Fault diagnosis method for oil-immersed core reactor based on multi-physical field coupling and feature map

Abstract

The oil-immersed core reactor fault diagnosis method based on multi-physical field coupling and feature map includes: constructing a multi-physical field coupling calculation model of the core reactor to obtain the simulation results of the electromagnetic-structure force field and the sound field of the core reactor; establishing a simulation model of the core reactor under typical defects to obtain the simulation results of the electromagnetic-structure force field and the sound field of the core reactor under different working conditions; analyzing the influence law of different defects on the vibration and noise of the core reactor; according to the vibration and noise distribution law, obtaining the best vibration and noise measuring point distribution, and extracting the vibration and noise signals of the core reactor under different working conditions; converting the extracted vibration and noise signals under different working conditions into gram angle and field characteristic map; inputting the gram angle and field characteristic map into AlexNet for transfer learning to realize fault diagnosis. The oil-immersed core reactor fault diagnosis method can more comprehensively and accurately identify the faults of the core reactor, and improve its operation safety and reliability.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology for iron core reactors, and specifically to a fault diagnosis method for oil-immersed iron core reactors based on multi-physics coupling and feature maps. Background Technology

[0002] As a crucial piece of equipment in power transmission and transformation systems, the safe and reliable operation of core reactors is paramount, necessitating timely fault diagnosis. Traditional fault diagnosis methods for oil-immersed core reactors include oil chromatography, winding DC resistance testing, and partial discharge detection. These methods typically rely on periodic inspections and tests, failing to provide real-time monitoring of the winding status. For instance, oil chromatography infers internal faults by analyzing dissolved gases in the reactor oil. While this method can reflect internal fault conditions, particularly high-temperature, discharge, and partial breakdown faults, and is simple to operate, it only reveals indirect symptoms and cannot precisely pinpoint the specific location of winding faults. It is also insensitive to small-scale, localized winding insulation damage. Winding DC resistance testing assesses winding health by measuring the DC resistance of each winding, but it can only detect static electrical faults, not dynamic ones. Furthermore, resistance changes may be slow, making early detection difficult. Therefore, it can only detect more severe faults and cannot accurately diagnose minor winding damage or early insulation problems. Partial discharge detection methods determine the insulation status of windings by detecting partial discharge signals (such as ultrasound, electrical pulses, etc.). For small, localized discharge phenomena, this method may require high-precision instruments and equipment, which increases equipment and maintenance costs. In addition, the discharge signals are easily interfered with by external noise, which may lead to misdiagnosis or missed diagnosis. For more serious faults, the discharge may have already occurred and is difficult to detect in time.

[0003] Therefore, traditional fault diagnosis methods for iron core reactors often cannot monitor the status of iron core reactors in real time, comprehensively and accurately, making it difficult to detect minor or complex faults in the early stages. Furthermore, some methods can only detect a certain type of fault, requiring the combination of other means for comprehensive diagnosis. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a fault diagnosis method for oil-immersed iron-core reactors based on multi-physics coupling and feature mapping. Through multi-physics coupling modeling, feature extraction, data analysis, and fault prediction, the method can more comprehensively and accurately identify faults in iron-core reactors, thereby improving their operational safety and reliability.

[0005] The technical solution adopted in this invention is as follows:

[0006] A fault diagnosis method for oil-immersed iron-core reactors based on multi-physics coupling and characteristic spectrum includes the following steps:

[0007] Step 1: Construct a multiphysics coupling calculation model of the iron-core reactor to obtain the simulation results of the electromagnetic-structural force field and acoustic field of the iron-core reactor;

[0008] Step 2: Establish a simulation model of the iron-core reactor under typical defects, including: iron core loosening, winding deformation, winding loosening and inter-turn short circuit, etc., and obtain the simulation results of the electromagnetic-structural force field and acoustic field of the iron-core reactor under different operating conditions.

[0009] Step 3: Based on the simulation results of Step 2, analyze the influence of different defects on the vibration and noise of the iron core reactor;

[0010] Step 4: Based on the vibration and noise distribution patterns, obtain the optimal distribution of vibration and noise measurement points, and extract the vibration and noise signals of the core reactor under different operating conditions;

[0011] Step 5: Convert the extracted vibration and noise signals under different working conditions into Gram angles and field feature maps;

[0012] Step 6: Input the Gram angle and field feature map into AlexNet for transfer learning to achieve fault diagnosis.

[0013] In step 1, a three-dimensional simulation model of the reactor core under test is established using finite element simulation software, as shown in Figure 2(a). The reactor structure mainly includes the core, windings, and tank. The three-dimensional simulation model of the reactor core is then meshed, material properties are set, and boundary and initial conditions are configured. The mesh is set to coarseness, and the meshing result is shown in Figure 2(b). The material property settings are shown in Table 1.

[0014] Table 1. Material Properties of Reactors

[0015]

[0016] Where: T is the temperature of the reactor oil;

[0017] The boundary conditions of the reactor are set as electromagnetic forces and magnetostrictive forces in the core and windings as loading conditions for the structural force field, and the initial pressure acoustic pressure is set to 0 Pa.

[0018] By adding multiphysics fields, including magnetic fields, structural force fields, and acoustic fields, simulation results of reactor vibration and acoustic fields are obtained.

[0019] The vibration acceleration distribution of the reactor is shown in Figure 3(a), with the maximum vibration acceleration being 12.6 Gal, located at the side yoke of the core; the sound pressure distribution of the reactor is shown in Figure 3(b), with the maximum sound pressure being 96 Pa, located at the top yoke of the core.

[0020] Equations for magnetic-structure coupling, structure-acoustic field coupling, and magnetic-acoustic field coupling are established. The general form of the magnetic-structure-acoustic field coupling equation is as follows:

[0021]

[0022] Equations (1) and (2) are the magnetic field equations, where H represents the magnetic field strength; B represents the magnetic flux density; and J represents the current density. This represents the Hamiltonian operator.

[0023] Equation (3) is the structural dynamics equation, where ρ represents the material density; u represents the structural displacement; σ represents the stress tensor; and f represents the external load.

[0024] Equation (4) is the sound field equation, where ρ c ρ represents the density of the acoustic medium; p represents the sound pressure. t represents the sound pressure vector field; t represents time.

[0025] In step 2, fault settings are performed on the three-dimensional simulation model of the iron core reactor from step 1:

[0026] a. Adjusting the core clamping force to simulate a core loosening fault;

[0027] b. Simulate winding deformation faults by setting axial bulges in the primary winding;

[0028] c. Simulation software is used to set the internal circuit of the winding and adjust the number of winding turns and fault impedance to simulate the inter-turn short circuit fault of the core reactor winding.

[0029] d. Simulate winding clamping loosening fault by adjusting the clamping force at the upper end of the winding.

[0030] In step 3, based on the multiphysics coupling calculation model and fault settings, the multiphysics coupling calculation results of the core reactor winding under different operating conditions are calculated. The multiphysics coupling simulation calculation results under normal operating conditions of the reactor are shown in Figure 3(a) and Figure 3(b); the simulation calculation results under inter-turn short circuit conditions are shown in Figure 4(a) and Figure 4(b); Figure 5(a) and Figure 5(b) are the simulation calculation results under the condition of loosening of the reactor winding clamping force.

[0031] In step 4, reasonable observation points are selected based on the vibration and noise distribution characteristics of the core reactor. The sound field observation point is located at the upper yoke of the reactor core, as shown in Figure 4(a); the vibration observation point is located on the upper wall of the reactor tank, as shown in Figure 4(b). Based on the multiphysics coupling calculation results of the core reactor under different operating conditions, vibration and noise signals under each operating condition are extracted from the observation points. Taking the signals extracted under normal operating conditions of the reactor as an example, Figure 5(a) shows the sound field signal extracted at the observation point; Figure 6(a) shows the extracted vibration signal.

[0032] In step 5, the principle of the Gram angle and the field is as follows:

[0033] S5.1: Normalize the signal to the interval [-1, 1], as shown in equation (6):

[0034]

[0035] In equation (6): This represents the normalized signal, where n represents the data sequence number; x i For time series signals; min(x) is the minimum sequence signal; max(x) is the maximum sequence signal;

[0036] S5.1: Map the normalized signal to the polar coordinate system, as shown in equation (7):

[0037]

[0038] In equation (7), r represents the polar radius in polar coordinates, and t i t represents the polar angle in polar coordinates. i Where N is the number of sampling points for the signal, and N is the length of the signal.

[0039] S5.3: Combining equations (6) and (7), the Gram angle and field (GASF) are defined as shown in equation (8):

[0040]

[0041] In equation (8): the subscripts i and j represent the serial numbers of different sampling points, respectively;

[0042] The feature signals extracted in step 4 are converted into two-dimensional feature images using Gram angle and field. Taking the sound field and vibration signals extracted under normal operating conditions of the reactor as an example, Figure 5(b) is the two-dimensional feature map obtained by converting the sound field signal in Figure 5(a); Figure 6(b) is the two-dimensional feature map obtained by converting the vibration signal in Figure 6.

[0043] In step 6, the basic structure of the AlexNet model includes an input layer, convolutional layers, pooling layers, fully connected layers, and an output layer. The input layer receives image data and preprocesses it; the convolutional layers progressively extract local features from the image; the pooling layers reduce the feature map size and enhance the model's robustness; the fully connected layers combine the extracted features and map them to the class space; and the output layer generates the final classification result. The fully connected layers use Dropout to prevent overfitting.

[0044] Step 6 includes the following steps:

[0045] S6.1: Use the Gram angle and field feature map set generated in step 5 as the target dataset and preprocess it, including resizing, normalization and data augmentation, to improve the generalization ability of the model.

[0046] S6.2: Load the pre-trained AlexNet model. In transfer learning, freeze the convolutional layers of AlexNet to avoid repeatedly training the learned features and accelerate the training process.

[0047] S6.3: Modify the fully connected layers of the AlexNet model according to the number of task categories;

[0048] S6.4: Set the loss function and optimizer;

[0049] S6.5: Based on the dataset, create a data loader to process the training and validation data;

[0050] S6.6: Train the model and validate it.

[0051] This invention provides a fault diagnosis method for oil-immersed iron-core reactors based on multi-physics coupling and feature maps, with the following technical advantages:

[0052] 1) This invention can accurately identify the fault type of the core reactor winding, thereby ensuring the reliability and safety of the power system, reducing downtime losses, extending equipment service life, optimizing operation and maintenance decisions, and promoting the intelligent and optimized design of core reactor equipment.

[0053] 2) This invention utilizes Gram angle and field to process acoustic and vibration signals, which helps to denoise the signals and compress the data, thereby improving the accuracy of fault diagnosis and reducing the time required for fault diagnosis.

[0054] 3) This invention, based on the acoustic field-structural force field, provides a method for diagnosing winding faults in core reactors. It combines the multi-physics characteristics of acoustics and mechanics, enabling early fault diagnosis and precise fault location. By collecting and analyzing vibration and acoustic signals during the operation of the core reactor, winding faults can be effectively identified. Attached Figure Description

[0055] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0056] Figure 1 This is a flowchart of a fault diagnosis method for iron core reactors.

[0057] Figure 2(a) shows the three-dimensional simulation model of the iron-core reactor established using finite element simulation software;

[0058] Figure 2(b) is a schematic diagram of the mesh generation of the three-dimensional simulation model of the iron core reactor.

[0059] Figure 3(a) shows the acoustic field simulation results of the iron core reactor under normal operating conditions;

[0060] Figure 3(b) shows the structural force field simulation results of the iron-core reactor under normal operating conditions.

[0061] Figure 4(a) shows the acoustic field simulation results under the deformation condition of the iron core reactor winding;

[0062] Figure 4(b) shows the structural force field simulation results under the deformation condition of the iron core reactor winding.

[0063] Figure 5(a) shows the acoustic field simulation results under the inter-turn short circuit condition of the iron core reactor;

[0064] Figure 5(b) shows the structural force field simulation results of the iron core reactor under the inter-turn short circuit condition.

[0065] Figure 6(a) shows the acoustic field simulation results under the condition of loose winding of iron core reactor;

[0066] Figure 6(b) shows the structural force field simulation results under the condition of loose winding of iron core reactor.

[0067] Figure 7(a) shows the location of the vibration signal observation point of the iron core reactor. Figure 1 ;

[0068] Figure 7(b) is a schematic diagram of the location of the acoustic field signal observation point of the iron core reactor.

[0069] Figure 8(a) is a schematic diagram of a random acoustic signature signal collected by a sensor simulated by simulation software.

[0070] Figure 8(b) shows the two-dimensional feature spectrum formed by the voiceprint signal through the Gram angle field.

[0071] Figure 9(a) is a schematic diagram of a random vibration signal collected by a sensor simulated by simulation software.

[0072] Figure 9(b) shows the two-dimensional feature spectrum formed by the vibration signal passing through the Gram angle field.

[0073] Figure 10(a) shows the results of inputting the feature map into the convolutional neural network (training set confusion matrix);

[0074] Figure 10(b) shows the results of inputting the feature map into the convolutional neural network (test set confusion matrix). Detailed Implementation

[0075] like Figure 1 As shown, the fault diagnosis process for iron-core reactors based on multi-physics coupling is as follows: A finite element simulation model of the iron-core reactor is constructed, the calculation methods for electromagnetic and structural force fields are clarified, and the operating state of the iron-core reactor windings under different fault conditions is simulated; based on the coupled simulation of acoustic and structural force fields, the performance of the iron-core reactor under normal and fault conditions is analyzed; acoustic patterns and vibration signals are selected as features, and the collected signals are converted into GASF image sets; the image sets are input into a convolutional neural network for training; the diagnostic model is initialized, and the training set is input into the model for transfer learning to complete the training process; the diagnostic effect is evaluated using test set data to complete the fault diagnosis.

[0076] Includes the following steps:

[0077] Step S1: Based on the parameters of the iron-core reactor, establish a physical model of the iron-core reactor, construct a multi-physics field coupling calculation model, and obtain the simulation results of the electromagnetic-structural force field and acoustic field of the reactor;

[0078] Step S2: Establish a simulation model of the iron-core reactor under typical defects, including: iron core loosening, winding deformation, winding loosening and inter-turn short circuit, and obtain the simulation results of the electromagnetic-structural force field and acoustic field of the reactor under different operating conditions;

[0079] Step S3: Based on the simulation results, analyze the influence of defects on vibration and noise;

[0080] Step S4: Based on the vibration and noise distribution patterns, obtain the optimal distribution of vibration and noise measurement points, and extract the vibration and noise signals of the core reactor under different operating conditions;

[0081] Step S5: Convert the extracted vibration and noise signals under different working conditions into Gram angle field feature maps;

[0082] Step S6: Input the Gram corner field feature map into AlexNet for transfer learning to achieve fault diagnosis.

[0083] In step S1, a three-dimensional simulation model of the reactor core under test is established using finite element simulation software. Multiphysics modules, including magnetic field, structural force field, and acoustic field modules, are added. Material parameters, mesh generation, boundary conditions, and initial conditions are then set for the model. Regarding material parameter settings, the hysteresis and magnetostriction characteristics of the core material are obtained from experimentally measured data of silicon steel sheets, while the reactor oil and windings are set according to actual parameters. For the mesh, a denser mesh is used at locations such as the core and windings, while a sparser mesh is used at locations far from the core and windings. Regarding boundary condition settings, since the bottom of the reactor is connected to the bottom of the oil tank, the bottom of the reactor core is set as a fixed constraint, i.e., zero displacement, a stationary wall.

[0084] In step S2, the magnetic-structural coupling and structural-acoustic coupling equations for the iron-core reactor are established. The magnetic field-structural force field coupling equation is as follows:

[0085]

[0086] Equations (1) and (2) are the magnetic field equations, where H is the magnetic field strength; B is the magnetic flux density; and J is the current density. Equation (3) is the structural dynamics equation, where ρ is the material density; u is the structural displacement; σ is the stress tensor; and f is the external load.

[0087] The structural force field-acoustic field coupling equation is as follows:

[0088]

[0089] Equation (4) is the sound field equation, where ρ c ρ is the density of the acoustic medium; p is the sound pressure.

[0090] In step S2, fault settings are performed on the core reactor model from step S1: Internal fault simulation of the oil-immersed core reactor mainly occurs in the core and windings. Regarding the core, core loosening faults are simulated by adjusting the core clamping force. The normal operating condition is set to 100% of the rated clamping force, and the clamping force is adjusted to 10%, 20%, 30%, and 40% of the rated value to simulate core loosening faults. Regarding the windings, radial and axial bulges are set on the primary winding to simulate winding deformation faults. Radial and axial bulges are set at different locations: radial and axial bulges are set in the upper, middle, and lower middle parts of the winding, respectively. The size of the bulges is controlled to... Simulate winding deformation faults under different fault degrees; simulate inter-turn short-circuit faults in iron-core reactor windings by setting the internal circuit of the winding and adjusting the number of winding turns and fault impedance using simulation software. Inter-turn short-circuit faults are set in the upper, middle, and lower middle parts of the winding, and different fault degrees are determined according to the proportion of the number of short-circuit turns in the winding under normal operating conditions, such as 1%, 2%, 3%, and 4%; simulate winding clamping force loosening faults by adjusting the clamping force at the upper end of the winding. Under normal operating conditions, the clamping force is set to 100% of the rated clamping force, and the clamping force is set to 10%, 20%, 30%, and 40% of the rated value to simulate winding clamping force loosening faults.

[0091] In step S3, based on the multiphysics coupling calculation model and fault settings described above, the multiphysics coupling calculation results of the core reactor winding under different operating conditions are calculated, and its vibration characteristics and noise distribution characteristics are compared and analyzed. Based on the simulation results, the magnetic flux density distribution under different paths of the core and winding positions is extracted and compared and analyzed; the stress distribution under different paths of the core and winding positions is extracted and compared and analyzed; the vibration displacement distribution under different paths of the core and winding positions is extracted and compared and analyzed; and the noise distribution under different paths around the core and winding positions is extracted and compared and analyzed.

[0092] In step S4, reasonable observation points are selected based on the vibration and noise distribution characteristics of the core reactor. The specific steps for measurement point optimization are as follows: Data from various operating conditions at different measurement points are sequentially extracted with a time interval of 0.2s and subjected to LMD decomposition to obtain multiple PF components. The correlation between each component and the original data is calculated, and the P components with high correlation are selected as effective components. Information entropy is calculated to construct a feature vector matrix. Keeping the same operating condition constant, correlation analysis is performed on the feature vector matrices X and Y of different measurement points, and the correlation between each measurement point under each operating condition is solved sequentially. The number of correlation indicators between each measurement point under each operating condition is counted, and those without correlation are retained to achieve the purpose of measurement point optimization. Based on the multiphysics coupling calculation results of the core reactor under different operating conditions, vibration and noise signals under each operating condition at the observation points are extracted.

[0093] In step S5, the feature signals extracted from different measurement points in step S4 are converted into a two-dimensional feature map using Gram angles and fields. The principle of Gram angles and fields is as follows:

[0094] A1: Normalize the signal to the interval [-1, 1], as shown in equation (6):

[0095]

[0096] In the formula This represents the normalized signal, and n represents the data sequence number.

[0097] A2: Map the normalized signal to the polar coordinate system, as shown in equation (7):

[0098]

[0099] In the formula, r represents the polar radius in polar coordinates, and t i t represents the polar angle in polar coordinates. i Where N is the number of sampling points for the signal, and N is the length of the signal.

[0100] A3: Combining equations (6) and (7), the Gram angle and field (GASF) are defined as shown in equation (8):

[0101]

[0102] In the formula, the subscripts i and j represent the serial numbers of different sampling points, respectively.

[0103] In step S6, the basic structure of the AlexNet model includes an input layer, convolutional layers, pooling layers, fully connected layers, and an output layer. The fully connected layers use Dropout technology to prevent overfitting. The specific fault classification steps are as follows:

[0104] B1: Use the GASF image set generated in step S5 as the target dataset and preprocess it, including resizing, normalization and data augmentation to improve the model's generalization ability.

[0105] B2: Load the pre-trained AlexNet model. In transfer learning, freeze the convolutional layers of AlexNet to avoid retraining already learned features and accelerate the training process.

[0106] B3: Modify the fully connected layers of AlexNet based on the number of task categories.

[0107] B4: Set the loss function and optimizer.

[0108] B5: Based on the dataset, create a data loader to process the training and validation data.

[0109] B6: Train the model and validate it.

[0110] Figures 2 and 3 show the model of the oil-immersed transformer established by the finite element simulation software and the simulation results after adding the acoustic field and structural force field. Their purpose is to simulate the core reactor in actual operation. The noise and vibration acceleration distribution patterns of the core reactor under different operating conditions are compared and analyzed to select the optimal measurement point locations, as shown in Figures 7(a) and 7(b). After the finite element simulation process is completed, the relationship curves between acoustic signature amplitude and time and vibration acceleration and time at the selected measurement points under various operating conditions are exported using the one-dimensional plotting group, as shown in Figures 8(a) and 9(a), reflecting the acoustic signature and vibration signals of the transformer. Since the extracted acoustic signature and vibration signals usually contain noise interfering with the useful parts of the original signal, this invention processes the extracted signals using a Gram angle field to convert the one-dimensional signal set into a two-dimensional feature spectrum, achieving noise reduction and signal compression, as shown in Figures 8(b) and 9(b).

[0111] As shown in Figures 10(a) and 10(b), this section demonstrates the accuracy of the algorithm after computation compared with the actual situation. Figure 10(a) shows the training set results, and Figure 10(b) shows the test set results. As can be seen from Figures 10(a) and 10(b), the accuracy of both is higher than 94%, indicating that the algorithm's results are accurate.

[0112] In practice, the specific operating steps are as follows:

[0113] Step 1: Establish a 3D simulation model of the reactor core under test using finite element simulation software, and perform mesh generation, material property settings, and boundary and initial condition settings. Add multiphysics fields, including magnetic field, structural force field, and acoustic field. Establish the magnetic-structural coupling, structural-acoustic coupling, and magnetic-acoustic coupling equations.

[0114] Step 2: Fault setting for the core reactor model: Adjust the core clamping force to simulate core loosening fault; axial bulging of the primary winding to simulate winding deformation fault; simulate inter-turn short circuit fault of the core reactor winding by setting the internal circuit of the winding and adjusting the number of winding turns and fault impedance through simulation software; simulate winding clamping force loosening fault by adjusting the clamping force at the upper end of the winding.

[0115] Step 3: Based on the multiphysics coupling calculation model and fault settings, calculate the multiphysics coupling calculation results of the iron core reactor winding under different operating conditions, and compare and analyze its vibration characteristics and noise distribution characteristics.

[0116] Step 4: Select appropriate observation points based on the vibration and noise distribution characteristics of the core reactor. Based on the multiphysics coupling calculation results of the core reactor under different operating conditions, extract the magnetic characteristic signals, vibration signals, and noise signals at each observation point under each condition. Then, use Gram angles and fields to convert the extracted feature signals into two-dimensional feature maps. Input the feature maps into the improved AlexNet convolutional neural network for transfer learning to complete fault diagnosis.

Claims

1. A fault diagnosis method for oil-immersed iron-core reactors based on multi-physics coupling and characteristic spectrum, characterized in that... Includes the following steps: Step 1: Construct a multiphysics coupling calculation model of the iron-core reactor to obtain the simulation results of the electromagnetic-structural force field and acoustic field of the iron-core reactor; Step 2: Establish a simulation model of the iron-core reactor under typical defects, including: iron core loosening, winding deformation, winding loosening and inter-turn short circuit, and obtain the simulation results of the electromagnetic-structural force field and acoustic field of the iron-core reactor under different operating conditions; Step 3: Based on the simulation results of Step 2, analyze the influence of different defects on the vibration and noise of the core reactor; Step 4: Based on the vibration and noise distribution patterns, obtain the optimal distribution of vibration and noise measurement points, and extract the vibration and noise signals of the core reactor under different operating conditions; Step 5: Convert the extracted vibration and noise signals under different working conditions into Gram angles and field feature maps; Step 6: Input the Gram angle and field feature map into AlexNet for transfer learning to achieve fault diagnosis; In step 4, reasonable observation points are selected based on the vibration distribution characteristics and noise distribution characteristics of the core reactor; the sound field observation point is located at the upper yoke of the reactor core, and the vibration observation point is located on the upper wall of the reactor tank; based on the multi-physics coupling calculation results of the core reactor under different operating conditions, the vibration signal and noise signal of each operating condition at the observation point are extracted.

2. The fault diagnosis method for oil-immersed iron-core reactors based on multi-physics coupling and feature maps according to claim 1, characterized in that: In step 1, a three-dimensional simulation model of the reactor under test is established using finite element simulation software. The reactor structure includes the core, windings, and tank. The three-dimensional simulation model of the reactor is then meshed, and material properties, boundary conditions, and initial conditions are set. The mesh is set to coarsening, and the reactor boundary conditions are set to electromagnetic forces and magnetostrictive forces in the core and windings as loading conditions for the structural force field. Multiphysics fields, including magnetic field, structural force field, and acoustic field, are added to obtain the simulation results of reactor vibration and acoustic field.

3. The fault diagnosis method for oil-immersed iron-core reactors based on multi-physics coupling and feature maps according to claim 2, characterized in that: Equations for magnetic-structure coupling, structure-acoustic field coupling, and magnetic-acoustic field coupling are established. The magnetic-structure-acoustic field coupling equation is expressed as follows: (1)； (2)； (3)； (4)； Equations (1) and (2) are the magnetic field equations, where H represents the magnetic field strength; B Indicates magnetic flux density; J Indicates current density; Represents the Hamiltonian operator; Equation (3) is the structural dynamics equation, where, ρ Indicates the density of the material; u Indicates structural displacement; σ Represents the stress tensor; f Represents the external load; Equation (4) is the sound field equation, where, ρ c Indicates the density of the acoustic medium; p Indicates sound pressure level; Represents the sound pressure vector field; Indicates time.

4. The fault diagnosis method for oil-immersed iron-core reactors based on multi-physics coupling and characteristic spectrum as described in claim 1, characterized in that: In step 2, fault settings are performed on the three-dimensional simulation model of the iron core reactor from step 1: a. Adjusting the core clamping force to simulate a core loosening fault; b. Simulate winding deformation faults by setting axial bulges in the primary winding; c. Simulation software is used to set the internal circuit of the winding and adjust the number of winding turns and fault impedance to simulate the inter-turn short circuit fault of the core reactor winding. d. Simulate winding clamping loosening fault by adjusting the clamping force at the upper end of the winding.

5. The fault diagnosis method for oil-immersed iron-core reactors based on multi-physics coupling and characteristic spectrum as described in claim 1, characterized in that: In step 5, the principle of the Gram angle and the field is as follows: S5.1: Normalize the signal to the interval [-1, 1], as shown in equation (6): (6)； In formula (6): Represents the normalized signal. n Indicates the data sequence number; It is a time series signal; It is a minimum sequence signal; It is the maximum sequence signal; S5.1: Map the normalized signal to the polar coordinate system, as shown in equation (7): (7)； In equation (7), r Represents the polar radius in polar coordinates. t i Represents the polar angle in polar coordinates. t i The number of sampling points for the signal. N The length of the signal; S5.3: Combining equations (6) and (7), the Gram angle and field (GASF) are defined as shown in equation (8): (8)； In equation (8): subscript i and j These represent the sequence numbers of different sampling points; The feature signals extracted in step 4 are converted into two-dimensional feature images using Gram angles and field.

6. The fault diagnosis method for oil-immersed iron-core reactors based on multi-physics coupling and feature maps according to claim 1, characterized in that: In step 6, the basic structure of the AlexNet model includes an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer. The input layer receives image data and performs preprocessing. The convolutional layer extracts local features of the image step by step. The pooling layer reduces the size of the feature map and enhances the robustness of the model. The fully connected layer combines the extracted features and maps them to the category space. The output layer generates the final classification result. The fully connected layer uses Dropout technology to prevent overfitting.

7. The fault diagnosis method for oil-immersed iron-core reactors based on multi-physics coupling and feature maps according to claim 6, characterized in that: Step 6 includes the following steps: S6.1: Use the Gram angle and field feature map set generated in step 5 as the target dataset and preprocess it, including resizing, normalization and data augmentation, to improve the generalization ability of the model. S6.2: Load the pre-trained AlexNet model and freeze the convolutional layers of AlexNet during transfer learning; S6.3: Modify the fully connected layers of the AlexNet model according to the number of task categories; S6.4: Set the loss function and optimizer; S6.5: Based on the dataset, create a data loader to process the training and validation data; S6.6: Train the model and validate it.

Images