. The load evaluation database 11 includes a load features hierarchy 12 having a number of load feature layers 14, which are discussed in conjunction with a load category evaluation method executed at step 112. The fire risk detection system 10 further comprises an analog to digital converter (ADC) 16 used to collect raw voltage and current samples during the method 100.

[0043] FIG. 5 further shows a building electrical system 20, with the building electricalsystem 20 comprising a main panel 21, a number of branch circuits 22, and a number of receptacles 24. Each receptacle 24 is hardwired to one of the branch circuits 22 and is structured to enable a number of load devices to receive power from the branch circuit 22. It will be appreciated that, in an actual building, there would be multiple receptacles 24 connected to most branch circuits 22, but for the sake of simplicity, each branch circuit 22 is depicted in FIG. 5 as having one receptacle 24 connected.

[0044] The method 100 is designed for use in monitoring the electrical system of an entire building such as the building electrical system 20, but is implemented at the branch circuit level such that, in FIG. 5, the fire risk detection system 10 simultaneously executes a separate instance of the method 100 for each branch circuit 22 that is to be monitored. As a non-limiting illustrative example, if a first branch circuit 22 shown in FIG. 5 is designated for the kitchen while a second branch is designated for the laundry room, then the fire risk detection system 10 would execute a first instance of the method 100 for the kitchen circuit and would execute a separate second instance of the method 100 for the laundry room. Hereinafter, the term “monitored circuit” is used to refer to a single branch circuit being monitored for fire risk via the method 100, such as one of the branch circuits 22 shown in FIG. 5, and the reference number 22 is used in conjunction with the term “monitored circuit”.

[0045] It is noted that the receptacles 24 only need to be standard receptacles (as opposed to smart receptacles, which can perform computing tasks and exchange data with other devices), as the method 100 is directed to performing a branch circuit-level analysis. That is, if the controller 10 determines that there is a problem in the monitored circuit 22 and all receptacles 24 in the circuit 22 are standard receptacles, then the controller 10 will alert the user to check all loads connected to the circuit 22 when a fault-imminent condition is detected in the circuit 22, as the controller 10 can identify that such a condition exists in the circuit 22 but not necessarily which specific receptacle 24 the fault condition is localized to. However, if any of the receptacles 24 connected to the monitored circuit 22 is a smart receptacle and a fault condition is present at the smart receptacle, then the controller 10 can additionally inform the user of the specific receptacle where the fault condition is present.

[0046] As an initial matter, it is noted that the method 100 is intended to be continuously run while the fire risk detection system 10 is monitoring the building electrical system 20. There is a START block at the beginning and a STOP block at the end of the flowchart shown in FIG.6, and it should be noted these START and STOP blocks are intended to refer to the start and stop of a single iteration of the method 100 run for a single voltage cycle in a single monitored circuit 22. That is, within a given monitored circuit 22, after a single iteration of the method 100 is complete, if the fire risk detection system 10 is still monitoring the monitored circuit 22, then the method 100 is performed again for the next voltage cycle, starting at step 101. The method 100 comprises a few dozen individual steps, and can also be segmented into stages, with each stage comprising either a plurality of consecutive individual steps or at least one step. As such, in addition to the reference numbers used to number each individual step in FIG. 6, there are 14 stages (Stage 1 - Stage 14) labeled in the figure, with the steps belonging to each stage being indicated with a bracket in the flow chart.

[0047] The method 100 will now be discussed in detail. Stage 1 includes steps 101-103 and is directed to collection of raw electrical data. At step 101, a calibration routine is run to calculate any DC offset (if any) that may be present in current and voltage readings due to the hardware of the building electrical system 20. That is, to the extent that the hardware of the main panel 21 introduces an inherent non-zero baseline current or voltage value, running the calibration routine enables the fire risk detection system 10 to separate those baseline values from the values of the actual loads connected to the monitored circuit 22. At step 102, raw ADC voltage and current samples are read from the monitored circuit 22. At step 103, if raw voltage and current data is not available, the method returns to step 102 to obtain ADC samples. If raw voltage and current data is available at step 103, the method proceeds to step 104 and Stage 2.

[0048] Stage 2 includes steps 104-107 and is directed toward detecting an operating mode for each load that is turned ON. At step 104, RMS voltage (Vrms) and RMS current (Irms) are calculated using the cycle raw voltage and current values available from Stage 1. In one non-limiting illustrative example, if 32 samples per voltage cycle were considered the minimum acceptable sampling rate, Vrms and Irms would be calculated once 32 samples are available. At step 105, mode detection algorithm features are extracted, and at step 106, a mode detection algorithm is run using the mode detection algorithm features extracted at step 105 in order to determine an operating mode of the load. U.S. Pat. Appl. Pub. No. 2016 / 0116508 Al discloses a method for extracting mode detection features that can be used at step 105 and also discloses a mode detection algorithm that can be used at step 106. The method used at step 105 and the algorithm used at step 106 are discussed in further detail later herein after the detaileddiscussion of the method 100. At Step 107, the controller 10 determines if any load is ON in the monitored circuit 22 based on the operating mode determined at step 106. If no load is ON, the method returns to step 102. If any load is determined to be ON at step 107, then the method proceeds to step 108 and Stage 3.

[0049] Stage 3 includes steps 108-109 and is directed toward steady state evaluation and event detection. It is noted that multiple loads can be connected to the monitored circuit 22 and that each load can be turned ON or OFF independently of every other load connected to the monitored circuit 22. In Stage 3, the term “event detection” refers to determining that at least one load connected to the monitored circuit 22 has turned from ON to OFF or from OFF to ON from an immediately preceding interval of time to the present interval of time. The term “interval of time” as used in the context of Stage 3 should be long enough that a stable state can be detected with reasonable certainty but short enough to detect with reasonable precision when an event has occurred. Given that the power supplied is at utility / mains frequency of 50 Hz or 60 Hz, it is suggested that using Irms data from one to two seconds’ worth of voltage cycles (i.e. 50-100 cyles for 50 Hz supply frequency and 60-120 cycles for 120 Hz supply frequency) would be sufficient for detecting a change in stable state and an event, in one non-limiting illustrative example. The interval of time will be described with more specificity in relation to each of steps 108 and 109 hereafter.

[0050] At step 108, the change in Irms between the immediately preceding interval of time and the present interval of time is calculated as a fractional or percentage change. That is, the mean Irms of the present interval of time is compared to the mean Irms of the immediately preceding interval of time, and the change between the mean Irms of the present interval and the mean Irms of the immediately preceding interval is determined as a fraction / percentage of the Irms of the immediately preceding interval. In a non-limiting illustrative example where the interval of time is defined as one second’s worth of voltage cycles for a power supply frequency of 60 Hz, step 108 constitutes the following four sub-steps: (1) finding the mean Irms of the 60 most recent voltage cycles (one second’s worth of voltage cycles for the present interval of time), (2) finding the mean Irms of the 60 voltage cycles preceding the 60 most recent cycles (one second’s worth of voltage cycles for the immediately preceding interval of time), (3) determining the difference between the Irms for the present interval and the Irms for the immediately preceding interval, and (4) expressing the difference as a fraction / percentage of the Irms for theimmediately preceding interval of time.

[0051] At step 109, the percentage change in Irms found at sub-step 4 of step 108 is compared to a predetermined event detection threshold to determine whether or not the monitored circuit 22 is in a steady state and if an event has occurred. If the percentage change in Irms calculated at step 108 is less than the event detection threshold, then the controller 10 determines at step 109 that the monitored circuit 22 is in a steady state, and if the change in Irms calculated at step 108 is greater than or equal to the event detection threshold, then the controller 10 determines at step 109 that the monitored circuit 22 is not in a steady state in the present interval and that an event has occurred between the previous interval and the present interval. In one non-limiting illustrative example, if the event detection threshold is 0.1 (i.e. 10%), then the controller 10 will determine that the monitored circuit 22 is in a steady state if the change in Irms between the immediately preceding interval and the present interval is less than 10%, and will determine that the monitored circuit 22 is not in a steady state (i.e. that an event has occurred) if the change in Irms is greater than or equal to 10%.

[0052] If the change in Irms was not greater than the event detection threshold at step 109, then the method proceeds to step 152 and Stage 13, detailed later herein. If the change in Irms was greater than the event detection threshold at step 109, then the method proceeds to step 110 and Stage 4. It is noted that, after the controller 10 initially commences executing the method 100, there may be insufficient data to determine the change in Irms at step 108 during the initial several iterations of the method 100. Returning briefly to the non-limiting illustrative example in which the interval of time at step 108 is defined as one second’s worth of voltage cycles for a power supply frequency of 60 Hz, it is noted that there will be insufficient voltage cycle data to determine the change in Irms at step 108 prior to there being 60 voltage cycles’ worth of data (i.e. one second’s worth of voltage cycle data). In this case, when the method 100 proceeds to step 109, the controller 10 will determine that no change in state greater than the event detection threshold has occurred, and the method will proceed to step 152 and Stage 13.

[0053] Stage 4 includes steps 110-111 and is directed toward finding the event current. At step 110, the event current is extracted by subtracting the previous state current from the present state current. FIG. 7 provides an illustrative example showing, for a present iteration of the method 100, what constitutes the present state current and the previous state current. The difference in the two states’ currents is attributed to a load, i.e. to a load either being turned ONor being turned OFF. It is noted that the method 100 is designed such that there is no need to perform a complete load disaggregation analysis or clustering per load analysis at step 110.

[0054] At step 111, the controller 10 newly stores the Irms of the present state (i.e. the present state Irms of steps 101-110) as being the previous state Irms for the next iteration of the method 100 that will be performed for the next voltage cycle. Stage 5 includes step 112 and is directed toward performing a load category evaluation of the event current. At step 112, a load category evaluation is performed. The method used for evaluating load category at step 112 is disclosed in U.S. Pat. Appl. Pub. No 2013 / 0138669 and in U.S. Patent Number 9,819,226. The method used at step 112 is discussed in further detail later herein after the detailed discussion of the method 100. It is noted that, because the analysis performed during the method 100 is performed at the level of the branch circuit 22, the load category evaluation at step 112 is determined based on whichever load(s) dominates the electrical signal being read from the monitored circuit 22. As a non-limiting illustrative example, if there is an inductive load component to the electrical signal read from the monitored circuit 22 and that inductive load component is greater than the components contributed by any resistive or electronic loads that are also connected to the monitored circuit 22, then the signal for the entire monitored circuit 22 will be categorized as inductive.

[0055] The method proceeds to step 113 and Stage 6, which includes steps 113-115 and is directed toward determining if excessive current is being drawn from the circuit. At step 113, the present cycle Irms is compared to the previous cycle Irms and the fault flag is checked. If the present cycle Irms is greater than the previous cycle Irms and the fault flag is set at zero, then the method proceeds to step 114. Otherwise, the method proceeds from step 113 to step 116 and Stage 7, detailed later herein. It is noted that if a set fault flag were detected during step 113, that would be due to the fault flag having been set during a previous iteration of the method performed for a previous voltage cycle. At step 114, the maximum value of Irms per load category from the signal read for the present voltage cycle is stored. At step 115, the RMS current for the latest / present voltage cycle is set as the present Irms.

[0056] The method proceeds to step 116 and Stage 7, which includes steps 116-121 and is directed toward an overload condition evaluation. At step 116, the controller 10 evaluates whether there has been a voltage drop. At step 117, if a voltage drop was not observed, then the method proceeds to step 122 and Stage 8, detailed later herein. At step 117, if a voltage dropwas observed, then the method proceeds to step 118. It should be noted that there will be a detectable voltage drop when a load draws excessive current as compared to the circuit capacity. When consistent voltage drop is seen, a fault flag will be set (as will be detailed later herein in connection with specific steps of the method 100). In this case, the maximum Irms value is not stored in the database, which is why the method proceeded from step 113 to step 116 if the fault flag had been set prior to step 113. Typically, overcurrent generates excessive heat which leads to arcing and then can generate fire. If the present Irms value exceeds a predetermined ceiling value, then the stored maximum Irms value will get checked, in order to determine if unattended or faulty loads are drawing excessive current. In the case of a non-dedicated circuit, a maximum tolerable value should be stored in the database for the non-dedicated circuit, since high power consuming loads like a vacuum cleaner or an iron can be connected and disconnected at any time.

[0057] At step 118, an inconsistent cycle count for voltage drop is incremented, i.e. to the extent that the voltage drop between two consecutive voltage cycles is greater than a predetermined voltage drop threshold, then the inconsistent cycle count for voltage drop is incremented by one, with the count being incremented by one each time a new voltage cycle starts and said new voltage cycle exhibits a voltage drop greater than the voltage drop threshold in comparison to the previous voltage cycle. At step 119, if the inconsistent cycle count for voltage drop is not greater than or equal to 50, then the method proceeds to step 122 and Stage 8, detailed later herein.

[0058] At step 119, if the inconsistent cycle count for voltage drop is greater than or equal to 50, then the method proceeds to step 120. When there is a voltage drop, the cause may be either an overloaded circuit or a utility undervoltage, and at step 120, the controller 10 sends an alert or notification to the user to inform that user that: (1) the circuit is overloaded or there is utility undervoltage, (2) advising the user to check any plugged in power strip, if a power strip is present, (3) a connected load is drawing current exceeding the capacity of the circuit and advising the user to check for an unattended ON load (i.e. appliance), and (4) advising the user to switch OFF the appliance. At step 121, the overload fault flag is set. The method then proceeds to step 122 and Stage 8. It is noted that, even though the overload fault flag is set at step 121, further conditions are evaluated in the subsequent steps of the method.

[0059] The method proceeds to step 122 and Stage 8, which includes steps 122-123 andis directed toward performing an overcurrent check when a fault flag was previously set. At step 122, the Irms is checked to see if it is greater than or equal to 1.5 times the maximum value stored in the real time database per load category. If the Irms is greater than or equal to 1.5 times the maximum value stored and the overload fault flag is still set at a value of 1 (i.e. from step 121), then the method proceeds to step 123. At step 123, if the Irms is greater than the predetermined threshold condition (1.5 times the maximum value stored in the real time database per load category), then the controller 10 sends an alert or notification to the user indicating that a load is using too much current and / or there is an overloaded circuit and / or there is an unattended load. If, however, the Irms is greater than or equal to 1.5 times the maximum value stored and the overload fault flag is no longer set at a value of 1 , then the method proceeds to step 124 and Stage 9.

[0060] Stage 9 includes step 124 and is directed toward performing a load categoryspecific evaluation, if applicable. As previously noted in connection with step 112, a load category evaluation is performed at step 112 only if the controller 10 determined at step 108 that the circuit 22 is in the steady state. As such, at step 124, if the load category is known (i.e. due to the circuit 22 being in the steady state at step 108 and the load category evaluation being performed at step 112), then the method can proceed through any applicable load categoryspecific evaluations at Stage 10 (starting with step 125), Stage 11 (starting with step 134), or Stage 12 (starting with step 143), as will be detailed later herein. However, if the load category is unknown at step 124 (i.e. due to the controller 10 determining at step 108 that the circuit 22 is not in the steady state), then the only risk condition assessments performed are the undervoltage (i.e. overload) check that was performed in Stage 8 and the overcurrent check that was performed in Stage 9.

[0061] When the load category is unknown at step 124, the method proceeds to step 125 to query whether the load category is known to be inductive / reactive. Since the answer is “no”, the method proceeds to step 134 to query whether the load category is known to be electronic. Since the answer is “no”, the method proceeds to step 143 to query whether the load category is known to be resistive. Since the answer is “no”, the method proceeds to step 152 and Stage 13, which includes steps 152-153 and is directed toward determining if any fault flags are set.

[0062] At step 152, if it is determined that no fault flag is set, then the method proceeds to step 153 to issue a notification to the user indicating that the monitored circuit is operatingnormally. If it is determined at step 152 that at least one fault flag is set, i.e. at least one of the overload fault flag or the arc fault flag, then the method proceeds to step 154 and Stage 14, which includes steps 154-155 and is directed toward determining whether or not a reset request has been received for any fault flags. It is noted that a reset request would be received after a user has attended to and resolved any earlier-identified faults. At step 154, if a reset fault flag request has not been received, then the present iteration of the method 100 is completed for the present voltage cycle. At step 154, if a reset fault flag request has been received, then the method proceeds to step 155, where any fault flag that is set at 1 is then reset to 0. After step 155, the present iteration of the method 100 is completed for the present voltage cycle.

[0063] As previously noted, after a single iteration of the method 100 is complete, if the fire risk detection system 10 is still monitoring the building electrical system 20, then the method 100 is performed again for the next voltage cycle, starting at step 101. It is worth noting that when the circuit is in a steady state and there is no load that has just been turned ON or OFF, the method proceeds to step 152 immediately after step 109, such that the present iteration of the method is essentially complete after step 109, since it would not be expected for there to be any fault flags set prior to step 109, and steps 152 and 154 are directed toward addressing the cause of any set fault flags and then resetting the fault flags.

[0064] As previously stated, when the load category is known at step 124, the method can proceed to an inductive / reactive load-specific evaluation that starts at step 125, an electronic load-specific evaluation that starts at step 134, or a resistive load-specific evaluation that starts at step 143. For the inductive / reactive load-specific evaluation, the method proceeds from step 125 to step 126, where cross correlation between the present current cycle and the previous current cycle is performed, then to step 127 where the DC offset and number of fluctuations in the cycle are determined. Regarding the cross correlation at step 126, it should be noted that during normal operation, the cross correlation value will be closer to 1. During an arc fault, the cross correlation value will be closer to 0, or less than a defined threshold range which indicates that adjacent cycles do not have the same Irms value. Regarding the DC offset determination at step 127, it should be noted that during normal operation, the DC offset is close to 0. During an arc fault, the DC bias increases or decreases depending on whether the voltage is in the positive or negative half-cycle, as positive and negative half-cycles do not exhibit similar behavior, and random noise may be generated during zero crossing.

[0065] At step 128, the controller 10 performs evaluation of membership functions based on the data gathered at steps 126-127. Specifically, performing evaluation of membership functions pertains to determining whether an arc is present in the monitored circuit 22. The membership functions include fuzzy sigmoid double sigmoid functions and are evaluated based on preaquired data obtained from various device types (i.e. layer 14B in FIG. 9) within the load category (i.e. layer 14A in FIG. 9). The preacquired data includes healthy data (corresponding to normal electrical behavior for the load category) and faulty data (corresponding to electrical behavior under arc conditions for the load category). A heuristic fuzzy approach for determining parameters of membership functions automatically is used. At step 129, if the fault probability is determined to be less than or equal to 70%, then the method proceeds to the electronic load determination at step 134. At step 129, if the fault probability is determined to be greater than 70%, then the method proceeds to step 130 to increment a counter for consecutive inconsistent cycles in the inductive / reactive load category. At step 131, if the inconsistent cycle count is less than 50, then the method proceeds to the electronic load determination at step 134. At step 131, if the inconsistent cycle count is greater than or equal to 50, then the method proceeds to step 132 where the controller 10 notifies the user that there is an arc fault and advises the user to check for a loose connection, broken wire, visual sparks, or noise. The method proceeds to 133, where an arc fault flag is set, and then proceeds to the electronic load determination step 134.

[0066] For the electronic load-specific evaluation, the method proceeds from step 134 to step 135, where the number of fluctuations observed in the cycle are calculated, then to step 136, where the DC offset for the cycle is calculated. Regarding the calculation of fluctuations at step 135, it should be noted that during normal operation, the number of fluctuations will be minimum or null. During an arc fault, the number of fluctuations will be more, i.e. non-trivial in number. Regarding the DC offset determination at step 136, it should be noted that during normal operation, the DC offset is close to 0. During an arc fault, the DC bias increases or decreases depending on whether the voltage is in the positive or negative half-cycle, as positive and negative half-cycles do not exhibit similar behavior, and random noise can be observed at any position of the waveform. In addition, positive and negative spikes will not match with one another.

[0067] At step 137, the controller 10 performs evaluation of membership functions based on the data gathered at steps 135-136 to determine whether an arc is present in the monitoredcircuit 22. The evaluation of membership functions performed at step 137 is the same as that previously discussed in connection with step 128. At step 138, if the fault probability is determined to be less than or equal to 70%, then the method proceeds to the resistive load determination at step 143. At step 138, if the fault probability is determined to be greater than 70%, then the method proceeds to step 139 to increment a counter for consecutive inconsistent cycles in the electronic load category.

[0068] At step 140, if the inconsistent cycle count is less than 50, then the method proceeds to the resistive load determination step at 143. At step 140, if the inconsistent cycle count is greater than or equal to 50, then the method proceeds to step 141 where the controller 10 notifies the user that there is an arc fault and advises the user to check for a loose connection, broken wire, visual sparks, or noise. The method proceeds to 142, where an arc fault flag is set, and then proceeds to the resistive load determination step 143.

[0069] For the resistive load- specific evaluation, the method proceeds from step 143 to step 144, where the voltage and current are normalized, then to step 145, where the Euclidean distance between voltage and current in the first half of the positive half cycle and Euclidean distance between voltage and current in the first half of the negative half cycle are calculated. Referring to FIGS. 8A-8C, it is noted that during normal operation for a resistive load, the voltage will always be in phase with the current. During an arc condition, the current ramp up time in both the positive and negative half-cycle gets delayed as the arc fault progresses.Regarding the normalization that is performed at step 144, it is noted that the voltage and current is normalized between values of - 1 and 1. Regarding the Euclidean distance that is calculated at step 145, during normal operation, the Euclidean distance between the voltage and current ramp up time will be closer to 0, whereas during an arc condition, a greater and greater Euclidean distance between the ramp up time of the voltage and the current will be observed in the positive half-cycle and the negative half-cycle of complete voltage cycles.

[0070] At step 146, the controller 10 performs evaluation of membership functions based on the data gathered at steps 144-145 to determine whether an arc is present in the monitored circuit 22. The evaluation of membership functions performed at step 146 is the same as that previously discussed in connection with step 128. At step 147, if the fault probability is determined to be less than or equal to 70%, then the method proceeds to the fault flag determination at step 152. At step 147, if the fault probability is determined to be greater than70%, then the method proceeds to step 148 to increment a counter for consecutive inconsistent cycles in the resistive load category.

[0071] At step 149, if the inconsistent cycle count is less than 50, then the method proceeds to the fault flag determination at step 152. At step 149, if the inconsistent cycle count is greater than or equal to 50, then the method proceeds to step 150 where the controller 10 notifies the user that there is an arc fault and advises the user to check for a loose connection, broken wire, visual sparks, or noise. The method proceeds to 151, where an arc fault flag is set, and then proceeds to the fault flag determination at step 152.Discussion of Mode Detection Feature Extraction Method Used for Step 105 and Mode Detection Algorithm Used for Step 106 Disclosed in U.S. Pat. Appl. Pub. No.2016 / 0116508A1:

[0072] A discussion of the method for extracting mode detection features used at step 105 and the mode detection algorithm used at step 106 will now be presented. As previously noted, U.S. Pat. Appl. Pub. No. 2016 / 0116508 Al discloses the mode detection feature extraction method used at step 105 and also discloses the mode detection algorithm that used at step 106. Six operating modes that the method used at step 105 and the algorithm used at step 106 are designed to detect include a load operating mode Ml (i.e. the load is powered on and being used for its intended purpose), a load low power mode M2 (e.g., without limitation, standby; hibernating; energy saving), a parasitic mode M3 (the load is locally switched off but is still electrically connected to utility / mains power and is still consuming a relatively small amount of power), and three modes that apply only to outlets of a power strip: a mode M4, a mode M0, and a mode MOO. That is, the modes M4, M0, and MOO pertain to a power strip that is connected to the branch circuit 22. The mode M4 occurs when a load is connected to the power strip and the load is powered ON, the mode M0 occurs when a load is connected to the power strip and the load is powered OFF, and the mode MOO occurs when there are no loads connected to the power strip. The M4, M0, and MOO modes are combined for the purpose of the condition evaluation.

[0073] For characterization of the operating modes, three features are calculated at step 105 per Equations 1-3:(Eq. 3) wherein:THD>7 is total harmonic distortion greater than the seventh harmonic;IRMS is RMS current;Ii is current at the first harmonic; l3_nomis nominal current at the third harmonic; l5_nomis nominal current at the fifth harmonic; l7_nomis nominal current at the seventh harmonic;Pavg is average power; n is an integer number of samples; k is an integer; v[k] is the k* voltage sample; i[k] is the k* current sample;A is area of a voltage-current (VI) plot;N is an integer number of samples in the VI plot; i is an integer; andXi and yi are the 1thnormalized voltage and 1thnormalized current samples, respectively, in the VI plot.

[0074] The mode detection algorithm main logic for Ml, M3 and M4 mode differentiation is as follows: if(mf5_l(real power) >= 0.7) mode ID = “Ml” else if (area < 0) / / negative area fnAreaMode3 = mf4_l(area)fnAreaModel = mf4_2(area) else / / positive area fnAreaMode3 = mf4_3(area) fnAreaModel = mf4_4(area) end

[0075] The probability of the mode being M4, M3 or Ml is calculated from respective Equations 5-7, with the yy[] array being sorted in descending order, and the mode with the highest probability being the winner: yy[0] = mfl_l(realPower) * mf2_l(THD>?)(Eq. 5) yy

[0001] = mf3_l(realPower) * fnAreaMode3(Eq. 6) yy[2] = mf2_2(THD>7) * fnAreaModel * mfl_2(realPower)(Eq. 7)

[0076] The end results are available in the yy[] array, where yy[0] stores the probability of M4, yy

[0001] stores the probability of M3, and yy[2] stores the probability of Ml. The Mode Type ID first winner is the mode with the highest probability in yy[], and the Mode Type ID second winner is the mode with the second highest probability in yy[]. The Probability difference = 1 - (probability of second winner / probability of first winner).

[0077] For M2 detection, M2 is always followed by Ml. Thus, in order to detect M2, the load has to go to Ml at least once after it powers ON. The only major difference between the two modes is the power level, with M2 power being less than Ml power. When the load goes from Ml to M2, the real power step down ratio is <0.5. The logic is: If the load is detected in ‘Ml’ per the mode detection algorithm main logic and there is a step down ratio of < 0.5, then the mode is assigned as ‘M2’.Discussion of Method For Evaluating Load Category Used at Step 112 Disclosed in U.S.Pat. Appl. Pub. No 2013 / 0138669 and in U.S. Patent Number 9,819,226:

[0078] The method used at step 112 to evaluate load category combines methodsdisclosed in U.S. Pat. Appl. Pub. No. 2013 / 0138669 and U.S. Patent Number 9,819,229 (which incorporates U.S. Pat. Appl. Pub. No. 2013 / 0138669 by reference). As previously noted, the load evaluation database 11 of the fire risk detection system 10 includes a load features hierarchy 12 comprising a number of layers 14 (non-limiting examples of which are shown in FIG. 9 as layers 14A and 14B). The layer 14A labeled Layer 1 in the table of FIG. 9 pertains to the general load categories of resistive, inductive / reactive, and electronic, while the layer 14B labeled Layer 2 in the table of FIG. 9 pertains to specific device or appliance types that can fall under Layer 1. It is noted that in the present application, there is no need during execution of the method 100 to identify the specific Layer 2 device or appliance type, and the layer 14B is provided in FIG. 9 solely to provide non-limiting illustrative examples of specific devices and appliances that fall under each Layer 1 load category. In the table in FIG. 9, “PFC” signifies power factor correction and TXM signifies a transformer.

[0079] The controller 10 is configured to determine, from the detected voltage and current, at least four different load features for a corresponding one of the different electric loads, and identify a load type for a given load from among the load types included in the load feature database 12 by relating the different load features of each load type to the hierarchical load feature database 4. Each layer 14 in FIG. 9 includes a corresponding load feature set, and for each layer 14, the load feature set corresponding to that layer 14 is different from the load feature set corresponding to the other layer 14.

[0080] FIG. 10 is a block diagram of the primary data acquisition functions and sequences 253 used during the load category evaluation performed at step 112. FIG. 10 corresponds to FIG. 4 from U.S. Patent Number 9,819,229. In FIG. 10, the voltage and current obtained from the monitored circuit 22 are represented by a voltage signal 248 and a current signal 250 output by the ADC 16 (FIG. 5). The inputs 248,250 are converted to floating point values at 254. The calculated floating point values (v, i) are stored in cycle data storage 256 and cumulative sums 258. The cycle data is stored for each voltage zero crossing as detected at 260. The cycle data storage 256 has a double buffer scheme. The two buffers 262,264 are switched if they are fully occupied. The buffers 262,264 have respective read / write access bits 266,268, which are used for buffer read / write access control. The cumulative sums 258 include: (1) an average power sum 270: the sum of the multiplication of the instantaneous samples of the voltage and current channels; and (2) an RMS current sum 272: the sum of the square of theinstantaneous current samples. The outputs of the data acquisition functions 253 include the cycle data storage buffers 262,264 and the sums 270,272.

[0081] FIG. 11 is a block diagram of a hierarchical load identification system architecture 222 used during the load category evaluation performed at step 112, and uses data obtained during the primary data acquisition functions and sequences 253 shown in FIG. 10. FIG. 11 corresponds to portions of FIG. 3 from U.S. Patent Number 9,819,229 that are relevant to the present invention.

[0082] In FIG. 11 , a data acquisition 246 inputs one voltage 248 and one current 250 from the ADC 16 (FIG. 5). The functions of the data acquisition 246 include block reads of the two digitally converted inputs 248,250, at an example rate of 1920 / 1600 Hz, and data acquisition and storage, as described above in conjunction with the primary data acquisition functions and sequences 253 shown in FIG. 10. During power on, the data acquisition 246 performs analog input offset calibration of the ADC 16.

[0083] The mode detection function 226 receives input from the cycle data buffers 262,264 (FIG. 10) and performs these actions: (1) executes at an example period of 80 ms / 100 ms (5 cycles); (2) reads a cycle of data from the cycle data buffers 262,264; and (3) provides mode feature extraction by calculating the features for mode identification including average power, total harmonic distortion (THD) greater than the 7th harmonic, and cycle area. The output 274 of the mode detection function 226 includes the mode ID result with confidence level for a given cycle.

[0084] The quantization function 282 inputs the cumulative sum of current and average power. The function 282 performs these actions: (1) calculates the average current and average power at the example period of 5 cycles; (2) performs quantization and generates QSS (quantized state sequence using RMS current) and power QSS (quantized state sequence using real power); and (3) calculates features (e.g., phase angle variation; average time difference) which are specific to the state level. The outputs include QSS 284 and power QSS 286.

[0085] Still referring to FIG. 11, the Level 1 ID function 228 inputs the cycle data buffers 262,264 (FIG. 10) and the detection of stable state 288 from the quantization function 282. It is noted that the Level 1 ID function 228 pertains to determining which of the Layer 1 load categories 14A (FIG. 9) the load being analyzed belongs to. The function 228 (on detection of the stable state 288) provides these actions: (1) reads the cycle data buffers 262,264; (2) extractsbinary VI features; (3) executes a Level 1 ID algorithm; (4) saves Level 1 ID results to QSS 284 and power QSS 286; and (5) based on the cycle Level 1 ID results, generates final Level 1 ID results. Binary VI features are derived from voltage and current trajectory. Non-limiting illustrative examples of binary VI features include: LHS (left hand side) Central Cell, Central Cell, Antidiagonal Line, Self Intersections, LHS Vertical Line, Antidiagonal Mean Line, Straight Line Between Current Peaks, and Central Horizontal Segment Mean Line. The output 290 includes the Level 1 ID and confidence level results for cycle, and final Level 1 ID results. The PQ (power quality) features function 314 inputs the cycle data, calculates PQ features from the cycle data, and outputs example PQ features, such as active and reactive power, THD, true and displacement PF, and cumulative energy to a PQ features database 316.

[0086] Some of the many advantages and benefits of the disclosed innovative system 10 and method 100 include the fact that the method 100 can be implemented on an edge device having limited resources and that data is processed in real time. Performing processing on an edge device means that only minimal data needs to be transferred on the cloud, thus significantly reducing the cost of data transfer, storage, and processing complex machine learning algorithms. Additional memory is not needed to store historic data on the edge device. In addition to providing an arc fault notification, the disclosed system 10 and method 100 also evaluates and sends user notifications for circuit overload, utility problems, and overcurrent conditions. That is, this method 100 gives specific recommendations based on current signature analysis and can point to specific potentials causes of fire. The disclosed system 10 and method 100 can also help to reduce overload when multiple loads are connected to a power strip in excess of its capacity.

[0087] Furthermore, the disclosed method 100 can handle real time complexities in terms of usage of appliances because it is agnostic of appliance combinations that can be connected to the circuit and agnostic of inherent load behaviors, and because computationally efficient features are defined for each load category. That is, the method 100 recognizes the characteristics that distinguish arc faults from overcurrent faults. For example, the presence of arc faults causes randomness in frequency content in the power line current due to jumping of electricity between conductors, leading to changes in current signature. This behavior is observed when there is a wire or insulation problem. In contrast, when a load draws overcurrent and there is no problem with any wire or insulation, it leads to an increase in current magnitude but not a change in the current signature pattern. This enables the method 100 to handle both thescenario of a rise in current due to inherent load behavior in a dedicated circuit as well as the scenario of a rise in current in non-dedicated circuits having connected loads such as a vacuum cleaner or iron.

[0088] While specific embodiments of the invention have been described in detail, it will be appreciated by those skilled in the art that various modifications and alternatives to those details could be developed in light of the overall teachings of the disclosure. Accordingly, the particular arrangements disclosed are meant to be illustrative only and not limiting as to the scope of disclosed concept which is to be given the full breadth of the claims appended and any and all equivalents thereof.

Claims

What is claimed is:

1. A method for detecting fire risk for a circuit in an electrical system of a building, the circuit being structured to be connected to a number of loads, the method being executable by a controller and comprising: at a Stage 1 of the method, collecting raw electrical data for one voltage cycle of the circuit; at a Stage 2 of the method that immediately follows Stage 1 once one voltage cycle of raw electrical data has been collected, detecting an operating mode for each load in the number of loads using data derived from the raw electrical data; at a Stage 3 of the method that immediately follows Stage 2 once at least one load in the number of loads is determined to be powered ON based on the operating mode detected for each load in the number of loads, evaluating whether the circuit is in a steady state and performing event detection for the circuit; at a Stage 4 of the method that immediately follows Stage 3 when an event is detected at Stage 3, extracting an event current for the event; at a Stage 5 of the method that immediately follows Stage 4, performing a load category evaluation of the event current to determine if the event current corresponds to a resistive load category, a reactive load category, or an electronic load category; at a Stage 6 of the method that immediately follows Stage 5, determining if excessive current is being drawn from the circuit; at a Stage 7 of the method that immediately follows Stage 6, performing an overload condition evaluation for the circuit; at a Stage 8 of the method that immediately follows Stage 7, performing an overcurrent check for the circuit; at a Stage 9 of the method that immediately follows Stage 8, making a decision to perform a load category-specific evaluation for each load in the number of loads whose load category was determined at Stage 5, and making a decision not to perform the load categoryspecific evaluation for any load in the number of loads whose load category was not determined at Stage 5; at a Stage 10 of the method that immediately follows Stage 9, performing a reactive loadspecific evaluation when the decision to perform a load category-specific evaluation was made atStage 9 and at least one load in the number of loads was determined to be in the reactive load category at Stage 5 ; at a Stage 11 of the method that immediately follows Stage 10, performing an electronic load-specific evaluation when the decision to perform a load category-specific evaluation was made at Stage 9 and at least one load in the number of loads was determined to be in the electronic load category at Stage 5; at a Stage 12 of the method that immediately follows Stage 11, performing a resistive load-specific evaluation when the decision to perform a load category-specific evaluation was made at Stage 9 and at least one load in the number of loads was determined to be in the resistive load category at Stage 5; at a Stage 13 of the method that immediately follows Stage 3 when no event is detected at Stage 3 and otherwise immediately follows Stage 12, determining if any fault flags are set in the controller; and at a Stage 14 of the method that immediately follows Stage 13, determining if the controller has received any reset request for the fault flags.

2. The method of claim 1 , wherein Stage 2 comprises: calculating an RMS voltage, Vrms, and an RMS current, Irms from, the raw electrical data; extracting mode detection features from the Vrms and Irms; and running a mode detection algorithm using the mode detection features to determine the operating mode for each load in the number of loads, wherein the operating mode for each load in the number of loads can be one of a load operating mode Ml, a load low power mode M2, a parasitic mode M3, and a no load power strip connected mode M4.

3. The method of claim 1, further comprising at Stage 7: incrementing an inconsistent cycle count for voltage drop when a voltage drop is observed;sending a first alert to a user stating that the circuit is overloaded or that there is a utility undervoltage when the inconsistent cycle count for voltage drop is 50 or greater; and setting an overload fault flag, wherein the first alert advises the user to: check any plugged in power strip, if a power strip is present; check for an unattended ON load drawing current exceeding the capacity of the circuit; and advising the user to switch off the unattended ON load.

4. The method of claim 3, further comprising at Stage 8: sending a second alert to the user stating that the load is using excessive current or the circuit is overloaded, after determining that Irms for the present voltage cycle exceeds a value equal to 1.5 * the maximum Irms stored for any load category present in the circuit and that the overload fault flag is set, wherein the second alert advises the user to: check for a wiring or insulation issue, and to check for an unattended load overheating.

5. The method of claim 4, wherein performing the reactive load-specific evaluation further comprises: performing cross correlation between the present current cycle and the previous current cycle; determining a DC offset and a number of fluctuations in the present current cycle; evaluating membership functions based on the cross correlation, the DC offset, and the number of fluctuations; incrementing a count for consecutive inconsistent cycles in the reactive load category when a fault probability performed during the evaluating of membership functions is greater than 70%; sending a third alert to the user stating that there is an arc fault when the count for consecutive inconsistent cycles in the reactive load category is 50 or greater; and setting an arc fault flag, and wherein the third alert advises the user to check for a loose connection, broken wire, visual sparks, or noise.

6. The method of claim 4, wherein performing the electronic load-specific evaluation further comprises: determining a number of fluctuations and a DC offset in the present current cycle; evaluating membership functions based on the number of fluctuations and the DC offset; incrementing a counter for consecutive inconsistent cycles in the electronic load category when a fault probability performed during the evaluating of membership functions is greater than 70%; sending a third alert to the user stating that there is an arc fault when the count for consecutive inconsistent cycles in the electronic load category is 50 or greater; and setting an arc fault flag, and wherein the third alert advises the user to check for a loose connection, broken wire, visual sparks, or noise.

7. The method of claim 4, wherein performing the resistive load- specific evaluation further comprises: normalizing voltage and current for the present voltage cycle; for the present voltage cycle, calculating Euclidean distance between ramp up in voltage and ramp up in current in the positive half-cycle and ramp up in the voltage signal and ramp up in the current cycle in the negative half-cycle; evaluating membership functions based on the normalized voltage and current and the calculated Euclidian distance; incrementing a counter for consecutive inconsistent cycles in the resistive load category when a fault probability performed during the evaluating of membership functions is greater than 70%; sending a third alert to the user stating that there is an arc fault when the count for consecutive inconsistent cycles in the electronic load category is 50 or greater; and setting an arc fault flag, and wherein the third alert advises the user to check for a loose connection, broken wire, visual sparks, or noise.

8. The method of claim 1, wherein Stage 13 further comprises sending a first alert to the user stating that the circuit is operating normally.

9. The method of claim 1, wherein Stage 14 further comprises resetting any set fault flags when a reset request is received.

10. A system for detecting fire risk in a building electrical system, the building electrical system including a main panel and a number of branch circuits, the system comprising: a load evaluation database, the load evaluation database including a load features hierarchy having a number of load feature layers; an analog to digital converter, ADC, the ADC being structured and configured to collect raw electrical data from a monitored circuit of the number of branch circuits, the raw electrical data including raw voltage and current samples; and a controller, the controller being configured to access the load evaluation database and to receive data from the ADC, wherein the monitored circuit is structured to be connected to a number of loads, wherein the controller is structured and configured to execute a method for detecting fire risk for the monitored circuit, the method comprising: at a Stage 1 of the method, collecting raw electrical data for one voltage cycle of the monitored circuit; at a Stage 2 of the method that immediately follows Stage 1 once one voltage cycle of raw electrical data has been collected, detecting an operating mode for each load in the number of loads using data derived from the raw electrical data; at a Stage 3 of the method that immediately follows Stage 2 once at least one load in the number of loads is determined to be powered ON based on the operating mode detected for each load in the number of loads, evaluating whether the monitored circuit is in a steady state and performing event detection for the monitored circuit; at a Stage 4 of the method that immediately follows Stage 3 when an event is detected at Stage 3, extracting an event current for the event;at a Stage 5 of the method that immediately follows Stage 4, performing a load category evaluation of the event current to determine if the event current corresponds to a resistive load category, a reactive load category, or an electronic load category; at a Stage 6 of the method that immediately follows Stage 5, determining if excessive current is being drawn from the monitored circuit; at a Stage 7 of the method that immediately follows Stage 6, performing an overload condition evaluation for the monitored circuit; at a Stage 8 of the method that immediately follows Stage 7, performing an overcurrent check for the monitored circuit; at a Stage 9 of the method that immediately follows Stage 8, making a decision to perform a load category-specific evaluation for each load in the number of loads whose load category was determined at Stage 5, and making a decision not to perform the load categoryspecific evaluation for any load in the number of loads whose load category was not determined at Stage 5; at a Stage 10 of the method that immediately follows Stage 9, performing a reactive loadspecific evaluation when the decision to perform a load category-specific evaluation was made at Stage 9 and at least one load in the number of loads was determined to be in the reactive load category at Stage 5 ; at a Stage 11 of the method that immediately follows Stage 10, performing an electronic load-specific evaluation when the decision to perform a load category-specific evaluation was made at Stage 9 and at least one load in the number of loads was determined to be in the electronic load category at Stage 5; at a Stage 12 of the method that immediately follows Stage 11, performing a resistive load-specific evaluation when the decision to perform a load category-specific evaluation was made at Stage 9 and at least one load in the number of loads was determined to be in the resistive load category at Stage 5; at a Stage 13 of the method that immediately follows Stage 3 when no event is detected at Stage 3 and otherwise immediately follows Stage 12, determining if any fault flags are set in the controller; and at a Stage 14 of the method that immediately follows Stage 13, determining if the controller has received any reset request for the fault flags.

11. The system of claim 10, wherein Stage 2 comprises: calculating an RMS voltage, Vrms, and an RMS current, Irms from, the raw electrical data; extracting mode detection features from the Vrms and Irms; and running a mode detection algorithm using the mode detection features to determine the operating mode for each load in the number of loads, and wherein the operating mode for each load in the number of loads can be one of a load operating mode Ml, a load low power mode M2, a parasitic mode M3, and a no load power strip connected mode M4.

12. The system of claim 10, wherein, at Stage 7, the method further comprises: incrementing an inconsistent cycle count for voltage drop when a voltage drop is observed; sending a first alert to a user stating that the monitored circuit is overloaded or that there is a utility undervoltage when the inconsistent cycle count for voltage drop is 50 or greater; and setting an overload fault flag, and wherein the first alert advises the user to: check any plugged in power strip, if a power strip is present; check for an unattended ON load drawing current exceeding the capacity of the monitored circuit; and advising the user to switch off the unattended ON load.

13. The system of claim 12, wherein, at Stage 8, the method further comprises: sending a second alert to the user stating that the load is using excessive current or the monitored circuit is overloaded, after determining that Irms for the present voltage cycle exceeds a value equal to 1.5 * the maximum Irms stored for any load category present in the monitored circuit and that the overload fault flag is set, andwherein the second alert advises the user to: check for a wiring or insulation issue, and to check for an unattended load overheating.

14. The system of claim 13, wherein performing the reactive load-specific evaluation further comprises: performing cross correlation between the present current cycle and the previous current cycle; determining a DC offset and a number of fluctuations in the present current cycle; evaluating membership functions based on the cross correlation, the DC offset, and the number of fluctuations; incrementing a count for consecutive inconsistent cycles in the reactive load category when a fault probability performed during the evaluating of membership functions is greater than 70%; sending a third alert to the user stating that there is an arc fault when the count for consecutive inconsistent cycles in the reactive load category is 50 or greater; and setting an arc fault flag, and wherein the third alert advises the user to check for a loose connection, broken wire, visual sparks, or noise.

15. The system of claim 13, wherein performing the electronic load-specific evaluation further comprises: determining a number of fluctuations and a DC offset in the present current cycle; evaluating membership functions based on the number of fluctuations and the DC offset; incrementing a counter for consecutive inconsistent cycles in the electronic load category when a fault probability performed during the evaluating of membership functions is greater than 70%; sending a third alert to the user stating that there is an arc fault when the count for consecutive inconsistent cycles in the electronic load category is 50 or greater; and setting an arc fault flag, andwherein the third alert advises the user to check for a loose connection, broken wire, visual sparks, or noise.

16. The system of claim 13, wherein performing the resistive load- specific evaluation further comprises: normalizing voltage and current for the present voltage cycle; for the present voltage cycle, calculating Euclidean distance between ramp up in voltage and ramp up in current in the positive half-cycle and ramp up in the voltage signal and ramp up in the current cycle in the negative half-cycle; evaluating membership functions based on the normalized voltage and current and the calculated Euclidian distance; incrementing a counter for consecutive inconsistent cycles in the resistive load category when a fault probability performed during the evaluating of membership functions is greater than 70%; sending a third alert to the user stating that there is an arc fault when the count for consecutive inconsistent cycles in the electronic load category is 50 or greater; and setting an arc fault flag, and wherein the third alert advises the user to check for a loose connection, broken wire, visual sparks, or noise.

17. The system of claim 10, wherein Stage 13 further comprises sending a first alert to the user stating that the monitored circuit is operating normally.

18. The system of claim 10, wherein Stage 14 further comprises resetting any set fault flags when a reset request is received.

Images