Reactor fault diagnosis method based on mechanism-data hybrid driving

Abstract

The invention relates to the field of electric reactor fault diagnosis, in particular to an electric reactor core column mechanical fault diagnosis method based on mechanism-data hybrid driving, which is characterized in that mechanism model driving and data model driving are cooperatively carried out, mechanism modeling is mainly carried out aiming at the vibration of an electric reactor core column, and fault features are extracted by using a mechanism model; and fault diagnosis is carried out by using the data driving model, and a mechanism-data dual-driving fault diagnosis model is established, so that fault diagnosis of mechanical faults of the core column of the electric reactor is realized. According to the method, different degrees of mechanical faults of the reactor iron core column can be accurately diagnosed, the accuracy of fault diagnosis is improved, and the diagnosis result is more interpretable.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of reactor fault diagnosis, and in particular to a reactor core column mechanical fault diagnosis method based on mechanism-data hybrid driving. BACKGROUND

[0002] As a vital reactive power compensation and current limiting device in the power system, the operation reliability of the reactor is directly related to the stability and safety of the power grid. The fault modes of the reactor can be generally divided into two categories: electrical faults and mechanical faults. Among them, mechanical faults are the main form of early faults of the reactor. Mechanical faults may be very weak in the early stage, and may not immediately cause obvious electrical parameter abnormalities. The device may still be able to continue to operate. However, if it is operated for a long time, it may cause insulation aging breakdown, winding inter-turn short circuit and other conditions, which seriously affect the smooth operation of the reactor. Therefore, through effective diagnosis, timely capture and elimination of faults are the key to avoiding serious power outages and improving the reliability and economy of the power grid.

[0003] The core column, as a special part of the magnetic circuit structure of the reactor, is subjected to greater electromagnetic force due to its position in the main magnetic flux path and its stacking of discrete core cakes, resulting in greater deformation risk. Once the core column fails, it may cause abnormal excitation current, excessive temperature rise and increased noise of the reactor, and may even cause coil insulation damage, magnetic circuit saturation and even overall burning. SUMMARY

[0004] In view of the lack of diagnosis scheme for core column faults in the prior art, and considering the modeling complexity of single mechanism driving and the lack of fault samples and poor interpretability of single data driving, a reactor core column fault diagnosis method based on mechanism-data hybrid driving is provided.

[0005] To achieve the above purpose, the present application adopts the following scheme:

[0006] A reactor core column mechanical fault diagnosis method based on mechanism-data hybrid driving is used to diagnose the fault of the high-voltage reactor core column, characterized in that it comprises a model construction method of a fault classification model, and the model construction method comprises the following steps:

[0007] S1: Constructing a reactor core column mechanical vibration model;

[0008] The reactor core column vibration model is represented by a reactor core column vibration acceleration peak value formula;

[0009] S2: Obtaining sample signals and labeling the corresponding sample signals with fault labels to obtain corresponding sample data;

[0010] The sample signals include corresponding sample vibration signals and sample electrical signals;

[0011] S3: fault feature extraction;

[0012] extracting fault features based on the sample signals, the fault features including equivalent damping coefficient, equivalent stiffness and magnetostriction force coefficient;

[0013] S4: model training;

[0014] training a model based on the obtained fault features and corresponding fault labels to obtain a corresponding fault classification model.

[0015] Further, the reactor core column mechanical vibration model constructed in step 1 is:

[0016]

[0017] wherein M is the equivalent core column mass, C is the equivalent damping coefficient, K is the equivalent stiffness of the model, z is the equivalent maximum displacement, ω is the angular frequency of alternating current, F c,max is the maximum amplitude of magnetostriction force, ω is the angular frequency of the power supply, is the phase angle, k is the magnetostriction force coefficient, and I is the operating current.

[0018] Further, the data acquisition and preprocessing steps in step 2 are as follows:

[0019] Step 2.1 simulates the operation state of the reactor under different conditions and collects the vibration waveform to obtain the corresponding vibration signal;

[0020] Step 2.2 processes the measured vibration signal using a fourth-order Butterworth low-pass filter to obtain the corresponding sample vibration signal.

[0021] wherein the cutoff frequency is set to 80% of the Nyquist frequency, and zero-phase filtering is used during filtering.

[0022] Further, in step 2, the fault is mainly classified based on the degree of fault, which is divided into three categories: "0" for normal state; "1" for slight looseness fault (looseness degree 10% ~ 40%); "2" for serious looseness fault (looseness degree more than 40%).

[0023] Further, the fault feature extraction steps in step 3 are as follows:

[0024] Step 3.1 determines the direct obtained quantity and the indirect obtained quantity, and takes the indirect obtained quantity as the fault feature.

[0025] Wherein: the directly measurable variables include: equivalent core mass, angular frequency of alternating current, phase angle, vibration acceleration, operating current; the indirectly obtained variables include: equivalent damping coefficient, equivalent stiffness, magnetostrictive force coefficient, which are inherent body characteristics of the reactor and can better reflect the true health status of the equipment.

[0026] Step 3.2 is based on the directly obtained variables, uses the mechanical vibration model of the reactor core, and iteratively calculates the values of each indirectly obtained variable in the corresponding vibration state by using the Gauss-Newton method to obtain the corresponding fault characteristics.

[0027] Further, the step of iteratively calculating the indirectly obtained variables in step 3 of the Gauss-Newton method is as follows:

[0028] Step 3.2.1 gives an initial parameter vector, sets a convergence threshold and a maximum number of iterations

[0029] Step 3.2.2 calculates the residual vector, Jacobian matrix and update direction according to the current parameter estimation.

[0030] Step 3.2.3 performs parameter updating and conditional constraints.

[0031] Wherein: the constraint condition is that the equivalent damping coefficient, equivalent stiffness and magnetostrictive force coefficient are all greater than 0.

[0032] Step 3.2.4 When the iteratively calculated result meets the following conditions, the result is output, otherwise the iteration is continued.

[0033] Wherein: the iteration condition meets:

[0034] ||Δθ (s) ||2<ε θ

[0035] |ΔJ|<ε J

[0036] ||g (s) ||2<ε g

[0037] Wherein: ||Δθ (s) ||2 is the Euclidean norm of the parameter update vector of the s-th iteration, ε θ is the preset parameter convergence tolerance; ΔJ is the improvement of the fitting error, ε J is the preset objective function convergence tolerance; ||g (s) ||2 is the Euclidean norm of the gradient vector of the s-th iteration, ε g is the preset gradient convergence tolerance.

[0038] Further, the specific process of model training and optimization in step 4 is as follows:

[0039] Step 4.1 corresponds the fault feature to the fault label to form a data set;

[0040] Step 4.2 selects the scarab optimization algorithm to optimize the fault classification algorithm, and the fault classification algorithm and the convolutional neural network-long short-term memory network, and the optimization object includes: learning rate, convolution kernel number, LSTM neuron number, batch size, and full connection layer neuron number;

[0041] Step 4.3 trains the optimized fault classification algorithm by using the data set;

[0042] Step 4.4 outputs the trained fault classification model.

[0043] Further, the input of the model in step 4 is the indirect acquisition quantity obtained in step 3, including: equivalent damping coefficient, equivalent stiffness, magnetostrictive force coefficient, and the output of the model is the fault label.

[0044] Further, the convolutional neural network-long short-term memory network method is used as the fault classification model in step 4, the local spatial features between the input quantities are extracted by the convolutional neural network model, and the long-term dependence relationship of the feature sequence in time is input to the long short-term memory network for learning, and finally the fault probability of each category is output through the full connection layer.

[0045] The beneficial effects of the present application are:

[0046] (1) The core column is a special magnetic circuit structure component of the electric reactor. Since it is in the main magnetic path and is stacked by discrete core cakes, it bears greater electromagnetic force, resulting in greater risk of mechanical failure. Once the core column fails mechanically, it will cause abnormal excitation current of the electric reactor, excessive temperature rise and noise increase, and in severe cases, it may cause coil insulation damage, magnetic circuit saturation and even overall burning. However, current fault diagnosis is only for ordinary core stack structure, and there is no diagnostic method for the fault of the special core column structure in the core. In view of the lack of existing technology, a mechanism-data hybrid driven electric reactor core column mechanical fault diagnosis method is proposed. Compared with the existing technology, the present technology analyzes the mechanism and diagnoses the fault for the special structure of the electric reactor core column, solves the problem of lack of fault diagnosis method for this aspect, and realizes effective diagnosis for the mechanical fault of the electric reactor core column.

[0047] (2) The current commonly used reactor fault diagnosis method is mostly a single data model driven method or a single mechanism model driven method. However, due to the lack of fault samples, the single data driven method lacks accuracy and physical basis, and at the same time, the mechanism analysis is too complex, so that the single mechanism driven method is difficult to realize accurate diagnosis. In view of the defects of the prior art, a mechanism-data hybrid driven reactor core column mechanical fault diagnosis method is proposed. Compared with the prior art, the mechanism-data hybrid driven method is introduced, which combines the advantages of mechanism driven model and data driven model, so that it can realize accurate fault diagnosis without a large number of samples, solve the problems of complex mechanism analysis of traditional mechanism model driven fault diagnosis method and the need for a large number of effective data and lack of interpretability of traditional data driven fault diagnosis method, realize the judgment of whether the core column mechanical fault occurs through a small amount of fault data samples, and the interpretability of the fault diagnosis result is stronger. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 is a flow chart of the reactor core column fault diagnosis method based on mechanism-data hybrid driving of the present application;

[0049] Figure 2 is a reactor core column vibration equivalent model diagram of the reactor core column fault diagnosis method based on mechanism-data hybrid driving of the present application. DETAILED DESCRIPTION

[0050] The present application will be further described in detail below in combination with examples. Reactor is a vital reactive power compensation and current limiting device in power system, and its operation reliability is directly related to the stability and safety of power grid. The core column is a special magnetic circuit structure component of the reactor, which is in the main magnetic path and is subjected to greater electromagnetic force due to the stacking of discrete core cakes, resulting in greater risk of deformation. However, in the prior art, fault diagnosis has not been carried out for the special structure of the reactor core, and at present, mechanism model driving and data driving have not been combined to realize double driving for fault diagnosis and improve the accuracy and interpretability of fault diagnosis.

[0051] In view of the limitations of the above-mentioned core column fault research, considering the modeling complexity of single mechanism driving and the lack of fault samples and poor interpretability of single data driving, the present application proposes a reactor core column fault diagnosis method based on mechanism-data hybrid driving, aiming to realize accurate fault positioning and parameter adaptive optimization. Mechanism model driving and data model driving are cooperated, mainly through mechanism modeling for reactor core column vibration, extracting finite elements by using mechanism model, and using data driven model for fault diagnosis, establishing a mechanism-data double driven fault diagnosis model, improving the accuracy of fault diagnosis, and making the diagnosis result more interpretable.

[0052] As Figure 1 shown, the application based on mechanism-data hybrid driving reactor core column fault diagnosis method includes the training method of fault classification model, and the specific training steps are as follows:

[0053] Step 1: Establish a reactor core column vibration model;

[0054] The reactor core column vibration model in this embodiment is expressed by the reactor core column vibration acceleration peak value formula.

[0055] The reactor core column is the core part of its magnetic circuit, which is pressed from silicon steel sheets. In order to analyze the vibration model, the reactor core column vibration model is equivalent to a single degree of freedom spring-mass-damping system, and the equivalent model is as shown in Figure 2 , Figure 2 In the formula, K represents the equivalent stiffness of the model, C represents the equivalent damping coefficient, and m represents the equivalent mass of a single core pie.

[0056] According to the second law of Newton, the motion equation is determined, and in this embodiment, the motion equation is:

[0057]

[0058] Where: M is the equivalent core column mass, which is the sum of the mass of multiple core pies, C is the equivalent damping coefficient, K is the equivalent stiffness of the model, which is related to the material elastic modulus and the stacking pre-tightening force, F(t) is the magnetostriction force, and z is the amplitude.

[0059] Under the condition of ignoring the order state decay, the steady-state displacement response is:

[0060]

[0061] Where: z(t) is the equivalent displacement and time function, z is the amplitude, ω is the angular frequency of the power supply, is the phase angle.

[0062] And at a certain time, the magnetostriction force can be expressed as:

[0063] F c (t)=F c,max sin(2ωt)

[0064] Where: ω is the angular frequency of alternating current, F c,max is the maximum amplitude of magnetostriction force.

[0065] The steady-state displacement response formula and the magnetostriction force expression above are brought into the motion equation, and the amplitude is:

[0066]

[0067] The reactor core column vibration acceleration a(t) is obtained as follows:

[0068]

[0069] The maximum amplitude frequency equivalent formula of magnetostrictive force is set as follows:

[0070] F c,max =k·I 2

[0071] Wherein, k is the magnetostrictive force coefficient, and I is the operating current.

[0072] Therefore, the peak value formula of the reactor core column vibration acceleration can be equivalent to:

[0073]

[0074] Step 2: Data acquisition and pretreatment:

[0075] 2.1, simulate the operating state of the reactor;

[0076] In this embodiment, the experimental platform is used to simulate the operating state of the reactor under different degrees of core column loosening and under each sample voltage input;

[0077] The voltage values of each sample voltage are different, and in this embodiment, the input voltage is adjusted by using a self-coupled contact voltage regulator, and the vibration signals of the reactor are collected in the voltage range of 100V to 260V with an interval of 20V.

[0078] The experimental platform includes a reactor vibration experimental platform including three-phase power supply, self-coupled contact voltage regulator, single-phase double-column dry-type shunt reactor and other power equipment, and a data acquisition system including three-axis vibration acceleration sensor, alternating current clamp, data acquisition card; those skilled in the art can directly use the existing published reactor vibration test platform and data acquisition system, and can also build the corresponding experimental platform according to the actual needs, and this embodiment does not limit it in detail.

[0079] 2.2, collect vibration signals:

[0080] The vibration signals of the reactor during operation are collected by the vibration acceleration sensor adsorbed on the surface of the iron yoke of the reactor.

[0081] 2.3, vibration signal pretreatment:

[0082] The measured vibration signals are filtered and denoised by using a fourth-order Butterworth low-pass filter, and the corresponding sample vibration signals are obtained, wherein the cutoff frequency is set to 80% of the Nyquist frequency;

[0083] In this embodiment, zero-phase filtering is used in the filtering process to avoid signal phase distortion that may be caused by ordinary filtering, thereby ensuring that the true position of the fault impact signal on the time axis is not offset.

[0084] 2.4, setting a fault label:

[0085] At the same time of collecting the vibration signal, the current sensor is used to collect the electric signal input to the reactor to obtain a corresponding sample electric signal;

[0086] The corresponding sample vibration signal and sample electric signal are taken as sample signals, and different looseness is labeled with a fault label for each group of sample signals. In this embodiment, "0" is a normal state (looseness degree 0%~10%); "1" is a slight looseness fault (looseness degree 10%~40%); and "2" is a serious looseness fault (looseness degree 40% or more).

[0087] Those skilled in the art can classify the looseness degree and set the corresponding fault label according to actual needs, and the present specification does not limit them in detail.

[0088] Step 3: fault feature extraction:

[0089] 3.1, analyzing the model and extracting variables:

[0090] The reactor core column vibration model is analyzed, and the involved variables are extracted, including direct acquisition quantities and indirect acquisition quantities. The direct acquisition quantities are variables that can be directly measured.

[0091] In this embodiment, the direct acquisition quantities include: the equivalent core column mass M obtained by searching the initial parameters of the reactor, the angular frequency ω of the alternating current and the operating current I obtained based on the electric signal, and the phase angle and the vibration acceleration a;

[0092] The indirect acquisition quantities include: the equivalent damping coefficient C, the equivalent stiffness K, and the magnetostrictive force coefficient k.

[0093] Note: those skilled in the art can measure the above direct acquisition quantities based on the existing publicly disclosed related technologies, and the present specification does not repeat them. The corresponding variable parameters can be obtained.

[0094] 3.2, using the direct acquisition quantities, the values of each indirect acquisition quantity in the corresponding vibration state are iteratively calculated based on the reactor core column vibration model by the Gauss-Newton method.

[0095] The specific steps are as follows:

[0096] 3.2.1, setting initial parameters, convergence threshold and maximum iteration number;

[0097] That is, given an initial parameter vector θ = [K, C, k] Τ , set a convergence threshold ∈ (for example, 10 -6 ) and the maximum number of iterations s max .

[0098] 3.2.2, calculate the residual matrix, the Jacobian matrix and the update direction;

[0099] In this embodiment, the residual corresponding to each measurement point is calculated, and the residual matrix is constructed based on the obtained residual, wherein the residual of the i-th measurement point is:

[0100]

[0101] In the formula: r i (s) (θ) is the residual of the i-th measurement point at the s-th iteration, A i is the amplitude of the vibration acceleration measured at the i-th measurement point.

[0102] In this embodiment, the Jacobian matrix is calculated based on the following formula:

[0103]

[0104] Where: J i1 (s) represents the partial derivative of the i-th residual with respect to the stiffness K at the s-th iteration, J i2 (s) represents the partial derivative of the i-th residual with respect to the damping C at the s-th iteration, J i3 (s) represents the partial derivative of the i-th residual with respect to the magnetostriction force coefficient k at the s-th iteration.

[0105] In this embodiment, the update direction is calculated using the following formula:

[0106] H (s) = J (s)T J (s)

[0107] g (s) = J (s)T r (s)

[0108] Δθ (s) = -[H (s) + μA] -1 g (s)

[0109] Where: H (s) represents the approximate Hessian matrix at the s-th iteration, g (s) represents the gradient vector at the s-th iteration, and Δθ (s)θs represents the parameter update quantity of the s-th iteration, and A represents a unit matrix.

[0110] 3.2.3, parameter update and condition constraint;

[0111] In this embodiment, the parameter update is performed by using the following formula:

[0112] θ (s+1) = θ (s) + α s Δθ (s)

[0113] wherein θ (s) represents the current iteration step, the target parameter to be updated, θ (s+1) represents the target parameter updated by the current iteration step, θ (s+1) will be used for the next iteration step; α s is a line search step length, and its range is (0, 1].

[0114] In this embodiment, the constraint condition is that the equivalent damping coefficient, the equivalent stiffness and the magnetostrictive force coefficient are all greater than 0.

[0115] 3.2.4, judging whether the condition constraint is met;

[0116] When the iteration result meets the following any condition, the result, i.e., the target parameter, can be output, otherwise the iteration is continued.

[0117] wherein the iteration condition is met:

[0118] ||Δθ (s) ||2< ε θ

[0119] |ΔJ| < ε J

[0120] ||g (s) ||2< ε g

[0121] wherein ||Δθ (s) ||2is the Euclidean norm of the parameter update vector of the s-th iteration, ε θ is a preset parameter convergence tolerance; ΔJ is an improvement quantity of the fitting error, ε J is a preset target function convergence tolerance; ||g (s) ||2is the Euclidean norm of the gradient vector of the s-th iteration, ε g is a preset gradient convergence tolerance.

[0122] Step 4: model training and optimization

[0123] 4.1, the fault features and the fault labels correspond to each other to form a data set;

[0124] That is, the indirect acquisition quantity calculated in step 3, i.e. the target parameter, is used as the fault feature, and it is matched with the corresponding fault label to form a dataset.

[0125] 4.2 Optimize the model using optimization algorithms;

[0126] That is, the fault classification model is optimized using model optimization algorithms.

[0127] Specifically, the existing publicly available convolutional neural network-long short-term memory (LSTM) method was chosen as the fault classification algorithm. The convolutional neural network model extracts local spatial features between input quantities, which are then fed into the LSM network to learn the long-term temporal dependencies of the feature sequences. Finally, the fault probability for each category is output through a fully connected layer. The existing publicly available dung beetle optimization algorithm was used to optimize the above fault classification algorithm. The optimization targets include: learning rate, number of convolutional kernels, number of LSTM neurons, batch size, and number of neurons in the fully connected layer.

[0128] 4.3 Training the fault classification model:

[0129] The obtained dataset is used to train the fault classification algorithm to obtain a fault classification model with fault features as input and fault labels as output.

[0130] 4.4 Output the fault classification model;

[0131] Once the model is trained, a mechanism-data hybrid-driven mechanical fault diagnosis model for reactor core columns can be obtained.

[0132] This application provides a mechanism-data hybrid-driven method for diagnosing reactor core column faults, which further includes a method for fault diagnosis based on the fault classification model trained above. Specifically, when a mechanical fault occurs in the core column of the target reactor, vibration and electrical signals of the target reactor are collected; fault features are extracted based on the vibration and electrical signals, including equivalent damping coefficient, equivalent stiffness, and magnetostrictive force coefficient; the fault features are input into the fault classification model to obtain the corresponding fault classification results.

[0133] This embodiment collects vibration signals from the reactor under test and inputs them into a diagnostic model. By judging the fault label, it can be determined whether a core column fault has occurred and the degree of the fault. Since the input variables of the fault diagnosis model are parameters directly affected by a mechanical fault in the core column, it can more effectively reflect the current condition of the reactor. This allows for the determination of whether a mechanical fault in the core column has occurred using only a small number of fault data samples, and the fault diagnosis results are highly interpretable.

[0134] The technical solution of the present invention has been described in detail above. Any content related to the present invention and its related scope is considered plagiarism.

Claims

1. A mechanism-data hybrid driving-based mechanical fault diagnosis method for a reactor core column, used for diagnosing faults of a high-voltage reactor core column, characterized in that, The model construction method comprises the following steps: S1: Constructing a mechanical vibration model of the reactor core column; The reactor core column vibration model is represented by a reactor core column vibration acceleration peak value formula; S2: Obtain sample signals and label corresponding sample signals with fault labels to obtain corresponding sample data; The sample signals include corresponding sample vibration signals and sample electrical signals; S3: Fault feature extraction; Based on the sample signals, fault features are extracted, including equivalent damping coefficient, equivalent stiffness, and magnetostriction force coefficient; S4: Model training; Based on the obtained fault features and corresponding fault labels, model training is performed to obtain a corresponding fault classification model.

2. The method according to claim 1, wherein the method is characterized by, The mechanical vibration model of the reactor core column is: where: M is the equivalent core mass, C is the equivalent damping coefficient, K is the equivalent stiffness of the model, z is the equivalent maximum displacement, ω is the angular frequency of the alternating current, F c,max is the maximum amplitude of magnetostrictive force, ω is the angular frequency of the power supply, is the phase angle, k is the magnetostrictive force coefficient, I is the operating current.

3. The method according to claim 1, wherein the method is characterized by, The method for obtaining sample vibration signals in step S2 is: Step 2.1: Simulate the operating state of the reactor under different conditions and collect vibration waveforms to obtain corresponding vibration signals; Step 2.2: Process the measured vibration signals using a fourth-order Butterworth low-pass filter to obtain corresponding sample vibration signals.

4. The method according to claim 3, wherein the method is characterized by: The fourth-order Butterworth low-pass filter cutoff frequency is set to 80% of the Nyquist frequency.

5. The method according to claim 1, wherein the method is characterized by, The fault labels include: "0" indicates a normal state; "1" indicates a slight looseness fault; "2" indicates a serious looseness fault.

6. The method of claim 1, wherein the method is characterized by: The specific steps of fault feature extraction in step S3 are: Step 3.1: Determine the directly obtained quantities and indirectly obtained quantities, and take the indirectly obtained quantities as fault features; Step 3.2: Based on the directly obtained quantities, use the mechanical vibration model of the reactor core column to iteratively calculate the values of each indirectly obtained quantity under the corresponding vibration state using the Gauss-Newton method to obtain corresponding fault features.

7. The method according to claim 6, wherein the method is characterized by: In the Gauss-Newton method iteration, the constraint condition for parameter updating is that the equivalent damping coefficient, equivalent stiffness, and magnetostriction force coefficient are all greater than 0.

8. The method of claim 1, wherein the method is characterized by: The input of the fault classification model is the fault feature, including: equivalent damping coefficient, equivalent stiffness, and magnetostriction force coefficient; The output is the fault label, and the fault classification model uses a convolutional neural network-long short-term memory network.

9. The method according to claim 8, wherein the method is characterized by, The specific steps of model training in step S4 are: Step 4.1: Correspond the fault features and fault labels to form a data set; Step 4.2: Select the dung beetle optimization algorithm to optimize the fault classification algorithm, and the optimization objects include: learning rate, number of convolution kernels, number of LSTM neurons, batch size, and number of fully connected layer neurons; Step 4.3: Train the optimized fault classification algorithm using the data set; Step 4.4: Output the trained fault classification model.

10. The method according to claim 1, wherein the method is characterized by, The method also includes a fault diagnosis method based on the trained fault classification model, comprising the following steps: Collect the vibration signals and electrical signals of the target reactor; Extract fault features based on the vibration signals and electrical signals, including equivalent damping coefficient, equivalent stiffness, and magnetostriction force coefficient; Input the fault features into the fault classification model to obtain the corresponding fault classification result.

Images