A method, system and equipment for diagnosing rotor bar breakage in a three-phase asynchronous motor

By preprocessing motor signals and applying deep learning models, combined with the fusion of temporal and spatial signal features, the problem of low accuracy in diagnosing broken rotor bars in three-phase asynchronous motors was solved. Real-time monitoring of motor status and fault prediction were achieved, improving diagnostic accuracy and production efficiency.

CN117289127BActive Publication Date: 2026-11-10CHN ENERGY DADU RIVER REPAIR & INSTALLATION CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202311328089.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-10-13
Publication Date
2026-11-10
Estimated Expiration
2043-10-13

AI Technical Summary

Technical Problem

The accuracy of fault diagnosis for broken rotor bars in existing three-phase asynchronous motors is low, resulting in untimely fault diagnosis and affecting production efficiency and safety.

Method used

By collecting raw signal data from the motor stator, data preprocessing is performed to remove noise, fault feature signals are extracted, and a deep learning model is used for fault diagnosis. By combining the feature fusion of time and spatial signals, a graph neural network is used to capture node features and dependencies.

Benefits of technology

It improves the accuracy of rotor bar breakage fault diagnosis, enables real-time monitoring and fault prediction of motors in unknown states, and reduces the frequency and cost of maintenance downtime.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117289127B_ABST
    Figure CN117289127B_ABST
Patent Text Reader

Abstract

This invention relates to a method, system, and device for diagnosing rotor bar breakage faults in a three-phase asynchronous motor. The method includes: acquiring raw signals from the motor stator as first signal data; preprocessing the first signal data to remove noise, resulting in second signal data; extracting fault feature signals based on the second signal data; and performing fault diagnosis based on the extracted fault feature signals. In this method, preprocessing the acquired motor signals removes noise, thereby improving the accuracy of subsequent fault diagnosis. The preprocessed fault diagnosis object undergoes both temporal and spatial signal processing, and the features of both are then fused. This allows for rotor bar breakage fault diagnosis of motors in unknown states, aiding in real-time monitoring of the motor's status. During spatial data processing, a graph neural network is used to capture the features of the nodes themselves and the dependencies between nodes, resulting in more accurate fault diagnosis results.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of deep learning technology, specifically to a method, system, and device for diagnosing rotor bar breakage faults in a three-phase asynchronous motor. Background Technology

[0002] Three-phase asynchronous motors, with their superior performance, are widely used in industrial production, driving various equipment and providing continuous power. When a motor fails, traditional reactive maintenance methods typically require motor shutdown, which means production line interruptions, worker idleness, and potential order delays. Furthermore, downtime incurs additional repair and maintenance costs, requiring more manpower and time to diagnose and repair the fault. Another approach is periodic maintenance, which is prone to over- or under-maintenance. Excessive maintenance can increase costs, while insufficient maintenance may fail to address potential problems promptly. To overcome the shortcomings of traditional maintenance methods, it is necessary to develop motor condition monitoring technology to monitor motor status in real time and provide targeted maintenance recommendations.

[0003] Rotor bar breakage is a common fault in three-phase asynchronous motors, which can severely impact motor performance. When the motor is running, broken rotor bars cause uneven magnetic field distribution, leading to abnormal vibration, noise, and increased temperature. In severe cases, it can even cause equipment damage and personal injury. When a rotor bar breakage fault occurs in a three-phase asynchronous motor, the fault component in the stator current increases; these fault components typically manifest as additional current harmonics.

[0004] A series of problems in the existing fault diagnosis of broken rotor bars result in low accuracy of fault diagnosis.

[0005] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0006] The technical problem to be solved by this application is to provide a method, system and equipment for diagnosing broken rotor bars of a three-phase asynchronous motor, which has the feature of higher accuracy in diagnosing broken rotor bars of a three-phase asynchronous motor.

[0007] In a first aspect, one embodiment provides a method for diagnosing rotor bar breakage faults in a three-phase asynchronous motor, comprising: The original signals from the motor stator are collected as the first signal data, including three-phase current, voltage, vibration, temperature and torque; The first signal data is preprocessed to remove noise and obtain the second signal data. Fault feature signals are extracted based on the second signal data; Fault diagnosis is performed based on the extracted fault feature signals.

[0008] Secondly, one embodiment provides a Samsung asynchronous motor rotor bar breakage fault diagnosis system, comprising: The raw signal acquisition module is used to acquire the raw signals of the motor stator as the first signal data. The raw signals include three-phase current, voltage, vibration, temperature and torque. The data preprocessing module is used to preprocess the first signal data, remove noise from the first signal data, and obtain the second signal data. The fault feature signal extraction module is used to extract fault feature signals based on the second signal data. The fault diagnosis module is used to perform fault diagnosis based on the extracted fault feature signals.

[0009] Thirdly, in one embodiment, a fault diagnosis device for broken rotor bars of a three-phase asynchronous motor is provided, including a memory and a processor. The memory stores a program, which includes a method for diagnosing broken rotor bars of a three-phase asynchronous motor that can be loaded by the processor and processed according to any of the above embodiments.

[0010] Fourthly, in one embodiment, a computer-readable storage medium is provided, the medium storing a program that can be loaded by a processor and executed as the three-phase asynchronous motor rotor bar breakage fault diagnosis method of any of the above embodiments.

[0011] The beneficial effects of this invention are: By preprocessing the acquired motor signals, noise is removed, thereby improving the accuracy of subsequent fault diagnosis. The preprocessed fault diagnosis object undergoes both temporal and spatial signal processing, and the features of both are then fused. This allows for rotor bar breakage fault diagnosis of motors in unknown states, facilitating real-time monitoring of motor status. Specifically, during spatial data processing, a graph neural network is used to capture the characteristics of individual nodes and the dependencies between them, resulting in more accurate fault diagnosis results. Attached Figure Description

[0012] Figure 1 This is a schematic flowchart of a fault diagnosis method according to an embodiment of this application; Figure 2 This application Figure 1 A schematic diagram of the method flow of one embodiment of step S20; Figure 3 This application Figure 1 A schematic diagram of the method flow of one embodiment of step S30; Figure 4 This application Figure 1 A schematic diagram of a method flow for one embodiment of step S40; Figure 5 This application Figure 4 A schematic diagram of the method flow of one embodiment of step S402; Figure 6 This application Figure 5 A schematic diagram of the method flow for one embodiment of step S4022. Detailed Implementation

[0013] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. Similar elements in different embodiments are referred to by associated similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of the present application. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to the present application are not shown or described in the specification. This is to avoid obscuring the core parts of the present application with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0014] Furthermore, the features, operations, or characteristics described in the specification can be combined in any suitable manner to form various embodiments. At the same time, the steps or actions in the method description can be rearranged or adjusted in a manner obvious to those skilled in the art. Therefore, the various orders in the specification and drawings are only for the clear description of a particular embodiment and do not imply a necessary order, unless otherwise stated that a particular order must be followed.

[0015] The serial numbers assigned to components in this document, such as "first" and "second," are used only to distinguish the described objects and have no sequential or technical meaning. The terms "connection" and "linkage" used in this application, unless otherwise specified, include both direct and indirect connections (linkages).

[0016] To facilitate the explanation of the inventive concept of this application, the following is a brief description of the fault diagnosis technology for broken rotor bars of a three-phase asynchronous motor.

[0017] When a rotor bar breakage fault occurs in a three-phase asynchronous motor, the fault component in the stator current will increase. These fault components usually manifest as additional current harmonic components.

[0018] Several issues need to be addressed in conducting fault diagnosis of rotor bar breakage: First, the collected three-phase stator current data of the motor contains noise, which may interfere with the extraction and diagnosis of fault features, so the original current data needs to be denoised; second, after removing the noise, fault features need to be extracted from the current data; finally, after extracting the fault features, the data obtained is input into a deep learning model for fault diagnosis.

[0019] Since this fault diagnosis method relies on a trained fault diagnosis network model, to more clearly illustrate the method, we will first explain the training method of the fault diagnosis network model, which includes: Raw signals from the motor stator were acquired and preprocessed for data denoising. Fault feature signals were then extracted, and a fault diagnosis model was trained based on these extracted signals. The raw data included three-phase current, voltage, vibration, temperature, and torque signals. The extracted fault feature signals were segmented into units equal to one cycle of the motor input voltage, with each segment being an integer multiple of the unit value. These segmented signals were used as training samples for the fault diagnosis network model, constructing a preliminary training network. A loss function and optimizer were selected, and labeled data was used to supervise the model's learning process. During training, the model's hyperparameters were adjusted based on test results on the test set until the model converged or achieved the expected performance.

[0020] The trained model is tested using a test set, and evaluation metrics are calculated based on the test results, including accuracy, precision, and recall. The calculation formulas are as follows:

[0021]

[0022]

[0023] in, It is the number of normal samples that are predicted as normal. It is the number of faulty samples that are predicted to be faulty. This is the number of normal samples that are predicted as faulty. It is the number of faulty samples that are predicted to be normal.

[0024] Adjust the model. Adjust and optimize the model according to the above evaluation metrics until the model meets the performance requirements.

[0025] Make predictions. After optimization, this model can be used for fault diagnosis of broken bars in motor rotors.

[0026] In the training of the aforementioned model, temporal and spatial signals are processed separately, and then their features are fused. This enables rotor bar breakage fault diagnosis for motors in unknown states, facilitating real-time monitoring of the motor's status. The use of graph neural networks for spatial data processing captures the characteristics of individual nodes and the dependencies between them, resulting in more accurate predictions.

[0027] Based on the modules learned through the above training, one embodiment of this application provides a method for diagnosing rotor bar breakage faults in a three-phase asynchronous motor. This method processes the input fault diagnosis object using both temporal and spatial signals, then fuses their features. This allows for rotor bar breakage fault diagnosis of motors in unknown states, aiding in real-time monitoring of the motor's status. Furthermore, using graph neural networks for spatial data processing captures the features of the nodes themselves and the dependencies between nodes, resulting in more accurate predictions. Please refer to... Figure 1 The method includes: Step S10: Acquire the raw signals from the motor stator as the first signal data. These raw signals include three-phase current, voltage, vibration, temperature, and torque.

[0028] In this step, data on the three-phase current, voltage, vibration, temperature, and torque of the motor stator are collected, and the data is preprocessed. The EEMD method is used to remove noise from the data, thereby reducing the impact of noise on fault diagnosis.

[0029] Step S20: Perform data preprocessing on the first signal data to remove noise from the first signal data and obtain the second signal data.

[0030] In one embodiment, please refer to Figure 2 Specific methods include: Step S201: Add white noise. Add the first signal data to the white noise signal to form a perturbation signal.

[0031] First, the original signal needs to be added to the white noise signal to form a perturbation signal. In one embodiment, this includes:

[0032] Where t represents time t, and I(t) represents the first signal data. Represents white noise signal, This indicates a disturbance signal.

[0033] Step S202: Perform EEMD decomposition. Apply EMD to the perturbation signal to obtain a set of intrinsic mode functions.

[0034] In one embodiment, it includes:

[0035] Where N represents the total number of intrinsic mode functions, and i represents the index of the intrinsic mode function, 1≤i≤N. Let i represent the i-th intrinsic mode function.

[0036] The resulting set of intrinsic mode functions includes N intrinsic mode functions.

[0037] Step S203: Remove noise. Calculate the correlation coefficient between each intrinsic mode function and the first signal data.

[0038] In one embodiment, it includes:

[0039] Where M represents the number of time samples for the i-th intrinsic mode function, m represents the index of the time sample for the i-th intrinsic mode function, and 1 ≤ m ≤ M. This represents the value of the i-th intrinsic mode function at the m-th time sample. Let represent the mean of the i-th intrinsic mode function. This represents the value of the m-th time sample of the first signal data. This represents the mean of the first signal data. This represents the correlation coefficient between the i-th intrinsic mode function and the first signal data.

[0040] Based on this, the correlation coefficient of each intrinsic mode function is obtained. According to the magnitude of the correlation coefficient, it can be determined whether a certain intrinsic mode function contains noise. If the correlation coefficient is less than a certain preset threshold or close to 0, it can be considered that the corresponding intrinsic mode function contains noise.

[0041] Step S204, reconstruct the signal. Remove the noisy intrinsic mode functions (IMFs), and reconstruct the signal from the remaining IMFs to obtain the denoised second signal data.

[0042] Based on the obtained correlation coefficient and the preset correlation coefficient threshold, intrinsic mode functions with correlation coefficients less than the preset correlation threshold are removed.

[0043] In one embodiment, the retained intrinsic mode functions are reconstructed to obtain denoised second signal data, including:

[0044] Where Q represents the number of retained intrinsic mode functions, q represents the index of the retained intrinsic mode function, and 1 ≤ q ≤ Q. This represents the q-th retained eigenmode function. This indicates the second signal data.

[0045] Step S30: Extract fault feature signals based on the second signal data.

[0046] While existing techniques can be used to extract fault feature signals, this application provides a novel method for such extraction. Please refer to [link / reference]. Figure 3 ,include: Step S301: Perform Hilbert transform on the second signal data to obtain the Hilbert transformed signal.

[0047] In one embodiment, the denoised original signals (including current, voltage, vibration, temperature, and torque signals) obtained in step S20 are represented as... Perform a Hilbert transform to obtain the Hilbert-transformed signal. :

[0048] Where t represents time t, This represents the time integral variable.

[0049] In step S302, the second signal data is used as the real part and the Hilbert transform signal is used as the imaginary part to construct an analytic signal.

[0050] In one embodiment, an analytic signal is constructed by using the second signal data as the real part and the Hilbert transform signal as the imaginary part, including: use As the real part, the Hilbert transform signal As the imaginary part, construct the analytic signal :

[0051] Where j represents the imaginary unit.

[0052] Step S303: Calculate the modulus of the analytical signal to obtain the fault characteristic signal.

[0053] In one embodiment, calculating the modulus of the analytical signal to obtain a fault characteristic signal includes:

[0054] in, Indicates fault characteristic signals, express The real part, express The imaginary part.

[0055] At this point, the modulus of the analytical signal no longer contains the fundamental component, but instead includes the DC component and some fault characteristic components.

[0056] Step S40: Perform fault diagnosis based on the extracted fault feature signals.

[0057] After step S30, a signal with fault characteristics is obtained. Subsequently, time network models and spatial network models were used for fault diagnosis.

[0058] In one embodiment, please refer to Figure 4 Fault diagnosis based on extracted fault feature signals includes: Step S401: Collect the fault diagnosis object from the fault characteristic signal.

[0059] In one embodiment, at least one unit of the fault characteristic signal is collected as the fault diagnosis object, taking one cycle of the motor input voltage as the unit.

[0060] Step S402: Perform fault diagnosis using the fault diagnosis object as input. Please refer to [link / reference]. Figure 5 ,include: Step S4021: The three-phase current signals in the fault diagnosis object are used to generate a time feature vector through a recurrent neural network.

[0061] The three-phase current signal consists of three current signals. Each current signal needs to be processed through a recurrent neural network to generate a corresponding feature vector. Finally, the overall time feature vector of the three-phase current signal is obtained.

[0062] In one embodiment, for any current signal x at any time step... The processing includes: Using the reset gate to control the discarding of past memories, including:

[0063] in, Indicates the current time step. This indicates the output of the reset gate at the current time step. This represents the Sigmoid function. This represents the weight matrix of the reset gate obtained through model training. This represents the hidden state of the previous time step.

[0064] Update the current candidate memory using update gate control, including:

[0065] in, This indicates the output of the update gate at the current time step. This represents the weight matrix of the update gate obtained through model training.

[0066] Candidate memories are obtained based on the reset gate and the input at the current time step, including:

[0067] in, This represents the candidate memory output for the current time step. This represents the weight matrix of the candidate memories obtained through model training. represents element-wise multiplication, and tanh represents the hyperbolic tangent function.

[0068] Update the hidden state based on the update door, the previous hidden state, and candidate memories, including:

[0069] Here, represents the current hidden state at the current time step.

[0070] The current hidden state is used as the output of the first layer of the recurrent neural network, and the output of each layer of the recurrent neural network is used as the input of the next layer of the recurrent neural network. The output of the last layer of the recurrent neural network is used as the final hidden state. All the final hidden states of the three-phase current signal in the fault diagnosis object are obtained, and the final output is the time feature vector T.

[0071] By processing based on the above-mentioned gating unit, the long-term dependencies in the step signal sequence can be effectively identified, and a time feature vector T can be generated, enabling more accurate fault identification.

[0072] Step S4022: Process the voltage, torque, vibration and temperature signals in the fault diagnosis object, extract time-independent features, construct a graph structure, process the graph structure using a graph neural network, and generate a spatial feature vector.

[0073] After step S4022, the spatial signals (voltage, torque, vibration and temperature signals) are processed to extract time-independent features and obtain spatial feature vectors.

[0074] In one embodiment, please refer to Figure 6 ,include: Step S40221, Feature Vector Extraction. The amplitude spectra of the Fourier transforms of the voltage, torque, and vibration signals in the fault diagnosis object are used as their respective feature vectors. The temperature signal in the fault diagnosis object is divided into multiple overlapping windows, the mean square energy of each window is calculated, and the mean square energy is used as the feature vector of the temperature signal.

[0075] For voltage, torque, and vibration signals, fault information is contained in their frequency domains; therefore, the amplitude spectrum of the Fourier transform is used as the corresponding eigenvector. For temperature absorption signals, short-time energy information is calculated by dividing the signal into multiple overlapping windows and calculating the mean square energy within each window; the result is used as its eigenvector.

[0076] Step S40222, Graph Structure Construction. Each feature vector extracted from the feature vectors is treated as a node, and the connection weights between nodes are calculated to obtain the graph structure.

[0077] A graph structure consists of nodes and connections. We extract each feature vector as a node to obtain all nodes, and then calculate the connection weights between the nodes to obtain the graph structure.

[0078] The calculation of connection weights between nodes can be achieved using existing techniques. This application provides a new calculation method, including:

[0079] in, Represents any two nodes and The connection weights, Represents a node The a-th element in the array, k and Both represent the index of the node, 1 ≤ ≤K, and k≠ K is the number of nodes. Represents a node The a-th element in the array, where A represents the node. The number of elements in the array, where nodes The number of elements and nodes in The number of elements in the two arrays is equal. Represents a node The mean of all elements in the set. Represents a node The mean of all elements in the dataset.

[0080] Step S40223, Graph Neural Network Processing. The graph structure is analyzed using a graph neural network to generate spatial feature vectors.

[0081] In one embodiment, a graph neural network is used to analyze the graph structure and generate spatial feature vectors, including: First, each node The initial features are represented as the initial feature vector, denoted as... .

[0082] Secondly, each node is updated through graph convolution operations. The features include:

[0083] in, Represents a node The feature representation at layer l, l≥1, Represents a node The combination of neighboring nodes, This represents the index of the neighboring node.

[0084] The graph convolution process considers updating the information of the nodes representing features and their neighboring nodes, making the final fault diagnosis results more accurate.

[0085] In one embodiment, the graph convolutional network includes three graph convolutional layers, each of which updates the node feature representation.

[0086] Finally, an attention mechanism is used for feature fusion, which merges the updated feature representations of all nodes in the graph structure into a single feature vector to form a spatial feature vector S.

[0087] In one embodiment, the method includes: calculating the attention weight of each node to other nodes based on the similarity of each node to other nodes; weighting and fusing the feature representations of each node to other nodes based on the calculated attention weights to generate a fused feature vector; and averaging the fused feature vectors of all nodes to obtain the spatial feature vector S.

[0088] In one embodiment, the attention weight between each node and other nodes is calculated based on the similarity between each node and other nodes, including:

[0089] in, Represents a node Attention weights Represents a node Feature representation, Represents a node The feature representations are: Attention represents the attention mechanism, softmax represents the activation function, and MLP represents the multilayer perceptron.

[0090] In one embodiment, the feature representations of each node are weighted and fused with those of other nodes based on the calculated attention weights to generate a fused feature vector, including:

[0091] in, Represents a node The fused feature vector.

[0092] Step S4023: Concatenate the temporal feature vector and the spatial feature vector to obtain the hidden state vector.

[0093] In this step, spatiotemporal data fusion is performed. In one embodiment, it includes:

[0094] in, Represents the hidden state vector. This indicates that two vectors are concatenated.

[0095] Step S4024: Map the hidden state vector to the probability distribution of normal and fault, and obtain the fault diagnosis result based on the probability distribution.

[0096] In one embodiment, the method includes: inputting the hidden state vector into a fully connected layer, and using a sigmoid activation function to map the hidden state to probability distributions for two categories: normal and faulty. In a specific embodiment, the method includes:

[0097]

[0098] in, This represents the weight matrix of the output layer, obtained through model training. The output layer bias term, representing the final fault output, is obtained through model training. This represents the probability distribution of belonging to the normal category. This represents the probability distribution of belonging to a fault category.

[0099] In the aforementioned fault diagnosis process, noise is removed from the acquired motor signals through preprocessing, thereby improving the accuracy of subsequent fault diagnosis. The preprocessed fault diagnosis object undergoes both temporal and spatial signal processing, and the features of both are then fused. This allows for rotor bar breakage fault diagnosis of motors in unknown states, aiding in real-time monitoring of the motor's status. Specifically, during spatial data processing, a graph neural network is used to capture the characteristics of the nodes themselves and the dependencies between nodes, resulting in more accurate fault diagnosis results.

[0100] One embodiment of this application provides a fault diagnosis system for broken rotor bars of a three-phase asynchronous motor, used to diagnose broken rotor bars of a three-phase asynchronous motor in conjunction with the aforementioned fault diagnosis method. The system includes a raw signal acquisition module, a data preprocessing module, a fault feature signal extraction module, and a fault diagnosis module. The raw signal acquisition module acquires raw signals from the motor stator as first signal data, including three-phase current, voltage, vibration, temperature, and torque. The data preprocessing module preprocesses the first signal data to remove noise, obtaining second signal data. The fault feature signal extraction module extracts fault feature signals based on the second signal data. The fault diagnosis module performs fault diagnosis based on the extracted fault feature signals.

[0101] The data preprocessing module includes: a white noise addition unit, used to add the first signal data to a white noise signal to form a perturbation signal; an EEMD decomposition unit, used to apply EMD to the perturbation signal to obtain a set of intrinsic mode functions; a noise removal unit, used to calculate the correlation coefficient between each intrinsic mode function and the first signal data; and a signal reconstruction unit, used to remove the noisy intrinsic mode functions and reconstruct the signal from the remaining intrinsic mode functions to obtain the denoised second signal data.

[0102] The fault feature signal extraction module includes: a Hilbert transform unit for performing a Hilbert transform on the second signal data to obtain a Hilbert transform signal; an analytical signal construction unit for constructing an analytical signal by using the second signal data as the real part and the Hilbert transform signal as the imaginary part; and a fault feature signal acquisition unit for calculating the modulus of the analytical signal to obtain the fault feature signal.

[0103] The fault diagnosis module includes: a fault diagnosis object acquisition unit, used to acquire fault diagnosis objects from the fault characteristic signals; and a fault diagnosis unit, used to perform fault diagnosis by taking the fault diagnosis objects as input.

[0104] The fault diagnosis unit further includes: a time feature vector generation subunit, used to generate time feature vectors from the three-phase current signals in the fault diagnosis object through a recurrent neural network; a spatial feature vector generation subunit, used to process the voltage, torque, vibration, and temperature signals in the fault diagnosis object, extract time-independent features, construct a graph structure, and process the graph structure through a graph neural network to generate spatial feature vectors; a hidden state vector acquisition subunit, used to concatenate the time feature vectors and spatial feature vectors to obtain hidden state vectors; and a fault diagnosis result acquisition subunit, used to map the hidden state vectors to the probability distributions of normal and fault conditions, and obtain the fault diagnosis result based on the probability distributions.

[0105] One embodiment of this application provides a fault diagnosis device for a three-phase asynchronous motor rotor bar breakage, including a memory and a processor. The memory stores a program that can be loaded by the processor and processed by the fault diagnosis method in any of the above embodiments.

[0106] One embodiment of this application provides a computer-readable storage medium storing a program, the stored program including a fault diagnosis method that can be loaded by a processor and processed in any of the above embodiments.

[0107] Those skilled in the art will understand that all or part of the functions of the various methods in the above embodiments can be implemented by hardware or by computer programs. When all or part of the functions in the above embodiments are implemented by computer programs, the program can be stored in a computer-readable storage medium, which may include: read-only memory, random access memory, disk, optical disk, hard disk, etc., and the program is executed by a computer to achieve the above functions. For example, the program can be stored in the memory of a device, and when the program in the memory is executed by the processor, all or part of the above functions can be achieved. In addition, when all or part of the functions in the above embodiments are implemented by computer programs, the program can also be stored in a server, another computer, disk, optical disk, flash drive, or external hard drive, etc., and can be downloaded or copied to the memory of a local device, or the system of the local device can be updated. When the program in the memory is executed by the processor, all or part of the functions in the above embodiments can be achieved.

[0108] The above examples illustrate the present invention only to aid in understanding it and are not intended to limit the scope of the invention. Those skilled in the art can make various simple deductions, modifications, or substitutions based on the principles of this invention.

Claims

1. A method for diagnosing rotor bar breakage faults in a three-phase asynchronous motor, characterized in that, include: The original signals from the motor stator are collected as the first signal data, including three-phase current, voltage, vibration, temperature and torque; The first signal data is preprocessed to remove noise and obtain the second signal data. Fault feature signals are extracted based on the second signal data; Fault diagnosis is performed based on the extracted fault feature signals; The process of preprocessing the first signal data to remove noise and obtain the second signal data includes: Add white noise by adding the first signal data to the white noise signal to form a perturbation signal; Perform EEMD decomposition and apply EMD to the disturbance signal to obtain a set of intrinsic mode functions; Remove noise and calculate the correlation coefficient between each intrinsic mode function and the first signal data; The signal is reconstructed by removing noisy intrinsic mode functions (IMFs), and the remaining IMFs are then used to reconstruct the signal to obtain the denoised second signal data. The step of adding the first signal data to a white noise signal to form a perturbation signal includes: Where t represents time t, and I(t) represents the first signal data. Represents white noise signal, Indicates a disturbance signal; The application of EMD to the disturbance signal yields a set of intrinsic mode functions, including: Where N represents the total number of intrinsic mode functions, and i represents the index of the intrinsic mode function, 1≤i≤N. Represents the i-th intrinsic mode function; The calculation of the correlation coefficient between each intrinsic mode function and the first signal data includes: Where M represents the number of time samples for the i-th intrinsic mode function, m represents the index of the time sample for the i-th intrinsic mode function, 1≤m≤M, IMF i (t) m IMF represents the value of the i-th intrinsic mode function at the m-th time sample. i Let i represent the i-th intrinsic mode function. Let represent the mean of the i-th intrinsic mode function. This represents the value of the m-th time sample of the first signal data. This represents the mean of the first signal data. This represents the correlation coefficient between the i-th intrinsic mode function and the first signal data; The removal of noisy intrinsic mode functions includes: Based on the obtained correlation coefficient and the preset correlation coefficient threshold, the intrinsic mode functions with correlation coefficients less than the preset correlation threshold are removed; The process of reconstructing the retained intrinsic mode functions to obtain the denoised second signal data includes: Where Q represents the number of retained intrinsic mode functions, q represents the index of the retained intrinsic mode function, and 1 ≤ q ≤ Q. This represents the q-th retained eigenmode function. Indicates the second signal data; The method of extracting fault features based on the second signal data includes: The second signal data is subjected to Hilbert transform to obtain the Hilbert transformed signal; The second signal data is used as the real part, and the Hilbert transform signal is used as the imaginary part to construct an analytic signal; Calculate the modulus of the analytical signal to obtain the fault characteristic signal; The process of performing a Hilbert transform on the second signal data to obtain a Hilbert-transformed signal includes: The second signal data is represented as Perform a Hilbert transform to obtain the Hilbert-transformed signal. : Where t represents time t, Represents the time integral variable; The method of constructing an analytic signal by using the second signal data as the real part and the Hilbert transform signal as the imaginary part includes: use As the real part, the Hilbert transform signal As the imaginary part, construct the analytic signal : Where j represents the imaginary part unit; The calculation of the modulus of the analytical signal to obtain the fault characteristic signal includes: in, Indicates fault characteristic signals, express The real part, express The imaginary part; The aforementioned fault diagnosis based on extracted fault feature signals includes: Collect the fault diagnosis object from the fault characteristic signals; Using the fault diagnosis object as input for fault diagnosis includes: The three-phase current signals in the fault diagnosis object are used to generate a time feature vector through a recurrent neural network; The voltage, torque, vibration, and temperature signals in the fault diagnosis object are processed to extract time-independent features, construct a graph structure, and then the graph structure is processed by a graph neural network to generate a spatial feature vector. The time feature vector and the spatial feature vector are concatenated to obtain the hidden state vector. The hidden state vector is mapped to the probability distributions of normal and faulty states, and the fault diagnosis result is obtained based on the probability distributions. The acquisition of the fault diagnosis object from the fault feature signal includes: acquiring at least one unit of the fault feature signal as the fault diagnosis object, taking one cycle of the motor input voltage as a unit. The step of generating a time feature vector from the three-phase current signal in the fault diagnosis object using a multilayer recurrent neural network includes: For any current signal x in the three-phase current signals of the fault diagnosis object, at any time step... : Using the reset gate to control the discarding of past memories, including: in, Indicates the current time step. This indicates the output of the reset gate at the current time step. This represents the Sigmoid function. This represents the weight matrix of the reset gate obtained through model training. This indicates the hidden state of the previous time step; Update the current candidate memory using update gate control, including: in, This indicates the output of the update gate at the current time step. This represents the weight matrix of the update gate obtained through model training; Candidate memories are obtained based on the reset gate and the input at the current time step, including: in, This represents the candidate memory output for the current time step. This represents the weight matrix of the candidate memories obtained through model training. tanh represents element-wise multiplication, and tanh represents the hyperbolic tangent function. Update the hidden state based on the update door, the previous hidden state, and candidate memories, including: Wherein, represents the current hidden state at the current time step; The current hidden state is used as the output of the first layer of the recurrent neural network, and the output of each layer of the recurrent neural network is used as the input of the next layer of the recurrent neural network. The output of the last layer of the recurrent neural network is obtained as the final hidden state. All the final hidden states of the three-phase current signal in the fault diagnosis object are obtained, and the final output is the time feature vector T. The voltage, torque, vibration, and temperature signals from the fault diagnosis object are processed to extract time-independent features, a graph structure is constructed, and the graph structure is processed by a graph neural network to generate spatial feature vectors, including: Feature vector extraction involves using the amplitude spectra of the Fourier transforms of the voltage, torque, and vibration signals in the fault diagnosis object as their respective feature vectors, dividing the temperature signal in the fault diagnosis object into multiple overlapping windows, calculating the mean square energy of each window, and using the mean square energy as the feature vector of the temperature signal. Graph structure construction involves treating each extracted feature vector as a node and calculating the connection weights between nodes to obtain the graph structure. Graph neural network processing utilizes graph neural networks to analyze graph structures and generate spatial feature vectors. The connection weights between the computing nodes include: in, Represents any two nodes and The connection weights, Represents a node The a-th element in the array, k and Both represent the index of the node, 1 ≤ ≤K, and k≠ K is the number of nodes. Represents a node The a-th element in the array, where A represents the node. The number of elements in the array, where nodes The number of elements and nodes in The number of elements in the two arrays is equal. Represents a node The mean of all elements in the set. Represents a node The mean of all elements in the dataset; The method of using graph neural networks to analyze graph structures and generate spatial feature vectors includes: Initialize node characteristics, and set each node The initial features are represented as the initial feature vector, denoted as... ; Graph convolution updates each node through graph convolution operations. The features include: in, Represents a node The feature representation at layer l, l≥1, Represents a node The combination of neighboring nodes, Indicates the index of the neighboring node; Represents the Sigmoid function; Represents a node Feature representation at layer l-1; Feature fusion uses an attention mechanism to fuse the updated feature representations of all nodes in the graph structure into a single feature vector to form a spatial feature vector S. The aforementioned feature fusion using an attention mechanism, which fuses the updated feature representations of all nodes in the graph structure into a single feature vector to form a spatial feature vector S, includes: The attention weights of each node and other nodes are calculated based on the similarity of each node to other nodes. The feature representations of each node are weighted and fused with those of other nodes based on the calculated attention weights to generate a fused feature vector. The average value of the feature vectors after fusion of all nodes is taken, and the sum is used to obtain the spatial feature vector S. The calculation of attention weights between each node and other nodes based on the similarity between each node and other nodes includes: in, Represents a node Attention weights Represents a node Feature representation, Represents a node The feature representations are: Attention represents the attention mechanism, softmax represents the activation function, and MLP represents the multilayer perceptron. The method of weighted fusing the feature representations of each node with those of other nodes based on the calculated attention weights to generate a fused feature vector includes: in, Represents a node The fused feature vector; The time feature vector and the spatial feature vector are concatenated to obtain the hidden state vector, including: in, Represents the hidden state vector. This indicates that two vectors are concatenated, where T is the time feature vector and S is the spatial feature vector; Mapping the hidden state vector to normal and fault probability distributions includes: The hidden state vector is input into a fully connected layer, and the sigmoid activation function is used to map the hidden state to the probability distribution of the normal and fault categories. The process of inputting the hidden state vector into a fully connected layer and using a sigmoid activation function to map the hidden state to probability distributions for normal and faulty categories includes: in, This represents the weight matrix of the output layer, obtained through model training. This represents the bias term of the output layer, obtained through model training. This represents the probability distribution of belonging to the normal category. This represents the probability distribution of belonging to a fault category.

2. A fault diagnosis system for broken rotor bars of a three-phase asynchronous motor, used to execute the fault diagnosis method for broken rotor bars of a three-phase asynchronous motor as described in claim 1, characterized in that, include: The raw signal acquisition module is used to acquire the raw signals of the motor stator as the first signal data. The raw signals include three-phase current, voltage, vibration, temperature and torque. The data preprocessing module is used to preprocess the first signal data, remove noise from the first signal data, and obtain the second signal data. The fault feature signal extraction module is used to extract fault feature signals based on the second signal data. The fault diagnosis module is used to perform fault diagnosis based on the extracted fault feature signals.

3. The three-phase asynchronous motor rotor bar breakage fault diagnosis system as described in claim 2, characterized in that, The data preprocessing module includes: The white noise addition unit is used to add the first signal data to the white noise signal to form a disturbance signal; The EEMD decomposition unit is used to apply EMD to the disturbance signal to obtain a set of intrinsic mode functions; The noise removal unit is used to calculate the correlation coefficient between each intrinsic mode function and the first signal data; The signal reconstruction unit is used to remove noisy intrinsic mode functions and reconstruct the signal from the remaining intrinsic mode functions to obtain the denoised second signal data.

4. The three-phase asynchronous motor rotor bar breakage fault diagnosis system as described in claim 2, characterized in that, The fault feature signal extraction module includes: The Hilbert transform unit is used to perform a Hilbert transform on the second signal data to obtain the Hilbert transformed signal. An analytic signal construction unit is used to construct an analytic signal by taking the second signal data as the real part and the Hilbert transform signal as the imaginary part. The fault feature signal acquisition unit is used to calculate the modulus of the analytical signal to obtain the fault feature signal.

5. The three-phase asynchronous motor rotor bar breakage fault diagnosis system as described in claim 2, characterized in that, The fault diagnosis module includes: The fault diagnosis object acquisition unit is used to acquire the fault diagnosis object in the fault characteristic signal; A fault diagnosis unit, used to perform fault diagnosis by taking the fault diagnosis object as input, includes: The time feature vector generation subunit is used to generate time feature vectors from the three-phase current signals in the fault diagnosis object through a recurrent neural network. The spatial feature vector generation subunit is used to process the voltage, torque, vibration and temperature signals in the fault diagnosis object, extract time-independent features, construct a graph structure, and perform graph neural network processing on the graph structure to generate spatial feature vectors. The hidden state vector acquisition subunit is used to concatenate the time feature vector and the spatial feature vector to obtain the hidden state vector. The fault diagnosis result acquisition subunit is used to map the hidden state vector to the probability distribution of normal and fault, and obtain the fault diagnosis result based on the probability distribution.

6. A fault diagnosis device for broken rotor bars of a three-phase asynchronous motor, characterized in that, It includes a memory and a processor, wherein the memory stores a program, the program including the method for diagnosing broken bars in a three-phase asynchronous motor as described in claim 1, which can be loaded and processed by the processor.

7. A computer-readable storage medium, characterized in that, The medium stores a program that can be loaded and executed by a processor as described in claim 1 for diagnosing a broken bar fault in a three-phase asynchronous motor rotor.

Citation Information

Patent Citations

  • Stacked noise reduction self-coding motor fault diagnosis method based on vibration and current signals

    CN107957551A

  • Methods and Apparatuses for Defect Diagnosis in a Mechanical System

    US20150233792A1