Apparatus for detecting arc occurrence on basis of neural network model
The apparatus uses a neural network model with a sensor and computation unit to accurately detect arcs, enhancing AFCI performance and reducing fire risks with minimal manufacturing changes.
Patent Information
- Application Number
- PCT/KR2026/001939
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-07
- Filing Date
- 2026-02-03
- Publication Date
- 2026-11-12
AI Technical Summary
Conventional arc fault circuit interrupters (AFCIs) struggle to accurately detect arcs due to their high frequency of malfunctions, posing a risk of electrical fires, as they rely on rule-based systems that are prone to errors in defining the arc phenomenon.
An apparatus utilizing a neural network model with a sensor unit, computation unit, and control unit to detect arc occurrences, incorporating a cell array with memory cells storing weights of a previously learned neural network model, and generating a blocking signal based on digital outputs.
Enables high-speed and accurate arc detection, allowing for high-performance AFCI functions to be implemented with minimal changes to existing circuit breakers, reducing the risk of electrical fires.
Smart Images

Figure KR2026001939_12112026_PF_FP_ABST
Abstract
Description
A device that detects arc generation based on a neural network model
[0001] The present disclosure relates to an apparatus for detecting arc generation based on a neural network model.
[0002] Conventional earth leakage circuit breakers have a problem in that they cannot respond to arc faults, which are the cause of electrical fires. Although Arc-Fault Circuit Interrupters (AFCIs) have been developed to detect arcs using current or voltage waveforms, high-frequency detection, or optical sensors, conventional AFCIs are unable to effectively detect arcs, which pose a high risk of accidents, as they are used with reduced sensitivity due to a high frequency of malfunctions.
[0003] It is known that arcs occur due to complex causes in various situations. Consequently, there is a clear limitation in that arc detection using rule-based systems is prone to frequent malfunctions, as it is difficult to provide a clear definition of the arc phenomenon.
[0004] Accordingly, there is a need for research and development of technology to detect arc occurrences with high accuracy by utilizing neural network models trained on a large amount of diverse data for arc detection.
[0005] The aforementioned background technology is technical information that the inventor possessed for the derivation of the present disclosure or acquired during the process of deriving the present disclosure, and it cannot be considered as prior art disclosed to the general public prior to the filing of the present disclosure.
[0006] The present disclosure provides an apparatus for detecting arc generation based on a neural network model. The problems to be solved by the present disclosure are not limited to those mentioned above, and other problems and advantages of the present disclosure not mentioned can be understood from the following description and will be more clearly understood by the embodiments of the present disclosure. Furthermore, it will be understood that the problems and advantages to be solved by the present disclosure can be realized by the means and combinations thereof set forth in the claims.
[0007] As a technical means for achieving the aforementioned technical problem, the first aspect of the present disclosure may provide an apparatus for detecting arc occurrence based on a neural network model, comprising: a sensor unit for acquiring sensor data for an electrical circuit; a computation unit including a cell array including a plurality of memory cells for storing weights of a previously learned neural network model; and a control unit for inputting a digital input configured based on the sensor data to the computation unit, acquiring a digital output of the computation unit corresponding to the digital input, and generating a blocking signal based on the digital output, wherein the digital output indicates whether an arc has occurred in the electrical circuit.
[0008] According to one embodiment, the device may further include: a digital-to-analog converter that converts the digital input into an analog input and inputs it to the cell array; and an analog-to-digital converter that converts the analog output of the cell array into the digital output.
[0009] According to one embodiment, the sensor unit comprises a first sensor unit for acquiring an alternating current; a second sensor unit for acquiring an alternating voltage; and a third sensor unit for acquiring a leakage current, and the device may provide a device comprising at least one amplifier for preprocessing the sensor data.
[0010] According to one embodiment, the device may include at least one amplifier that converts the alternating current into a plurality of alternating currents amplified to different magnitudes.
[0011] According to one embodiment, the first sensor unit may provide a device comprising an offset adding circuit that preserves the waveform of the alternating current by adding a preset offset to the original waveform.
[0012] According to one embodiment, the first sensor unit may provide a device comprising a Hall sensor that measures the alternating current.
[0013] According to one embodiment, a device may be provided in which communication between the control unit and the operation unit is performed based on synchronous communication.
[0014] According to one embodiment, the sensor unit includes a zero crossing detection unit that detects a point where the alternating voltage for the electrical circuit indicates a zero crossing, and the control unit transmits the digital input to the calculation unit during a valid period excluding a communication restricted period including the point indicating the zero crossing, and the calculation unit transmits the digital output to the control unit during the valid period.
[0015] According to one embodiment, a device may be provided in which the transmission of the digital input and the transmission of the digital output are repeatedly performed at every first period, and the first period is set as an integer multiple of the half-period of the alternating voltage.
[0016] According to one embodiment, the control unit determines whether to repeatedly generate the cutoff signal every second period, and determines whether to generate the cutoff signal based on a plurality of digital outputs repeatedly acquired every first period, wherein the second period is set as an integer multiple of the half-period of the alternating voltage, and the plurality of digital outputs includes a preset number of digital outputs including the most recently acquired digital output.
[0017] A second aspect of the present disclosure provides a method for detecting arc occurrence based on a neural network model, comprising: inputting a digital input configured based on sensor data for an electrical circuit to a processing unit; obtaining a digital output of the processing unit corresponding to the digital input; and generating a blocking signal based on the digital output, wherein the processing unit includes a cell array comprising a plurality of memory cells storing weights of a previously learned neural network model, and the digital output indicates whether an arc has occurred in the electrical circuit.
[0018] Other aspects, features, and advantages other than those described above will become clear from the following drawings, claims, and detailed description of the invention.
[0019] According to the means for solving the problem of the present disclosure described above, a neural network-based arc detection operation can be performed using a cell array comprising a plurality of memory cells that store weights of a previously learned neural network model, so that the occurrence of an arc can be detected at a high speed and with high accuracy.
[0020] In addition, according to the means for solving the problem of the present disclosure, high-performance AFCI functions can be implemented by installing additional modules on existing circuit breakers, such as earth leakage circuit breakers, so changes to the manufacturing process can be minimized.
[0021] The effects of the embodiments of the present disclosure are not limited to the effects mentioned above, and other unmentioned effects will be clearly understood by those skilled in the art from the description in this specification.
[0022] The following drawings attached to this specification are intended to illustrate at least one embodiment according to the present disclosure and serve to further enhance understanding of the technical concept of the present disclosure together with the detailed description of the invention set forth below; therefore, the present disclosure should not be interpreted as being limited only to the matters described in the drawings.
[0023] FIG. 1 is a block diagram of a device for detecting arc generation based on a neural network model according to one embodiment.
[0024] Figure 2 is an exemplary diagram illustrating the implementation of a neuromorphic chip.
[0025] FIGS. 3a and 3b are exemplary drawings for comparing a von Neumann structure and an in-memory computing structure.
[0026] FIG. 4 is an exemplary drawing for explaining a calculation unit according to one embodiment.
[0027] FIGS. 5a and 5b are drawings for explaining a method of operating a cell array according to one embodiment.
[0028] FIGS. 6a and 6b are drawings for comparing vector-matrix multiplication and operations performed in a cell array according to one embodiment.
[0029] FIG. 7 is a diagram schematically illustrating the process of operation of a device for detecting arc generation based on a neural network model according to one embodiment.
[0030] FIG. 8 is an exemplary diagram illustrating the process of preprocessing sensor data according to one embodiment.
[0031] FIG. 9 is an exemplary drawing for explaining the process of obtaining alternating current according to one embodiment.
[0032] FIG. 10 is an exemplary drawing for explaining the process of transmitting digital input and digital output according to one embodiment.
[0033] FIG. 11 is an exemplary drawing for explaining a communication restriction section and a valid section according to one embodiment.
[0034] FIG. 12 is an exemplary drawing for illustrating an analog area and a high-speed communication area according to one embodiment.
[0035] FIG. 13 is an exemplary drawing for explaining a process for determining whether to repeatedly generate a blocking signal at each preset period according to one embodiment.
[0036] FIG. 14 is an example of how a device for detecting arc generation based on a neural network model operates.
[0037] FIGS. 15a and 15b are schematic side views illustrating an example of a device for detecting arc generation based on a neural network model according to one embodiment.
[0038] FIG. 16 is an exemplary drawing for explaining a blocking member according to one embodiment.
[0039] The present disclosure relates to an apparatus and method for detecting arc generation based on a neural network model. According to one embodiment of the present disclosure, the apparatus may provide an apparatus for detecting arc generation based on a neural network model, comprising: a sensor unit for acquiring sensor data for an electrical circuit; a computation unit including a cell array including a plurality of memory cells for storing weights of a previously learned neural network model; and a control unit for inputting a digital input configured based on the sensor data to the computation unit, acquiring a digital output of the computation unit corresponding to the digital input, and generating a blocking signal based on the digital output, wherein the digital output indicates whether an arc has occurred in the electrical circuit.
[0040] The advantages and features of the present disclosure and the methods for achieving them will become clear by referring to the embodiments described in detail together with the accompanying drawings. However, the present disclosure is not limited to the embodiments presented below, but can be implemented in various different forms and should be understood to include all modifications, equivalents, and substitutions that fall within the spirit and scope of the present disclosure. The embodiments presented below are provided to make the present disclosure complete and to fully inform those skilled in the art of the scope of the invention. In describing the present disclosure, detailed descriptions of related prior art are omitted if it is determined that such detailed descriptions may obscure the essence of the present disclosure.
[0041] The terms used herein are used merely to describe specific embodiments and are not intended to limit the disclosure. Unless otherwise defined, all terms used herein have the same meaning as generally understood by those skilled in the art to which this disclosure pertains.
[0042] In this specification, singular expressions include plural expressions unless the context clearly indicates otherwise. Furthermore, terms such as "comprising" or "having" are intended to specify the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.
[0043] Additionally, terms including ordinal numbers, such as "first" or "second" as used herein, may be used to describe various components, but the components should not be limited by the terms. The terms are used solely for the purpose of distinguishing one component from another.
[0044] Phrases such as "in one embodiment," "according to one embodiment," "related to one embodiment," or "according to an implementation of one embodiment" in this specification do not necessarily refer to the same embodiment. Furthermore, throughout this specification, "examples" are arbitrary distinctions to facilitate the description of the present disclosure, and each embodiment does not need to be mutually exclusive. For example, configurations mentioned for the description of one embodiment may be applied and / or implemented in other embodiments, and may be modified and applied and / or implemented to the extent that they do not depart from the scope of the present disclosure.
[0045] Some embodiments of the present disclosure may be represented by functional block configurations and various processing steps. Some or all of these functional blocks may be implemented by various numbers of hardware and / or software configurations that perform specific functions. For example, the functional blocks of the present disclosure may be implemented by one or more microprocessors or by circuit configurations for a specific function.
[0046] For example, the functional blocks of the present disclosure may be implemented in various programming or scripting languages. The functional blocks may be implemented as algorithms executed on one or more processors. Additionally, the present disclosure may employ prior art for electronic configuration, signal processing, and / or data processing, etc. Terms such as "mechanism," "element," "means," and "configuration" may be used broadly and are not limited to mechanical and physical configurations. Furthermore, terms such as "-part," "-module," etc. refer to a unit that processes at least one function or operation, which may be implemented in hardware or software, or as a combination of hardware and software.
[0047] Furthermore, the connecting lines or connecting members between the components depicted in the drawings are merely illustrative of functional connections and / or physical or circuit connections. In the actual device, connections between components may be represented by various alternative or added functional connections, physical connections, or circuit connections.
[0048] In addition, some components in the drawings may be depicted with their size or proportions slightly exaggerated. Also, components depicted in one drawing may not be depicted in another drawing.
[0049] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the attached drawings.
[0050] FIG. 1 is a block diagram of a device for detecting arc generation based on a neural network model according to one embodiment.
[0051] Referring to FIG. 1, a device (hereinafter referred to as the "device") (100) for detecting arc generation based on a neural network model according to one embodiment of the present disclosure is provided in an electrical circuit and can detect the generation of an arc in the electrical circuit. At this time, the electrical circuit may include residential electrical wiring (e.g., electrical wiring exhibiting an alternating voltage in the range of 50Hz-60Hz), but is not limited thereto. The generation of an arc detected by the device (100) can be used for various purposes, such as preventing a fire caused by arc generation by interrupting the current of the electrical circuit, or analyzing the degree of deterioration of the electrical circuit.
[0052] A device (100) according to one embodiment may be implemented as an AFCI (Arc Fault Circuit Breaker) and may detect the occurrence of an arc in an electrical circuit and interrupt the current in the electrical circuit based on the result of the detection. In addition, the device (100) implemented as an AFCI may be integrated with a leakage circuit breaker and / or an overload circuit breaker according to the prior art to perform both current interruption due to leakage and / or overload and current interruption due to arc occurrence.
[0053] According to the embodiments of the present disclosure described below, high-performance AFCI functions can be implemented by mounting additional modules on existing circuit breakers, thereby minimizing changes to the manufacturing process.
[0054] In one embodiment, the device (100) may include a sensor unit (110), a control unit (120), and a calculation unit (130).
[0055] The sensor unit (110) can acquire sensor data for an electrical circuit equipped with the device (100). The sensor unit (110) may include at least one sensor used to measure current, voltage, resistance, or temperature, and may acquire current or voltage, etc. as sensor data. Here, the sensor data may include data acquired as an analog signal and / or data acquired as a digital signal.
[0056] Meanwhile, an example of a specific process for the device (100) to acquire sensor data will be described later with reference to FIGS. 7 to 9, etc.
[0057] The control unit (120) can control input and output for the operation unit (130). The control unit (120) may be implemented as a single MCU (Main Controller Unit; Microcontroller Unit), or it may be implemented by distinguishing between an input control unit that controls input to the operation unit (130) and an output control unit that controls output to the operation unit (130). For example, the input control unit may include an analog-to-digital converter for converting sensor data into a digital input and a communication unit for transmitting the digital input to the operation unit (130). Additionally, for example, the output control unit may include a communication unit that obtains a digital output from the operation unit (130) and a blocking signal generation unit that generates a blocking signal.
[0058] Meanwhile, the control unit (120) may be implemented including a general-purpose processor, a central processing unit (CPU), a microprocessor, a digital signal processor (DSP), etc., but is not limited thereto.
[0059] In one embodiment, the control unit (120) can generate a digital input based on sensor data. In this case, the sensor data may include various data regarding current, voltage, resistance and / or temperature, etc., for an electrical circuit.
[0060] For example, data regarding current in an electrical circuit may include data regarding the waveform of the current or specific frequency components of the current (e.g., high-frequency components, etc.), and data regarding voltage may include data regarding the waveform of the voltage or specific frequency components (e.g., high-frequency components, etc.), but is not limited thereto.
[0061] In addition, for example, data regarding resistance to an electrical circuit may include resistance or changes in resistance, and data regarding temperature may include temperature or changes in temperature, but are not limited thereto.
[0062] In one embodiment, the control unit (120) can generate a digital input by converting the entire sensor data into a digital form as is, or by converting preprocessed sensor data generated by preprocessing at least a portion of the sensor data into a digital form.
[0063] In one embodiment, the control unit (120) may input a digital input configured based on sensor data for an electrical circuit to the calculation unit (130). In addition, in one embodiment, the control unit (120) may obtain a digital output of the calculation unit (130) corresponding to the digital input. At this time, the digital output may indicate whether an arc has occurred in the electrical circuit.
[0064] Meanwhile, an example of a specific process in which the control unit (120) and the operation unit (130) exchange digital input and digital output will be described later with reference to FIGS. 10 to 11, etc.
[0065] The operation unit (130) can generate a digital output corresponding to a digital input based on a digital input received from the control unit (120) using a previously trained neural network model. In one embodiment, the operation unit (130) may include a cell array comprising a plurality of memory cells that store weights of the previously trained neural network model. At this time, the operation unit (130) can generate a digital output based on the digital input using the cell array.
[0066] In one embodiment, the digital output may indicate whether an arc has occurred in the electric circuit. That is, the calculation unit (130) can generate a digital output by performing a calculation to predict the occurrence of an arc in the electric circuit based on a digital input regarding the electric circuit.
[0067] Meanwhile, specific examples of the operation unit (130) and specific examples of operations performed in the operation unit (130) will be described later with reference to FIGS. 4, FIGS. 5a to 5b and FIGS. 6a to 6b, etc.
[0068] In one embodiment, the control unit (120) and the operation unit (130) may each be implemented as separate hardware that performs individual functions. For example, the control unit (120) and the operation unit (130) may each be implemented as separate chips. As an example, the control unit (120) may be implemented as an MCU chip, and the operation unit (130) may be implemented as a neuromorphic chip including a cell array.
[0069] In another embodiment, the control unit (120) and the operation unit (130) may be integrated into a single piece of hardware. For example, the control unit (120) may be implemented as a single chip, and the operation unit (130) may be provided on the control unit (120) implemented as a chip as an operation module including a cell array.
[0070] At this time, unlike as illustrated in FIG. 1, the control unit (120) may be understood to include an operation unit (130). As an example, the control unit (120) may be implemented as an MCU chip equipped with an operation unit (130) implemented as an operation module including a cell array.
[0071] Although not shown in FIG. 1, the device (100) may include other components, such as general-purpose components. For example, the device (100) may further include a blocking unit (not shown) that blocks the current of an electrical circuit.
[0072] In one embodiment, the device (100) may generate a cutoff signal based on a digital output. In this case, the cutoff signal may be a signal indicating that it is necessary to cut off the current of the electrical circuit as an arc occurs in the electrical circuit. For example, the control unit (120) may generate a cutoff signal based on a digital output obtained from the calculation unit (130).
[0073] Figure 2 is an exemplary diagram illustrating the implementation of a neuromorphic chip.
[0074] Referring to FIG. 2, a previously learned neural network model (210) and a neural network device (220) in which the neural network model (210) is implemented can be seen.
[0075] The fact that the neural network model (210) has been trained means that the weights constituting each layer of the neural network model (210) have been determined by a large amount of training data. If the weights, which are the results of the training of the neural network model (210), are stored in a central cloud server, a cloud computing device using the neural network model (210) can communicate with the central cloud server to transmit input values to the neural network model (210) and receive output values. In this case, even if the neural network model (210) is very complex or large-scale, the cloud computing device can use the output values without any problems.
[0076] However, in the case where the neural network device (220) is an on-device computing device that processes data within the neural network device (220) itself without communicating with a central cloud server, the weights of the neural network model (210) determined through learning are stored in the neural network device (220), which is actual hardware, and specifically, can be stored in a plurality of memory cells constituting a cell array of the neural network device (220). In this case, the neural network device (220) can be implemented as a neuromorphic chip.
[0077] Neuromorphic chips are hardware that mimics human brain functions by creating circuits that imitate the shape of neurons. In other words, neuromorphic chips refer to any type of chip that mimics the structure of the nervous system. For example, neuromorphic chips can include CIM (computing in memory)-based AI chips. As an example, a neuromorphic chip can be implemented as an analog CIM-based AI chip configured to enable analog computation using analog signals such as voltage or current.
[0078] As another example, neuromorphic chips may be implemented as digital CIM-based AI chips or as Spiking Neural Network-based AI chips, but are not limited thereto.
[0079] Neuromorphic chips can be composed solely of the circuits necessary for neural network computation, offering advantages in terms of power, area, and speed. By mimicking the workings of the brain, neuromorphic chips allow structures connecting neurons and synapses to be configured in parallel.
[0080] The Von Neumann architecture, which is primarily adopted in existing computer architectures, processes data sequentially upon input, making it excellent for executing precisely written programs; however, it has problems such as limitations in power consumption and low efficiency in areas like pattern recognition and real-time recognition.
[0081] On the other hand, neuromorphic chips are not limited to digital methods that process data as 0s and 1s, but can also process data through analog operations using various analog states. The artificial neurons constituting neuromorphic chips can operate in an event-driven manner that does not require clock operation. Furthermore, neuromorphic chips can be used to efficiently process unstructured text, voice, and images that are difficult for conventional computers to intuitively recognize.
[0082] In one embodiment, when input data such as images, voice, or electromagnetic waves is input to a neuromorphic chip, a predetermined output data may be output from the neuromorphic chip through internal operations on the input data. At this time, the data input to the neuromorphic chip is not limited to the aforementioned images, voice, or electromagnetic waves, and may include various forms of data such as various signals, images, and text.
[0083] A neural network device (220) according to one embodiment may be implemented as an on-device AI chip. On-device AI refers to a technology that executes AI algorithms on a hardware device using on-device computing based on data generated by a system. AI processing is primarily carried out in cloud-based data centers requiring massive computing capacity, resulting in high server dependency. On the other hand, using on-device AI allows the execution of AI algorithm computations to be performed locally, thereby reducing dependency on the cloud (server) and consequently lowering communication costs. Furthermore, sensitive personal information is not transmitted to the cloud, thus protecting privacy. Therefore, by configuring the neuromorphic chip as an on-device AI chip, not only are cost reductions and security enhancements achieved, but a system with high responsiveness can also be implemented as computations are processed immediately within the same hardware.
[0084] Meanwhile, the state values corresponding to each weight of the neural network model (210) can be very diverse (e.g., 128 states, etc.), and the memory cells of the cell array implemented in the neural network device (220) can be implemented as multibit memory cells that store state values corresponding to the weights using multibits (e.g., 7 bits or 8 bits, etc.).
[0085] Meanwhile, various methods are being proposed to reduce the chip area and increase the accuracy and speed of computation when manufacturing the neural network device (220) in the form of a chip.
[0086] In the following description, the computation unit (130) according to one embodiment of the present disclosure may include the neural network device (220) described above. In one embodiment, the computation unit (130) may include a neural network device (220) implemented as a neuromorphic chip, but is not limited thereto.
[0087] Since rapid blocking is very important for arc detection, the computation unit (130) is implemented as an on-device AI chip, so that the occurrence of an arc can be responded to in a short period of time without being affected by network delays or communication failures.
[0088] FIGS. 3a and 3b are exemplary drawings for comparing a von Neumann structure and an in-memory computing structure.
[0089] Referring to Fig. 3a, the von Neumann architecture is a computer architecture proposed by John von Neumann, and is a stored-program computer architecture consisting of a typical three-level structure of main memory, central processing unit, and input / output device.
[0090] The von Neumann architecture can provide high versatility by changing only the software (programs) without modifying the hardware configuration (such as wiring) when changing tasks performed on a computing device. However, limitations arise in terms of computational speed and energy efficiency due to structural characteristics that require sequential instruction execution and repetitive data transfer between the computing unit and memory. This structural constraint is known as the 'von Neumann bottleneck' phenomenon.
[0091] To resolve the von Neumann bottleneck, alternatives such as the Harvard architecture, which separates memory into instruction and data storage areas, or the CIM architecture, which directly performs data operations as well as data storage in memory, are being proposed.
[0092] Referring to FIG. 3b, it can be seen that the in-memory computing structure includes memory having computing functions. The in-memory computing structure illustrated in FIG. 3b is composed of a processing unit (e.g., a processor, etc.) and memory having computing functions.
[0093] Unlike the conventional von Neumann architecture, where all data within memory is moved to the processing unit for computation, the in-memory computing architecture performs at least a portion of the computation within memory upon receiving an instruction from the processing unit and then transmits the resulting data to the processing unit. This eliminates the movement of large amounts of data, thereby effectively resolving the aforementioned von Neumann bottleneck. Additionally, it offers the advantage of significantly reduced power consumption.
[0094] Meanwhile, the CIM structure is a structure available for implementing an in-memory computing structure, representing a structure in which analog or digital operations are performed directly within the memory cell itself.
[0095] An apparatus (100) according to one embodiment of the present disclosure can perform a specific operation using a CIM-based AI chip based on a CIM structure. For example, the apparatus (100) can perform a CIM-based operation to derive the operation result of a neural network model using a CIM-based AI chip.
[0096] Since CIM-based AI chips can perform self-contained computations using memory cells without exchanging data with separate memory, they can eliminate the bottlenecks caused by data transfer between memory and the processing unit that occur in the traditional Von Neumann architecture. This enables CIM-based AI chips to achieve high performance and energy efficiency.
[0097] According to one embodiment, the computation unit (130) may be implemented as a CIM-based AI chip. For example, the computation unit (130) may be implemented as an analog CIM-based AI chip. As an example, the computation unit (130) may be implemented as a CIM-based AI chip that performs parallel computations through a cell array, and since data storage (e.g., weight storage, etc.) and data computation (e.g., vector-matrix multiplication, etc.) are performed integrally within the cell array itself, data transfer to a separate memory can be minimized. Through this, the computation unit (130) can have high energy efficiency.
[0098] A cell array constituting an operation unit (130) according to one embodiment may be composed of a plurality of memory cells capable of multi-bit implementation. That is, each memory cell included in the cell array may include multi-bit memory cells capable of multi-bit implementation.
[0099] For example, a cell array may be composed of memory cells capable of implementing 7 bits (128 analog memory states) or 8 bits (256 analog memory states). In this case, a memory cell capable of implementing n-bits may indicate that the memory cell can store any one of 2 to the power of n distinct state values.
[0100] Through this, unlike general CIM-based AI chips which exhibit problems such as increased power consumption or performance degradation as neural network models become more complex and the amount of data to be processed increases, the cell array constituting the computation unit (130) according to one embodiment of the present disclosure can perform processing of a vast amount of data with low power and high performance even during long-term use.
[0101] FIG. 4 is an exemplary drawing for explaining an operation unit according to one embodiment. The operation unit (400) shown in FIG. 4 may correspond to the operation unit (130) shown in FIG. 1.
[0102] Referring to FIG. 4, the computation unit (400) may be implemented as a dedicated AI accelerator mounted on the device (100), and may be implemented as an AI accelerator such as a Neural Processing Unit (NPU), Tensor Processing Unit (TPU), or Neural Engine, which is a dedicated module for neural network computation, but is not limited thereto. Additionally, the computation unit (400) according to one embodiment may be implemented as a CIM-based AI chip as described above with reference to FIG. 2, etc.
[0103] Hereinafter, an example of the specific structure of a computation unit (400) implemented with an analog CIM-based AI chip according to one embodiment will be described in detail.
[0104] An operation unit (400) according to one embodiment may include a digital-to-analog converter (410), a crossbar array (420), and an analog-to-digital converter (430). Only components related to the embodiments of the present disclosure are shown in the operation unit (400) illustrated in FIG. 4, and it is obvious to a person skilled in the art that the operation unit (400) may include other general-purpose components in addition to the components illustrated in FIG. 4.
[0105] In one embodiment, the operation unit (400) can convert a digital input into an analog input and input it into a cell array. For example, the operation unit (400) may include a digital-to-analog converter (410) that converts a digital input into an analog input and inputs it into a cell array.
[0106] A digital-to-analog converter (410) according to one embodiment can convert a digital input representing a digital value into an analog input. For example, the analog input may be an analog signal represented as voltage or current. That is, the digital-to-analog converter (410) can convert a digital signal into an analog signal, either voltage or current. As an example, the digital-to-analog converter (410) can receive a digital voltage composed of multiple bits, convert it into an analog voltage corresponding to the number of bit lines, and apply it to a plurality of bit lines.
[0107] An operation unit (400) according to one embodiment may include a crossbar array (420). The crossbar array (420) may include a plurality of word lines and a plurality of bit lines, and may include a cell array comprising a plurality of memory cells arranged along the plurality of word lines and the plurality of bit lines.
[0108] A plurality of word lines of a crossbar array (420) according to one embodiment are connected to a digital-to-analog converter (410) to receive an analog input converted from a digital input from the digital-to-analog converter (410).
[0109] In one embodiment, the crossbar array (420) may include a cell array comprising a plurality of memory cells that store weights of a previously trained neural network model. A neural network model according to one embodiment may be trained to generate an output predicting whether an arc occurs based on data of a pre-designed form.
[0110] In this case, the data input to the neural network model may be designed to include various indicators regarding measured AC current, measured AC voltage and / or measured leakage current, measured resistance, or measured temperature in an electrical circuit supplying AC power. For example, the data input to the neural network model may be designed to include data processed through various preprocessing steps, such as high-frequency signals of measured current or voltage, which are primarily measured data.
[0111] Subsequently, the weights constituting the neural network model (e.g., weights constituting each layer of the neural network model) can be stored in a cell array through a process of being transferred to memory cells. That is, the weights of the previously trained neural network model can be stored in the cell array constituting the crossbar array (420).
[0112] In one embodiment, the digital-to-analog converter (410) converts the digital input into an analog input and can input the analog input into a cell array that stores the weights of a previously trained neural network model through a plurality of word lines of a crossbar array (420).
[0113] In one embodiment, when an analog input from a digital-to-analog converter (410) is input through each of a plurality of word lines of a crossbar array (420), a Multiply and Accumulate (MAC) operation is performed with the weights of a neural network model stored in a plurality of memory cells of a cell array, and an analog output can be output through each of a plurality of bit lines as a result of the operation. At this time, similar to the analog input, the analog output can be represented as a signal of either current or voltage.
[0114] Additionally, the operation unit (400) according to one embodiment may include an analog-to-digital converter (430). The analog-to-digital converter (430) according to one embodiment is connected to a plurality of bit lines of a crossbar array (420) and can receive an analog output reflecting the weights of a cell array.
[0115] An analog-to-digital converter (430) according to one embodiment can convert a received analog output into a digital output representing a digital value. That is, the analog-to-digital converter (430) can convert an analog output represented as either voltage or current into a digital output. As an example, the analog-to-digital converter (430) can receive an analog voltage output from a plurality of bit lines and convert it into a digital output having a predetermined number of bits.
[0116] Since rapid interruption is crucial for arc detection, the operation of a crossbar array using analog input / output enables a response to arc occurrence in a short period, even when performing complex neural network calculations.
[0117] FIGS. 5a and 5b are drawings for explaining a method of operating a cell array according to one embodiment.
[0118] Referring to FIG. 5a, the cell array may include a plurality of memory cells (530). The memory cells (530) may be implemented as devices whose electrical characteristics, such as conductance, capacitance, or magnetoresistance, change according to an electrical signal (e.g., voltage or current) applied to both ends. Such changes in electrical characteristics may be used to store or represent the weights of a neural network model in an analog or digital manner.
[0119] For example, each memory cell (530) can be implemented as ReRAM (Resistive RAM), FeRAM (Ferroelectric RAM), PRAM (Phase-change RAM), MRAM (Magnetoresistive RAM), or NAND / NOR flash memory, and each memory cell (530) can be implemented as a multi-level memory in which a single memory cell represents multiple bits. Additionally, a cell array composed of multiple memory cells (530) can implement RCA (Resistive Crossbar memory Arrays). Additionally, a cell array composed of multiple memory cells (530) can be implemented in the form of a crossbar array in which multiple memory cells (530) are arranged along mutually orthogonal wiring.
[0120] In one embodiment, the cell array may be provided with a wiring (512) extending in a first direction (e.g., horizontal direction) and a wiring (522) extending in a second direction (e.g., vertical direction) that intersects the first direction. Hereinafter, for convenience of explanation, the wiring (512) extending in the first direction will be referred to as a row line, and the wiring (522) extending in the second direction will be referred to as a column line. A plurality of memory cells (530) may be arranged at each intersection of the row line (512) and the column line (522) to connect the corresponding row line (512) and the corresponding column line (522) to each other.
[0121] A memory cell (530) according to one embodiment may have various characteristics, such as exhibiting analog behavior in which conductivity changes gradually according to the number of input electrical pulses without abrupt resistance changes during programming and erasing. Specifically, a weight control circuit for setting weights may apply a varied voltage to the memory cell (530). Accordingly, the weight control circuit may cause the resistance value or weight of the memory cell (530) to change gradually.
[0122] The operation of the cell array described above is explained below with reference to FIG. 5b. For convenience of explanation, the row wiring (512) shown in FIG. 5b is named, in order from the top, as the first row wiring (512A), the second row wiring (512B), the third row wiring (512C), and the fourth row wiring (512D), and the column wiring (522) is named, in order from the left, as the first column wiring (522A), the second column wiring (522B), the third column wiring (522C), and the fourth column wiring (522D).
[0123] Referring to FIG. 5b, in the initial state, all of the multiple memory cells (530) may be in a state of relatively low conductivity, that is, a high resistance state. If at least some of the multiple memory cells (530) are in a low resistance state, an initialization operation to make them into a high resistance state may be additionally required. Each of the multiple memory cells (530) may have a predetermined threshold value required for a change in resistance and / or conductivity. More specifically, when a voltage or current smaller than a predetermined threshold value is applied across each memory cell (530), the conductivity of the memory cell (530) does not change, and when a voltage or current larger than a predetermined threshold value is applied to the memory cell (530), the conductivity of the memory cell (530) may change.
[0124] In this state, in order to perform the operation of outputting specific data as a result of a specific column wiring (522), an input signal corresponding to the specific data (or an analog input in which the input signal is converted) may be input to the row wiring (512). For example, the input signal may be represented by the application of an electrical pulse to each row wiring (512). Additionally, the column wiring (522) may be driven with an appropriate voltage or current for output.
[0125] In one example, if a column wiring (522) to output specific data is already determined, this column wiring (522) is driven so that a memory cell (530) located at the intersection with a row wiring (512) corresponding to '1' receives a voltage greater than or equal to the voltage required for a set operation (hereinafter, set voltage), and the remaining column wiring (522) can be driven so that the remaining memory cell (530) receives a voltage smaller than the set voltage. For example, if the magnitude of the set voltage is Vset and the column wiring (522) to output data of '0011' is set to the third column wiring (522C), the magnitude of the electrical pulse applied to the third and fourth row wiring (512C, 512D) may be greater than Vset and the voltage applied to the third column wiring (522C) may be 0V, so that the first and second memory cells (530A, 530B) located at the intersection of the third column wiring (522C) and the third and fourth row wiring (512C, 512D) receive a voltage greater than or equal to Vset. Accordingly, the first and second memory cells (530A, 530B) may be in a low-resistance state. The conductivity of the first and second memory cells (530A, 530B) in the low-resistance state may gradually increase as the number of electrical pulses increases. The size and width of the applied electrical pulse may be substantially constant. In order for the remaining memory cells (530), excluding the first and second memory cells (530A, 530B), to receive a voltage smaller than Vset, the voltage applied to the remaining column wiring, i.e., the first, second, and fourth column wirings (522A, 522B, 522D), may have a value between 0V and Vset, for example, 1 / 2Vset. Accordingly, the resistance state of the remaining memory cells (530), excluding the first and second memory cells (530A, 530B), may not change.
[0126] As another example, the column wiring (522) to output specific data may not be determined. In this case, by applying an electrical pulse corresponding to the specific data to the row wiring (512) and measuring the current flowing through each column wiring (522), the column wiring (522) that first reaches a predetermined threshold current, such as the third column wiring (522C), may become the column wiring (522) that outputs the specific data.
[0127] By the method described above, different data can be output to different column wirings (522).
[0128] Meanwhile, the row wiring (512) of the cell array described above can represent a word line, and the column wiring (522) of the cell array can represent a bit line.
[0129] FIGS. 6a and 6b are drawings for comparing vector-matrix multiplication and operations performed in a cell array according to one embodiment.
[0130] First, referring to FIG. 6a, operations between input data and weight data can be performed through vector-matrix multiplication, which allows for parallel processing. For example, input data can be represented as matrix X (610), and weight data can be represented as matrix W (611). At this time, the result of the multiplication operation between matrix X (610) and matrix W (611) can be represented as output data, matrix Y (612).
[0131] Referring to FIG. 6b, a vector multiplication operation can be performed using a plurality of memory cells constituting a cell array. For example, the vector-matrix multiplication illustrated in FIG. 6a can be performed using a plurality of memory cells illustrated in FIG. 5a to FIG. 5b, etc.
[0132] As described in comparison with FIG. 6b, input data can be input into a cell array, and the input data can be expressed as voltage (620). Additionally, weight values can be stored in synapses within the neuromorphic core, that is, in memory cells constituting the cell array, and the weight values stored in the memory cells can be expressed as conductance (621). Accordingly, the output of the cell array can be expressed as current (622), which is the result of a multiplication operation between voltage (620) and conductance (621).
[0133] Meanwhile, for convenience of explanation, FIG. 6 is illustrated as having input data of a cell array as voltage (620) and output of a cell array as current (622), but input data of a cell array may be voltage (620) or current, and likewise output of a cell array may also be current (622) or voltage.
[0134] FIG. 7 is a diagram schematically illustrating the process of operation of a device for detecting arc generation based on a neural network model according to one embodiment.
[0135] The device (700) illustrated in FIG. 7 is an example of the device (100) illustrated in FIG. 1. The first sensor unit (711), the second sensor unit (712), and the third sensor unit (713) illustrated in FIG. 7 may correspond to the sensor unit (110) illustrated in FIG. 1, and the control unit (720) and the operation unit (730) illustrated in FIG. 7 may correspond to the control unit (120) or the operation unit (130) illustrated in FIG. 1, respectively.
[0136] Referring to FIG. 7, the first sensor unit (711), the second sensor unit (712), and the third sensor unit (713) can be used to collect sensor data. In one embodiment, the sensor data collected using the first sensor unit (711), the second sensor unit (712), and the third sensor unit (713) may include at least one of an alternating current, an alternating voltage, and a leakage current of an electrical circuit.
[0137] In one embodiment, the device (700) may include a first sensor unit (711) for acquiring an alternating current of an electric circuit, a second sensor unit (712) for acquiring an alternating voltage of an electric circuit, and a third sensor unit (713) for acquiring a leakage current of an electric circuit.
[0138] Meanwhile, the device (700) may include at least one sensor unit classified into fewer or more numbers than shown in FIG. 7. For example, the device (700) may further include a fourth sensor unit (not shown) for collecting sensor data regarding resistance and / or a fifth sensor unit (not shown) for collecting sensor data regarding temperature.
[0139] In one embodiment, the first sensor unit (711) is connected to the input power line (701) of the electrical circuit and can measure the waveform of the alternating current flowing through the input power line (701). The first sensor unit (711) according to one embodiment may include at least one current sensor, such as a CT (Current Transformer) or a Hall sensor, but is not limited thereto. The waveform of the alternating current measured by the first sensor unit (711) can be transmitted to the control unit (720) in real time.
[0140] In addition, in one embodiment, the second sensor unit (712) is connected to the input power line (701) and input neutral line (702) of the electric circuit to measure the waveform of the alternating voltage applied to the electric circuit. The second sensor unit (712) according to one embodiment may include at least one voltage sensor, such as a resistive sensor or a capacitive sensor, but is not limited thereto. The waveform of the alternating voltage measured by the second sensor unit (712) can be transmitted to the control unit (720) in real time.
[0141] Additionally, in one embodiment, the third sensor unit (713) is connected to the input power line (701) and output power line (703) of the electrical circuit, and to the input neutral line (702) and output neutral line (704) to measure the waveform of the leakage current generated in the electrical circuit. The third sensor unit (713) according to one embodiment may include, but is not limited to, a zero-phase current transformer (ZCT). The waveform of the leakage current measured by the third sensor unit (713) can be transmitted to the control unit (720) in real time.
[0142] Meanwhile, the sensor data received by the control unit (720) may include preprocessed sensor data, and specific examples of preprocessing of sensor data will be described later with reference to FIG. 8, etc.
[0143] In one embodiment, the control unit (720) can configure a digital input based on sensor data for an electrical circuit collected in real time. For example, the control unit (720) can receive sensor data from a first sensor unit (711), a second sensor unit (712), and a third sensor unit (713).
[0144] Subsequently, the control unit (720) can configure the sensor data into a single digital input. For example, the control unit (720) can configure a digital input for the operation unit (730) by converting a waveform signal into a digital signal. Subsequently, the control unit (720) can input the digital input configured based on the sensor data for the electrical circuit to the operation unit (730).
[0145] At this time, the digital input configured by the control unit (720) may correspond to a constant time period set based on the period of the AC power applied to the electric circuit. For example, the digital input corresponding to the first time point may represent the waveform of the AC current, the waveform of the AC voltage and the waveform of the leakage current, the high frequency of the AC current, and the high frequency of the AC voltage for the time interval corresponding to the first time point. Specifically, the constant time period set based on the period of the AC power may represent a half period or a full period of the AC power, but is not limited thereto.
[0146] Subsequently, the control unit (720) can obtain a digital output of the operation unit (730) corresponding to the digital input. For example, the operation unit (730) can convert the received digital input into an analog input for the crossbar array using a digital-to-analog converter, generate an analog output as a result of operation on the analog input, and convert the analog output of the crossbar array into a digital output using an analog-to-digital converter. Since the operation of the operation unit (730) that receives the digital input and generates the digital output is the same as described above with reference to FIG. 4, etc., a specific example of the operation of the operation unit (730) that receives the digital input and generates the digital output is omitted below.
[0147] In one embodiment, the digital output may indicate whether an arc has occurred in the electrical circuit. For example, the operation unit (730) may include a cell array that stores weights of a neural network model that predicts arc occurrence based on input data regarding the electrical circuit, and since the digital output can be generated as a prediction result of arc occurrence using the cell array, the digital output may indicate whether an arc has occurred in the electrical circuit.
[0148] For example, if the digital input is configured for one cycle of the alternating current power applied to the electric circuit, the calculation unit (730) generates a digital output of the predicted result of arc occurrence based on sensor data collected during the time interval of one cycle, so the digital output can indicate whether an arc occurred during that time interval.
[0149] Subsequently, the control unit (720) can generate a cutoff signal based on the digital output. In one embodiment, the control unit (720) can determine whether to generate a cutoff signal based on the digital output. The control unit (720) can generate a cutoff signal based on the digital output using a preset standard, and the preset standard may be set considering the possibility of an accident, such as a fire, occurring according to the size and frequency of an arc generated in the electrical circuit.
[0150] At this time, the digital output used to determine whether the control unit (720) generates a blocking signal may be a single digital output corresponding to a single digital input, but is not limited thereto. For example, the control unit (720) may determine whether to generate a blocking signal based on a plurality of digital outputs corresponding to each of a plurality of digital inputs.
[0151] Meanwhile, the specific process for determining whether to generate a blocking signal based on multiple digital outputs will be described later with reference to FIG. 13, etc.
[0152] In one embodiment, the control unit (720) can generate a blocking signal and apply the blocking signal to a blocking unit (740) that blocks the current of an electrical circuit based on the blocking signal. Through this, accidents such as fire caused by an arc can be prevented.
[0153] FIG. 8 is an exemplary diagram illustrating the process of preprocessing sensor data according to one embodiment.
[0154] Referring to FIG. 8, the device (100) may further include at least one amplifier. For example, the device (100) may include a first amplifier (811) that amplifies an alternating current obtained through a first sensor unit (810), a second amplifier (821) that amplifies an alternating voltage obtained through a second sensor unit (820), and a third amplifier (831) that amplifies a leakage current obtained through a third sensor unit (830).
[0155] In one embodiment, the digital input may be configured based on sensor data preprocessed using at least one amplifier. For example, the first amplifier (811), the second amplifier (821), and the third amplifier (831) can preprocess sensor data by amplifying the waveform of an alternating current, the waveform of an alternating voltage and the waveform of a leakage current, the high frequency of the alternating current, the high frequency of the alternating voltage, etc.
[0156] Sensor data collected by at least one of the first sensor unit (810), the second sensor unit (820), and the third sensor unit (830) may be a signal that is too small or too large to predict arc occurrence. By converting the sensor data to an appropriate scale through preprocessing using at least one amplifier, the control unit (840) can obtain a signal waveform of an effective and appropriate size for predicting arc occurrence as preprocessed sensor data.
[0157] Additionally, in one embodiment, the preprocessed sensor data may include a plurality of alternating currents amplified to different sizes. For example, the device (100) may include a multi-amplifier. As an example, the first amplifier (811) may be implemented as a multi-amplifier that converts an alternating current into a plurality of alternating currents amplified to different sizes.
[0158] In one embodiment, the first amplifier (811) can acquire an alternating current from the first sensor unit (810) and generate a preset number of amplified alternating currents. In one embodiment, the first amplifier (811) may have an input channel and a plurality of output channels, and can output a plurality of amplified alternating currents by amplifying the alternating current acquired through the input channel to a level corresponding to each of the plurality of output channels.
[0159] For example, the first amplifier (811) can generate a plurality of alternating currents amplified to different sizes, such as an average of 1A, an average of 5A, an average of 30A, and an average of 500A. As another example, the first amplifier (811) can generate a plurality of alternating currents amplified to different sizes, such as a 1x, 5x, 30x, and 500x scale. Subsequently, the first amplifier (811) can transmit the generated alternating currents to the control unit (840).
[0160] Through this, the control unit (840) can configure a digital input that reflects all of the various ranges of current waveforms, and can configure a digital input that reflects all of the characteristics of microcurrents and / or large currents that are not adequately represented by only the fixed scale of a single current channel. In addition, as a result, the accuracy of the digital output can be improved, thereby reducing the possibility of false arc detection.
[0161] FIG. 9 is an exemplary drawing for explaining the process of obtaining alternating current according to one embodiment. The offset addition circuit (910) and current sensor (920) shown in FIG. 9 are examples of the first sensor unit (810) shown in FIG. 8, etc.
[0162] Referring to FIG. 9, the first sensor unit (810) for measuring alternating current of an electric circuit may include an offset addition circuit (910) and a current sensor (920).
[0163] Generally, when a current sensor (920) measures alternating current, it collects signals in both the positive signal region (+) and the negative signal region (-). In order to perform data processing, such as arc detection, using the collected signals, the prior art used a full-wave rectifier circuit. However, the full-wave rectifier circuit uses a diode element to convert the signal waveform appearing in the negative signal region into the positive signal region, and since the diode element has a physical characteristic of having a turn-on voltage, there is a problem that a small signal of a magnitude close to zero may be lost.
[0164] In one embodiment, the first sensor unit (810) may include an offset adding circuit (910) that adds a preset offset (940) to the original waveform appearing on the input power line (901), and with reference to the graph shown in FIG. 9, the current sensor (920) may measure an alternating current (930) representing the waveform preserved by the offset adding circuit (910).
[0165] In one embodiment, the offset addition circuit (910) can generate an offset (940) by generating a preset DC potential using an operational amplifier or a reference voltage circuit, etc.
[0166] In one embodiment, the offset (940) may be set so that the entire waveform of the alternating current (930) to which the offset (940) is added is measured in the positive signal region. For example, the offset (940) may be set so that the minimum value of the alternating current (930) to which the offset (940) is added has a value greater than 0. As an example, the offset (940) may be set to half the level of the system power (Vcc) (e.g., 1.65V or 2.5V, etc.), but is not limited thereto.
[0167] Through this, the loss of the alternating current signal can be prevented, and the current sensor (920) can obtain an alternating current waveform in which the characteristics of arc generation appearing at a position close to zero relative to the original waveform of the alternating current are preserved.
[0168] Additionally, in one embodiment, the current sensor (920) may include a Hall sensor that measures alternating current. That is, the alternating current included in the sensor data may be measured by the Hall sensor.
[0169] When the current sensor (920) is implemented as a CT, if a current exceeding a preset measurable range is applied, the iron core of the CT reaches a state of magnetic saturation, which may cause distortion throughout the output signal. In contrast, when the current sensor (920) is implemented as a Hall sensor, even if a current exceeding a preset measurable range is applied, relative linearity is maintained in the remaining measurement range excluding the excess range, so that overall distortion of the output signal may not occur.
[0170] According to one embodiment, by implementing the current sensor (920) as a Hall sensor, the first sensor unit (810) can measure an alternating current (930) without distortion. Through this, the reliability of arc detection can be improved, and since a highly reliable digital input can be configured even if the current sensor (920) is implemented as a single sensor, the device (100) can be miniaturized.
[0171] FIG. 10 is an exemplary drawing for explaining the process of transmitting digital input and digital output according to one embodiment.
[0172] Referring to FIG. 10, the device (100) may further include a zero crossing detection unit (1010). In one embodiment, the sensor unit (110) may include a zero crossing detection unit (1010) that detects a point (1040) where the alternating voltage (1030) for the electrical circuit indicates a zero crossing.
[0173] A zero crossing detection unit (1010) according to one embodiment is connected to an input power line (1001) and an input neutral line (1002) of an electric circuit, and can detect a point (1040) where the alternating voltage (1030) of the electric circuit indicates a zero crossing. In one embodiment, the zero crossing detection unit (1010) may be included in a second sensor unit (712) shown in FIG. 7, etc., but may also be implemented as at least one sensor distinct from the second sensor unit (712).
[0174] In an electrical circuit, an arc is likely to occur around the zero crossing point (1040), and if an arc occurs, communication interference may be caused by noise from the arc.
[0175] Accordingly, the control unit (1020) can exchange digital input and digital output while avoiding the zero crossing point (1040), the zero crossing detection unit (1010) transmits the detection result to the control unit (1020), and the control unit (1020) can exchange digital input and digital output while avoiding the zero crossing point (1040) by using the detection result of the zero crossing detection unit (1010).
[0176] Meanwhile, a specific example of the process in which the control unit (1020) exchanges digital input and digital output while avoiding the zero crossing point (1040) using the detection result of the zero crossing detection unit (1010) will be described later with reference to FIG. 11.
[0177] Through this, even if the device (100) is miniaturized, communication interference due to the arc can be minimized, so that a highly reliable prediction result of arc occurrence can be generated.
[0178] FIG. 11 is an exemplary drawing for explaining a communication restriction section and a valid section according to one embodiment.
[0179] Referring to FIG. 11, the control unit (1110) can exchange digital input (1101) and digital output (1102) with the operation unit (1120).
[0180] In one embodiment, the control unit (1110) can transmit a digital input (1101) to the calculation unit (1120) during the valid period (S1), excluding the communication restriction period (S2) which includes the point where the alternating voltage for the electric circuit indicates a zero crossing. Additionally, the control unit (1110) can obtain a digital output (1102) from the calculation unit (1120) during the valid period (S1), excluding the communication restriction period (S2).
[0181] Through this, even if the device (100) is miniaturized, communication interference due to the arc can be minimized, so that a highly reliable prediction result of arc occurrence can be generated.
[0182] For example, the communication restriction period (S2) may represent a time interval having a preset range based on a point indicating a zero crossing (e.g., a range of 1ms before and after a point indicating a zero crossing). Since the AC power applied to the electrical circuit has periodicity, the effective period (S1) and the communication restriction period (S2) may also be switched periodically, and the actual communication period (S'1) in which communication between the control unit (1110) and the calculation unit (1120) is performed during the effective period (S1) may also occur periodically.
[0183] In one embodiment, the control unit (1110) transmitting a digital input (1101) to the operation unit (1120) and the operation unit (1120) transmitting a digital output (1102) to the control unit (1110) can be performed repeatedly at every first cycle.
[0184] At this time, the first period can be set to an integer multiple of the half-period of the alternating voltage. The communication restriction period (S2) appears twice for every period (P) of the alternating voltage appearing in the electrical circuit. Accordingly, the effective period (S1) also appears twice for every period (P) of the alternating voltage, and the transmission of the digital input (1101) and the transmission of the digital output (1102) can be performed repeatedly with a time period having an integer multiple of half of the period (P) of the alternating voltage in order to be performed during the effective period (S1).
[0185] As an example, the first period may be half the period (P) of the alternating voltage. In this case, the actual communication interval (S'1) appears as the same number as the number of effective intervals (S1). As another example, the first period may be the period (P) of the alternating voltage. In this case, the actual communication interval (S'1) appears as half the number of effective intervals (S1).
[0186] Through this, the control unit (1110) can repeatedly induce calculations of the calculation unit (1120) based on sensor data collected in real time, and can repeatedly obtain calculation results of the calculation unit (1120) corresponding to the sensor data collected in real time.
[0187] That is, the control unit (1110) can input a plurality of digital inputs (1101) repeatedly to the operation unit (1120) with periodicity and obtain a plurality of digital outputs (1102) repeatedly from the operation unit (1120) with periodicity.
[0188] In one embodiment, the device (100) may generate a blocking signal based on a plurality of digital outputs (1102) corresponding to a plurality of digital inputs (1101) rather than generating a blocking signal based on a single digital output (1102) corresponding to a single digital input (1101) in order to resolve the risk indicated in the predicted result of arc generation in a more appropriate manner.
[0189] Meanwhile, the specific process of generating a blocking signal based on multiple digital outputs (1102) will be described later with reference to FIG. 13, etc.
[0190] FIG. 12 is an exemplary drawing for illustrating an analog area and a high-speed communication area according to one embodiment.
[0191] Referring to FIG. 12, the area between the sensor unit (1210) and the control unit (1220) can be defined as an analog area (1201), and the area between the control unit (1220) and the operation unit (1230) can be defined as a high-speed communication area (1202).
[0192] In one embodiment, the device (100) including the sensor unit (1210), the control unit (1220), and the computation unit (1230) may be implemented as a miniaturized module. In this case, when performing communication between the control unit (1220) and the computation unit (1230), it is necessary to maximize the distance between the hardware elements constituting the analog area (1201) and the hardware elements constituting the digital area (1202) in order to minimize data loss and errors caused by communication interference.
[0193] For example, the hardware elements constituting the sensor unit (1210) may be provided on a different PCB that is distinct from the PCB (Printed Circuit Board) on which the hardware elements constituting the computation unit (1230) are provided.
[0194] Through this, even if the device (100) is miniaturized, when the control unit (1220) and the computation unit (1230) communicate in the high-speed communication area (1202), data loss and errors caused by communication interference due to signal transmission between the sensor unit (1210) and the control unit (1220) can be minimized.
[0195] Meanwhile, in one embodiment, communication between the control unit (1220) and the operation unit (1230) can be performed based on various communication methods such as synchronous communication, asynchronous communication, half-duplex communication, full-duplex communication, serial communication, or parallel communication.
[0196] In one embodiment, communication between the control unit (1220) and the operation unit (1230) may be performed based on asynchronous communication such as UART (Universal Asynchronous Receiver / Transmitter) or RS-485.
[0197] In another embodiment, communication between the control unit (1220) and the operation unit (1230) may be performed based on synchronous communication, such as I2C (Inter Integrated Circuit) communication or SPI (Serial Peripheral Interface) communication. For example, digital input may be transmitted from the control unit (1220) to the operation unit (1230) based on synchronous communication, and digital output may be transmitted from the operation unit (1230) to the control unit (1220) based on synchronous communication.
[0198] Through this, the control unit (1220) can input a digital input to the calculation unit (1230) during a valid period and obtain a digital output from the calculation unit (1230). Meanwhile, the control unit (1220) inputs a digital input to the calculation unit (1230) during a predetermined valid period, and is not required to obtain a digital output corresponding to the digital input input during the predetermined valid period from the calculation unit (1230).
[0199] For example, the control unit (1220) can input a first digital input corresponding to a first time point preceding the second time point, and a second digital input corresponding to the second time point, to the operation unit (1230) during the first effective period and the second effective period, respectively.
[0200] At this time, the digital output obtained from the operation unit (1230) during the second valid period may be the second digital output as the result of the operation of the operation unit (1230) for the second digital input according to the design, but may also be the first digital output as the result of the operation of the operation unit (1230) for the first digital input, or it may be a plurality of digital outputs accumulated and stored in a memory device separately provided in the operation unit (1230).
[0201] In one embodiment, synchronous communication may include SPI communication. Through this, the control unit (1220) and the operation unit (1230) can transmit multiple bits at once and exchange a large amount of data over a short communication period, thereby enabling a high-speed communication environment.
[0202] FIG. 13 is an exemplary diagram illustrating a process for determining whether to repeatedly generate a blocking signal at preset cycles according to one embodiment. FIG. 13 shows the waveform (1310) of an alternating voltage applied to an electric circuit, an actual communication interval (1320), and a digital output analysis range (1330).
[0203] Referring to FIG. 13, the waveform (1310) of the alternating voltage is composed of unit periods, and the unit period represents half of the period of the alternating voltage waveform (1310). For example, in the alternating voltage waveform (1310), the first period is composed of a first unit period (C11) and a second unit period (C12), and the second period is composed of a first unit period (C21) and a second unit period (C22).
[0204] Meanwhile, the actual communication interval (1320) may appear at every first period set as an integer multiple of the unit period. The actual communication interval (1320) illustrated in FIG. 13 is illustrated on the premise that a digital output is obtained at every unit period. For example, in the first unit period (C11) of the first period, a first digital output may be obtained during the first communication interval (O11) of the first period, and in the second unit period (C12) of the first period, a second digital output may be obtained during the second communication interval (O12) of the first period.
[0205] That is, the control unit (120) can repeatedly obtain a digital output from the operation unit (130) at each first cycle. In one embodiment, the control unit (120) determines whether to repeatedly generate a blocking signal at each second cycle, and may determine whether to generate a blocking signal based on a plurality of digital outputs repeatedly obtained at each first cycle. At this time, the second cycle may be set as an integer multiple of the unit cycle.
[0206] For example, the control unit (120) can determine whether to generate a cutoff signal at each unit period. For example, the control unit (120) can determine whether to generate a cutoff signal at the first unit period (C11) of the first period in the waveform (1310) of the alternating voltage, and then determine again whether to generate a cutoff signal at the second unit period (C12) of the first period.
[0207] In one embodiment, a plurality of digital outputs used by the control unit (120) to determine whether to generate a cutoff signal may include a preset number of digital outputs including a recently acquired digital output. For example, the control unit (120) may use a preset number of digital outputs including a first digital output acquired in the first communication section (O11) of the first cycle to determine whether to generate a cutoff signal in the first unit period (C11) of the first cycle of the AC voltage waveform (1310), and may use a preset number of digital outputs including a second digital output acquired in the second communication section (O12) of the first cycle to determine whether to generate a cutoff signal in the second unit period (C12) of the first cycle.
[0208] The digital output analysis range (1330) illustrated in FIG. 13 is illustrated on the premise that the digital output analysis range (1330), which is used to determine whether the control unit (120) will generate a cutoff signal, is set to three cycles of the alternating voltage waveform (1310).
[0209] For example, the control unit (120) can determine whether to generate a blocking signal corresponding to the second unit period (C32) of the third period in the AC voltage waveform (1310) by using six digital outputs including a second digital output obtained in the second communication section (O32) of the third period. Additionally, for example, the control unit (120) can determine whether to generate a blocking signal corresponding to the first unit period (C41) of the fourth period in the AC voltage waveform (1310) by using six digital outputs including a first digital output obtained in the first communication section (O41) of the fourth period.
[0210] In one embodiment, the control unit (120) can determine whether to generate a blocking signal based on the number of arc occurrences, frequency, intensity, and probability, etc., shown in a plurality of digital outputs included in the analysis range.
[0211] For example, the control unit (120) can generate a blocking signal based on the fact that among a plurality of digital outputs included in the analysis range, the number of digital outputs exceeding a preset number (e.g., exceeding 3 out of 6, etc.) or the number of digital outputs exceeding a preset ratio relative to the total (e.g., exceeding 60%) indicates that an arc has occurred.
[0212] As another example, the control unit (120) can generate a blocking signal based on the fact that the sum of the arc occurrence probabilities appearing in each of the multiple digital outputs included in the analysis range exceeds a preset threshold.
[0213] By generating a cutoff signal using multiple digital outputs rather than a single digital output, the malfunction of the device (100) due to false arc detection can be prevented. For example, even if at least one digital output is omitted from the analysis range due to a communication failure or communication interference, the remaining digital outputs included in the analysis range can be used to generate a cutoff signal, thereby preventing the failure of current interruption in situations where current interruption is required. Additionally, for example, if a low-risk arc occurs at a low frequency, current interruption can be prevented to ensure a stable power supply.
[0214] FIG. 14 is an example of how a device for detecting arc generation based on a neural network model operates.
[0215] Referring to FIG. 14, in step 1410, the device (100) can input a digital input configured based on sensor data for an electrical circuit to the operation unit (130). For example, the device (100) may include a control unit (120), and the control unit (120) can input a digital input to the operation unit (130).
[0216] At this time, the operation unit (130) may include a cell array comprising a plurality of memory cells that store weights of a previously learned neural network model.
[0217] In one embodiment, the device (100) may include a sensor unit (110). The sensor unit (110) according to one embodiment may acquire sensor data for an electrical circuit.
[0218] In one embodiment, the sensor data may include alternating current, alternating voltage and leakage current, alternating current high frequency, and alternating voltage high frequency. For example, the sensor unit (110) may include a first sensor unit for acquiring alternating current, a second sensor unit for acquiring alternating voltage, and a third sensor unit for acquiring leakage current.
[0219] In one embodiment, the digital input may be configured based on sensor data preprocessed using at least one amplifier. For example, the device (100) may include at least one amplifier, and at least one amplifier may preprocess the sensor data.
[0220] Additionally, in one embodiment, the preprocessed sensor data may include a plurality of alternating currents amplified to different sizes. For example, the device (100) may include a multi-amplifier, and the multi-amplifier may convert the alternating current into a plurality of alternating currents amplified to different sizes.
[0221] In one embodiment, the alternating current may represent a preserved waveform by adding a preset offset to the original waveform. For example, the first sensor unit may include an offset adding circuit, and the offset adding circuit may preserve the waveform of the alternating current by adding a preset offset to the original waveform.
[0222] Additionally, in one embodiment, the alternating current may be measured by a Hall sensor. For example, the first sensor unit may include a Hall sensor that measures the alternating current.
[0223] In one embodiment, the operation unit (130) can convert a digital input into an analog input and input it into a cell array. For example, the operation unit (130) may further include a digital-to-analog converter that converts a digital input into an analog input and inputs it into a cell array.
[0224] In step 1420, the device (100) can obtain a digital output of the operation unit (130) corresponding to the digital input. For example, the control unit (120) can obtain a digital output of the operation unit (130) corresponding to the digital input. At this time, the digital output may indicate whether an arc has occurred in the electrical circuit.
[0225] In one embodiment, the operation unit (130) can convert the analog output of the cell array into a digital output. For example, the operation unit (130) may further include an analog-to-digital converter that converts the analog output of the cell array into a digital output.
[0226] In one embodiment, each of the digital input and digital output may be transmitted to or from the operation unit (130) based on synchronous communication.
[0227] For example, communication between the control unit (120) and the operation unit (130) can be performed based on synchronous communication. As an example, a digital input can be transmitted from the control unit (120) to the operation unit (130) or from the operation unit (130) to the control unit (120).
[0228] In one embodiment, steps 1410 to 1420 may be performed during an effective period excluding a communication restricted period that includes a point where the alternating voltage for the electric circuit indicates a zero crossing.
[0229] For example, the sensor unit (110) may include a zero-crossing detection unit that detects a point where the alternating voltage of the electrical circuit indicates a zero crossing. At this time, the control unit (120) may transmit a digital input to the calculation unit (130) during the valid period, excluding the communication restricted period that includes the point indicating the zero crossing. Additionally, the calculation unit (130) may transmit a digital output during the valid period to the control unit (120).
[0230] In one embodiment, steps 1410 to 1420 may be performed repeatedly at every first cycle. For example, the transmission of digital input and the transmission of digital output may be performed repeatedly at every first cycle. In this case, the first cycle may be set as an integer multiple of the half-cycle of the alternating voltage.
[0231] In step 1430, the device (100) can generate a cutoff signal based on the digital output of the calculation unit (130). For example, the control unit (120) can generate a cutoff signal based on the digital output of the calculation unit (130). At this time, the digital output of the calculation unit (130) may indicate whether an arc has occurred in the electrical circuit.
[0232] In one embodiment, the device (100) determines whether to repeatedly generate a blocking signal every second period, and may determine whether to generate a blocking signal based on a plurality of digital outputs repeatedly obtained every first period. For example, the control unit (120) determines whether to repeatedly generate a blocking signal every second period, and may determine whether to generate a blocking signal based on a plurality of digital outputs repeatedly obtained every first period.
[0233] At this time, the second period may be set as an integer multiple of the half-period of the alternating voltage, and the plurality of digital outputs may include a preset number of digital outputs including the most recently acquired digital output.
[0234] FIGS. 15a and 15b are schematic side views illustrating an example of a device for detecting arc generation based on a neural network model according to one embodiment.
[0235] Referring to FIGS. 15a and 15b, the device (100) may include a sensor unit (1501), a power supply unit (1502), a calculation unit (1503), and a blocking unit (1504). In one embodiment, the device (100) may include a first PCB (1510) equipped with a sensor unit (1501), a second PCB (1520) equipped with a power supply unit (1502) and a blocking unit (1504), and a third PCB (1530) equipped with a calculation unit (1503).
[0236] As described above with reference to FIG. 7, etc., the sensor unit (1501) may include at least one of a first sensor unit for measuring alternating current of an electrical circuit equipped with a device (100), a second sensor unit for measuring alternating voltage, and a third sensor unit for measuring leakage current.
[0237] Meanwhile, the power supply unit (1502) can supply driving power for driving at least one component constituting the device (100), such as the control unit (120) and the calculation unit (130). For example, the power supply unit (1502) may include an AC-DC converter that converts AC power into DC driving power.
[0238] Additionally, as described above with reference to FIG. 4, etc., the operation unit (1503) may include a digital-to-analog converter, a crossbar array including a cell array, and an analog-to-digital converter.
[0239] Additionally, the blocking unit (1504) may include a solenoid and a solenoid control circuit that perform a blocking operation.
[0240] In one embodiment, the first PCB (1510) and the second PCB (1520) may be connected vertically to the third PCB (1530). For example, each of the first PCB (1510) and the second PCB (1520) may be connected vertically to the third PCB (1530) so as to be parallel to each other.
[0241] In one embodiment, the sensor portion (1501) may be placed on one side of the first PCB (1510) opposite to the second PCB (1520). Additionally, the first PCB (1510) and the second PCB (1520) may be spaced apart by a predetermined gap (1540). For example, the predetermined gap (1540) may be designed to be in the range of 5-15 mm, but may be changed as needed, such as for specifications.
[0242] Through this, the sensor unit (1501) can be separated from the power supply unit (1502) by a certain distance, and air insulation breakdown phenomena such as switching noise or instantaneous overvoltage caused by the operation of the switching circuit of the power supply unit (1502) can be prevented, and sensor data can be measured stably.
[0243] Meanwhile, in one embodiment, the operation unit (1503) may be placed on one side of the third PCB (1530) opposite to the power supply unit (1502).
[0244] Through this, the influence of switching noise or momentary overvoltage generated in the power supply unit (1502) on the communication and neural network operations of the operation unit (1503) can be minimized, so the operation unit (1503) can stably transmit and receive digital input and output and perform neural network operations based on analog input and output.
[0245] Additionally, the blocking unit (1504) can generate a relatively large magnetic field by means of a solenoid, and the magnetic field can cause distortion in the measurement of the sensor unit (1501), such as a current sensor.
[0246] In one embodiment, the blocking portion (1504) may be placed on one side of the second PCB (1520) opposite to the sensor portion (1501).
[0247] Through this, the first PCB (1510) and the second PCB (1520) can provide a shielding effect, minimize the effect of the magnetic field caused by the blocking part (1504) on the sensor part (1501), and maximize the measurement performance of the sensor part (1501).
[0248] FIG. 16 is an exemplary drawing for explaining a blocking member according to one embodiment.
[0249] Referring to FIG. 16, the blocking unit (1600) may include a solenoid (1610). The solenoid (1610) may control a circuit breaker switch, and the blocking unit (1600) may further include a field effect transistor (FET) (1620) for controlling the circuit breaker switch control operation of the solenoid (1610).
[0250] According to the prior art, a silicon controlled rectifier was used to control the circuit breaker switch control operation of the solenoid (1610). However, the silicon controlled rectifier has the problem that the required power is fixed to the output power line, and once it starts operating, it can only stop operating when the power supply to the load is cut off, which causes malfunction. Additionally, there is a risk of damage and fire in the event of reverse wiring.
[0251] According to one embodiment, the blocking unit (1600) may be implemented by including a solenoid (1610) and a field-effect transistor (1620) that controls the blocking switch control operation of the solenoid (1610).
[0252] By using a field-effect transistor (1620) that allows for easy on-off control, safety against malfunction can be ensured, and since the input power line or output power line can be freely selected using a connecting line (1601), an efficient layout design of the device (100) can be achieved.
[0253] Unless explicitly stated otherwise, the steps constituting the method according to the present disclosure may be performed in a suitable order. The present disclosure is not necessarily limited by the order in which the steps are described. The use of any examples or exemplary terms (e.g., etc.) in the present disclosure is merely for the purpose of describing the present disclosure in detail and, unless limited by the claims, the scope of the present disclosure is not limited by such examples or exemplary terms. Furthermore, those skilled in the art will understand that various modifications, combinations, and changes may be made according to design conditions and factors within the scope of the claims or equivalents to which they are added.
[0254] Accordingly, the scope of the present disclosure should not be limited to the embodiments described above, and all scopes equivalent to or equivalently modified from the claims set forth below, as well as the claims set forth below, shall be considered to fall within the scope of the scope of the present disclosure.
Claims
1. A sensor unit for acquiring sensor data for an electrical circuit; A computation unit comprising a cell array including a plurality of memory cells for storing weights of a previously trained neural network model; and A control unit that inputs a digital input configured based on the sensor data to the processing unit, obtains a digital output of the processing unit corresponding to the digital input, and generates a blocking signal based on the digital output; Includes, The above digital output is a device for detecting arc generation based on a neural network model, which indicates whether an arc has occurred in the electrical circuit.
2. In Paragraph 1, The above operation unit is, A digital-to-analog converter that converts the above digital input into an analog input and inputs it to the cell array; and An analog-to-digital converter that converts the analog output of the cell array into the digital output; A device that further includes 3. In Paragraph 1, The sensor unit above is, A first sensor unit for acquiring alternating current; A second sensor unit for acquiring alternating current voltage; and A third sensor unit for acquiring leakage current; Includes, The above device is, At least one amplifier for preprocessing the sensor data; A device including 4. In Paragraph 3, The device, wherein at least one amplifier comprises a multi-amplifier that converts the alternating current into a plurality of alternating currents amplified to different magnitudes.
5. In Paragraph 3, The above first sensor unit is a device comprising an offset addition circuit that preserves the waveform of the alternating current by adding a preset offset to the original waveform.
6. In Paragraph 3, The above first sensor unit is a device comprising a Hall sensor that measures the alternating current.
7. In Paragraph 1, A device in which communication between the control unit and the operation unit is performed based on synchronous communication.
8. In Paragraph 1, The sensor unit above is, A zero crossing detection unit that detects a point where the alternating voltage for the above electrical circuit indicates a zero crossing; Includes, The control unit transmits the digital input to the calculation unit during a valid period excluding a communication restriction period including a point indicating the zero crossing, and The above-mentioned operation unit is a device that transmits the digital output to the control unit during the above-mentioned valid interval.
9. In Paragraph 8, The transmission of the digital input and the transmission of the digital output are performed repeatedly at every first period, and A device in which the first period is set as an integer multiple of the half-period of the alternating voltage.
10. In Paragraph 9, The control unit determines whether to repeatedly generate the blocking signal every second period, and determines whether to generate the blocking signal based on a plurality of digital outputs repeatedly obtained every first period. The above second period is set as an integer multiple of the half-period of the above alternating voltage, and The above plurality of digital outputs comprises a preset number of digital outputs, including a recently acquired digital output.
11. A step of inputting a digital input configured based on sensor data for an electrical circuit to a processing unit; A step of obtaining a digital output of the processing unit corresponding to the digital input; and A step of generating a blocking signal based on the above digital output; Includes, The above operation unit includes a cell array comprising a plurality of memory cells that store weights of a previously learned neural network model, and A method for detecting arc generation based on a neural network model, wherein the digital output above indicates whether an arc has occurred in the electrical circuit.
12. In Paragraph 11, The above-described operation unit converts the digital input into an analog input and inputs it to the cell array, and converts the analog output of the cell array into the digital output.
13. In Paragraph 11, The above sensor data includes alternating current, alternating voltage, and leakage current, and A method in which the above digital input is configured based on sensor data preprocessed using at least one amplifier.
14. In Paragraph 13, A method in which the above-mentioned preprocessed sensor data includes a plurality of alternating currents amplified to different sizes.
15. In Paragraph 13, A method in which the above alternating current represents a preserved waveform by adding a preset offset to the original waveform.
16. In Paragraph 13, The above alternating current is a method in which the current is measured by a Hall sensor.
17. In Paragraph 11, A method in which each of the above digital input and the above digital output is transmitted to the above processing unit or transmitted from the above processing unit based on synchronous communication.
18. In Paragraph 11, The above input step to the above acquisition step is, A method performed during a valid section excluding a communication restricted section that includes a point where the alternating voltage for the above electrical circuit indicates a zero crossing.
19. In Paragraph 18, The above input step to the above acquisition step is, It is performed repeatedly in every first cycle, and A method in which the first period is set as an integer multiple of the half-period of the alternating voltage.
20. In Paragraph 19, The step of generating the above is, A step of determining whether to repeatedly generate the blocking signal every second period, wherein the determination of whether to generate the blocking signal based on a plurality of digital outputs repeatedly acquired every first period; Includes more, The above second period is set as an integer multiple of the half-period of the above alternating voltage, and A method in which the plurality of digital outputs includes a predetermined number of digital outputs, including a recently acquired digital output.
