Digital twinning system for diagnosing turn-to-turn short circuit fault of permanent magnet synchronous motor
By constructing a digital twin model of a permanent magnet synchronous motor and combining signal processing and machine learning algorithms, accurate diagnosis and real-time early warning of inter-turn short-circuit faults in permanent magnet synchronous motors were achieved, solving the problem of the lack of a digital twin system for diagnosing inter-turn short-circuit faults in permanent magnet synchronous motors in existing technologies.
Patent Information
- Application Number
- CN202411002296.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-25
- Publication Date
- 2024-11-19
AI Technical Summary
Current technology lacks a digital twin system for diagnosing inter-turn short-circuit faults in permanent magnet synchronous motors.
A digital twin model of a permanent magnet synchronous motor is constructed, combining a signal acquisition layer, a signal processing layer, and a fault diagnosis and classification layer. Fault features are extracted using variational mode decomposition, maximum correlation kurtosis deconvolution, and particle swarm optimization methods, and fault identification is performed using the BP-Adaboost model.
It achieves effective diagnosis of inter-turn short circuit faults in permanent magnet synchronous motors, with an average diagnosis error controlled at 3.6% and an optimal diagnosis error rate of 1.75%. It also has real-time early warning capabilities, improving the accuracy and efficiency of diagnosis.
Smart Images

Figure CN118980926A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of permanent magnet synchronous motor digital twin systems, and in particular to a permanent magnet synchronous motor turn-to-turn short circuit fault diagnosis digital twin system. Background Art
[0002] The deep integration of new-generation information and digital models has attracted great attention from industry and academia to Digital Twin (DT) technology. DT technology is known as dynamic mapping from physical space to virtual space, which can realize closed-loop interaction between people, machines and the environment throughout the product life cycle. The technology aims to synchronize the real-time operating status of physical entities through massive data mirroring, allowing analysis, deduction and control through data-driven models in digital space, thereby achieving more accurate modeling, performance optimization, reasonable decision-making and improved production efficiency.
[0003] Permanent magnet synchronous motors are key components of modern industry and are essential to improving the efficiency and stability of mechanical systems. However, turn-to-turn short circuit faults are a common potential problem during their operation, which may be caused by electromagnetic and thermal stresses, leading to changes in the internal materials of the motor, which in turn affects performance and system safety. By analyzing its current components, the operating status and potential problems of the motor can be revealed.
[0004] At present, some scholars have tried to combine digital twin technology with equipment fault diagnosis. For example, in the literature (Zhang Xi, Liao Yulan, Li Qinyi, et al. Digital twin fault diagnosis of automotive motor bearings under safe driving [J]. Journal of Automotive Safety and Energy Conservation, 2023, 14(02): 232-238.), Zhang Xi et al. combined deep learning with digital twin technology to map motor bearing properties to virtual space, realizing real-time diagnosis of motor bearing faults.
[0005] In the literature (Luo Changliang, Mao Ya, Chen Zuobing, et al. Research on fault warning of firing system driven by fusion of digital twin and neural network [J / OL]. Mechanical Science and Technology, 1-8 [2024-03-26].]), Luo Changliang et al. proposed a fault warning method for firing system that integrates digital twin and ELM neural network, providing new ideas for fault diagnosis and prediction of firing system.
[0006] In the literature (Cao Jinhua, Hong Yingjie, Zhou Jie. Research on fault diagnosis of the swing arm mechanism of a space launch tower based on digital twin [J]. Journal of Ordnance Equipment Engineering, 2024, 45(01): 194-200.), Cao Jinhua et al. proposed a high-precision fault diagnosis method using digital twin technology and UG modeling, which was successfully applied to fault detection and maintenance solution verification of the cable swing arm system of the launch tower.
[0007] In the literature (Jiang Lun, Wang Dajiang, Sun Wenlei, et al. Research on transformer fault diagnosis method based on digital twin [J / OL]. Journal of System Simulation, 1-15 [2024-03-26].), Jiang Lun et al. proposed a power transformer fault diagnosis method combined with digital twin technology. By monitoring the oil and gas content and temperature changes, real-time monitoring of the transformer operating status is achieved, and maintenance efficiency and optimization of maintenance plans are improved.
[0008] However, the relevant research on digital twins and permanent magnet synchronous motor fault diagnosis is still in its infancy, and there is currently no digital twin system for permanent magnet synchronous motor turn-to-turn short-circuit fault diagnosis. Summary of the invention
[0009] The present invention provides a permanent magnet synchronous motor turn-to-turn short-circuit fault diagnosis digital twin system to solve the problem that the prior art lacks a permanent magnet synchronous motor turn-to-turn short-circuit fault diagnosis digital twin system.
[0010] In order to achieve the above object, the technical solution adopted by the present invention is:
[0011] A permanent magnet synchronous motor turn-to-turn short-circuit fault diagnosis digital twin system includes a permanent magnet synchronous motor digital twin model, a signal acquisition layer, a signal processing layer, and a fault diagnosis classification layer; the signal acquisition layer collects three-phase stator current signal data when the motor is running from the permanent magnet synchronous motor digital twin model and sends it to the signal processing layer;
[0012] The signal processing layer first uses a variational mode decomposition method to perform VMD decomposition on the stator current signal data, thereby separating the noise component data and the fault characteristic component data; then the signal processing layer uses a maximum correlation kurtosis deconvolution method to enhance the impact characteristic data in the fault characteristic component data; and the signal processing layer uses a particle swarm optimization method to optimize the fault characteristic component data enhanced by the maximum correlation kurtosis deconvolution method to obtain the optimal fault characteristic data, and the signal processing layer sends the obtained optimal fault characteristic data to the fault diagnosis classification layer;
[0013] The fault diagnosis classification layer adopts the trained BP-Adaboost model as the turn-to-turn short circuit fault diagnosis model, and processes the optimal fault feature data through the turn-to-turn short circuit fault diagnosis model to obtain the turn-to-turn short circuit fault identification and classification result.
[0014] Furthermore, the permanent magnet synchronous motor digital twin model includes a shape module, an attribute module, a behavior module, and a criterion module, wherein the shape module displays the physical characteristics of the motor; the attribute module reflects the damage parameters of each component of the motor under high load or fatigue operation; the behavior module fits the training or test of the turn-to-turn short-circuit fault diagnosis model in space; the criterion module integrates historical data rules, implicit knowledge experience and industry standards. These rules improve themselves over time, giving the digital twin model the ability to judge, evaluate, optimize and predict in real time.
[0015] Furthermore, in the fault diagnosis classification layer, when training the BP-Adaboost model, multiple training results are averaged, and the BP-Adaboost model parameters are adjusted and optimized according to the average result.
[0016] The present invention innovatively combines digital twin technology, signal processing technology and machine learning algorithms to achieve effective diagnosis of inter-turn short-circuit faults in permanent magnet synchronous motors. A digital twin model of a permanent magnet synchronous motor is constructed using digital twin technology, which not only realizes real-time interaction of motor data, but also can provide fault warning. At the same time, by collecting three-phase stator current data during motor operation, combining the particle swarm optimization algorithm PSO to optimize the variational mode decomposition VMD-maximum correlation kurtosis deconvolution MCKD signal analysis method and the BP-Adaboost machine learning algorithm, the data is analyzed online, the running state of the motor is effectively monitored, and the inter-turn short-circuit fault is accurately diagnosed. The average diagnostic error is controlled at about 3.6%, and the optimal diagnostic error rate reaches 1.75%. The present invention can push early warning prompts to users in a timely manner through the digital twin system before and after the fault occurs, which is more targeted and effective than traditional regular maintenance or post-maintenance. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Figure 1 This is a diagram of the digital twin model of the permanent magnet synchronous motor constructed in Example 1 of the present invention.
[0018] Figure 2 It is a process diagram of virtual-reality control in the first embodiment of the present invention.
[0019] Figure 3 It is an equivalent circuit diagram of the B-phase inter-turn short circuit in the first embodiment of the present invention.
[0020] Figure 4 This is a schematic diagram of the signal processing layer in the first embodiment of the present invention.
[0021] Figure 5 It is a data processing process diagram of the BP-Adaboost model in the first embodiment of the present invention.
[0022] Figure 6It is the optimal diagnostic error map in the second embodiment of the present invention, wherein: (a) is the time domain mapped frequency domain error map, (b) is the time domain feature error map, and (c) is the time domain + mapped frequency domain feature error map.
[0023] Figure 7 It is a fault warning effect and equipment management diagram of the digital twin system in the second embodiment of the present invention. DETAILED DESCRIPTION
[0024] The present invention is further described below in conjunction with the accompanying drawings and embodiments.
[0025] Embodiment 1
[0026] This embodiment discloses a permanent magnet synchronous motor turn-to-turn short-circuit fault diagnosis digital twin system, including a permanent magnet synchronous motor digital twin model, a signal acquisition layer, a signal processing layer, and a fault diagnosis classification layer, which are specifically described as follows.
[0027] (1) Permanent Magnet Synchronous Motor Digital Twin Model
[0028] Digital twin technology can accurately convert the detailed properties of real equipment into virtual space, thereby building a mirror entity with high fidelity and confidence corresponding to the real equipment. This technology has shown its unique advantages and value in building a digital twin model of permanent magnet synchronous motors. The construction formula of the digital twin five-dimensional model is as follows:
[0029] M DT =(PE,VE,Ss,DD,CN) (1)
[0030] In formula (1): PE is physical entity; VE is virtual entity (digital twin model); Ss is digital twin service platform; DD is twin data; CN is the connection between components.
[0031] Referring to the above five-dimensional model, the permanent magnet synchronous motor digital twin model (VE) constructed in this embodiment is as follows: Figure 1 As shown, the shape of the motor (G V ), Attributes(P V ), behavior (B V ), criteria (R V ) and other four basic modules, as shown in formula (2):
[0032] VE=(G V ,P V ,B V ,R V ) (2)
[0033] Among them, the shape module can accurately display the size, structure and orientation of the motor and other physical characteristics; the attribute module can reflect the damage parameters of each component of the motor under high load or fatigue operation in real time, providing strong data support for the motor's state monitoring and predictive maintenance; the behavior module can fit the inter-turn short-circuit fault characteristics required for training or testing in space, satisfying the various virtual simulation tests of the motor without interfering with the actual system operation. This can not only obtain more complete data samples for optimizing design solutions and improving system efficiency, but also save a lot of testing costs and time; the criterion module integrates historical data rules, implicit knowledge experience and industry standards. These rules are self-perfected over time, giving the digital twin model the ability to judge, evaluate, optimize and predict in real time. It not only integrates existing knowledge, but also uses machine learning to mine new rules, which can not only guide the operation of physical entities, but also assist in the correction and consistency analysis of digital twins.
[0034] For the permanent magnet synchronous motor digital twin model constructed above, this embodiment also establishes the virtual-real control process of the digital twin model as follows: Figure 2 As shown, it includes the following parts:
[0035] (a) Building a virtual image: The first task is to create a virtual image, which combines the three-dimensional representation of computer graphics with the simulation technology of mathematical models to form a digital twin model that can accurately simulate the appearance, behavior and properties of physical entities.
[0036] (b) Design control strategy: Based on the virtual image, a control strategy is designed through the integration of programming and interface design. This strategy covers multiple components such as sensors, actuators, and control algorithms, aiming to comprehensively monitor and control the status and behavior of the transformer.
[0037] (c) Realize virtual-real interaction: connect the virtual image with the physical entity, which is mainly achieved through devices such as sensors and actuators. Sensors are responsible for capturing the status information of the transformer, while actuators are responsible for regulating the actions and behaviors of the transformer.
[0038] (d) Real-time monitoring and control: With the algorithms and logic in the control strategy, the motor status can be instantly understood and corresponding control operations can be performed according to needs. These operations may include replacing transformer parts, daily maintenance and repairs, etc.
[0039] (e) Feedback optimization: Continuously adjust and optimize the control strategy based on the feedback information from the motor. The feedback information collected by the sensor can be used to adjust the control algorithm and parameters to achieve more accurate and stable control effects.
[0040] At the same time, for the permanent magnet synchronous motor digital twin model established above, this embodiment also establishes a turn-to-turn short-circuit fault circuit model through motor mechanism model analysis. Specifically, taking the B-phase stator winding of the permanent magnet synchronous motor as an example, when a turn-to-turn short-circuit fault occurs in the B-phase stator winding of the permanent magnet synchronous motor, an additional loop consisting of a short-circuit resistor Rf is added to the winding, resulting in the B phase being divided into a healthy part (b1) and a faulty part (b2), as shown in FIG. Figure 3 shown.
[0041] In order to obtain the stator three-phase current, without considering the motor magnetic field saturation, the voltage equation of the permanent magnet synchronous motor under the condition of inter-turn short circuit fault is constructed as shown in formula (3):
[0042]
[0043] In formula (3):
[0044] [u abcf ]=[u a u b u c u f ] T ,u a 、u b 、u c is the three-phase voltage of the winding, u f is the fault winding voltage.
[0045] R a , R b1 , R b2 , R c is the stator winding resistance, R f is the fault winding resistance.
[0046] [i abcf ]=[i a i b i c i f ] T ,i a 、i b 、i c is the stator winding phase current, i f is the stator fault phase current.
[0047] L a , L b1 , L b1 , L c is the three-phase self-inductance of the stator winding, M kj is the inductance between stator windings.
[0048] [eabcf ]=[e a e b e c e f ] T is the back electromotive force matrix, which are the induced electromotive force on the three phases and the fault phase respectively.
[0049] [u 0abcf ]=[u a0 u b0 u co 0] T is the zero-sequence voltage matrix.
[0050] The inter-turn short-circuit ratio of the permanent magnet synchronous motor is μ, as shown in formula (4):
[0051]
[0052] In formula (4), n is the number of short-circuit turns of the coil, and N is the total number of turns.
[0053] Under normal circumstances, the resistance and inductance of the three-phase winding of the motor are equal, that is, the self-inductance and mutual inductance are equal. Based on this feature and the relationship between the short-circuit turns ratio and the inductance, the resistance and inductance matrix can be replaced as shown in equations (5) and (6):
[0054]
[0055] (2) In this embodiment, a signal acquisition layer, a signal processing layer, and a fault diagnosis classification layer are constructed in the digital twin system under the machine learning architecture. By collecting the three-phase stator current signal in the permanent magnet synchronous motor digital twin model as input data, the abnormal changes of the three-phase stator current are captured to identify the inter-turn short circuit fault.
[0056] When a motor has a turn-to-turn short circuit, the short-circuit current will affect the magnetic field distribution inside the motor, thereby generating specific harmonics in the stator current. The frequency of these harmonics is related to the fundamental frequency and the number of pole pairs of the motor. By analyzing these characteristic harmonics in the stator current, it can be determined whether the motor has a turn-to-turn short circuit fault.
[0057] Therefore, in this embodiment, the three-phase stator current signal data of the permanent magnet synchronous motor during operation is collected from the digital twin model of the permanent magnet synchronous motor through the signal acquisition layer and sent to the signal processing layer.
[0058] In the signal processing layer, the variational mode decomposition (VMD)-maximum correlated kurtosis deconvolution (MCKD) method is used to extract fault features from current data, and particle swarm optimization (PSO) is used to optimize and find the best fault feature data under the state. The signal processing layer combines the advantages of three technologies: PSO, VMD and MCKD: PSO is an optimization algorithm that simulates the behavior of natural groups, and it searches for the best parameter combination by iteratively searching the solution space in the space. In signal processing, the PSO algorithm is used to optimize the parameter selection of VMD to maximize the effect of signal denoising and feature extraction.
[0059] Specific as Figure 4 As shown, the signal processing layer first uses the VMD algorithm to decompose the complex stator current signal data into multiple modal components with different frequencies and amplitudes. Through VMD decomposition, the noise cost data and fault feature component data in the stator current signal data can be separated. Then, the MCKD algorithm is used in the signal processing layer to enhance the impact feature data in the fault feature component data to improve the recognition ability of the fault feature. The MCKD will further act on the optimal modal component obtained after VMD decomposition to extract the impact features related to the fault. In addition, the signal processing layer uses PSO to optimize the fault feature component data enhanced by the MCKD method to obtain the optimal fault feature data. Finally, the optimal fault feature data obtained is sent to the fault diagnosis classification layer by the signal processing layer.
[0060] The trained BP-Adaboost model is used as the turn-to-turn short circuit fault diagnosis model in the fault diagnosis classification layer. The data processing process of the BP-Adaboost model is as follows: Figure 5 As shown in the figure, the BP neural network is used as a weak classifier to predict sample output through training, and these weak classifiers are integrated into strong classifiers through the Adaboost algorithm. In this process, each BP neural network weak classifier is given a weight according to its performance in the training process, so as to adjust its contribution in the final model. The BP-Adaboost model combines the nonlinear fitting ability of the BP neural network and the integrated learning advantage of the Adaboost algorithm to improve the accuracy and stability of the model. At the same time, it also has a faster training speed and stronger generalization ability, and can effectively solve complex nonlinear problems.
[0061] In this embodiment, the PSO is used to optimize the VMD-MCKD method in the signal processing layer to decompose and reconstruct the stator current signal, effectively remove noise and interference, extract useful fault feature information, and obtain the optimal fault feature data. Then, the BP-Adaboost model is used to process the optimal fault feature data in the fault diagnosis classification layer, thereby completing accurate fault identification and classification. The BP neural network in the BP-Adaboost model can model complex nonlinear relationships, thereby extracting fault-related features. The AdaBoost algorithm improves the overall classification performance by iteratively training and adjusting the weights of weak classifiers, and ultimately obtains more accurate and reliable classification results.
[0062] Embodiment 2
[0063] This embodiment is an experimental analysis of turn-to-turn short circuit fault diagnosis on the digital twin system described in Embodiment 1.
[0064] The data used in this embodiment is the stator current data collected from the permanent magnet synchronous motor by the Korea Advanced Institute of Science and Technology. Seven inter-turn short circuit fault conditions with different fault degrees and normal conditions are selected to synthesize eight types of fault data. The labels are set to 1 (normal state) and -1 (inter-turn short circuit) respectively. The motor parameter summary and fault labels are shown in Table 1 below.
[0065] Table 1 Summary of motor parameters (a) and fault label (b)
[0066]
[0067] (a)
[0068]
[0069] (b)
[0070] High-dimensional raw signals should not be directly input into the neural network, and feature extraction can reduce the dimension and optimize the input. In order to improve the accuracy and stability of fault diagnosis, this embodiment adopts the method of averaging multiple trainings when training the BP-Adaboost model. Specifically, the fault diagnosis model is first trained using a specific data set, and the model performance is improved by continuously adjusting the model parameters and the optimization algorithm. In order to ensure the reliability of the results, the model is trained multiple times, and each training uses the same data set but different initialization parameters or random seeds. After multiple trainings, the diagnostic results obtained by training are averaged to reduce random errors and obtain more stable results. In order to comprehensively evaluate the reliability and accuracy of the method, three different network input parameters are designed: the first one only uses time domain feature parameters as input; the second one uses the frequency domain feature parameters obtained by time domain mapping; the third one is to use the time domain features and the mapped frequency domain feature parameters as a joint input. According to the methods proposed in Sections 2.1 and 2.2, the first 1,200,000 data under eight working conditions are extracted for fault analysis.
[0071] The average error rate of fault diagnosis is shown in Table 2 below:
[0072] Table 2 Mean error rate of fault diagnosis
[0073]
[0074] As can be seen from Table 2, the diagnosis results are good and the error rate is low when only the time-frequency characteristic parameters are used. The best diagnostic error rate of the data processed by VMD-MCKD optimized by PSO and processed by BP-Adaboost algorithm for fault diagnosis is 0.017 5, and the average error rate of fault diagnosis is 0.035 565, which are 0.002 9 and 0.001 85 lower than the error rate of fault diagnosis using BP neural network, and the diagnosis is more accurate.
[0075] The method of averaging multiple trainings has achieved good results in practical applications. By comparing the results of single training and averaging multiple trainings, it can be found that the latter has obvious advantages in diagnostic accuracy, stability and robustness. Multiple trainings can effectively reduce the risk of overfitting and improve the model's generalization ability for unknown data. At the same time, averaging the results can reduce the influence of random factors, making the diagnostic results more reliable and stable. The optimal diagnostic error is as follows: Figure 6 shown.
[0076] In the digital twin system of this embodiment, by constructing a permanent magnet synchronous motor fault diagnosis prediction model trained by historical data, the generated permanent magnet synchronous motor status data is uploaded in real time, and the BP-Adaboost model is used for fault analysis. The model can process these real-time data and analyze the possible faults or abnormal trends of the motor through complex algorithms and calculations. Once the model predicts that the permanent magnet synchronous motor may fail, it will issue an abnormal warning through the constructed motor digital twin, simulate the corresponding fault situation in the digital twin, and issue a warning to the staff through the panel. Figure 7 The fault warning effect and equipment management of the system are demonstrated.
[0077] The preferred embodiments of the present invention are described in detail above in conjunction with the accompanying drawings. The embodiments described in the present invention are merely descriptions of the preferred embodiments of the present invention, and do not limit the concept and scope of the present invention. The various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction, and such combinations should also be regarded as the contents disclosed in the present disclosure as long as they do not violate the concept of the present invention. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.
[0078] The present invention is not limited to the specific details of the above-mentioned embodiments. Within the technical concept of the present invention and without departing from the design concept of the present invention, various modifications and improvements made to the technical solution of the present invention by technical personnel in this field should fall within the protection scope of the present invention. The technical contents for which protection is sought in the present invention have been fully recorded in the claims.
Claims
1. A digital twin system for diagnosis of inter-turn short-circuit faults in permanent magnet synchronous motors, characterized in that: It includes a permanent magnet synchronous motor digital twin model, as well as a signal acquisition layer, a signal processing layer, and a fault diagnosis classification layer; the signal acquisition layer collects three-phase stator current signal data when the motor is running from the permanent magnet synchronous motor digital twin model and sends it to the signal processing layer; The signal processing layer first uses a variational mode decomposition method to perform VMD decomposition on the stator current signal data, thereby separating the noise component data and the fault characteristic component data; then the signal processing layer uses a maximum correlation kurtosis deconvolution method to enhance the impact characteristic data in the fault characteristic component data; and the signal processing layer uses a particle swarm optimization method to optimize the fault characteristic component data enhanced by the maximum correlation kurtosis deconvolution method to obtain the optimal fault characteristic data, and the signal processing layer sends the obtained optimal fault characteristic data to the fault diagnosis classification layer; The fault diagnosis classification layer adopts the trained BP-Adaboost model as the turn-to-turn short circuit fault diagnosis model, and processes the optimal fault feature data through the turn-to-turn short circuit fault diagnosis model to obtain the turn-to-turn short circuit fault identification and classification result.
2. The permanent magnet synchronous motor turn-to-turn short circuit fault diagnosis digital twin system according to claim 1 is characterized in that: The permanent magnet synchronous motor digital twin model includes a shape module, an attribute module, a behavior module, and a criterion module, wherein the shape module displays the physical characteristics of the motor; The attribute module reflects the damage parameters of each component of the motor under high load or fatigue operation; the behavior module fits the training or test of the turn-to-turn short-circuit fault diagnosis model in space; the criterion module integrates historical data rules, implicit knowledge experience and industry standards. These rules improve themselves over time, giving the digital twin model the ability to judge, evaluate, optimize and predict in real time.
3. The permanent magnet synchronous motor turn-to-turn short circuit fault diagnosis digital twin system according to claim 1 is characterized in that: In the fault diagnosis classification layer, the BP-Adaboost model is trained by averaging multiple training results, and adjusting and optimizing the BP-Adaboost model parameters according to the average result.
Citation Information
Cited By
Open-circuit fault diagnosis method for five-phase motor
CN119881640A
Multi-fault intelligent diagnosis method and system for permanent magnet synchronous motor
CN119986363A
Three-phase permanent magnet synchronous motor turn-to-turn short circuit fault severity and phase diagnosis method
CN120993266A
A stator dynamic inter-turn short circuit fault diagnosis method based on digital twinning
CN122860212A
