Apparatus and method for diagnosing state of permanent magnet synchronous motor

Abstract

An apparatus and method for diagnosing the state of a Permanent Magnet Synchronous Motor (PMSM) are disclosed. A method for diagnosing the state of a Permanent Magnet Synchronous Motor according to one embodiment of the present invention may include: (a) collecting a current signal associated with the PMSM using a current sensor; (b) converting the current signal into a spectrogram in the time-frequency domain; and (c) inputting the converted spectrogram into an artificial intelligence-based analysis model of a pre-trained Binary-Weighted Transformer (BWT) structure to derive analysis information regarding the state of the PMSM.

Description

Technology Field

[0001] The present invention relates to an apparatus and method for diagnosing the condition of a Permanent Magnet Synchronous Motor (PMSM). For example, the present invention relates to a Prognostics and Health Management (PHM) technique for a binary-weighted transformer-based PMSM. Background Technology

[0002] Permanent Magnet Synchronous Motors (PMSMs) are widely used in industrial settings due to their high efficiency and power density. For the stable operation of PMSMs, a Prognostics and Health Management (PHM) system that performs fault diagnosis and predictive maintenance is essential.

[0003] Conventionally, methods analyzing current signals have been used to diagnose PMSM failures, and techniques have been developed to determine normal and faulty states through frequency component analysis. In particular, research utilizing deep learning technology to classify motor conditions has recently been actively underway.

[0004] However, conventional deep learning-based fault diagnosis methods require a large number of parameters, resulting in large model sizes and high-performance processors such as high-performance central processing units (CPUs) or graphics processing units (GPUs). Consequently, there is a limitation in that real-time processing is difficult on resource-constrained edge devices used in actual industrial settings.

[0005] Furthermore, while existing Transformer-based models achieve high classification accuracy, their implementation in embedded systems is practically difficult due to the use of parameters exceeding several hundred megabytes (MB). Moreover, systems implemented solely in software suffer from slow processing speeds, which limits real-time monitoring and immediate fault diagnosis.

[0006] Therefore, lightweight artificial intelligence models and hardware acceleration technologies are required to diagnose the status of PMSMs in real time while maintaining high accuracy, even in resource-constrained edge device environments.

[0007] The technology forming the background of the present invention is disclosed in Korean Published Patent Application No. 10-2022-0156037. The problem to be solved

[0008] The present invention aims to solve the problems of the aforementioned conventional technology by providing an apparatus and method for diagnosing the state of a permanent magnet synchronous motor (PMSM) that can diagnose the state of the PMSM with high accuracy even in resource-constrained edge devices by utilizing a lightweight binary-weighted transformer (BWT) structure.

[0009] The present invention aims to solve the problems of the aforementioned conventional technology by providing an apparatus and method for diagnosing the condition of a permanent magnet synchronous motor that enables real-time fault diagnosis with significantly improved processing speed compared to software-only implementation through software / hardware co-design using a matrix multiplication accelerator.

[0010] The present invention aims to solve the problems of the aforementioned conventional technology by providing an apparatus and method for diagnosing the condition of a permanent magnet synchronous motor that can accurately classify various failure types, including demagnetization and unbalanced states as well as normal states, by analyzing the time-frequency characteristics of a current signal.

[0011] However, the technical problems that the embodiments of the present invention aim to solve are not limited to the technical problems described above, and other technical problems may exist. means of solving the problem

[0012] As a technical means for achieving the above-mentioned technical task, a method for diagnosing the state of a permanent magnet synchronous motor according to one embodiment of the present invention may include: (a) collecting a current signal associated with the PMSM using a current sensor; (b) converting the current signal into a spectrogram in the time-frequency domain; and (c) inputting the converted spectrogram into an artificial intelligence-based analysis model of a pre-trained Binary-Weighted Transformer (BWT) structure to derive analysis information regarding the state of the PMSM.

[0013] In addition, the above analysis model may be a model in which weights are quantized into binary values.

[0014] Additionally, the analysis model may include a class embedding layer for extracting classification information from the spectrogram, a positional encoding layer for encoding positional information of the spectrogram, a source fully connected layer for converting the spectrogram into a feature vector, an encoder for processing the feature vector to extract state features of the PMSM, and a last fully connected layer for generating analysis information based on the extracted state features.

[0015] Additionally, the above step (c) may include a step of processing matrix multiplication operations for the operation of the analysis model using a matrix multiplication accelerator.

[0016] In addition, the matrix multiplication accelerator may be composed of a systolic array structure.

[0017] In addition, the matrix multiplication accelerator may be configured to perform matrix multiplication operations between binary weights and floating-point values.

[0018] In addition, the matrix multiplication accelerator may include a processing element that performs a multiplication operation using a sign converter including a logic inverter and a multiplexer.

[0019] Additionally, the above step (c) may derive information regarding the result of classifying the state of the PMSM into any one of a plurality of predefined states, including at least one of a normal state, a fault state, a demagnetization state, and an unbalanced state, as the analysis information.

[0020] In addition, the current sensor may include a three-phase current sensor.

[0021] In addition, step (b) above can convert the current signal into the spectrogram using a Short-Time Fourier Transform (STFT).

[0022] Meanwhile, a device for diagnosing the state of a permanent magnet synchronous motor according to one embodiment of the present invention may include a data collection unit that collects a current signal associated with the PMSM using a current sensor, a data conversion unit that converts the current signal into a spectrogram in the time-frequency domain, and an analysis execution unit that inputs the converted spectrogram into an artificial intelligence-based analysis model of a pre-trained Binary-Weighted Transformer (BWT) structure to derive analysis information regarding the state of the PMSM.

[0023] In addition, the analysis execution unit can process the matrix multiplication operation for the operation of the analysis model using a matrix multiplication accelerator.

[0024] In addition, the analysis performing unit may derive information regarding the result of classifying the state of the PMSM into any one of a plurality of predefined states including at least one of a normal state, a fault state, a demagnetization state, and an unbalanced state as the analysis information.

[0025] In addition, the data conversion unit can convert the current signal into the spectrogram using a Short-Time Fourier Transform (STFT).

[0026] The means for solving the problem described above are merely exemplary and should not be interpreted as intended to limit the present invention. In addition to the exemplary embodiments described above, additional embodiments may exist in the drawings and the detailed description of the invention. Effects of the invention

[0027] According to the means for solving the problem of the present invention described above, it is possible to provide an apparatus and method for diagnosing the state of a permanent magnet synchronous motor (PMSM) that can diagnose the state of the PMSM with high accuracy even in resource-constrained edge devices by utilizing a lightweight binary-weighted transformer (BWT) structure.

[0028] According to the means for solving the problem of the present invention described above, it is possible to provide an apparatus and method for diagnosing the condition of a permanent magnet synchronous motor capable of real-time fault diagnosis with significantly improved processing speed compared to software-only implementation through software / hardware co-design using a matrix multiplication accelerator.

[0029] According to the means for solving the problem of the present invention described above, it is possible to provide an apparatus and method for diagnosing the condition of a permanent magnet synchronous motor that can accurately classify various types of failures, including demagnetization and unbalanced conditions as well as normal conditions, by analyzing the time-frequency characteristics of a current signal.

[0030] According to the solution to the problem of the present invention described above, there is an advantage in that efficient diagnosis is possible even on resource-constrained edge devices by achieving high classification accuracy while reducing parameter size through a Binary-Weighted Transformer (BWT) structure.

[0031] According to the solution to the problem of the present invention described above, through software / hardware collaborative design utilizing a matrix multiplication accelerator based on a systolic array structure, significantly improved processing speed compared to software-only implementation is achieved, enabling real-time fault diagnosis and predictive maintenance.

[0032] According to the solution to the problem of the present invention described above, by accurately classifying various conditions including normal, faulty, demagnetized, and unbalanced states through spectrogram analysis of current signals, it is possible to contribute to improving motor reliability and reducing maintenance costs in industrial sites.

[0033] However, the effects obtainable from this invention are not limited to those described above, and other effects may exist. Brief explanation of the drawing

[0034] FIG. 1 is a schematic diagram of a prognostics and health management (PHM) system according to one embodiment of the present invention. FIG. 2 is a conceptual diagram illustrating the overall operation flow of a device for diagnosing the state of a permanent magnet synchronous motor (PMSM) according to one embodiment of the present invention. FIG. 3 illustrates a data acquisition environment according to one embodiment of the present invention, wherein (a) is a Permanent Magnet Synchronous Motor (PMSM) for Urban Air Mobility (UAM) and (b) is a diagram exemplarily illustrating a current sensor. FIG. 4 is a conceptual diagram showing the structure of a Binary-Weighted Transformer (BWT) according to one embodiment of the present invention. FIG. 5 shows the hardware structure of a matrix multiplication accelerator according to one embodiment of the present invention, wherein (a) is the overall structure of the matrix multiplication accelerator and (b) is an exemplary diagram showing the structure of a processing element (PE) within the matrix multiplication accelerator. FIG. 6 is a confusion matrix showing the classification performance of a Binary-Weighted Transformer (BWT) according to one embodiment of the present invention. FIG. 7 is a schematic diagram of a device for diagnosing the condition of a permanent magnet synchronous motor (PMSM) according to one embodiment of the present invention. FIG. 8 is an operation flowchart for a method of diagnosing the state of a permanent magnet synchronous motor (PMSM) according to one embodiment of the present invention. Specific details for implementing the invention

[0035] Embodiments of the present invention are described below with reference to the attached drawings to enable those skilled in the art to easily implement the invention. However, the present invention may be embodied in various different forms and is not limited to the embodiments described herein. Furthermore, in order to clearly explain the present invention in the drawings, parts unrelated to the explanation have been omitted, and similar parts throughout the specification are denoted by similar reference numerals.

[0036] Throughout this specification, when a part is described as being "connected" to another part, this includes not only cases where they are "directly connected," but also cases where they are "electrically connected" or "indirectly connected" with other elements interposed between them.

[0037] Throughout the entire specification, when a component is described as being located "on," "on top," "on top," "under," "on bottom," or "on bottom" of another component, this includes not only cases where the component is in contact with the other component but also cases where another component exists between the two components.

[0038] Throughout this specification, when a part is described as "comprising" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components.

[0039] The present invention relates to an apparatus and method for diagnosing the condition of a Permanent Magnet Synchronous Motor (PMSM). For example, the present invention relates to a Prognostics and Health Management (PHM) technique for a binary-weighted transformer-based PMSM.

[0040] FIG. 1 is a schematic diagram of a prognostics and health management (PHM) system according to one embodiment of the present invention.

[0041] Referring to FIG. 1, a prognostics and health management (PHM) system (10) according to one embodiment of the present invention may include a device (100) for diagnosing the condition of a permanent magnet synchronous motor (PMSM) according to one embodiment of the present invention (hereinafter referred to as the 'diagnostic device (100)'), a current sensor (200), a user terminal (300), and a database (400).

[0042] The diagnostic device (100), current sensor (200), user terminal (300), and database (400) can communicate with each other through a network (20). The network (20) refers to a connection structure that enables information exchange between each node, such as terminals and servers. Examples of such a network (20) include, but are not limited to, a 3GPP (3rd Generation Partnership Project) network, an LTE (Long Term Evolution) network, a 5G network, a WIMAX (World Interoperability for Microwave Access) network, the Internet, a LAN (Local Area Network), a Wireless LAN (Wireless Local Area Network), a WAN (Wide Area Network), a PAN (Personal Area Network), a Wi-Fi network, a Bluetooth network, a satellite broadcasting network, an analog broadcasting network, and a DMB (Digital Multimedia Broadcasting) network.

[0043] The user terminal (300) can be any type of wireless communication device, such as a smartphone, smartpad, tablet PC, PCS (Personal Communication System), GSM (Global System for Mobile communication), PDC (Personal Digital Cellular), PHS (Personal Handyphone System), PDA (Personal Digital Assistant), IMT (International Mobile Telecommunication)-2000, CDMA (Code Division Multiple Access)-2000, W-CDMA (W-Code Division Multiple Access), Wibro (Wireless Broadband Internet) terminal.

[0044] Additionally, in the description of the embodiments of the present invention, the database (400) can store and manage various data generated or utilized during the process in which the diagnostic device (100) performs a state diagnosis of a permanent magnet synchronous motor (PMSM). For example, the database (400) can store raw data of current signals collected through a current sensor (200). For another example, the database (400) can store spectrogram data generated through a Short-Time Fourier Transform (STFT). For yet another example, the database (400) can store analysis information derived from an artificial intelligence-based analysis model with a Binary-Weighted Transformer (BWT) structure, and said analysis information may include the state classification result of the permanent magnet synchronous motor (PMSM), classification probability value, confidence score, etc.

[0045] Additionally, the database (400) can store learning parameters, weight information, model structure information, etc. of a pre-trained artificial intelligence-based analysis model. The database (400) can be implemented in various forms, such as a relational database, an object-oriented database, or a NoSQL database, and can be configured in the form of cloud-based storage or a local storage device.

[0046] Additionally, data stored in the database (400) can be used as training data for the diagnostic device (100), or for purposes such as tracking long-term motor status through history analysis, statistical analysis, and detection of abnormal patterns.

[0047] FIG. 2 is a conceptual diagram illustrating the overall operation flow of a device for diagnosing the state of a permanent magnet synchronous motor (PMSM) according to one embodiment of the present invention.

[0048] Referring to FIG. 2, the diagnostic device (100) disclosed herein can collect a current signal from a permanent magnet synchronous motor (PMSM) and convert it into a spectrogram in the time-frequency domain. The converted spectrogram can be input into a lightweight artificial intelligence-based analysis model with a Binary-Weighted Transformer (BWT) structure and used to classify the state of the motor.

[0049] Meanwhile, the diagnostic device (100) disclosed herein can be designed to enable real-time processing even in an edge device environment with limited resources, such as a Field-Programmable Gate Array (FPGA), and can implement high-speed computation through software / hardware co-design utilizing a matrix multiplication accelerator, and as described in detail below, the diagnostic device (100) can classify the normal state, fault state, demagnetization state, and unbalanced state of a permanent magnet synchronous motor (PMSM) with high accuracy.

[0050] FIG. 3 illustrates a data acquisition environment according to one embodiment of the present invention, wherein (a) is a Permanent Magnet Synchronous Motor (PMSM) for Urban Air Mobility (UAM) and (b) is a diagram exemplarily illustrating a current sensor.

[0051] Referring to FIG. 3(a), the permanent magnet synchronous motor (PMSM) disclosed herein may, for example, be a component of a propulsion system for Urban Air Mobility (UAM). In particular, the permanent magnet synchronous motor (PMSM) may be suitable as a propeller drive motor for a UAM system with Vertical Take-Off and Landing (VTOL) capabilities due to its high power density, excellent efficiency, and low noise characteristics, but is not limited thereto.

[0052] For reference, UAM systems generally adopt a multi-rotor structure, and each rotor can be driven by an independent permanent magnet synchronous motor (PMSM). In multi-rotor UAM systems, even an imbalance in a single propeller can significantly degrade the overall system stability, which can induce motor eccentricity and have a critical impact on flight safety.

[0053] In addition, the demagnetization of permanent magnets can cause serious problems such as torque reduction or overcurrent, so real-time motor condition diagnosis is essential to meet strict safety requirements to protect passengers and assets during UAM operation.

[0054] For example, the permanent magnet synchronous motor (PMSM) disclosed herein may have a three-phase (U, V, W) winding structure, and the normal operation of the motor can be determined by monitoring the current signal of each phase. For another example, the permanent magnet synchronous motor (PMSM) may have operating specifications such as rated output, rated voltage, rated current, and rotational speed, and the normal range of the current signal may be determined according to these specifications.

[0055] Additionally, referring to FIG. 3(b), the current sensor (200) disclosed herein may be a sensor module capable of measuring the current flowing through the three-phase windings (U, V, W) of a permanent magnet synchronous motor (PMSM) in real time. For example, the current sensor (200) may be implemented in various ways, such as a Hall Effect-based current sensor, a Shunt Resistor-based current sensor, or a Current Transformer (CT)-based current sensor.

[0056] As another example, the current sensor (200) may be a high-precision current measurement sensor and may have high sensitivity characteristics capable of detecting minute current changes in the milliampere (mA) range. The current sensor (200) may sample the current signal at a constant sampling frequency, and the sampling frequency may be set to more than twice the maximum frequency of the signal to be measured according to the Nyquist Theorem, thereby preventing aliasing and ensuring accurate reconstruction of the signal.

[0057] Meanwhile, the current signal obtained through the current sensor (200) can be converted into a digital signal through an analog-to-digital converter (ADC), and the converted digital signal can be transmitted to a diagnostic device (100) and used for subsequent signal processing and analysis.

[0058] In addition, according to one embodiment of the present invention, current sensors (200) can be independently placed for each phase (U, V, W) of a permanent magnet synchronous motor (PMSM), and the three current sensors can be synchronized to measure current values ​​simultaneously, thereby allowing for the analysis of the interrelationship of three-phase current signals.

[0059] Below, the specific functions and operations of the diagnostic device (100) will be described in detail.

[0060] The diagnostic device (100) can collect a current signal associated with the PMSM using a current sensor (200).

[0061] For example, the current sensor (200) may be a three-phase current sensor, and the current signal associated with the PMSM obtained through the current sensor (200) is an analog signal that changes over time and may contain important information that reflects the operating state of the permanent magnet synchronous motor (PMSM).

[0062] Under normal operating conditions, the current signal may exhibit a constant pattern depending on the motor's rotational speed, load conditions, power supply voltage, etc. In contrast, if a fault or abnormal condition occurs in the motor, the amplitude, frequency, and phase of the current signal may change, and these changes can display different characteristics depending on the type and severity of the fault.

[0063] For example, when demagnetization of a permanent magnet occurs, the back electromotive force (Back-EMF) of the motor decreases, and consequently, certain harmonic components of the current signal may increase. As another example, when imbalance occurs in the propeller, periodic load fluctuations occur in the motor, which may appear in the form of amplitude modulation of the current signal.

[0064] As another example, when an electrical fault occurs, such as a short circuit or open circuit in the winding, an imbalance between the three-phase current signals may appear. While current signals can be analyzed directly in the time domain, it may be advantageous to convert them to the frequency domain or time-frequency domain to effectively capture minute changes in frequency components caused by the fault.

[0065] Additionally, the diagnostic device (100) can convert the collected current signal into a spectrogram in the time-frequency domain. In this regard, the spectrogram visualizes the change in the frequency component of the signal over time in the form of a two-dimensional image, where the horizontal axis represents time, the vertical axis represents frequency, and the color or brightness represents the signal strength (magnitude or power) at the corresponding time-frequency point.

[0066] Such spectrograms can be particularly useful for analyzing non-stationary signals and can effectively capture frequency characteristics that change over time during motor operation. For example, in a steady-state permanent magnet synchronous motor (PMSM), harmonic components corresponding to the fundamental frequency and its multiples may appear in a consistent pattern on the spectrogram.

[0067] As another example, in a demagnetization state, patterns in which signal strength increases or decreases in specific frequency bands may be observed. As yet another example, in an unbalanced state, patterns in which additional frequency components appear in the low-frequency band or the frequency modulates over time may be observed. Since spectrograms are data in the form of images, they can be directly used as input for deep learning models such as Convolutional Neural Networks (CNNs) and Transformers, and the models can automatically classify the state of the motor by learning the patterns of the spectrograms.

[0068] For example, the diagnostic device (100) can convert a current signal into a spectrogram using a Short-Time Fourier Transform (STFT).

[0069] More specifically, according to one embodiment of the present invention, the Short-Time Fourier Transform (STFT) may be a method for dividing a signal in the time domain into small time intervals (windows) and performing a Fourier Transform on each interval to extract frequency components in the corresponding time interval. As an example, a Hamming window, a Hann window, a Gaussian window, etc., may be used as the window function, and the window size and hop length may be determined by considering the trade-off between time resolution and frequency resolution.

[0070] As another example, a larger window size improves frequency resolution but may degrade temporal resolution, while conversely, a smaller window size improves temporal resolution but may degrade frequency resolution. The spectrogram generated by the Short-Time Fourier Transform (STFT) can be in the form of complex number data, which can be converted into magnitude or power spectrograms for use.

[0071] In addition, the magnitude spectrogram represents the signal magnitude at each time-frequency point, and the power spectrogram can represent the square of the magnitude. The generated spectrogram can undergo a normalization process to adjust pixel values ​​to a constant range (e.g., between 0 and 1 or between -1 and 1), which can contribute to the learning stability and performance improvement of AI-based analysis models.

[0072] Additionally, the diagnostic device (100) can derive analysis information regarding the state of the PMSM by inputting the converted spectrogram into an artificial intelligence-based analysis model of a pre-trained Binary-Weighted Transformer (BWT) structure.

[0073] FIG. 4 is a conceptual diagram showing the structure of a Binary-Weighted Transformer (BWT) according to one embodiment of the present invention.

[0074] Referring to FIG. 4, the artificial intelligence-based analysis model disclosed herein is a model in which weights are quantized into binary values, and may be composed of a hierarchical structure including a class embedding layer for extracting classification information from a spectrogram, a positional encoding layer for encoding positional information of the spectrogram, a source fully connected layer for converting the spectrogram into a feature vector, an encoder for processing the feature vector to extract state features of the PMSM, and a last fully connected layer for generating analysis information based on the extracted state features.

[0075] More specifically, referring to FIG. 4, the artificial intelligence-based analysis model of the Binary-Weighted Transformer (BWT) structure disclosed herein may have a multilayer neural network structure for classifying the state of a permanent magnet synchronous motor (PMSM) from an input spectrogram. First, a class embedding layer may generate a special token for classification tasks and add it to the input data, which can perform the role of aggregating global information in the transformer structure.

[0076] Next, the Positional Encoding layer can encode and add positional information for each patch or token of the spectrogram, enabling the model to learn spatial positional relationships in the time-frequency domain. Positional encoding can be implemented using a fixed encoding method or a learnable encoding method utilizing sine and cosine functions.

[0077] Additionally, the Source Fully Connected Layer can convert the input spectrogram into a high-dimensional feature vector, which can be provided as input to the encoder. The encoder can be composed of a multi-layered self-attention mechanism and a feed-forward neural network, and each layer may include residual connections and layer normalization.

[0078] In addition, the self-attention mechanism can extract global features by learning the relationships between all locations in the input sequence, and can simultaneously learn information from various representation subspaces through Multi-Head Self-Attention (MHSA).

[0079] In addition, the Last Fully Connected Layer can take the encoder's output feature vector as input and ultimately output a probability distribution for each state class, and can provide normalized probability values ​​through a Softmax activation function.

[0080] Meanwhile, according to one embodiment of the present invention, the artificial intelligence-based analysis model of the Binary-Weighted Transformer (BWT) structure disclosed herein is not merely a model that applies binary weights to an existing Transformer structure, but may be a model specifically designed and optimized for the application of real-time predictive maintenance and prognostics and health management (PHM) in resource-constrained edge device environments.

[0081] In particular, the analysis model disclosed herein can significantly reduce the parameter size of the model compared to existing floating-point based models by quantizing weights into binary values, which drastically reduces memory usage and storage space requirements, thereby facilitating deployment in embedded systems.

[0082] Furthermore, the analysis model disclosed herein can replace complex floating-point multiplication operations during matrix multiplication with simple sign conversion and addition operations by using binary weights, which has the advantage of reducing computational complexity and enabling the efficient implementation of hardware accelerators.

[0083] In addition, the analysis model can effectively learn changes in frequency characteristics due to motor failures or abnormal conditions by using time-frequency domain spectrograms as input, and can simultaneously extract local and global features through the Transformer's self-attention mechanism.

[0084] In addition, the analysis model disclosed herein can achieve high classification accuracy with only a small number of parameters, which means that lightweighting and high performance of the model have been achieved simultaneously. In particular, the analysis model is implemented on a Field-Programmable Gate Array (FPGA) platform and can provide real-time processing performance through software / hardware co-design.

[0085] Meanwhile, the diagnostic device (100) can derive information as analysis information regarding the result of classifying the state of the PMSM into one of a plurality of predefined states, including at least one of a normal state, a fault state, a demagnetization state, and an unbalanced state.

[0086] Specifically, among multiple predefined states, the Healthy state (HE) refers to a state in which the Permanent Magnet Synchronous Motor (PMSM) operates normally according to its designed specifications and operating conditions. In the Healthy state, the current signal exhibits a constant pattern, and the fundamental frequency and harmonic components appear in a predictable form on the spectrogram. In the Healthy state, vibration, noise, and heat generation are within normal ranges, and stable operation over a long period is possible.

[0087] In contrast, a fault can refer to a condition in which normal operation is hindered due to mechanical or electrical defects occurring in a permanent magnet synchronous motor (PMSM). A fault can be caused by various factors, such as bearing wear, short or open circuits in the windings, or rotor damage, and abnormal patterns in the amplitude, frequency, and phase of the current signal may be observed. If the fault persists, problems such as reduced motor efficiency, overheating, and increased noise may occur, and it can ultimately lead to the complete failure of the motor.

[0088] Furthermore, demagnetization refers to a state in which the permanent magnets of a Permanent Magnet Synchronous Motor (PMSM) gradually lose their magnetic force due to external factors. Demagnetization can occur due to prolonged operation in high-temperature environments, the application of a reverse magnetic field caused by overcurrent, or mechanical shock. As the magnetic force of the permanent magnets decreases, the motor's torque may drop and the back-EMF may decrease. In the demagnetization state, specific harmonic components of the current signal change, and patterns distinct from the normal state may appear on the spectrogram. As demagnetization progresses, problems such as reduced motor output, decreased efficiency, and control instability may occur.

[0089] Finally, imbalance (SI) refers to a condition in which there is an imbalance in mass distribution within a load connected to a Permanent Magnet Synchronous Motor (PMSM), such as a propeller or rotor. In Urban Air Mobility (UAM) systems, imbalance can occur if the center of mass of the propeller deviates from the center of rotation due to damage or poor attachment of a single blade. Under imbalance conditions, periodic load fluctuations occur in the motor, which can manifest as amplitude modulation of the current signal and an increase in low-frequency components. If imbalance persists, it can shorten the lifespan of the motor and bearings, increase vibration and noise, and degrade the flight stability of the UAM system.

[0090] In this regard, the artificial intelligence-based analysis model of the diagnostic device (100) can perform a multi-class classification task that classifies the state of a permanent magnet synchronous motor (PMSM) into one of a plurality of predefined states based on a feature vector extracted from an input spectrogram.

[0091] Specifically, the feature vectors passed through the encoder are input into the last fully connected layer, which can convert the feature vectors into logit values ​​corresponding to each state class. The logit values ​​can be converted into a probability distribution through a softmax function, and the sum of the probability values ​​for each class can be 1. The model can output the class with the highest probability value as the final classification result, and along with this, it can provide the probability values ​​for each class to indicate the confidence of the classification result.

[0092] For example, if a model outputs a probability distribution for a specific spectrogram of 95% for a steady state, 3% for a broken state, 1% for a potato state, and 1% for an unbalanced state, the final classification result is a steady state and can be interpreted as having a high confidence level of 95%. As another example, if a model outputs similar probability values ​​for multiple classes, this may imply that the input data is located at the boundary between multiple states or has ambiguous characteristics, and in such cases, additional verification or judgment by a human expert may be required.

[0093] Meanwhile, the classification performance of the analysis model can be evaluated through indicators such as accuracy, precision, recall, and F1-score, and the classification results between each class can be visually analyzed through the confusion matrix.

[0094] In addition, according to one embodiment of the present invention, the diagnostic device (100) can process matrix multiplication operations for the operation of the analysis model using a matrix multiplication accelerator.

[0095] FIG. 5 shows the hardware structure of a matrix multiplication accelerator according to one embodiment of the present invention, wherein (a) is the overall structure of the matrix multiplication accelerator and (b) is an exemplary diagram showing the structure of a processing element (PE) within the matrix multiplication accelerator.

[0096] Referring to FIG. 5, the matrix multiplication accelerator disclosed herein is composed of a systolic array structure and can be configured to perform matrix multiplication operations between binary weights and floating-point values.

[0097] Specifically, according to one embodiment of the present invention, the Matrix Multiplication Accelerator disclosed herein can be designed based on a Systolic Array structure and can efficiently process matrix multiplication operations of a Binary-Weighted Transformer (BWT). A Systolic Array is a parallel computing structure in which regular and repetitive Processing Elements (PEs) are arranged in a grid form, and operations can be performed as data flows through the PEs in a pipeline manner.

[0098] For example, a matrix multiplication accelerator can be composed of 16×16 processing elements (PEs), each PE capable of independently performing multiplication and accumulation operations. At the input of the matrix multiplication accelerator, an input RAM (Random Access Memory, RAM) storing input data in Floating-Point 16-bit (FP16) format and a weight RAM storing binarized weights can be placed. Since binarized weights have a value of +1 or -1, each weight can be represented by 1 bit, which can reduce memory usage to 1 / 32 compared to 32-bit floating-point.

[0099] During the matrix multiplication process, input data can be broadcast to PEs corresponding to each row, and weight data can be transmitted to PEs corresponding to each column. Each PE can perform a multiplication operation using the input data and weight data, accumulate the result, and transmit it to the next PE.

[0100] Due to the characteristics of the systolic array, all PEs perform operations simultaneously, allowing for high throughput, and the pipeline structure minimizes computational latency. An output RAM for storing computation results may be placed at the output of the matrix multiplication accelerator, and the output data may be used as input for subsequent layers or output as final analysis information.

[0101] For example, the matrix multiplication accelerator disclosed herein can be designed using a Hardware Description Language (HDL) such as Verilog-HDL and can be synthesized and implemented on a Field-Programmable Gate Array (FPGA) platform.

[0102] Additionally, referring to FIG. 5(b), the matrix multiplication accelerator may include a processing element that performs multiplication operations using a sign converter including a logic inverter and a multiplexer.

[0103] Each processing element (PE) constituting a matrix multiplication accelerator can be designed to perform multiplication operations between binary weights and floating-point values. While a standard floating-point multiplier requires complex logic circuitry and significant hardware resources, using binary weights can simplify multiplication operations.

[0104] Specifically, when the binary weight is +1, the multiplication result is identical to the input value, and when the binary weight is -1, the multiplication result can be the input value with its sign inverted. Utilizing these characteristics, each processing element (PE) can use a sign converter composed of a logic inverter and a multiplexer instead of a complex multiplier. Depending on the value of the binary weight, the sign converter can invert or pass the sign bit of the input floating-point value as is, which can be a very simple operation that can be completed within a single clock cycle.

[0105] Each processing element (PE) can calculate a partial sum by adding the value generated through a code converter to an accumulator, and this partial sum can be passed to the next processing element (PE) or stored as a final output. The hardware implementation of the processing elements (PE) can be optimized to minimize field programmable gate array (FPGA) resources such as look-up tables (LUTs), flip-flops, and block RAMs, thereby freeing up space to deploy more processing elements (PEs) or implement other functions within limited hardware resources.

[0106] FIG. 6 is a confusion matrix showing the classification performance of a Binary-Weighted Transformer (BWT) according to one embodiment of the present invention.

[0107] Referring to FIG. 6, a confusion matrix is ​​shown that evaluates the classification performance of an AI-based analysis model of a Binary-Weighted Transformer (BWT) structure disclosed herein. The confusion matrix is ​​a table that visualizes the classification accuracy between each class by comparing the model's prediction results with the actual labels, where the rows represent the actual classes and the columns represent the predicted classes.

[0108] The diagonal elements of the confusion matrix represent the number of correctly classified samples, and the off-diagonal elements represent the number of incorrectly classified samples. In this embodiment, the test dataset consists of a total of 1,080 samples and may include 270 samples evenly distributed for each state. As a result of analyzing the confusion matrix, for the healthy state (HE), all 270 samples out of 270 were correctly classified, achieving 100% accuracy.

[0109] For the Fault status, 269 out of 270 were correctly classified and 1 was misclassified as a different status, resulting in an accuracy of approximately 99.63%. For Demagnetization, 269 out of 270 were correctly classified, achieving an accuracy of approximately 99.63%. For Imbalance (SI), all 270 were correctly classified, demonstrating 100% accuracy.

[0110] Overall, 1,078 out of 1,080 test samples were correctly classified, achieving an overall classification accuracy of approximately 99.81%. These results demonstrate that the AI-based analysis model of the Binary Weighted Transformer (BWT) structure disclosed herein maintains very high classification performance despite significantly reducing the parameter size.

[0111] Additionally, the diagnostic device (100) can output the derived analysis information through a user terminal (300), etc. Specifically, the diagnostic device (100) can visually display the derived analysis information on the display screen of the user terminal (300), and the current motor status, classification probability, reliability score, history data, etc., can be provided through a graphical user interface (GUI).

[0112] As another example, if the analysis information indicates an abnormal state (failure, deterioration, imbalance, etc.), the diagnostic device (100) can automatically generate a warning alert and send it to the user terminal (300), and the alert can be provided in the form of a visual display, an auditory signal, vibration, a text message, an email, etc.

[0113] As another example, the analysis information can be stored in a database (400) and used for long-term motor condition history management, and through time-series analysis, the motor's performance degradation trend, failure occurrence pattern, and predicted remaining useful life (RUL) can be analyzed.

[0114] In addition, the analysis information can be linked with a higher-level control system (e.g., a flight control system for UAM) and utilized for automated decision-making; for instance, if an anomaly is detected in a specific motor, protective measures such as limiting the output of that motor or redistributing the load to other motors can be automatically performed.

[0115] In addition, the analysis information can be linked with a Maintenance Management System to automatically establish predictive maintenance schedules, prepare necessary parts in advance, and support the deployment of maintenance personnel. Finally, the analysis information can be transmitted in real time to manufacturers or service providers via a Remote Monitoring System, enabling integrated management and statistical analysis of large-scale UAM fleets.

[0116] FIG. 7 is a schematic diagram of a device for diagnosing the condition of a permanent magnet synchronous motor (PMSM) according to one embodiment of the present invention.

[0117] Referring to FIG. 7, the diagnostic device (100) may include a data collection unit (110), a data conversion unit (120), an analysis execution unit (130), and a result output unit (140).

[0118] The data collection unit (110) can collect a current signal associated with the PMSM using a current sensor (200).

[0119] The data conversion unit (120) can convert the collected current signal into a spectrogram in the time-frequency domain.

[0120] For example, the data conversion unit (120) can convert the current signal into a spectrogram using a Short-Time Fourier Transform (STFT).

[0121] The analysis execution unit (130) can derive analysis information about the state of the PMSM by inputting the converted spectrogram into an artificial intelligence-based analysis model of a pre-trained Binary-Weighted Transformer (BWT) structure.

[0122] For example, the analysis performing unit (130) can derive information as analysis information regarding the result of classifying the state of the PMSM into one of a plurality of predefined states including at least one of a normal state, a fault state, a demagnetization state, and an unbalanced state.

[0123] In addition, according to one embodiment of the present invention, the analysis execution unit (130) can process the matrix multiplication operation for the operation of the analysis model using a matrix multiplication accelerator.

[0124] The result output unit (140) can output the derived analysis information through a user terminal (300), etc.

[0125] Below, based on the details described above, we will briefly examine the operation flow of the present invention.

[0126] FIG. 8 is an operation flowchart for a method of diagnosing the state of a permanent magnet synchronous motor (PMSM) according to one embodiment of the present invention.

[0127] The method for diagnosing the condition of a permanent magnet synchronous motor (PMSM) illustrated in FIG. 8 can be performed by the diagnostic device (100) described above. Therefore, even if the details are omitted below, the description of the diagnostic device (100) can be equally applied to the description of the method for diagnosing the condition of a permanent magnet synchronous motor (PMSM).

[0128] Referring to FIG. 8, in step S11, the data collection unit (110) can collect a current signal associated with the PMSM using a current sensor (200).

[0129] Next, in step S12, the data conversion unit (120) can convert the collected current signal into a spectrogram in the time-frequency domain.

[0130] For example, in step S12, the data conversion unit (120) can convert the current signal into a spectrogram using a Short-Time Fourier Transform (STFT).

[0131] Next, in step S13, the analysis execution unit (130) can derive analysis information about the state of the PMSM by inputting the converted spectrogram into an AI-based analysis model of a pre-trained Binary-Weighted Transformer (BWT) structure.

[0132] For example, in step S13, the analysis performing unit (130) can derive information as analysis information regarding the result of classifying the state of the PMSM into one of a plurality of predefined states including at least one of a normal state, a fault state, a demagnetization state, and an unbalanced state.

[0133] In addition, according to one embodiment of the present invention, in step S13, the analysis execution unit (130) can process the matrix multiplication operation for the operation of the analysis model using a matrix multiplication accelerator.

[0134] Next, in step S14, the result output unit (140) can output the derived analysis information through a user terminal (300), etc.

[0135] In the description above, steps S11 through S14 may be further divided into additional steps or combined into fewer steps according to an embodiment of the present invention. Additionally, some steps may be omitted as necessary, and the order of the steps may be changed.

[0136] A method for diagnosing the condition of a Permanent Magnet Synchronous Motor (PMSM) according to one embodiment of the present invention may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., either individually or in combination. The program instructions recorded on the medium may be those specifically designed and configured for the present invention, or they may be those known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, and flash memory. Examples of program instructions include machine code, such as that generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. The above-described hardware device may be configured to operate as one or more software modules to perform the operation of the present invention, and vice versa.

[0137] In addition, the method for diagnosing the condition of the aforementioned Permanent Magnet Synchronous Motor (PMSM) can also be implemented in the form of a computer program or application executed by a computer and stored on a recording medium.

[0138] The foregoing description of the present invention is for illustrative purposes only, and those skilled in the art will understand that other specific forms can be easily modified without altering the technical concept or essential features of the present invention. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. For example, each component described as a single unit may be implemented in a distributed manner, and components described as distributed may likewise be implemented in a combined form.

[0139] The scope of the present invention is defined by the claims set forth below rather than by the detailed description above, and all modifications or variations derived from the meaning and scope of the claims and the concept of equivalents thereof should be interpreted as being included within the scope of the present invention. Explanation of the symbols

[0140] 10: Prognostics and Health Management (PHM) System 100: Device for diagnosing the condition of a Permanent Magnet Synchronous Motor (PMSM) 110: Data Collection Unit 120: Data conversion unit 130: Analysis Execution Unit 140: Result output section 200: Current sensor 300: User terminal 400: Database 20: Network

Claims

Claim 1 A method for diagnosing the state of a Permanent Magnet Synchronous Motor (PMSM) comprises: (a) collecting a current signal associated with the PMSM using a current sensor; (b) converting the current signal into a spectrogram in the time-frequency domain; and (c) inputting the converted spectrogram into an analysis model, which is an AI-based analysis model with a pre-trained Binary-Weighted Transformer (BWT) structure in which weights are quantized into binary values, to derive analysis information regarding the state of the PMSM, wherein the analysis model comprises: a class embedding layer for extracting classification information from the spectrogram; a positional encoding layer for encoding positional information of the spectrogram; a source fully connected layer for converting the spectrogram into a feature vector; and an encoder for processing the feature vector to extract state features of the PMSM. A method comprising a last fully connected layer for generating analysis information based on the extracted state features. Claim 2 delete Claim 3 delete Claim 4 A method according to claim 1, wherein step (c) comprises the step of processing a matrix multiplication operation for the operation of the analysis model using a matrix multiplication accelerator. Claim 5 A method according to claim 4, wherein the matrix multiplication accelerator is configured as a systolic array structure and is configured to perform matrix multiplication operations between binary weights and floating-point values. Claim 6 A method according to claim 5, wherein the matrix multiplication accelerator comprises a processing element that performs a multiplication operation using a sign converter including a logic inverter and a multiplexer. Claim 7 A method according to claim 1, wherein step (c) derives information regarding the result of classifying the state of the PMSM into any one of a plurality of predefined states including at least one of a normal state, a fault state, a demagnetization state, and an unbalanced state as the analysis information. Claim 8 A method according to claim 1, wherein the current sensor comprises a three-phase current sensor. Claim 9 A method according to claim 1, wherein step (b) is characterized by converting the current signal into the spectrogram using a Short-Time Fourier Transform (STFT). Claim 10 A device for diagnosing the state of a Permanent Magnet Synchronous Motor (PMSM) comprises: a data collection unit that collects a current signal associated with the PMSM using a current sensor; a data conversion unit that converts the current signal into a spectrogram in the time-frequency domain; and an analysis execution unit that inputs the converted spectrogram into an analysis model, which is an AI-based analysis model with a pre-trained Binary-Weighted Transformer (BWT) structure in which weights are quantized into binary values, and derives analysis information regarding the state of the PMSM, wherein the analysis model comprises: a class embedding layer for extracting classification information from the spectrogram; a positional encoding layer for encoding position information of the spectrogram; a source fully connected layer for converting the spectrogram into a feature vector; and an encoder for processing the feature vector to extract state features of the PMSM. A device comprising a last fully connected layer for generating analysis information based on the extracted state features. Claim 11 delete Claim 12 In item 10, the above analysis performing unit processes matrix multiplication operations for the operation of the above analysis model using a matrix multiplication accelerator, the device. Claim 13 In claim 12, the matrix multiplication accelerator is formed in a systolic array structure and performs a matrix multiplication operation between binary weights and floating-point values, wherein the matrix multiplication accelerator includes a processing element that performs a multiplication operation using a sign converter including a logic inverter and a multiplexer. Claim 14 In claim 10, the device wherein the analysis performing unit derives information regarding the result of classifying the state of the PMSM into any one of a plurality of predefined states including at least one of a normal state, a fault state, a demagnetization state, and an unbalanced state as the analysis information. Claim 15 In claim 10, the device is characterized in that the data conversion unit converts the current signal into the spectrogram using a Short-Time Fourier Transform (STFT).

Images