A low-voltage motor adaptive comprehensive protection method and system and a storage medium thereof

Abstract

The application discloses a low-voltage motor adaptive comprehensive protection method and system and a storage medium thereof, and comprises the following steps: constructing a four-channel parallel flow processing hardware architecture to obtain current motor operation state information; transmitting the current motor operation state information to a distributed motor group cooperative protection system for cloud cooperative processing; performing feature fusion after the distributed motor group cooperative protection system acquires data of current channels and vibration channels and extracts special features; simultaneously mapping to a high-dimensional kernel space according to a similarity algorithm analysis of a deployment core vector machine (CVM); performing incremental update judgment of multi-dimensional feature quantity dynamic core vector data; and generating global decision processing data based on a local correlation relationship of the solution processing. The application realizes four-channel ADC flow acquisition, multi-domain feature fusion, incremental standardization, CVM online learning and global correlation motor fault diagnosis, realizes full-time domain and multi-dimensional continuous perception of the motor operation state, and improves accuracy.

Description

Technical Field

[0001] This invention relates to the field of motor condition monitoring and protection technology, and in particular to an adaptive integrated protection method, system and storage medium for low-voltage motors. Background Technology

[0002] As a fundamental power execution unit in industrial systems, the reliable operation of low-voltage motors is directly related to the continuity of production lines and the safety of personnel and equipment. Traditional protection devices have widely adopted a microcontroller plus fixed threshold criterion architecture, which has the advantages of maturity and low cost in the identification of typical faults such as overcurrent, short circuit, and stall, providing a basic guarantee for the stable operation of industrial systems. Existing solutions primarily employ relay-type or digital motor protectors using a fixed threshold comparison method. After sampling the current / voltage, the sample undergoes low-pass filtering and RMS value calculation, then is statically compared with preset thresholds, such as a 1.2Ie overload threshold and a 0.85Ue undervoltage threshold, to trigger tripping logic. Some high-end devices incorporate FFT spectrum analysis modules, but these are only used for offline diagnostic report generation and do not participate in real-time protection decisions. The resulting drawbacks are: inability to respond to threshold mismatches caused by operating condition drift, such as a slow increase in rated current due to motor aging; lack of joint criteria for multi-physical quantity coupled faults, such as early signs of bearing wear accompanied by increased vibration and sudden increases in current harmonics; and inability to identify new fault modes not pre-configured in the factory knowledge base, such as specific pulse groups superimposed with insulation degradation characteristics caused by IGBT failure under variable frequency drive, leading to high false negative rates and reliance on manual firmware upgrades for model updates. Summary of the Invention

[0003] In view of the problems existing in the current motor condition monitoring and protection, the present invention is proposed.

[0004] Therefore, one of the objectives of this invention is to provide an adaptive integrated protection method, system, and storage medium for low-voltage motors. The method designs a four-channel ADC module to synchronously acquire current and vibration data of three-phase motors, constructs a distributed motor group collaborative protection system, and improves the recognition accuracy, reduces redundancy, enhances the discriminative expression ability of fault features, and significantly reduces the false alarm rate by using a CVM incremental online learning algorithm and updates after multi-dimensional time-frequency domain feature fusion processing.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: On the one hand, the present invention provides an adaptive comprehensive protection method for low-voltage motors, including constructing a four-channel parallel stream processing hardware architecture to obtain the current motor operating status information, that is, being configured to synchronously collect low-voltage motor data and transmit it to a distributed motor group collaborative protection system for cloud collaborative processing through a communication interface circuit, and performing feedback adjustment through a motor protection circuit for motor fault start-stop processing; The distributed motor group collaborative protection system acquires data and extracts dedicated features from the current channel and vibration channel, performs frequency domain feature calculation and time-frequency domain feature calculation for the current channel and vibration channel, and then performs feature fusion. After time-frequency domain feature fusion of the frequency domain features of the current channel and vibration channel and Z-score data standardization, a dynamic core vector set of multi-dimensional feature quantities is generated. After receiving the dynamic core vector set of multi-dimensional feature quantities after data standardization, the system analyzes the similarity algorithm of the deployed core vector machine (CVM) to generate feature vector values ​​of the current motor operating status. The system matches the corresponding fault category of the motor according to the fault category label of the motor, and updates the confidence weight value of the current fault category of the motor according to the feature vector values. In the similarity algorithm analysis of Core Vector Machine (CVM), incremental data processing is performed. Specifically, after receiving the dynamic core vector set, the dynamic core vector set of multi-dimensional features is mapped to a high-dimensional kernel space, the minimum bounding sphere (MEB) is solved, and the incremental update judgment of the dynamic core vector data of multi-dimensional features is performed. Normal samples are automatically retained after compact aggregation, and faulty samples are automatically deleted after deviating from the sphere boundary. In response to the current fault category of the motor, the fault features of the current fault category of the motor are processed by neural network convolution to obtain the local correlation between the motor fault features and the solution processing. Based on the local correlation of the solution processing, the fault solution strategy of the current fault category of the motor is mined by using an attention mechanism, and then the global decision processing data of the current fault category of the motor is generated. Finally, the abnormal detection report of the motor is generated and output.

[0006] As a preferred embodiment of the present invention, the current channel and vibration channel frequency domain feature calculation, time frequency domain feature calculation and Z-score normalization processing are performed. The current channel performs time domain feature calculation including effective value RMS, peak value PEAK, peak factor CF, and waveform distortion rate THD time domain scalar feature. The frequency domain feature uses 256-point real number FFT to output a single-sided spectrum and calculate the normalized energy ratio of each time domain scalar feature. The time-frequency domain feature calculation uses db4 wavelet packet decomposition to the 3rd layer to extract the energy entropy of 16 sub-frequency bands; The frequency domain analysis of the current channel includes 0-100Hz, 100-500Hz, 500-1kHz, 1-3kHz, and 3-5kHz, while the frequency domain analysis of the vibration channel includes the 10-10kHz band; after Z-score normalization, they are spliced ​​into a 32-dimensional feature vector.

[0007] In a preferred embodiment of the present invention, the dynamic core vector set is represented as follows: Each dynamic core vector Feature vector containing the current motor operating state , , The 32-dimensional feature vector is represented by the fault category label of the motor, which is a set of fault category labels for the motor. ; Based on the dynamic core vector set analysis, local decision-making feature vectors are selected. Specifically, a cosine similarity model is used for similarity judgment, and the resulting vectors are selected as shown in the following formula: ; in, For the first A set of dynamic core vectors and feature vector values; For the first The set of feature vector values ​​for the fault category labels of each motor; For the first The number of terms with numerical values. The similarity value is... It is the cosine value of the feature vector of the dynamic core vector set and the feature vector of the fault category label of the motor.

[0008] In a preferred embodiment of the present invention, the cosine value is mapped onto a preset local decision interval, and the cosine value within the interval is set as the feature vector of the current motor operating state, i.e., the local decision judgment feature; based on the feature vector of the current motor operating state, a threshold judgment for online learning is performed, and if it is determined to be significantly abnormal, online learning is initiated, and the feature vector of the current motor operating state is added to the candidate set; specifically as follows: Set the feature vector of the current motor operating status With any dynamic core vector in the dynamic core vector set The difference threshold of the cosine similarity values ​​is used to determine the matching of the previous dynamic core vector based on the result of the difference threshold. The fault category of the output motor The feature vector of the current motor operating state is added to the dynamic core vector set as a new core vector, and the confidence weights are updated.

[0009] In a preferred embodiment of the present invention, the incremental processing of data during the Core Vector Machine (CVM) similarity algorithm analysis involves transforming the dynamic core vector set into a quadratic programming transformation of the CVM. Specifically, the initial samples are used to train the CVM core vector set to obtain the core set. The center of the minimum bounding sphere MEB in high-dimensional space and radius If no point falls outside the sphere, the algorithm terminates; otherwise, the core set is... Find a new minimum bounding sphere (MEB). Through the above steps, an offline training model can be obtained.

[0010] In a preferred embodiment of the present invention, the four-channel parallel stream processing hardware architecture specifically integrates four independent high-speed ADC modules in an embedded electronic device. The sampling rate of each ADC module is ≥256kSPS, and the effective number of bits (ENOB) is ≥16bit. Each ADC module is connected to the three-phase current of the motor and an IEPE sensor used to acquire the vibration acceleration signal on the motor housing. The output data stream of each ADC module is sliced ​​into 100ms sliding time windows with an overlap rate of 50% between windows.

[0011] In a preferred embodiment of the present invention, the fault category label is divided into electrical, mechanical, and operating condition categories; The electrical categories include overload, phase loss, grounding, minor short circuit in windings, phase loss, current imbalance, and increased harmonic distortion. The mechanical categories include bearing wear, rotor eccentricity, base loosening, and coupling misalignment. The operating condition categories include sudden load changes, speed drift, long-term fatigue degradation, and early latent faults.

[0012] As a preferred embodiment of the present invention, the distributed motor group cooperative protection system further includes adjustments for local optimization of incremental processing data, including offline redundancy removal, online parameter tuning, and kernel cache reuse.

[0013] On one hand, the present invention provides a system for an adaptive integrated protection method for low-voltage motors, the system comprising: The edge-side multi-channel acquisition module is used to build a four-channel parallel stream processing hardware architecture to obtain the current motor operating status information. It is configured to synchronously acquire low-voltage motor data and transmit it to the distributed motor group collaborative protection system for cloud collaborative processing through the communication interface circuit. A distributed motor group collaborative protection system, comprising a feature extraction unit, a feature fusion unit, a fault analysis unit, an incremental processing unit, a decision processing unit, and an optimization unit; The feature extraction unit is used for data acquisition and dedicated feature extraction of the current channel and vibration channel, and performs frequency domain feature calculation and real-time frequency domain feature calculation of the current channel and vibration channel. The feature fusion unit is used to fuse the time-frequency domain features of the current channel and vibration channel and to generate a dynamic core vector set of multidimensional feature quantities after Z-score data normalization. The fault analysis unit is used to receive the dynamic core vector set of multidimensional feature quantities after data standardization, calculate and generate the feature vector value of the current motor operating status according to the deployed similarity algorithm, match the corresponding motor fault category according to the fault category label of the motor, and update the confidence weight value of the current motor fault category according to the feature vector value. The incremental processing unit receives the dynamic core vector set, maps the dynamic core vector set of multidimensional features to a high-dimensional kernel space, solves the minimum bounding sphere (MEB), performs incremental update judgment of the dynamic core vector data of multidimensional features, automatically retains normal samples after compact aggregation, and automatically deletes faulty samples after they deviate from the sphere boundary. The decision processing unit is used to respond to the fault category to which the motor belongs, perform neural network convolution processing on the fault features of the fault category to which the motor belongs, obtain the local correlation between the motor fault features and the solution processing, and, based on the local correlation of the solution processing, use an attention mechanism to mine the fault solution strategy for the fault category to which the motor belongs, generate global decision processing data for the fault category to which the motor belongs, and generate and output the abnormal detection report of the motor. An optimization unit is used for adjusting the local optimization of the incremental processing unit, and includes an offline redundancy removal module, an online parameter tuning module, and a kernel cache reuse module. It also includes an edge-side node motor controller and a motor protection circuit for motor fault start-stop processing, wherein the edge-side node motor controller is connected to the edge-side multi-channel acquisition module and the motor protection circuit.

[0014] On one hand, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of a low-voltage motor adaptive integrated protection method.

[0015] The beneficial effects of this invention are as follows: This invention designs a four-channel ADC module to synchronously acquire current and vibration data of a three-phase motor, constructs a distributed motor group collaborative protection system, and utilizes multi-dimensional time-frequency domain feature fusion processing and CVM incremental online learning algorithm and updates to achieve full-time domain, multi-dimensional continuous perception of motor operating status. This improves the accuracy of identifying common faults such as overload and phase loss, while enhancing the discriminative expression capability of fault features. It also improves the separation degree of classification boundaries of similar fault modes by the CVM engine, significantly reduces the false alarm rate, and reduces redundancy, perfectly matching the low latency, small storage, and nonlinear modeling requirements of the high-speed data stream of the four-channel ADC. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a schematic diagram of the adaptive integrated protection method for low-voltage motors according to the present invention; Figure 2 This is a flowchart of the low-voltage motor adaptive integrated protection method of the present invention; Figure 3 This is a modular schematic diagram of the low-voltage motor adaptive integrated protection system of the present invention; The diagram is labeled as follows: 10. Edge-side multi-channel acquisition module; 20. Distributed motor group collaborative protection system; 101. Edge-side node motor controller; 102. Motor protection circuit connection; 201. Feature extraction unit; 202. Feature fusion unit; 203. Fault analysis unit; 204. Incremental processing unit; 205. Decision processing unit; 206. Optimization unit. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.

[0018] Example 1 Reference Figure 1 and 2 As one embodiment of the present invention, this embodiment provides an adaptive integrated protection method for low-voltage motors, as follows: A four-channel parallel stream processing hardware architecture is constructed to obtain the current motor operating status information. It is configured to synchronously collect low-voltage motor data and transmit it to the distributed motor group collaborative protection system for cloud collaborative processing through the communication interface circuit, and to the motor protection circuit for feedback adjustment through motor fault start-stop processing. The distributed motor group collaborative protection system acquires data and extracts dedicated features from the current and vibration channels. It then performs frequency domain and time-frequency domain feature calculations for both channels, followed by feature fusion. After time-frequency domain feature fusion and Z-score data standardization, a dynamic core vector set of multidimensional features is generated. Upon receiving this standardized dynamic core vector set, a similarity algorithm based on a deployed Core Vector Machine (CVM) is used to generate feature vector values ​​for the current motor's operating status. The system matches the motor's fault category label to the corresponding fault category and updates the confidence weight value of the current fault category based on the feature vector values. Specifically, during the CVM similarity algorithm analysis, incremental data processing is performed. After receiving the dynamic core vector set, it maps the multidimensional feature set to a high-dimensional kernel space, solves the minimum bounding sphere (MEB), and performs incremental updates of the dynamic core vector data. Normal samples are automatically retained after compact aggregation, while fault samples deviating from the sphere boundary are automatically deleted. In response to the current fault category of the motor, the fault features of the current fault category are processed by neural network convolution to obtain the local correlation between the motor fault features and the solution processing. Based on the local correlation of the solution processing, the fault solution strategy of the current fault category is mined by using an attention mechanism, and then the global decision processing data of the current fault category of the motor is generated. Finally, the abnormal detection report of the motor is generated and output.

[0019] This embodiment designs a four-channel ADC module to synchronously acquire current and vibration data of a three-phase motor, constructs a distributed motor group collaborative protection system, and uses multi-dimensional time-frequency domain feature fusion processing and CVM incremental online learning algorithm and update to continuously evaluate the motor health status of the low-voltage motor, improve the adjustment of adaptive local optimization, reduce redundancy, and perfectly match the low latency, small storage and nonlinear modeling requirements of the four-channel ADC high-speed data stream.

[0020] In this embodiment, the frequency domain feature calculation specifically involves extracting the normalized energy percentages of five frequency bands within the 0–5kHz range (0-100Hz, 100-500Hz, 500-1kHz, 1-3kHz, and 3-5kHz) after a 256-point FFT. It is important to emphasize that the frequency domain feature calculation uses a 256-point real-valued FFT, outputting a single-sided spectrum, and calculating the normalized energy percentage of each sub-band. First, the sampling parameters are standardized, including the sampling frequency. Sampling frame length The original sequence of a single frame is The algorithms for the current channel and the vibration channel are completely identical. Taking the time-domain features as an example, the time-domain features are divided into four time-domain scalar features: RMS, PEAK, CF, and THD. The root mean square (RMS) value is expressed as: ; Peak PEAK is represented as ; The peak factor CF is expressed as the quotient of the peak value PEAK and the root mean square effective value RMS. Waveform distortion rate (THD) is defined as the total effective value of harmonics or the effective value of the fundamental frequency. An FFT is performed on a 256-point sequence to separate the fundamental frequency. Harmonics The fundamental effective value is The total effective value of the fundamental frequency is That is, the waveform distortion rate (THD) is... ; In the time-frequency domain feature calculation, db4 wavelet packet decomposition is used to the third level to extract the energy entropy of 16 sub-bands. First, the parameters of the three levels are decomposed, then the energy of each sub-band is calculated. After processing the wavelet packet energy entropy, 16 independent energy entropies for each sub-band are output. Further, this technical solution uses fixed-point quantization of pre-stored db4 decomposition / reconstruction filter coefficients in the 3-level db4 wavelet packet decomposition. The 3-level decomposition is serialized and pipelined, outputting 16 sub-band coefficients. Energy and entropy calculations use fixed-point arithmetic to avoid floating-point overhead.

[0021] The vibration channel executes the same algorithm set, but the frequency domain analysis focuses on the 10–10kHz frequency band. All features are Z-score normalized and then concatenated into a 32-dimensional feature vector. Each channel retains a portion of the features: time domain 4 plus frequency domain 5 plus wavelet entropy 7, i.e., 16 dimensions per channel, and the two channels are concatenated into 32 dimensions. The technical solution of this embodiment adopts analysis after multi-domain feature fusion processing, which can simultaneously identify the sudden increase signal of current harmonics and vibration data information, thereby effectively and accurately identifying the degree of vibration wear of motor bearings. This avoids bearing wear caused by vibration aggravation accompanied by sudden increase in current harmonics, and other operating condition-related faults, including load changes, speed drift, long-term fatigue degradation, and early latent faults.

[0022] Furthermore, the four-channel parallel stream processing hardware architecture of this embodiment specifically integrates four independent high-speed ADC modules in the embedded electronic device. The sampling rate of the ADC module is ≥256kSPS, the effective number of bits ENOB is ≥16bit, and each ADC module is connected to the three-phase current of the motor and the IEPE sensor used to acquire the vibration acceleration signal on the motor housing. The output data stream of each ADC module is sliced ​​according to a 100ms sliding time window, and the overlap rate between windows is 50%.

[0023] Specifically, in this embodiment, a dual feature vector stream is input into an incremental core vector machine for online learning and classification: this engine is deployed on a dual-core processor, and a dynamic core vector set is maintained in memory. Each dynamic core vector Feature vector containing the current motor operating state , , Represents a 32-dimensional feature vector; the fault category label of the motor is the fault category label of the motor, represented as a set. ; In this embodiment, it should be noted that the fault category labels are divided into electrical, mechanical, and operating condition categories. The electrical category includes overload, phase loss, grounding, minor short circuit in windings, phase loss, current imbalance, and increased harmonic distortion. The mechanical category includes bearing wear, rotor eccentricity, base looseness, and coupling misalignment. The operating condition category includes sudden load changes, speed drift, long-term fatigue degradation, and early latent faults.

[0024] It is important to emphasize in this embodiment that the local decision-making feature vectors are selected based on the dynamic core vector set analysis. Specifically, the cosine similarity model is used to perform similarity judgment and then the vectors are selected, as shown in the following formula: in, For the first A set of dynamic core vectors and feature vector values; For the first The set of feature vector values ​​for the fault category labels of each motor; For the first The number of terms with numerical values. The similarity value is... It is the cosine value of the feature vector of the dynamic core vector set and the feature vector of the fault category label of the motor.

[0025] In this embodiment, the cosine value is mapped onto a preset local decision interval. The cosine value within the interval is set as the feature vector of the current motor operating state, i.e., the local decision judgment feature. Based on the feature vector of the current motor operating state, a threshold judgment is performed for online learning. If it is determined to be significantly abnormal, online learning is initiated, and the feature vector of the current motor operating state is added to the candidate set. Specifically, as follows: Set the feature vector of the current motor operating status With any dynamic core vector in the dynamic core vector set The difference threshold of the cosine similarity values ​​is used to determine the matching of the previous dynamic core vector. The fault category of the output motor The feature vector of the current motor operating state is added to the dynamic core vector set as a new core vector, and the confidence weights are updated. The confidence weights are between 0.1 and 1. The newly added feature vectors of the current motor operating state are then processed. The engine executes the process, calculating the Euclidean distance according to the Euclidean distance formula. If the calculated minimum distance value is less than or equal to a threshold (set to 0.35), then the most recent dynamic core vector is matched. Output the fault category label and update the confidence weight; if the calculated minimum distance value is less than or equal to the threshold of 0.72, it is judged as a significant anomaly and online learning is started; when the threshold of online learning is met for three consecutive windows, and the cosine similarity value between the feature vector and any dynamic core vector in the dynamic core vector set is less than 0.4, the feature vector of the current motor operating state is added to the dynamic core vector set as a new core vector, and the fault category label set of the motor is initialized to the "unknown anomaly" state.

[0026] Simultaneously, the implementation also includes incremental updates to newly incoming feature vectors. In this embodiment, the Core Vector Machine (CVM) similarity algorithm performs incremental data processing during analysis, transforming the dynamic core vector set into a quadratic programming transformation of the CVM. Specifically, the initial samples are used to train the CVM core vector set, resulting in the core set. The center of the minimum bounding sphere MEB in high-dimensional space and radius If no point falls outside the sphere, the algorithm terminates; otherwise, the core set is... Find a new minimum bounding sphere (MEB). Through the above steps, an offline training model can be obtained.

[0027] In this embodiment, the distributed motor group cooperative protection system further includes adjustments for local optimization of incremental processing data. These optimization adjustments include offline redundancy removal, online parameter tuning, and kernel cache reuse. Offline redundancy removal involves calculating the distance from the sample to the MEB center during iteration, permanently deleting samples that will not become supports, thus accelerating MEB solving. Online parameter tuning involves updating the classifier using only the old core set and new misclassified samples. Kernel cache reuse involves storing only the core set portion of the historical kernel matrix Gram, and calculating only the kernel values ​​of new samples and the core set during incremental processing.

[0028] The technical solution in this embodiment employs multi-domain feature fusion analysis, which can simultaneously identify sudden increases in current harmonic signals and vibration data information. This allows for effective and accurate identification of the vibration and wear degree of the motor bearing, avoiding bearing wear caused by increased vibration accompanied by sudden increases in current harmonics. The unique advantages of this solution are: Adaptation to progressive faults: Incremental updates adapt to the slow degradation of the motor, avoiding misjudgment of drift under normal operating conditions; Sensitivity to early faults: Weak changes in wavelet entropy and frequency band energy can be accurately captured by the MEB boundary, outperforming traditional threshold methods.

[0029] Example 2 Reference Figure 3 This embodiment provides a system for an adaptive integrated protection method for low-voltage motors. The system includes: The edge-side multi-channel acquisition module 10 is used to construct a four-channel parallel stream processing hardware architecture to obtain the current motor operating status information. It is configured to synchronously acquire low-voltage motor data and transmit it to the distributed motor group collaborative protection system for cloud collaborative processing through the communication interface circuit. The distributed motor group collaborative protection system 20 includes a feature extraction unit 201, a feature fusion unit 202, a fault analysis unit 203, an incremental processing unit 204, a decision processing unit 205, and an optimization unit 206. The feature extraction unit 201 is used for data acquisition and dedicated feature extraction of the current channel and vibration channel, and performs frequency domain feature calculation and real-time frequency domain feature calculation of the current channel and vibration channel. The feature fusion unit 202 is used to fuse the time-frequency domain features of the current channel and the vibration channel and to generate a dynamic core vector set of multidimensional feature quantities after Z-score data standardization. The fault analysis unit 203 receives the dynamic core vector set of multidimensional feature quantities after data standardization, calculates the feature vector value of the current motor operating status according to the deployed similarity algorithm, matches the corresponding motor fault category according to the fault category label of the motor, and updates the confidence weight value of the current motor fault category according to the feature vector value. The incremental processing unit 204 is used to receive the dynamic core vector set, map the dynamic core vector set of multidimensional features to a high-dimensional kernel space, solve the minimum bounding sphere (MEB), and make incremental update judgments on the dynamic core vector data of multidimensional features. Normal samples are automatically retained after compact aggregation, and faulty samples are automatically deleted after deviating from the sphere boundary. The decision processing unit 205 is used to respond to the fault category to which the current motor belongs, perform neural network convolution processing on the fault features of the fault category to which the current motor belongs, obtain the local correlation between the motor fault features and the solution processing, and based on the local correlation of the solution processing, use the attention mechanism to mine the fault solution strategy of the fault category to which the current motor belongs, generate global decision processing data of the fault category to which the current motor belongs, and generate an abnormality detection report of the motor and output it. The optimization unit 206 is used to adjust the local optimization of the incremental processing unit 204, and includes an offline redundancy removal module, an online parameter tuning module, and a kernel cache reuse module. It also includes an edge-side node motor controller 101 and a motor protection circuit 102 for motor fault start-stop processing. The edge-side node motor controller 101 is connected to the edge-side multi-channel acquisition module 10 and the motor protection circuit 102.

[0030] The system of the present invention adopts an embodiment that realizes a fully closed-loop embedded intelligent system for motor fault diagnosis, which includes four-channel ADC stream acquisition, multi-domain feature fusion, incremental standardization, CVM online learning, and global correlation. It solves the pain points of traditional motor diagnosis, such as inaccurate fixed thresholds, inability to update models, misjudgment due to operating condition drift, and missed detection of early faults.

[0031] Furthermore, the system of the low-voltage motor adaptive integrated protection method of the present invention can be installed in an electronic device. According to the scheme in Embodiment 1, the electronic device stores a computer program on a computer-readable storage medium according to the functions to be implemented. When the computer program is executed by the processor, it implements the steps of the low-voltage motor adaptive integrated protection method.

[0032] In summary, this invention designs a four-channel ADC module to synchronously acquire current and vibration data of a three-phase motor, constructing a distributed motor group collaborative protection system. After multi-dimensional time-frequency domain feature fusion processing, and using the CVM incremental online learning algorithm and updates, it achieves full-time-domain, multi-dimensional continuous perception of the motor's operating status. This improves the accuracy of identifying common faults such as overload and phase loss, while enhancing the discriminative representation of fault features. It also improves the separation of classification boundaries for similar fault modes by the CVM engine, significantly reducing the false alarm rate and redundancy. This perfectly matches the low latency, small storage, and nonlinear modeling requirements of the four-channel ADC's high-speed data stream.

[0033] Non-volatile memory may include read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory may include random access memory (RAM), which serves as an external cache. By way of example, but not limitation, many forms of RAM are available. For example, static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM).

[0034] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another.

[0035] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0036] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.

[0037] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process. Furthermore, the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functionality involved.

[0038] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus or device (such as a computer-based system, a processor-included system or other system that can fetch and execute instructions from, an instruction execution system, apparatus or device).

[0039] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. All or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware, the program being stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiments.

[0040] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in this application, and these should all be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for adaptive integrated protection of low-voltage motors, characterized in that, include: A four-channel parallel stream processing hardware architecture is constructed to obtain the current motor operating status information. It is configured to synchronously collect low-voltage motor data and transmit it to the distributed motor group collaborative protection system for cloud collaborative processing through the communication interface circuit, and to the motor protection circuit for feedback adjustment through motor fault start-stop processing. The distributed motor group collaborative protection system acquires data and extracts dedicated features from the current channel and vibration channel, performs frequency domain feature calculation and time-frequency domain feature calculation for the current channel and vibration channel, and then performs feature fusion. After time-frequency domain feature fusion of the frequency domain features of the current channel and vibration channel and Z-score data standardization, a dynamic core vector set of multi-dimensional feature quantities is generated. After receiving the dynamic core vector set of multi-dimensional feature quantities after data standardization, the system analyzes the similarity algorithm of the deployed core vector machine (CVM) to generate feature vector values ​​of the current motor operating status. The system matches the corresponding fault category of the motor according to the fault category label of the motor, and updates the confidence weight value of the current fault category of the motor according to the feature vector values. In the similarity algorithm analysis of Core Vector Machine (CVM), incremental data processing is performed. Specifically, after receiving the dynamic core vector set, the dynamic core vector set of multi-dimensional features is mapped to a high-dimensional kernel space, the minimum bounding sphere (MEB) is solved, and the incremental update judgment of the dynamic core vector data of multi-dimensional features is performed. Normal samples are automatically retained after compact aggregation, and faulty samples are automatically deleted after deviating from the sphere boundary. In response to the current fault category of the motor, the fault features of the current fault category of the motor are processed by neural network convolution to obtain the local correlation between the motor fault features and the solution processing. Based on the local correlation of the solution processing, the fault solution strategy of the current fault category of the motor is mined by using an attention mechanism, and then the global decision processing data of the current fault category of the motor is generated. Finally, the abnormal detection report of the motor is generated and output.

2. The adaptive integrated protection method for low-voltage motors as described in claim 1, characterized in that, The current channel and vibration channel frequency domain feature calculation, time frequency domain feature calculation and Z-score normalization processing are described. The current channel performs time domain feature calculation including effective value RMS, peak value PEAK, peak factor CF, and waveform distortion rate THD time domain scalar feature. The frequency domain feature uses 256-point real number FFT to output a single-sided spectrum and calculate the normalized energy ratio of each time domain scalar feature. The time-frequency domain feature calculation uses db4 wavelet packet decomposition to the 3rd layer to extract the energy entropy of 16 sub-frequency bands; The frequency domain analysis of the current channel includes 0-100Hz, 100-500Hz, 500-1kHz, 1-3kHz, and 3-5kHz, while the frequency domain analysis of the vibration channel includes the 10-10kHz band; after Z-score normalization, they are spliced ​​into a 32-dimensional feature vector.

3. The adaptive integrated protection method for low-voltage motors as described in claim 1, characterized in that, The dynamic core vector set is represented as: Each dynamic core vector Feature vector containing the current motor operating state , , The 32-dimensional feature vector is represented by the fault category label of the motor, which is a set of fault category labels for the motor. ; Based on the dynamic core vector set analysis, local decision-making feature vectors are selected. Specifically, a cosine similarity model is used for similarity judgment, and the resulting vectors are selected as shown in the following formula: ; in, For the first A set of dynamic core vectors and feature vector values; For the first The set of feature vector values ​​for the fault category labels of each motor; For the first The number of terms with numerical values. The similarity value is... It is the cosine value of the feature vector of the dynamic core vector set and the feature vector of the fault category label of the motor.

4. The adaptive integrated protection method for low-voltage motors as described in claim 3, characterized in that, The cosine value is mapped onto a preset local decision interval. The cosine value within the interval is set as the feature vector of the current motor operating state, i.e., the local decision judgment feature. Based on the feature vector of the current motor operating state, an online learning threshold judgment is performed. If it is determined to be significantly abnormal, online learning is initiated, and the feature vector of the current motor operating state is added to the candidate set. Specifically, as follows: Set the feature vector of the current motor operating status With any dynamic core vector in the dynamic core vector set The difference threshold of the cosine similarity values ​​is used to determine the matching of the previous dynamic core vector based on the result of the difference threshold. The fault category of the output motor The feature vector of the current motor operating state is added to the dynamic core vector set as a new core vector, and the confidence weights are updated.

5. The adaptive integrated protection method for low-voltage motors as described in claim 1, characterized in that, The Core Vector Machine (CVM) similarity algorithm performs incremental data processing during analysis. This involves transforming the dynamic core vector set into a quadratic programming model of the CVM. Specifically, the initial samples are used to train the CVM core vector set, resulting in the core set. The center of the minimum bounding sphere MEB in high-dimensional space and radius If no point falls outside the sphere, the algorithm terminates; otherwise, the core set is... Find a new minimum bounding sphere (MEB). Through the above steps, an offline training model can be obtained.

6. The adaptive integrated protection method for low-voltage motors as described in claim 1, characterized in that, The four-channel parallel stream processing hardware architecture specifically integrates four independent high-speed ADC modules in an embedded electronic device. The sampling rate of each ADC module is ≥256kSPS, and the effective number of bits (ENOB) is ≥16bit. Each ADC module is connected to the three-phase current of the motor and an IEPE sensor used to acquire the vibration acceleration signal on the motor housing. The output data stream of each ADC module is sliced ​​into 100ms sliding time windows with an overlap rate of 50% between windows.

7. The adaptive integrated protection method for low-voltage motors as described in claim 1, characterized in that, The fault category labels are divided into electrical, mechanical, and operating condition categories; The electrical categories include overload, phase loss, grounding, minor short circuit in windings, phase loss, current imbalance, and increased harmonic distortion. The mechanical categories include bearing wear, rotor eccentricity, base loosening, and coupling misalignment. The operating condition categories include sudden load changes, speed drift, long-term fatigue degradation, and early latent faults.

8. The adaptive integrated protection method for low-voltage motors as described in claim 1, characterized in that, The distributed motor group collaborative protection system also includes adjustments for local optimization of incremental processing data, including offline redundancy removal, online parameter tuning, and kernel cache reuse.

9. A system applied to the adaptive integrated protection method for low-voltage motors as described in claim 8, characterized in that, The system includes: The edge-side multi-channel acquisition module is used to build a four-channel parallel stream processing hardware architecture to obtain the current motor operating status information. It is configured to synchronously acquire low-voltage motor data and transmit it to the distributed motor group collaborative protection system for cloud collaborative processing through the communication interface circuit. A distributed motor group collaborative protection system, comprising a feature extraction unit, a feature fusion unit, a fault analysis unit, an incremental processing unit, a decision processing unit, and an optimization unit; The feature extraction unit is used for data acquisition and dedicated feature extraction of the current channel and vibration channel, and performs frequency domain feature calculation and real-time frequency domain feature calculation of the current channel and vibration channel. The feature fusion unit is used to fuse the time-frequency domain features of the current channel and vibration channel and to generate a dynamic core vector set of multidimensional feature quantities after Z-score data normalization. The fault analysis unit is used to receive the dynamic core vector set of multidimensional feature quantities after data standardization, calculate and generate the feature vector value of the current motor operating status according to the deployed similarity algorithm, match the corresponding motor fault category according to the fault category label of the motor, and update the confidence weight value of the current motor fault category according to the feature vector value. The incremental processing unit receives the dynamic core vector set, maps the dynamic core vector set of multidimensional features to a high-dimensional kernel space, solves the minimum bounding sphere (MEB), performs incremental update judgment of the dynamic core vector data of multidimensional features, automatically retains normal samples after compact aggregation, and automatically deletes faulty samples after they deviate from the sphere boundary. The decision processing unit is used to respond to the fault category to which the motor belongs, perform neural network convolution processing on the fault features of the fault category to which the motor belongs, obtain the local correlation between the motor fault features and the solution processing, and, based on the local correlation of the solution processing, use an attention mechanism to mine the fault solution strategy for the fault category to which the motor belongs, generate global decision processing data for the fault category to which the motor belongs, and generate and output the abnormal detection report of the motor. An optimization unit is used for adjusting the local optimization of the incremental processing unit, and includes an offline redundancy removal module, an online parameter tuning module, and a kernel cache reuse module. It also includes an edge-side node motor controller and a motor protection circuit for motor fault start-stop processing, wherein the edge-side node motor controller is connected to the edge-side multi-channel acquisition module and the motor protection circuit.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the low-voltage motor adaptive integrated protection method as described in any one of claims 1-8.

Images