Asynchronous motor fault diagnosis device based on model residual error and machine learning integration

Abstract

The invention discloses an asynchronous motor fault diagnosis device based on model residual error and machine learning integration, and relates to the technical field of industrial equipment state monitoring. According to the scheme of the invention, based on a motor state space model, a current estimation value is generated according to an input voltage signal, state space model parameters are updated by using a subspace identification algorithm, a difference value between an actually measured current and an estimated current is converted into a three-phase residual signal for fast Fourier transform, and the three-phase residual signal is converted into a frequency domain signal. According to a fault characteristic frequency theory of the asynchronous motor, extracting a characteristic frequency amplitude in the residual frequency spectrum, calculating a statistical threshold value of each characteristic frequency, when the obtained characteristic frequency amplitude exceeds the statistical threshold value, outputting a fault identification signal, constructing a fault characteristic frequency database by utilizing fault characteristic frequency data, and taking the residual characteristic as an input to determine the fault characteristic frequency of the asynchronous motor. And constructing a neural network model for fault classification by taking the corresponding fault type label as output, and outputting the fault type of the fault identification signal.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial equipment condition monitoring, and in particular to an asynchronous motor fault diagnosis device based on the integration of model residual and machine learning. Background Art

[0002] Induction motors are widely used in various industrial equipment, and fault diagnosis is crucial for ensuring equipment operation and reducing downtime and maintenance costs. Traditional fault diagnosis methods rely primarily on motor current signature analysis (MCSA), which detects fault signature frequencies in the current spectrum to determine the fault type. However, this approach has several limitations. First, fault signature frequencies are typically small in magnitude within the current spectrum and are easily overwhelmed by harmonics and noise, resulting in a low signal-to-noise ratio (SNR), which compromises diagnostic accuracy. Second, acquiring comprehensive fault datasets for training machine learning classifiers is both time-consuming and costly. Furthermore, in practical applications, fault datasets are often limited and fail to cover all possible fault types. This makes classifiers susceptible to failure when encountering unseen faults, potentially leading to serious equipment failures and safety hazards. Therefore, a fault diagnosis device for asynchronous motors based on the integration of model residuals and machine learning is proposed. Summary of the Invention

[0003] The main purpose of this invention is to provide an asynchronous motor fault diagnosis device based on the integration of model residuals and machine learning. By extracting model-based residual signals and dynamically updating the machine learning classifier, this device achieves high signal-to-noise ratio feature extraction and rapid and accurate fault diagnosis. This effectively addresses the problems mentioned in the previous article.

[0004] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0005] The fault diagnosis process of an asynchronous motor fault diagnosis device based on the integration of model residual and machine learning includes the following steps:

[0006] Step 1: Obtain the input voltage signal and input current signal during the operation of the motor to be tested, and generate the current estimation value according to the input voltage signal based on the state space model in the two-phase stationary coordinate system The measured current I abc With estimated current The difference Convert to three-phase residual signal;

[0007] Step 2: The residual signal R at time k abc (k) Perform fast Fourier transform to convert it into frequency domain signal R(f). According to the fault characteristic frequency theory of asynchronous motor, extract the characteristic frequency amplitude f in the residual spectrum fault ;

[0008] Step 3: Collect the historical normal data of the characteristic frequency, calculate the statistical threshold Thresholdf of each characteristic frequency, and when the characteristic frequency amplitude f is obtained fault When the statistical threshold Thresholdf is exceeded, a fault identification signal is output;

[0009] Step 4: Use the fault feature frequency data to build a fault feature frequency database, extract the residual features of the fault feature frequency data, use the residual features as input and the corresponding fault type label as output to build a neural network model for fault classification, train the neural network model, adjust the neural network model parameters according to the training results until its classification accuracy meets the set expected value, and use the trained neural network model to output the fault type of the fault identification signal.

[0010] An asynchronous motor fault diagnosis device based on the integration of model residual and machine learning includes:

[0011] Residual generation module, including:

[0012] A signal acquisition unit for acquiring input voltage signals and input current signals during the operation of the motor to be tested;

[0013] A generator mathematical model unit for generating a current estimation value based on a state space model of the asynchronous motor and an input voltage signal;

[0014] A data updating unit for updating state space model parameters using a subspace identification algorithm;

[0015] a residual calculation unit for converting the difference between the measured current and the estimated current into a three-phase residual signal;

[0016] A spectrum conversion unit for converting the residual signal from the time domain to the frequency domain and extracting the amplitude of the fault characteristic frequency;

[0017] Threshold detection module, including:

[0018] A dynamic threshold unit for calculating the statistical threshold of each frequency amplitude based on historical normal data;

[0019] A fault trigger unit for outputting a fault identification signal when the characteristic frequency amplitude exceeds a threshold;

[0020] Classifier integration module, including:

[0021] A feature storage unit for storing a fault feature frequency database;

[0022] A classifier training unit for constructing a neural network model for fault classification using residual features as input and corresponding fault type labels as output, and for training the neural network model;

[0023] An online update unit for automatically collecting data and updating the classifier model when a fault is detected.

[0024] The apparatus further includes a memory, a processor, and a computer program stored in the memory and executable on the processor.

[0025] Furthermore, in step 1, the state space model of the motor to be tested is described as:

[0026] x(k+1)=A(Θ)x(k)+B(Θ)v αβ (k)

[0027] i αβ (k)=Cx(k)+Dv αβ (k)

[0028] Where x(k+1) and x(k) are the state vectors of the motor to be tested at time k+1 and time k, respectively; wherein the state vectors include the stator current and the rotor flux;

[0029] The state vector is expressed as: x(k) = [i αs (k),i βs (k),λ αr (k),λ βr (k)] T ,i αs (k), i βs (k) is the stator current of the motor to be tested in the α and β coordinate systems at time k; λ αr (k), λ βr (k) are the rotor flux of the motor to be tested in the α and β coordinate systems at time k;

[0030] v αβ (k) is the three-phase voltage signal after Clarke transformation at time k, expressed as: v αβ (k)=[v αs (k),v βs (k)] T ;v αs (k), v βs (k) are the voltages of the motor to be tested at time k in the α and β coordinate systems;

[0031] i αβ (k) is the three-phase current signal after Clarke transformation at time k, expressed as: i αβ (k)=[i αs(k),i βs (k)] T ;i αs (k), i βs (k) are the currents of the motor to be tested at time k in the α and β coordinate systems;

[0032] A(Θ), B(Θ), C, and D are all state-space matrices.

[0033] Furthermore, in step 2, the characteristic frequencies include a broken rotor bar characteristic frequency, a bearing fault characteristic frequency, and a shaft misalignment characteristic frequency.

[0034] Furthermore, in step 1, the conversion process of the three-phase residual signal is:

[0035] Get the measured current I abc With estimated current The difference

[0036] Apply Clarke transformation to convert the difference R abc Converted to the current I in the two-phase stationary coordinate system αβ , where the current I in the α coordinate system is α =2 / 3(R a -1 / 2R b -1 / 2R c ); Current in β coordinate system R a 、R b and R c The measured current I abc With estimated current The difference between phase a, phase b and phase c;

[0037] The current I in the two-phase stationary coordinate system αβ Inverse Clarke transform is used to obtain the three-phase residual signal.

[0038] Furthermore, in step 3, the calculation formula of the statistical threshold Thresholdf is:

[0039] Thresholdf=μ f +3σ f

[0040] Where μ f is the mean value of the characteristic frequency amplitude in the collected historical normal data; σ f is the standard deviation of the characteristic frequency amplitude in the collected historical normal data.

[0041] Furthermore, in step 4, the fault type labels include a broken rotor bar fault type, a bearing fault type, a shaft misalignment fault type, and an unknown fault type.

[0042] Furthermore, when the fault identification signal identified by the trained neural network model does not belong to any of the rotor bar broken fault type, the bearing fault type and the shaft misalignment fault type, the fault identification signal is classified as an unknown fault type.

[0043] Furthermore, in step 4, the expected value is calculated as follows:

[0044]

[0045] Among them, E(Y) represents the expected value of classification accuracy; Q represents the input sample size of the neural network model; f(X k ) is represented as the output function of the neural network model; X k Represented as the kth output sample of the neural network model.

[0046] The present invention has the following beneficial effects:

[0047] Compared with the existing technology, the technical solution of the present invention effectively filters out irrelevant frequencies and improves the significance of fault characteristics through residual signal extraction based on the state space model.

[0048] Compared with the existing technology, the technical solution of the present invention converts the residual signal into a frequency domain signal by using fast Fourier transform, further extracts the amplitude of the fault characteristic frequency, and improves the characteristic quality. It has strong automatic update capability.

[0049] Compared to existing technologies, the present invention automatically labels data and updates the classifier after detecting new faults through threshold detection, reducing manual intervention. When a new fault is detected, data is automatically collected and added to the training set, retraining the classifier model to improve the classifier's adaptability and accuracy.

[0050] Compared with the existing technology, the technical solution of the present invention has strong compatibility, adapts to generators of different powers, and supports deployment in complex industrial environments.

[0051] Compared with existing technologies, the present invention utilizes a subspace identification algorithm based on the motor state-space model to periodically update the state-space model parameters. This adapts to changing motor operating conditions, effectively detecting early failures of asynchronous motors, and reducing equipment downtime and repair costs. This improves equipment reliability and safety, indirectly ensuring production efficiency and economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1Schematic diagram of the overall structure of the asynchronous motor fault diagnosis device based on the integration of model residual and machine learning of the present invention;

[0053] Figure 2 The figure is a schematic diagram of the fault diagnosis process of the asynchronous motor fault diagnosis device based on the integration of model residual and machine learning of the present invention. DETAILED DESCRIPTION

[0054] The present invention will be further described below in conjunction with specific embodiments. The accompanying drawings are only for illustrative purposes and are schematic diagrams rather than actual drawings. They should not be understood as limiting the present invention. In order to better illustrate the specific embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product.

[0055] The specific implementation process of the technical solution of the present invention includes the following steps:

[0056] Step 1: Obtain the input voltage signal and input current signal of the motor to be tested during operation.

[0057] Step 2: Generate current estimates based on the input voltage signal based on the state space model in the two-phase stationary coordinate system The state space model is described as:

[0058] x(k+1)=A(Θ)x(k)+B(Θ)v αβ (k)

[0059] i αβ (k)=Cx(k)+Dv αβ (k)

[0060] Where x(k+1) and x(k) are the state vectors of the motor to be tested at time k+1 and time k, respectively; the state vector includes the stator current and the rotor flux;

[0061] The state vector is expressed as: x(k) = [i αs (k),i βs (k),λ αr (k),λ βr (k)] T ,i αs (k), i βs (k) is the stator current of the motor to be tested in the α and β coordinate systems at time k; λ αr (k), λ βr (k) are the rotor flux of the motor to be tested in the α and β coordinate systems at time k;

[0062] v αβ (k) is the three-phase voltage signal after Clarke transformation at time k, expressed as: vαβ (k)=[v αs (k),v βs (k)] T ;v αs (k), v βs (k) are the voltages of the motor to be tested at time k in the α and β coordinate systems;

[0063] i αβ (k) is the three-phase current signal after Clarke transformation at time k, expressed as: i αβ (k)=[i αs (k),i βs (k)] T ;i αs (k), i βs (k) are the currents of the motor to be tested at time k in the α and β coordinate systems;

[0064] A(Θ), B(Θ), C, and D are all state-space matrices.

[0065] Step 3: Set the measured current I abc With estimated current The difference Converted into a three-phase residual signal; the conversion process is:

[0066] Get the measured current I abc With estimated current The difference

[0067] Apply Clarke transformation to convert the difference R abc Converted to the current I in the two-phase stationary coordinate system αβ , where the current I in the α coordinate system is α =2 / 3(R a -1 / 2R b -1 / 2R c ); Current in β coordinate system R a 、R b and R c The measured current I abc With estimated current The difference between phase a, phase b and phase c;

[0068] The current I in the two-phase stationary coordinate system αβ Inverse Clarke transform is used to obtain the three-phase residual signal.

[0069] Step 4: Use the subspace identification algorithm to update the state space model parameters. The specific process is as follows:

[0070] Step S41: setting the signal acquisition period to T, and acquiring the input voltage signal and the input current signal in the acquisition period T;

[0071] Step S42: using a data structure such as MATLAB's timetable or Python's pandas.DataFrame to store the collected input voltage signal and input current signal in a time series format;

[0072] Step S43: constructing a Hankel matrix based on the collected signal data, wherein the number of rows of the Hankel matrix is ​​usually selected to be 2k, k is a design parameter; the number of columns is N=nT-2k+1; nT is the total number of data points;

[0073] Step S44: Split the Hankel matrix into past input / output matrices U p , Y p and the future input / output matrix U f , Y f ;

[0074] Step S45: performing singular value decomposition on the constructed Hankel matrix to extract estimated values ​​of system matrices A, B, C, and D;

[0075] Step S46: Estimate the state sequence based on the singular value decomposition result and further calculate the parameters of the state space model;

[0076] Step S47: Set a timed task, and repeat the above steps S41 to S46 every other collection period T, and repeat this process.

[0077] The subspace identification algorithm is updated every acquisition period T to adapt to changes in the motor operating conditions.

[0078] Step 5: The residual signal R at time k abc (k) Perform fast Fourier transform to convert it into a frequency domain signal R(f).

[0079] Fast Fourier transform (FFT) is an efficient discrete Fourier transform algorithm that can quickly convert time domain signals into frequency domain signals, facilitating subsequent fault feature extraction.

[0080] Step 6: According to the fault characteristic frequency theory of asynchronous motor, extract the characteristic frequency amplitude f in the residual spectrum fault ;

[0081] It should be noted that the theory of asynchronous motor fault characteristic frequencies is based on the principle that under different fault conditions, specific frequency components will appear in the motor's current or vibration signal. The following is an analysis of some common asynchronous motor faults and their characteristic frequencies:

[0082] Bearing failure

[0083] Bearing failure is one of the most common faults in asynchronous motors, accounting for 41% of motor failures. Bearing failure can cause increased motor vibration and generate specific frequency components in the current or vibration signal. The characteristic frequencies of bearing failure mainly include:

[0084] Inner ring fault frequency: related to motor speed and bearing structural parameters, and can usually be calculated using a formula;

[0085] Outer race fault frequency: Similar to the inner race fault frequency, but the specific value depends on the fault location and motor operating status;

[0086] Rolling element failure frequency: related to the rotation speed and number of rolling elements.

[0087] Broken rotor bar fault

[0088] Broken rotor bars are another common fault of asynchronous motors, accounting for 10% of motor failures. Broken rotor bars can lead to reduced motor efficiency and torque fluctuations, and produce a specific frequency component in the current signal. Its characteristic frequency is: f fault =(1±s)f s ; Among them, s is the motor slip rate, f s is the power supply frequency.

[0089] Stator winding fault

[0090] Stator winding faults primarily include turn-to-turn short circuits and phase-to-phase short circuits, accounting for 37% of motor faults. These faults can cause motor current imbalance and generate specific frequency components in the current signal. For example, a stator winding turn-to-turn short circuit introduces a negative-sequence component in the current, the characteristic frequency of which is related to the motor's operating status and the severity of the fault.

[0091] Air gap eccentricity fault

[0092] Air gap eccentricity is divided into static eccentricity and dynamic eccentricity. Static eccentricity is caused by uneven air gap during motor manufacturing or installation, while dynamic eccentricity is caused by rotor deviation during operation. The characteristic frequency of air gap eccentricity fault can be calculated by the following formula: ec =1 / 2(f s ±f r ); where f s is the supply frequency, f r is the rotor mechanical frequency.

[0093] The theory of fault characteristic frequencies for asynchronous motors provides an effective means for motor fault diagnosis. By analyzing specific frequency components in the motor current or vibration signal, early detection and diagnosis of motor faults can be achieved. Different fault types have different characteristic frequencies, which can be determined through theoretical calculation and experimental verification. Common fault types include bearing faults, broken rotor bars, stator winding faults, and air gap eccentricity. Each fault has its own unique characteristic frequency. The following is a method for calculating the characteristic frequencies of common fault types:

[0094] The characteristic frequency of the broken rotor bar fault is: f fault =(1±s)f s ; Where s is the motor slip rate; f s is the power supply frequency;

[0095] The characteristic frequency of stator winding inter-turn short circuit fault is: f fault =|n±2k(1-s)f s |, where n and k are positive integers; f s is the power supply frequency;

[0096] The characteristic frequency of the air gap eccentricity fault is: f fault =f s ±kf r ;f s is the power supply frequency; f r is the rotor mechanical frequency;

[0097] The characteristic frequency of bearing fault is:

[0098] Inner race fault frequency: f inner =1 / 2f r (1+dp / db×cosβ);

[0099] Outer ring fault frequency: f outer =1 / 2f r (1-dp / db×cosβ);

[0100] Ball failure frequency: f ball =dp / 2db×f r [1-(dp / db×cosβ) 2 ];

[0101] Cage failure frequency: fcage = 1 / 2f r (1-dp / db×cosβ);

[0102] Among them, f r is the motor rotation frequency, db is the bearing rolling element diameter; dp is the cage diameter; β is the contact angle.

[0103] Step 7: Collect historical normal data of characteristic frequencies and calculate the statistical threshold Thresholdf of each characteristic frequency. When the characteristic frequency amplitude f is obtained fault When the statistical threshold Thresholdf is exceeded, a fault identification signal is output; wherein, the calculation formula of the statistical threshold Thresholdf is: Thresholdf=μ f +3σ f Where μ f is the mean value of the characteristic frequency amplitude in the collected historical normal data; σ f is the standard deviation of the characteristic frequency amplitude in the collected historical normal data.

[0104] After new faults are discovered through threshold detection, data is automatically labeled and the classifier is updated to reduce manual intervention.

[0105] Step 8: Use the fault feature frequency data to build a fault feature frequency database, extract the residual features of the fault feature frequency data, use the residual features as input and the corresponding fault type label as output to build a neural network model for fault classification, train the neural network model, and adjust the neural network model parameters according to the training results until its classification accuracy meets the set expected value. The expected value is calculated as follows:

[0106]

[0107] Among them, E(Y) represents the expected value of classification accuracy; Q represents the input sample size of the neural network model; f(X k ) is represented as the output function of the neural network model; X k Represented as the kth output sample of the neural network model.

[0108] When the classification model is constructed using a support vector machine (SVM), the construction process is as follows:

[0109] Model construction: Select a suitable kernel function (such as radial basis function RBF) to build the SVM model;

[0110] Parameter optimization: Use optimization algorithms (such as grid search, particle swarm optimization, etc.) to optimize the SVM parameters (such as penalty parameter C and kernel function parameters).

[0111] Training process: The extracted residual features are used as input and the corresponding fault labels are used as output. The SVM model is trained until its classification accuracy meets the set expected value.

[0112] When the constructed classification model is an artificial neural network ANN, the construction process is:

[0113] Network structure design: Design an ANN structure suitable for fault diagnosis, such as a multi-layer perceptron (MLP) or convolutional neural network (CNN);

[0114] Training algorithm selection: Select an appropriate training algorithm (such as backpropagation algorithm) and optimizer (such as Adam, SGD);

[0115] Training process: The residual features are input into the ANN for training. By adjusting the network weights and biases, the network can learn the mapping relationship between fault features and fault labels.

[0116] Model evaluation and validation

[0117] Performance evaluation: Use the test set to evaluate the trained model and calculate the model's performance indicators such as accuracy, recall rate, and F1 score. When the performance indicators meet the set expectations, the trained artificial neural network model is obtained.

[0118] Step 9: Use the trained neural network model to output the fault type of the fault identification signal. The fault type labels include rotor bar broken fault type, bearing fault type, shaft misalignment fault type, and unknown fault type.

[0119] It should be noted that when the fault identification signal identified by the trained neural network model does not belong to any of the rotor bar broken fault type, bearing fault type and shaft misalignment fault type, the fault identification signal is classified as an unknown fault type.

[0120] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. Asynchronous motor fault diagnosis device based on integration of model residual and machine learning, characterized in that: The troubleshooting process includes the following steps: Step 1: Obtain the input voltage signal and input current signal during the operation of the motor to be tested, and generate the current estimation value according to the input voltage signal based on the state space model in the two-phase stationary coordinate system The state space model parameters are updated using the subspace identification algorithm, and the measured current I abc With estimated current The difference Convert to three-phase residual signal; Step 2: The residual signal R at time k abc (k) Perform fast Fourier transform to convert it into frequency domain signal R(f). According to the fault characteristic frequency theory of asynchronous motor, extract the characteristic frequency amplitude f in the residual spectrum fault ; Step 3: Collect the historical normal data of the characteristic frequency, calculate the statistical threshold Thresholdf of each characteristic frequency, and when the characteristic frequency amplitude f is obtained fault When the statistical threshold Thresholdf is exceeded, a fault identification signal is output; Step 4: Use the fault feature frequency data to build a fault feature frequency database, extract the residual features of the fault feature frequency data, use the residual features as input and the corresponding fault type label as output to build a neural network model for fault classification, train the neural network model, adjust the neural network model parameters according to the training results until its classification accuracy meets the set expected value, and use the trained neural network model to output the fault type of the fault identification signal.

2. The asynchronous motor fault diagnosis device based on integration of model residual and machine learning according to claim 1, characterized in that: In step 1, the state space model of the motor to be tested is described as: x(k+1) A(Θ)x(k)+B(Θ)v αβ (k) i αβ (k)=Cx(k)+Dv αβ (k) Where x(k+1) and x(k) are the state vectors of the motor to be tested at time k+1 and time k, respectively; wherein the state vectors include the stator current and the rotor flux; The state vector is expressed as: x(k) = [i αs (k),i βs (k),λ αr (k),λ βr (k)] T ,i αs (k), i βs (k) is the stator current of the motor to be tested in the α and β coordinate systems at time k; λ αr (k), λ βr (k) are the rotor flux of the motor to be tested in the α and β coordinate systems at time k; v αβ (k) is the three-phase voltage signal after Clarke transformation at time k, expressed as: v αβ (k)=[v αs (k),v βs (k)] T ;v αs (k), v βs (k) are the voltages of the motor to be tested at time k in the α and β coordinate systems; i αβ (k) is the three-phase current signal after Clarke transformation at time k, expressed as: i αβ (k)=[i αs (k),i βs (k)] T ;i αs (k), i βs (k) are the currents of the motor to be tested at time k in the α and β coordinate systems; A(Θ), B(Θ), C, and D are all state-space matrices.

3. The asynchronous motor fault diagnosis device based on integration of model residual and machine learning according to claim 1, characterized in that: In step 2, the characteristic frequencies include a broken rotor bar characteristic frequency, a bearing fault characteristic frequency, and a shaft misalignment characteristic frequency.

4. The asynchronous motor fault diagnosis device based on integration of model residual and machine learning according to claim 1, characterized in that: In step 1, the conversion process of the three-phase residual signal is: Get the measured current I abc With estimated current The difference Apply Clarke transformation to convert the difference R abc Converted to the current I in the two-phase stationary coordinate system αβ , where the current I in the α coordinate system is α =2 / 3(R a -1 / 2R b -1 / 2R c ); Current I in the β coordinate system β =2 / 3 R a 、R b and R c The measured current I abc With estimated current The difference between phase a, phase b and phase c; The current I in the two-phase stationary coordinate system αβ Inverse Clarke transform is used to obtain the three-phase residual signal.

5. The asynchronous motor fault diagnosis device based on integration of model residual and machine learning according to claim 1, characterized in that: In step 3, the calculation formula of the statistical threshold Thresholdf is: Threshold f=μ f +3s f Where μ f is the mean value of the characteristic frequency amplitude in the collected historical normal data; σ f is the standard deviation of the characteristic frequency amplitude in the collected historical normal data.

6. The asynchronous motor fault diagnosis device based on integration of model residual and machine learning according to claim 1, characterized in that: In step 4, the fault type labels include a broken rotor bar fault type, a bearing fault type, a shaft misalignment fault type, and an unknown fault type.

7. The asynchronous motor fault diagnosis device based on integration of model residual and machine learning according to claim 6, characterized in that: When the fault identification signal identified by the trained neural network model does not belong to any of the rotor bar broken fault type, the bearing fault type and the shaft misalignment fault type, the fault identification signal is classified as an unknown fault type.

8. The asynchronous motor fault diagnosis device based on integration of model residual and machine learning according to claim 1, characterized in that: In step 4, the expected value is calculated as: Among them, E(Y) represents the expected value of classification accuracy; Q represents the input sample size of the neural network model; f(X k ) is represented as the output function of the neural network model; X k Represented as the kth output sample of the neural network model.

9. The asynchronous motor fault diagnosis device based on integration of model residual and machine learning according to claim 1, characterized in that: include: Residual generation module, including: A signal acquisition unit for acquiring input voltage signals and input current signals during the operation of the motor to be tested; A generator mathematical model unit for generating a current estimation value based on a state space model of the asynchronous motor and an input voltage signal; A data updating unit for updating state space model parameters using a subspace identification algorithm; a residual calculation unit for converting the difference between the measured current and the estimated current into a three-phase residual signal; A spectrum conversion unit for converting the residual signal from the time domain to the frequency domain and extracting the amplitude of the fault characteristic frequency; Threshold detection module, including: A dynamic threshold unit for calculating the statistical threshold of each frequency amplitude based on historical normal data; A fault trigger unit for outputting a fault identification signal when the characteristic frequency amplitude exceeds a threshold; Classifier integration module, including: A feature storage unit for storing a fault feature frequency database; A classifier training unit for constructing a neural network model for fault classification using residual features as input and corresponding fault type labels as output, and for training the neural network model; An online update unit for automatically collecting data and updating the classifier model when a fault is detected.

10. The asynchronous motor fault diagnosis device based on integration of model residual and machine learning according to claim 9, characterized in that: The device further includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor can implement the steps of any one of claims 1 to 8 when executing the program.

Images