CN118013400A - Motor fault diagnosis method, device, electronic equipment and storage medium - Google Patents

Motor fault diagnosis method, device, electronic equipment and storage medium Download PDF

Info

Publication number
CN118013400A
CN118013400A CN202410413720.4A CN202410413720A CN118013400A CN 118013400 A CN118013400 A CN 118013400A CN 202410413720 A CN202410413720 A CN 202410413720A CN 118013400 A CN118013400 A CN 118013400A
Authority
CN
China
Prior art keywords
training
model
data
fault
sample
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202410413720.4A
Other languages
Chinese (zh)
Inventor
段心林
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
GAC Aion New Energy Automobile Co Ltd
Original Assignee
GAC Aion New Energy Automobile Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by GAC Aion New Energy Automobile Co Ltd filed Critical GAC Aion New Energy Automobile Co Ltd
Priority to CN202410413720.4A priority Critical patent/CN118013400A/en
Publication of CN118013400A publication Critical patent/CN118013400A/en
Pending legal-status Critical Current

Links

Landscapes

  • Testing And Monitoring For Control Systems (AREA)

Abstract

The application provides a motor fault diagnosis method, a device, electronic equipment and a storage medium, wherein the motor fault diagnosis method comprises the steps of obtaining a target sample, wherein the target sample is a sample to be subjected to fault identification; extracting features of the target sample based on a feature extraction model; and inputting the characteristics of the target sample into a classifier model to output the fault category of the target sample based on the classifier model, wherein the data expansion model is constructed based on a quantum coding mechanism. The application can solve the problems of high data acquisition cost and rare samples in the fault diagnosis of the motor of the new energy automobile.

Description

Motor fault diagnosis method, device, electronic equipment and storage medium
Technical Field
The present application relates to the field of motor fault diagnosis, and in particular, to a motor fault diagnosis method, apparatus, electronic device, and storage medium.
Background
With the increasing severity of global energy crisis and environmental pollution problems, new energy automobiles are rapidly growing in market demand and technical development as an effective solution. The new energy automobile, especially the electric automobile, plays an important role in the process of converting the traditional internal combustion engine automobile into electric and intelligent. However, with the acceleration of such transformation, the reliability and maintenance requirements of the motor for one of the core components of the electric vehicle are also increased. The motor is used as the heart of the electric automobile, and the performance of the motor directly influences the running efficiency, the safety and the service life of the whole automobile.
In the prior art, motor fault diagnosis mainly depends on a traditional monitoring method and a fault detection technology. Most of these methods are based on preset failure modes and empirical rules, often not accurately reflecting the actual operating state of the motor, especially in the face of complex failure conditions or rare failure types. In addition, with the continuous progress of motor technology and the increasing complexity of electric vehicle systems, conventional fault diagnosis methods are struggling when dealing with large amounts of high-dimensional and nonlinear data. These methods often require extensive manual intervention, which not only increases maintenance costs, but also reduces diagnostic efficiency and accuracy.
The Chinese patent application No. CN202210665659.3 proposes a three-phase asynchronous motor fault diagnosis method and system based on mixed CNN-LSTM, the diagnosis method comprises the following steps: collecting motor state information and fault type data sets, and correspondingly processing the numerical value of the fault type data sets and the collecting time; dividing fault type data into a training sample and a test sample, and setting a label of the training sample; traversing a data sequence in a training sample, mining deep features and outputting a result; and constructing a CNN-LSTM deep learning model according to the output result, training the time sequence data in the training sample, and confirming model parameters of the CNN-LSTM deep learning model. The diagnostic system includes an acquisition module, a sample grouping module, a calculation module, a modeling module, and a prediction module. The invention can realize the effect of automatically and comprehensively extracting the characteristics of the motor, considers the front-back dependency relationship of the characteristic information, solves the problem of gradient disappearance, and further improves the accuracy of motor fault diagnosis results.
The invention patent of China with the application number of CN202311331451.9 provides a motor fault self-diagnosis method, a device and a storage medium, wherein the method relies on the motor fault self-diagnosis device to collect model parameters of a motor and comprises the following steps: stator resistance observation value R-s, rotor resistance observation value R-R, motor mutual inductance observation value L-m, motor leakage inductance observation value L- (sigma); and judging whether the motor is in a normal error range or not according to the motor model parameter observation value, and describing the motor abnormality and reporting the motor fault if the motor is out of the error range by calculating a stator resistance theoretical value R-s to (- '), a rotor resistance theoretical value R- (R) to (-'), a first motor flux linkage theoretical value psi-1, a rated motor flux linkage observation value psi-2 and a second motor flux linkage theoretical value psi-3. The method can realize that whether the motor has faults or not is judged by detecting and analyzing the state parameters of the motor under the condition of on-load operation or not without other hardware equipment, saves cost and has high precision, and can realize the on-line self-diagnosis of the motor faults.
The Chinese patent application No. CN202310925718.0 proposes a power-on diagnosis control method and device for a motor drive circuit and a vehicle, wherein the method comprises the following steps: after the electronic brake system is initialized, a polling multitasking state is entered; the multiple phases of the power-on diagnosis of the motor driving circuit are respectively executed in different polling periods. After the electronic braking system is initialized, the invention enters a polling multitasking state, and a plurality of stages of power-on diagnosis of the motor driving circuit are respectively executed in different polling periods, so that other polling tasks of the electronic braking system are prevented from being blocked during the power-on diagnosis of the motor driving circuit, and the stability of the electronic braking system is improved.
However, the prior art has the following drawbacks:
1. In the prior art, training of fault diagnosis models may be limited by the number and diversity of available data samples, particularly in rare or complex fault situations.
2. The prior art may have shortcomings in extracting key features from high-dimensional, non-linear, and complex data, which limits the performance of the fault diagnosis model.
3. Due to limitations of training data and feature extraction methods, existing fault diagnosis models may not perform well in terms of generalization ability and robustness, especially in the face of unseen fault types.
4. Classification algorithms used in the prior art may suffer from drawbacks in terms of accuracy, computational efficiency, or both, affecting the overall performance of the fault diagnosis.
Disclosure of Invention
The embodiment of the application aims to provide a motor fault diagnosis method, a motor fault diagnosis device, electronic equipment and a storage medium, which are used for solving the problems of high data acquisition cost and rare samples in the motor fault diagnosis of a new energy automobile.
In a first aspect, the present invention provides a motor fault diagnosis method, wherein the method includes:
Obtaining a target sample, wherein the target sample is a sample to be subjected to fault identification;
extracting features of the target sample based on a feature extraction model;
And inputting the characteristics of the target sample into a classifier model to output the fault category of the target sample based on the classifier model, wherein the classifier model is obtained by training a training sample generated based on a data expansion model, and the data expansion model is constructed based on a quantum coding mechanism.
According to the method, the target sample is a sample to be subjected to fault identification, so that the characteristics of the target sample can be extracted based on the characteristic extraction model, the characteristics of the target sample can be input into the classifier model, the fault category of the target sample can be output based on the classifier model, and the data expansion model is constructed based on a quantum coding mechanism. Compared with the prior art, the method has the advantages that the quantum coding mechanism is combined, the searching capability of the generator in a data space is enhanced, the superposition and entanglement characteristics of quantum states are utilized, the diversity and the authenticity of data generation are improved, the problems of high data acquisition cost and sample scarcity in the fault diagnosis of the motor of the new energy automobile are solved, and richer and real fault data samples are generated, so that the accuracy and the efficiency of the fault diagnosis are improved, the model can adapt to wider running conditions and fault types, the dependence on real fault data is reduced, and the data acquisition cost is reduced.
In an optional implementation manner, the input of the data expansion model is real new energy automobile motor fault data and random noise data, the data expansion model comprises a generator and a discriminator, and the generator is constructed based on the quantum coding mechanism;
and the training process of the data expansion model comprises the following steps:
Initializing parameters of the generator and initializing parameters of the arbiter;
causing the generator to generate the training samples based on the random noise data;
training the discriminator based on the training sample and the real new energy automobile motor fault data, so that the discriminator learns to distinguish the real new energy automobile motor fault data from the training sample and output feedback;
Updating parameters of the generator based on feedback of the discriminator and cycling the iterative training process until the generator and the discriminator reach dynamic balance.
According to the alternative implementation mode, parameters of the generator and parameters of the discriminator can be initialized, the generator can be enabled to generate the training sample based on the random noise data, the discriminator can be trained based on the training sample and the real new energy automobile motor fault data, the discriminator can learn to distinguish the real new energy automobile motor fault data and the training sample and output feedback, and the parameters of the generator can be updated based on the feedback of the discriminator and the iterative training process can be circulated until the generator and the discriminator reach dynamic balance.
In an alternative embodiment, the method further comprises:
optimizing the training process of the data expansion model based on chebyshev polynomials.
This alternative embodiment may optimize the training process of the data expansion model based on chebyshev polynomials.
In an alternative embodiment, the method further comprises:
taking the training sample as input of the feature extraction model, and initializing structural parameters of a neural network so that the neural network extracts features of the training sample based on a Markov chain Monte Carlo algorithm;
dynamically adjusting structural parameters of the neural network and parameters of the Markov chain Monte Carlo algorithm based on the characteristics of the training samples and actual fault diagnosis results;
and iteratively training the neural network based on continuous training periods to obtain the feature extraction model.
The optional implementation manner can extract the characteristics of the training sample based on a Markov chain Monte Carlo algorithm, dynamically adjust the structural parameters of the neural network and the parameters of the Markov chain Monte Carlo algorithm based on the characteristics of the training sample and the actual fault diagnosis result, and further iteratively train the neural network based on continuous training periods to obtain the characteristic extraction model, wherein the Markov chain Monte Carlo method is combined to extract key characteristics from complex data, so that the recognition and processing capacity of the model to nonlinear and high-dimensional characteristics are enhanced.
In an alternative embodiment, the method further comprises:
the feature extraction model is tested and validated based on accuracy.
This alternative embodiment is able to test and verify the feature extraction model based on accuracy.
In an alternative embodiment, the method further comprises:
Taking the characteristics of the training sample as the input of an extreme learning machine;
decomposing the features of the training sample based on a non-negative matrix decomposition algorithm to obtain decomposed features;
Training an extreme learning machine based on decomposition, wherein the weight and bias of a hidden layer in the extreme learning machine are randomly initialized and remain unchanged in the whole training process;
And iteratively training the extreme learning machine based on the training precision decision model to obtain the classifier model.
According to the method, the characteristics of the training samples can be used as input of the extreme learning machine, the characteristics of the training samples can be decomposed based on a non-negative matrix factorization algorithm to obtain decomposed characteristics, the extreme learning machine can be trained based on decomposition, wherein the weights and the biases of the hidden layers in the extreme learning machine are initialized randomly and remain unchanged in the whole training process, and the extreme learning machine can be trained iteratively based on a training precision decision model to obtain the classifier model.
In an alternative embodiment, the method further comprises:
and optimizing the output weight of the extreme learning machine based on quadratic programming.
This alternative embodiment may optimize the output weights of the extreme learning machine based on quadratic programming.
In a second aspect, the present invention provides a motor failure diagnosis apparatus, wherein the apparatus includes:
The acquisition module is used for acquiring a target sample, wherein the target sample is a sample to be subjected to fault identification;
The extraction module is used for extracting the characteristics of the target sample based on the characteristic extraction model;
the fault identification module is used for inputting the characteristics of the target sample into a classifier model to output the fault category of the target sample based on the classifier model, wherein the classifier model is obtained by training a training sample generated based on a data expansion model, and the data expansion model is constructed based on a quantum coding mechanism.
According to the device disclosed by the application, the target sample is a sample to be subjected to fault identification, so that the characteristics of the target sample can be extracted based on the characteristic extraction model, and further the characteristics of the target sample can be input into the classifier model to output the fault category of the target sample based on the classifier model, wherein the data expansion model is constructed based on a quantum coding mechanism. Compared with the prior art, the method has the advantages that the quantum coding mechanism is combined, the searching capability of the generator in a data space is enhanced, the superposition and entanglement characteristics of quantum states are utilized, the diversity and the authenticity of data generation are improved, the problems of high data acquisition cost and sample scarcity in the fault diagnosis of the motor of the new energy automobile are solved, and richer and real fault data samples are generated, so that the accuracy and the efficiency of the fault diagnosis are improved, the model can adapt to wider running conditions and fault types, the dependence on real fault data is reduced, and the data acquisition cost is reduced.
In a third aspect, the present invention provides an electronic device comprising:
a processor; and
A memory configured to store machine readable instructions that, when executed by the processor, perform the motor fault diagnosis method of any of the preceding embodiments.
The electronic apparatus of the third aspect of the present application is capable of obtaining a target sample by executing a motor failure diagnosis method, wherein the target sample is a sample to be subjected to failure recognition, and further capable of extracting features of the target sample based on a feature extraction model, and the data expansion model can be input into a classifier model based on the characteristics of the target sample to output the fault class of the target sample based on the classifier model, wherein the data expansion model is constructed based on a quantum coding mechanism. Compared with the prior art, the method has the advantages that the quantum coding mechanism is combined, the searching capability of the generator in a data space is enhanced, the superposition and entanglement characteristics of quantum states are utilized, the diversity and the authenticity of data generation are improved, the problems of high data acquisition cost and sample scarcity in the fault diagnosis of the motor of the new energy automobile are solved, and richer and real fault data samples are generated, so that the accuracy and the efficiency of the fault diagnosis are improved, the model can adapt to wider running conditions and fault types, the dependence on real fault data is reduced, and the data acquisition cost is reduced.
In a fourth aspect, the present invention provides a storage medium storing a computer program that is executed by a processor to perform the motor fault diagnosis method according to any one of the foregoing embodiments.
The storage medium of the fourth aspect of the present application further enables to obtain a target sample by executing a motor fault diagnosis method, wherein the target sample is a sample to be subjected to fault recognition, and further enables to extract features of the target sample based on a feature extraction model, and further enables to input features of the target sample to a classifier model to output a fault class of the target sample based on the classifier model, wherein the data expansion model is constructed based on a quantum coding mechanism. Compared with the prior art, the method has the advantages that the quantum coding mechanism is combined, the searching capability of the generator in a data space is enhanced, the superposition and entanglement characteristics of quantum states are utilized, the diversity and the authenticity of data generation are improved, the problems of high data acquisition cost and sample scarcity in the fault diagnosis of the motor of the new energy automobile are solved, and richer and real fault data samples are generated, so that the accuracy and the efficiency of the fault diagnosis are improved, the model can adapt to wider running conditions and fault types, the dependence on real fault data is reduced, and the data acquisition cost is reduced.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the embodiments of the present application will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and should not be considered as limiting the scope, and other related drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of a motor fault diagnosis method disclosed in an embodiment of the application;
FIG. 2 is a schematic diagram of a training process of a data expansion model according to an embodiment of the present application;
Fig. 3 is a schematic structural view of a motor fault diagnosis apparatus according to an embodiment of the present application;
Fig. 4 is a schematic structural diagram of an electronic device according to an embodiment of the present application.
Detailed Description
The technical solutions in the embodiments of the present application will be described below with reference to the accompanying drawings in the embodiments of the present application.
Example 1
Referring to fig. 1, fig. 1 is a flow chart of a motor fault diagnosis method disclosed in an embodiment of the application, and as shown in fig. 1, the method in the embodiment of the application includes the following steps:
101. obtaining a target sample, wherein the target sample is a sample to be subjected to fault identification;
102. Extracting features of the target sample based on the feature extraction model;
103. The characteristics based on the target sample are input into a classifier model to output fault categories of the target sample based on the classifier model, wherein the classifier model is obtained through training based on training samples generated by a data expansion model, and the data expansion model is constructed based on a quantum coding mechanism.
According to the method, the target sample is obtained, the target sample is the sample to be subjected to fault recognition, the characteristics of the target sample can be extracted based on the characteristic extraction model, the characteristics of the target sample can be input into the classifier model, the fault category of the target sample is output based on the classifier model, and the data expansion model is constructed based on a quantum coding mechanism. Compared with the prior art, the method has the advantages that the quantum coding mechanism is combined, the searching capability of the generator in a data space is enhanced, the superposition and entanglement characteristics of quantum states are utilized, the diversity and the authenticity of data generation are improved, the problems of high data acquisition cost and sample scarcity in the fault diagnosis of the motor of the new energy automobile are solved, and richer and real fault data samples are generated, so that the accuracy and the efficiency of the fault diagnosis are improved, the model can adapt to wider running conditions and fault types, the dependence on real fault data is reduced, and the data acquisition cost is reduced.
In an embodiment of the present application, the training samples are derived from various sensors and control systems of the new energy automobile motor, and these data include, but are not limited to, parameters such as the rotational speed, temperature, voltage, current, etc. of the motor, wherein each data point includes a plurality of attributes, and in one embodiment, the attributes are as follows:
: motor speed
: Motor temperature
: Motor voltage
: Motor current
: Motor vibration intensity
: Noise level
: Torque moment
: Output power
: Efficiency of
: Fault code
Wherein each attribute isIs one dimension in the time series that describes a particular characteristic of the motor at a particular point in time. In this example, 5 pieces of data are listed in table 1, wherein table 1 is a data detail table included in the training sample:
TABLE 1
In the invention, the labeling of the data is to accurately distinguish the normal running state and various fault states of the motor, and the labeling process is manually labeled, so that the data is manually labeled according to the performance parameters and fault records of the motor. For example, ifExceeds a set threshold and accompanies/>An increase may be noted as an overheat fault.
In one embodiment, the labeling categories are classified into "normal operation" and various fault types, such as "overheat", "overload", "voltage anomaly", and the like. In this embodiment, taking a new energy automobile motor as an example, the collected data includes a rotation speedTemperature/>Etc. At some point in time, an increase in motor temperature to/>, is observedAt the same time vibration intensity/>The data point is artificially marked as "overheat" failure.
The labeled samples form an original training data set for training of a subsequent model.
It should be emphasized that in the complex tasks of diagnosing motor faults of new energy vehicles, the number of data attributes involved may far exceed the number of data attributes in the embodiment of the present invention. These attributes may include sensor data, operating state parameters, environmental factors, etc. of various details, each of potential importance to fault diagnosis. The addition of the attributes can provide more comprehensive information for the model, thereby improving the accuracy and reliability of diagnosis.
In the embodiment of the application, as an optional implementation manner, the input of the data expansion model is real new energy automobile motor fault data and random noise data, the data expansion model comprises a generator and a discriminator, and the generator is constructed based on a quantum coding mechanism;
referring to fig. 2, fig. 2 is a schematic diagram of a training flow of a data expansion model according to an embodiment of the application. As shown in fig. 2, the training process of the data expansion model includes the following steps:
initializing parameters of a generator and parameters of an initialization discriminator;
Causing a generator to generate training samples based on the random noise data;
Training a discriminator based on the training sample and the real new energy automobile motor fault data so that the discriminator learns to distinguish the real new energy automobile motor fault data from the training sample and outputs feedback;
Updating parameters of the generator based on feedback of the discriminator and cycling the iterative training process until the generator and the discriminator reach dynamic balance.
According to the alternative implementation mode, parameters of the generator and parameters of the discriminator are initialized, the generator can generate training samples based on random noise data, the discriminator can be trained based on the training samples and real new energy automobile motor fault data, the discriminator can learn and distinguish the real new energy automobile motor fault data and the training samples and output feedback, and the parameters of the generator can be updated based on the feedback of the discriminator and the iterative training process is circulated until the generator and the discriminator achieve dynamic balance.
For the above alternative embodiments, in particular:
Initializing: first initializing parameters of a quantum code generator and a discriminator, including quantum code generator parameters And discriminant parameter/>. In one embodiment, the quantum code generator parameter/>The initialization mode of (a) is random initialization, and the parameters of the discriminator/>The initialization mode of (2) is random initialization.
Quantum coded data generation: in each iteration, the quantum code generator generates new data samples. With the superposition state of the qubits, each generated sample has a high diversity. Specifically, the process of quantum encoded data generation can be expressed as:
Wherein, Is a quantum code generator.
In one embodiment, a quantum code generatorThe specific implementation of (2) can be expressed as:
Wherein, Representation/>(1 /)Quantum state,/>Is the corresponding complex amplitude, and/>. Superposition of quantum states results in/>High-dimensional data can be expressed.
Training a discriminator: the discriminator receives the data and the real data from the quantum code generator, performs classification training, and learns and distinguishes the generated data and the real data. Specifically, during training of the arbiter, the loss function thereof can be expressed as:
Wherein, Is a discriminator.
Updating a generator: and updating parameters of the generator by using a Chebyshev polynomial optimization method according to feedback of the discriminator, so that the generated data is more similar to real data distribution. Specifically, the generator loss function optimized using chebyshev polynomials can be expressed as:
Wherein, Is chebyshev polynomial,/>Is a regularization parameter. In one embodiment, regularization parameters
Further, chebyshev polynomialsCan be specifically developed as follows:
,/> Is the order of the polynomial. For generator parameters/> The polynomial computes its contribution in higher order terms, adding non-linear adjustments in the optimization process.
Further, the improved generation of parameters of the countermeasure network is updated, and the manner of updating the parameters of the discriminator can be expressed as follows:
the manner in which the generator parameters are updated can be expressed as:
Wherein, And/>Learning rates of the respective discriminators and generators. In one embodiment, the learning rate,/>. In another embodiment, the learning rate is determined in a manner that is dynamically adaptively adjusted, which may be expressed as:
Wherein, And/>Respectively at the/>The learning rate of the arbiter and the generator in the multiple iterations. /(I)AndRepresenting the amount of change in the discriminator and generator loss functions, respectively, in the last iteration. /(I)Is a function of adjusting learning rate based on loss variation,/>Is a preset threshold.
Further, the learning rate adjustment functionThe calculation of (2) can be expressed as:
Wherein, Is a learning rate adjustment factor for controlling the magnitude of increase or decrease in learning rate, manually preset, in one embodiment,/>, of
Further, gradientAnd/>The calculation of (2) can be expressed as:
Wherein, And the derivative of the Chebyshev polynomial on the generator parameter is represented, and a nonlinear adjustment term is added to the algorithm.
And (3) loop iteration: the steps are repeated until the generator and the discriminator reach a dynamic balance, namely the discriminator cannot effectively distinguish the real data from the generated data. During initial training, generated fault dataPossibly with real data/>There is a large difference, but as training proceeds, the generated data gradually approaches the distribution of the real data. The classification accuracy of the discriminator is higher in the early stage, but with the improvement of the generator, the accuracy gradually decreases, reflecting that dynamic balance is achieved between the generator and the discriminator.
Finally, the output of the improved generation countermeasure network algorithm is:
: synthetic fault data generated by the quantum code generator;
By combining quantum coding and chebyshev theory, the invention obviously improves the efficiency and quality of data generation. In the application of the motor fault diagnosis of the new energy automobile, the improved generation countermeasure network can generate more diversified and real fault data samples, the data samples not only comprise common fault types, but also simulate rare or complex fault conditions, the robustness and the generalization capability of the fault diagnosis model are greatly enhanced, and the generalization capability and the accuracy of the fault diagnosis model are improved. Particularly, under the condition of scarce samples, the method can effectively supplement training data and improve the robustness of the model.
In an embodiment of the present application, as an optional implementation manner, the method of the embodiment of the present application further includes the following steps:
and optimizing a training process of the data expansion model based on the chebyshev polynomials.
This alternative embodiment may optimize the training process of the data expansion model based on chebyshev polynomials.
In an embodiment of the present application, as an optional implementation manner, the method of the embodiment of the present application further includes the following steps:
Taking the training sample as the input of a feature extraction model, and initializing structural parameters of the neural network so that the neural network extracts the features of the training sample based on a Markov chain Monte Carlo algorithm;
Based on the characteristics of the training samples and the actual fault diagnosis result, dynamically adjusting the structural parameters of the neural network and the parameters of the Markov chain Monte Carlo algorithm;
and iteratively training the neural network based on continuous training periods to obtain a feature extraction model.
The method and the device can extract the characteristics of the training sample based on the Markov chain Monte Carlo algorithm, dynamically adjust the structural parameters of the neural network and the parameters of the Markov chain Monte Carlo algorithm based on the characteristics of the training sample and the actual fault diagnosis result, further iteratively train the neural network in a continuous training period to obtain a characteristic extraction model, wherein the Markov chain Monte Carlo method is combined to extract key characteristics from complex data, and the recognition and processing capacity of the model to nonlinear and high-dimensional characteristics are enhanced.
Aiming at the optional implementation mode, the data feature extraction is carried out on the expanded new energy automobile motor fault data set by combining with the improvement of the Markov chain Monte Carlo method. The neural network algorithm based on structured learning can extract key features from complex new energy automobile motor data, and the network structure can more effectively identify and process nonlinear and high-dimensional features by learning the inherent structural relationship of the data. In addition, the algorithm fuses the improvement of the Markov chain Monte Carlo method to optimize probability inference in the feature extraction process, so that the stability and accuracy of the algorithm are improved. Specifically:
In an embodiment of the present application, as an optional implementation manner, the method of the embodiment of the present application further includes the following steps:
and (5) testing and verifying the feature extraction model based on the accuracy.
This alternative embodiment can test and verify feature extraction models based on accuracy.
The inputs of the neural network algorithm based on structured learning are:
the expanded new energy automobile motor data set: comprises running parameters and fault indexes of a new energy automobile motor, such as motor temperature, rotating speed, vibration signals and the like, and each data sample Is regarded as feature vector/>And corresponding failure tags/>
Further, the training flow of the neural network algorithm based on structured learning is as follows:
Initializing: and initializing structural parameters of the neural network according to specific fault diagnosis requirements of the new energy automobile motor. In one embodiment, a set of data sets for a new energy vehicle motor Wherein/>Represents the number of samples, each/>Comprises motor operation parameters and fault indexes. Meanwhile, initializing network parameters includes weighting matrix/>And bias/>Is performed in the initialization of the (c). In the present embodiment, the weight/>And bias/>Random initialization by standard normal distribution.
Feature learning and optimization: and carrying out deep analysis on the data through a structural chemistry neural network, and extracting key features. In the process, probability inference is carried out by using a Markov chain Monte Carlo method so as to ensure that the extracted characteristics can accurately reflect the running state and potential faults of the motor. Specifically, the output function of the structured learning neural network is defined as. For each feature/>Computing their characteristic representations by neural networks
Further, according to Markov chain Monte Carlo inference theory, defining the state transition probability asWherein/>Is a subsequent state, using the Monte Carlo method to estimate/>Is a distribution of (a).
And (3) self-adaptive adjustment: according to the extracted characteristics and the actual fault diagnosis result, the network parameters and key parameters in the Markov chain Monte Carlo method, such as transition probability and state updating rules, are dynamically adjusted so that the model can be self-optimized and adapt to the continuously changed data characteristics. Specifically, it is provided withAs a loss function, is used to evaluate the performance of the current model. In updating the parameters of the neural network, the manner of updating the weights of the neural network according to the gradient descent can be expressed as:
The manner in which the bias is updated can be expressed as:
Wherein, Is the learning rate. In one embodiment, the learning rate/>Set to 0.01 to ensure a stable learning process.
Further, the method comprises the steps of,Is the gradient of the loss function to the weight, and the calculation mode can be expressed as:
Is the gradient of the loss function versus bias, and the calculation method can be expressed as:
further according to And/>Bias-adaptive adjustment of state transition probabilities/>. Specifically, according to the state transition probability/>, in Markov chain Monte Carlo inferenceThe adjustment is performed, and the calculation mode can be expressed as follows:
Wherein, Is an energy function used to evaluate the probability density of the feature states. In one embodiment, the energy function is the inverse of the loss function.
Iteration and feedback: the feature learning and adaptive adjustment process is repeated continuously during successive training cycles, each iteration being modified based on the previous results. Meanwhile, a feedback mechanism is introduced, and algorithm parameters are further optimized according to the performance of the model in actual fault diagnosis. Specifically, it is provided withFor iteration number,/>And/>Respectively is/>Feature representation and loss after a number of iterations. During the iterative loop, for/>Perform update/>And/>Operating while adjustingAnd/>. In one embodiment, the number of iterations/>Set to 100.
Further, after each update, a performance evaluation is performed, and the overall loss value is calculated, and the calculation mode can be expressed as:
further, the loss function The calculation of (2) can be expressed as:
Wherein, Representing the euclidean norm.
In the training process, as the iteration number increases, the characteristic representationAnd the motor is gradually stable, and the running state and potential faults of the motor are better reflected. In addition, in the training process, the model performance index is the accuracy/>Calculated by a preset Softmax classification function, the accuracy/>The training performance is gradually improved, and the characteristic extraction model is shown to enhance the extraction effect of the new energy automobile motor data.
Model evaluation: finally, through a series of tests and verification, the accuracy and stability of the model are evaluated, and the extracted characteristics are ensured to meet the actual requirements of the motor fault diagnosis of the new energy automobile. Specifically, in the case of performing accuracy and stability evaluation of the final model, the manner of accuracy calculation may be expressed as:
Wherein, Is an indicator function.
Further, in the stability evaluation, the manner of calculating the characteristic representation change rate of different iteration rounds can be expressed as:
Stability index Will gradually decrease as the iteration proceeds, indicating that the model is more and more robust to changes in the input data.
The output of the neural network algorithm based on structured learning is:
Feature representation : The extracted feature representation of each data sample feature typically has a higher dimension.
The algorithm of the invention not only can effectively process the data of the new energy automobile motor and extract key characteristics, but also can remarkably improve the accuracy and stability of the subsequent fault diagnosis through the self-adaptive adjustment and iterative feedback mechanism.
In an embodiment of the present application, as an optional implementation manner, the method of the embodiment of the present application further includes the following steps:
taking the characteristics of the training sample as the input of the extreme learning machine;
decomposing the features of the training sample based on a non-negative matrix decomposition algorithm to obtain decomposed features;
Training the extreme learning machine based on decomposition, wherein the weight and bias of a hidden layer in the extreme learning machine are randomly initialized and kept unchanged in the whole training process;
and (5) iteratively training the extreme learning machine based on the training precision decision model to obtain the classifier model.
According to the method and the device, the characteristics of the training samples can be used as input of the extreme learning machine, the characteristics of the training samples can be decomposed based on a non-negative matrix factorization algorithm to obtain decomposed characteristics, the extreme learning machine can be trained based on decomposition, wherein the weights and the biases of the hidden layers in the extreme learning machine are initialized randomly and remain unchanged in the whole training process, and the extreme learning machine can be trained iteratively based on a training precision decision model to obtain a classifier model.
In an embodiment of the present application, as an optional implementation manner, the method of the embodiment of the present application further includes the following steps:
And optimizing the output weight of the extreme learning machine based on the quadratic programming.
This alternative embodiment may optimize the output weights of the limit learning machine based on quadratic programming.
For the above optional embodiment, specifically, S401, the input of the non-negative matrix factorization-based extreme learning machine classification algorithm is:
Feature data : And extracting the obtained characteristic data by the characteristic extraction model, wherein the data comprise the running state and possible fault characteristics of the motor of the new energy automobile.
Further, the training flow of the extreme learning machine classification algorithm based on the non-negative matrix factorization is as follows:
And (3) feature decomposition: and decomposing the data after feature extraction by utilizing non-negative matrix factorization, and extracting a key mode. Let the feature extracted data set be The process of non-negative matrix factorization can be expressed as:
Wherein, ,/>For the number of samples,/>Is the characteristic number/>And/>,/>Is the cardinality of the non-negative matrix factorization. In one embodiment, the cardinality/>, of the non-negative matrix factorizationSet to 10, i.e., 10 important patterns or components are extracted from the original features.
Further, the method comprises the steps of,And/>Obtained by optimizing the following objective function:
Wherein, Indicating the Frobenius norm.
Further, supplement: the optimization problem is updated by iterationAnd/>To solve,/>And/>The update rule of (2) can be expressed as:
;/>
Wherein, For learning rate, preset by human, in one embodiment, set/>
Training an extreme learning machine: and training an extreme learning machine classifier by using the decomposed features. In extreme learning machines, the weights and biases of the hidden layers are randomly initialized and remain unchanged throughout the training process. Specifically, the manner in which the hidden layer weights and biases are randomly initialized can be expressed as:
Wherein, Is the hidden layer neuron number. In one embodiment, the weights and biases of the hidden layer are randomly initialized from a uniform distribution, the number of hidden layer neurons/>Set to 50.
Further, hiding the output of the layerCan be expressed as:
Output weight optimization: the output weights of the limit learning machine are optimized using quadratic programming, with the objective of minimizing classification errors while taking into account regularization terms to prevent overfitting. Specifically, the objective function may be expressed as:
Wherein, For the output layer weight,/>Is a label matrix,/>Is a regularization parameter. In one embodiment, regularization parameters/>Set to 0.01 to balance model complexity and risk of overfitting.
Further, the optimization problem can be further converted into a standard quadratic programming problem and solved using Lagrangian multiplier optimization algorithm, i.e., the solution of the problemThis can be obtained by solving the following equation:
Wherein, Is an identity matrix.
During training, the data will be transformed from the original high-dimensional features to a form that is more directly related to the motor fault type by a combination of non-negative matrix factorization and an extreme learning machine.
Model evaluation: finally, the performance of the classifier is evaluated, and whether the model stops iterating or not is determined by training accuracy on the training samples. In one embodiment, when the training accuracy of the model is greater than 95%, the iteration is stopped, that is, the classifier model training is completed.
Through nonnegative matrix factorization, the algorithm can reveal deeper data structures and modes in characteristic representation, and is particularly important for fault diagnosis of a motor of a complex new energy automobile. The rapid learning capability of the extreme learning machine is combined with the characteristic expression capability of non-negative matrix factorization, so that the classification efficiency and accuracy are remarkably improved, the output weight of the extreme learning machine is further optimized, various potential faults in the motor can be accurately identified, early prevention and timely maintenance are facilitated, and the classifier can accurately identify motor faults of different types.
The output of the extreme learning machine classification algorithm based on non-negative matrix factorization is:
Classification result: the model outputs a classification of the motor fault type for identifying the specific faults that may exist in the motor. In one embodiment, the fault categories may be overheating, vibration anomalies, and the like.
Example two
Referring to fig. 3, fig. 3 is a schematic structural diagram of a motor fault diagnosis device according to an embodiment of the present application, and as shown in fig. 3, the device according to the embodiment of the present application includes the following functional modules:
an obtaining module 201, configured to obtain a target sample, where the target sample is a sample to be subjected to fault identification;
an extraction module 202, configured to extract features of the target sample based on the feature extraction model;
The fault recognition module 203 is configured to input the characteristics of the target sample to a classifier model based on the characteristics of the target sample, so as to output a fault class of the target sample based on the classifier model, where the classifier model is obtained by training a training sample generated based on a data expansion model, and the data expansion model is constructed based on a quantum coding mechanism.
According to the device provided by the embodiment of the application, the target sample is a sample to be subjected to fault recognition, so that the characteristics of the target sample can be extracted based on the characteristic extraction model, and further the characteristics of the target sample can be input into the classifier model to output the fault category of the target sample based on the classifier model, wherein the data expansion model is constructed based on a quantum coding mechanism. Compared with the prior art, the method has the advantages that the quantum coding mechanism is combined, the searching capability of the generator in a data space is enhanced, the superposition and entanglement characteristics of quantum states are utilized, the diversity and the authenticity of data generation are improved, the problems of high data acquisition cost and sample scarcity in the fault diagnosis of the motor of the new energy automobile are solved, and richer and real fault data samples are generated, so that the accuracy and the efficiency of the fault diagnosis are improved, the model can adapt to wider running conditions and fault types, the dependence on real fault data is reduced, and the data acquisition cost is reduced.
Example III
Referring to fig. 4, fig. 4 is a schematic structural diagram of an electronic device according to an embodiment of the present application, and as shown in fig. 4, the electronic device according to the embodiment of the present application includes:
A processor 301; and
A memory 302 configured to store machine readable instructions that, when executed by a processor, perform a motor fault diagnosis method as in any of the preceding embodiments.
According to the electronic equipment provided by the embodiment of the application, the target sample can be obtained by executing the motor fault diagnosis method, wherein the target sample is the sample to be subjected to fault identification, the characteristics of the target sample can be extracted based on the characteristic extraction model, the characteristics of the target sample can be input into the classifier model, the fault category of the target sample can be output based on the classifier model, and the data expansion model is constructed based on the quantum coding mechanism. Compared with the prior art, the method has the advantages that the quantum coding mechanism is combined, the searching capability of the generator in a data space is enhanced, the superposition and entanglement characteristics of quantum states are utilized, the diversity and the authenticity of data generation are improved, the problems of high data acquisition cost and sample scarcity in the fault diagnosis of the motor of the new energy automobile are solved, and richer and real fault data samples are generated, so that the accuracy and the efficiency of the fault diagnosis are improved, the model can adapt to wider running conditions and fault types, the dependence on real fault data is reduced, and the data acquisition cost is reduced.
Example IV
An embodiment of the present application provides a storage medium storing a computer program that is executed by a processor to perform the motor failure diagnosis method according to any one of the foregoing embodiments.
The storage medium of the embodiment of the application can further acquire the target sample by executing the motor fault diagnosis method, wherein the target sample is the sample to be subjected to fault identification, further can extract the characteristics of the target sample based on the characteristic extraction model, further can be input into the classifier model based on the characteristics of the target sample so as to output the fault category of the target sample based on the classifier model, and the data expansion model is constructed based on the quantum coding mechanism. Compared with the prior art, the method has the advantages that the quantum coding mechanism is combined, the searching capability of the generator in a data space is enhanced, the superposition and entanglement characteristics of quantum states are utilized, the diversity and the authenticity of data generation are improved, the problems of high data acquisition cost and sample scarcity in the fault diagnosis of the motor of the new energy automobile are solved, and richer and real fault data samples are generated, so that the accuracy and the efficiency of the fault diagnosis are improved, the model can adapt to wider running conditions and fault types, the dependence on real fault data is reduced, and the data acquisition cost is reduced.
In the embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other manners. The above-described apparatus embodiments are merely illustrative, for example, the division of units is merely a logical function division, and there may be other manners of division in actual implementation, and for example, multiple units or components may be combined or integrated into another system, or some features may be omitted, or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be through some communication interface, device or unit indirect coupling or communication connection, which may be in electrical, mechanical or other form.
Further, the units described as separate units may or may not be physically separate, and units displayed as units may or may not be physical units, may be located in one place, or may be distributed over a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
Furthermore, functional modules in various embodiments of the present application may be integrated together to form a single portion, or each module may exist alone, or two or more modules may be integrated to form a single portion.
It should be noted that the functions, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method of the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM) random access Memory (Random Access Memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
In this document, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions.
The above embodiments of the present application are only examples, and are not intended to limit the scope of the present application, and various modifications and variations will be apparent to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims (10)

1. A method of diagnosing a motor fault, the method comprising:
Obtaining a target sample, wherein the target sample is a sample to be subjected to fault identification;
extracting features of the target sample based on a feature extraction model;
And inputting the characteristics of the target sample into a classifier model to output the fault category of the target sample based on the classifier model, wherein the classifier model is obtained by training a training sample generated based on a data expansion model, and the data expansion model is constructed based on a quantum coding mechanism.
2. The method of claim 1, wherein the input of the data expansion model is real new energy automobile motor fault data and random noise data, the data expansion model comprises a generator and a discriminator, and the generator is constructed based on the quantum encoding mechanism;
and the training process of the data expansion model comprises the following steps:
Initializing parameters of the generator and initializing parameters of the arbiter;
causing the generator to generate the training samples based on the random noise data;
training the discriminator based on the training sample and the real new energy automobile motor fault data, so that the discriminator learns to distinguish the real new energy automobile motor fault data from the training sample and output feedback;
Updating parameters of the generator based on feedback of the discriminator and cycling the iterative training process until the generator and the discriminator reach dynamic balance.
3. The method of claim 2, wherein the method further comprises:
optimizing the training process of the data expansion model based on chebyshev polynomials.
4. The method of claim 2, wherein the method further comprises:
taking the training sample as input of the feature extraction model, and initializing structural parameters of a neural network so that the neural network extracts features of the training sample based on a Markov chain Monte Carlo algorithm;
dynamically adjusting structural parameters of the neural network and parameters of the Markov chain Monte Carlo algorithm based on the characteristics of the training samples and actual fault diagnosis results;
and iteratively training the neural network based on continuous training periods to obtain the feature extraction model.
5. The method of claim 4, wherein the method further comprises:
the feature extraction model is tested and validated based on accuracy.
6. The method of claim 4, wherein the method further comprises:
Taking the characteristics of the training sample as the input of an extreme learning machine;
decomposing the features of the training sample based on a non-negative matrix decomposition algorithm to obtain decomposed features;
Training an extreme learning machine based on decomposition, wherein the weight and bias of a hidden layer in the extreme learning machine are randomly initialized and remain unchanged in the whole training process;
And iteratively training the extreme learning machine based on the training precision decision model to obtain the classifier model.
7. The method of claim 6, wherein the method further comprises:
and optimizing the output weight of the extreme learning machine based on quadratic programming.
8. A motor fault diagnosis apparatus, characterized in that the apparatus comprises:
The acquisition module is used for acquiring a target sample, wherein the target sample is a sample to be subjected to fault identification;
The extraction module is used for extracting the characteristics of the target sample based on the characteristic extraction model;
the fault identification module is used for inputting the characteristics of the target sample into a classifier model to output the fault category of the target sample based on the classifier model, wherein the classifier model is obtained by training a training sample generated based on a data expansion model, and the data expansion model is constructed based on a quantum coding mechanism.
9. An electronic device, comprising:
a processor; and
A memory configured to store machine readable instructions that, when executed by the processor, perform the motor fault diagnosis method of any of claims 1-7.
10. A storage medium storing a computer program that is executed by a processor to perform the motor failure diagnosis method according to any one of claims 1 to 7.
CN202410413720.4A 2024-04-08 2024-04-08 Motor fault diagnosis method, device, electronic equipment and storage medium Pending CN118013400A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202410413720.4A CN118013400A (en) 2024-04-08 2024-04-08 Motor fault diagnosis method, device, electronic equipment and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202410413720.4A CN118013400A (en) 2024-04-08 2024-04-08 Motor fault diagnosis method, device, electronic equipment and storage medium

Publications (1)

Publication Number Publication Date
CN118013400A true CN118013400A (en) 2024-05-10

Family

ID=90943434

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202410413720.4A Pending CN118013400A (en) 2024-04-08 2024-04-08 Motor fault diagnosis method, device, electronic equipment and storage medium

Country Status (1)

Country Link
CN (1) CN118013400A (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111540193A (en) * 2020-03-13 2020-08-14 华南理工大学 Traffic data restoration method for generating countermeasure network based on graph convolution time sequence
CN113887136A (en) * 2021-10-08 2022-01-04 东北大学 Improved GAN and ResNet based electric vehicle motor bearing fault diagnosis method
CN115830383A (en) * 2022-12-08 2023-03-21 昆明理工大学 Motor bearing fault diagnosis model training method and device under small sample condition
CN117391188A (en) * 2023-11-02 2024-01-12 江苏新巢天诚智能技术有限公司 Internet of things model training method based on federal AI calculation

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111540193A (en) * 2020-03-13 2020-08-14 华南理工大学 Traffic data restoration method for generating countermeasure network based on graph convolution time sequence
CN113887136A (en) * 2021-10-08 2022-01-04 东北大学 Improved GAN and ResNet based electric vehicle motor bearing fault diagnosis method
CN115830383A (en) * 2022-12-08 2023-03-21 昆明理工大学 Motor bearing fault diagnosis model training method and device under small sample condition
CN117391188A (en) * 2023-11-02 2024-01-12 江苏新巢天诚智能技术有限公司 Internet of things model training method based on federal AI calculation

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
YUXIAN ZHANG ET AL.: "Fuzzy rule-based classification system using multi-population quantum evolutionary algorithm with contradictory rule reconstruction", 《APPLIED INTELLIGENCE 》, 21 May 2019 (2019-05-21), pages 1 - 7 *
李敏 等: "基于样本扩充的二维卷积网络轴承故障诊断", 《信息技术与信息化》, 31 December 2022 (2022-12-31), pages 1 - 8 *

Similar Documents

Publication Publication Date Title
Wu et al. An integrated ensemble learning model for imbalanced fault diagnostics and prognostics
CN106980822B (en) A kind of rotary machinery fault diagnosis method based on selective ensemble study
CN111914873A (en) Two-stage cloud server unsupervised anomaly prediction method
Yan et al. Deep order-wavelet convolutional variational autoencoder for fault identification of rolling bearing under fluctuating speed conditions
Ren et al. A systematic review on imbalanced learning methods in intelligent fault diagnosis
CN114297936A (en) Data anomaly detection method and device
CN113312447B (en) Semi-supervised log anomaly detection method based on probability label estimation
Liu et al. Wind turbine fault detection based on deep residual networks
Zhang et al. A novel data-driven method based on sample reliability assessment and improved CNN for machinery fault diagnosis with non-ideal data
CN111241673A (en) Health state prediction method for industrial equipment in noisy environment
CN113962259A (en) Multi-mode double-layer fault diagnosis method for fuel cell system
CN114676742A (en) Power grid abnormal electricity utilization detection method based on attention mechanism and residual error network
Zhang et al. End-to-end unsupervised fault detection using a flow-based model
Jiang et al. Fault diagnosis of planetary gearbox based on motor current signal analysis
CN113376516A (en) Medium-voltage vacuum circuit breaker operation fault self-diagnosis and early-warning method based on deep learning
Liu et al. Causal graph attention network with disentangled representations for complex systems fault detection
CN112966770B (en) Fault prediction method and device based on integrated hybrid model and related equipment
Maurya et al. Intelligent hybrid scheme for health monitoring of degrading rotary machines: An adaptive fuzzy c-means coupled with 1-D CNN
CN114358116A (en) Oil-immersed transformer fault diagnosis method and system and readable storage medium
CN113762967A (en) Risk information determination method, model training method, device, and program product
CN116821730A (en) Fan fault detection method, control device and storage medium
CN112595918A (en) Low-voltage meter reading fault detection method and device
CN115891741B (en) Remote fault early warning method and device suitable for electric automobile charging process
CN117131022A (en) Heterogeneous data migration method of electric power information system
CN118013400A (en) Motor fault diagnosis method, device, electronic equipment and storage medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination