CN113935460B - Intelligent diagnosis method for mechanical faults under unbalanced-like data set - Google Patents
Intelligent diagnosis method for mechanical faults under unbalanced-like data set Download PDFInfo
- Publication number
- CN113935460B CN113935460B CN202111136682.5A CN202111136682A CN113935460B CN 113935460 B CN113935460 B CN 113935460B CN 202111136682 A CN202111136682 A CN 202111136682A CN 113935460 B CN113935460 B CN 113935460B
- Authority
- CN
- China
- Prior art keywords
- data
- fault
- model
- training
- encoder
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
- G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The application discloses an intelligent diagnosis method for mechanical faults under unbalanced data sets, which comprises the following steps: step (1), data preprocessing: converting the mechanical vibration signal to the frequency domain and normalizing the amplitude to the [0,1] range; step (2), building a model: combining an automatic encoder with a generation countermeasure network, and building a data generation model; step (3), model training: training the data generation model by using fault data according to a preset loss function and an optimization algorithm; step (4), data generation: the low-dimensional characteristics of fault data learned in training by the data generation model are utilized, and fault data of corresponding classes are generated after repeated interpolation and noise addition, so that various data balances are realized; step (5), fault diagnosis: training a preset fault diagnosis model by using the class balance data set, and performing intelligent diagnosis on mechanical faults by using the trained fault diagnosis model. And (3) utilizing an automatic encoder to generate the combination of the countermeasure network to realize mechanical fault diagnosis.
Description
Technical Field
The application relates to the field of intelligent fault diagnosis, in particular to an intelligent mechanical fault diagnosis method under an unbalanced data set.
Background
As rotary mechanical devices continue to develop in an intelligent, precise, and complex direction, the structure of the mechanical devices is becoming more complex and compact. In the service process of the mechanical equipment, once a certain part fails, the operation of the whole mechanical equipment can be influenced, and even safety accidents are caused. In order to ensure healthy operation of mechanical equipment, deep learning theory is gradually applied to intelligent diagnosis of mechanical faults as the latest research results in the fields of pattern recognition and machine learning. Compared with the traditional fault diagnosis method, the intelligent diagnosis model based on deep learning utilizes the deep network model to adaptively extract effective fault characteristics from signals, has high diagnosis efficiency and does not depend on the signal processing experience of operators, and is widely paid attention to.
Models currently in common use in intelligent diagnosis of mechanical faults include Convolutional Neural Networks (CNNs), deep Belief Networks (DBNs), residual networks (res net), and the like. In the training process of the models, a large amount of historical data sets are often required to be input as training samples, so that the corresponding relation between the data and the health state category is established. Although the mechanical equipment fails to bring great potential safety hazard to the operation of the equipment, the failure is an occasional event, and the equipment cannot operate for a long time in a failure state, so that the normal state data are more, the failure state data are less, and the class imbalance problem of a data set is caused. The imbalance of the normal class and the fault class presents great difficulty and challenges for identifying the mechanical health state, and the unbalanced class data set easily causes the performance of the diagnosis model to be reduced, namely, the model is easy to be over-fitted to normal signals with a large number of samples, and under-fit to fault signals with a small number of samples. In addition, due to the rarity of fault samples, the model can easily learn some redundant and even irrelevant features in the process of extracting fault data features, and the features reduce the generalization capability of the model.
In order to solve the problem of performance degradation of the intelligent diagnosis model of mechanical faults caused by unbalance of the class, a dynamic weight method and a data generation method are commonly used. The dynamic weighting method gives more attention to a smaller number of faulty samples by adjusting the weight parameters in the network, thereby improving the under-fitting problem for the faulty samples. The data generation rule is to generate a new sample of the same type by using a small amount of fault data to expand the fault sample, balance the fault data with normal data, and train an intelligent diagnosis model by using the balanced data set. The traditional data generation method includes a few types of synthesis upsampling technology (SMOTE), adaptive synthesis sampling (ADASYN) and the like.
The traditional technology has the following technical problems:
in the intelligent diagnosis of mechanical faults under an actual unbalanced data set, the dynamic weighting method needs to dynamically adjust the weights according to the unbalance rate between normal and fault samples, so the method is suitable for being applied to the condition that the unbalance rate is known. And when the data is in extreme class unbalance, the dynamic weighting method is easily interfered by redundant features in a small number of fault samples, so that the model is over-fitted, and the accuracy of fault diagnosis is reduced. The data generation method enables various types in the data set to reach balance through upsampling a small number of fault signals, and fundamentally solves the phenomenon of unbalanced types. However, the mechanical structure is complex, the vibration signal of the vibration sensor is nonlinear, the vibration signal of the vibration sensor has stronger background noise under actual working conditions, and the vibration sensor shows obvious nonstationary characteristics under fault conditions. The traditional data generation method does not learn the distribution characteristics of the data, directly generates signals in time domain signals through interpolation technology, is easily interfered by measurement noise components, has low quality of the generated data, and is also easy to cause performance degradation of an intelligent diagnosis model.
Disclosure of Invention
Aiming at the problems that the application scene of a dynamic weight method is limited, the traditional data generation method is easy to be interfered by noise and the quality of generated data is low, the application provides a novel data generation method, which is based on a deep neural network, learns the low-dimensional distribution characteristic of fault signals through the combination of an automatic encoder and a generation countermeasure network, generates fault data according to the low-dimensional characteristic, thereby avoiding the interference of noise, obtaining high-quality generated data and realizing high-performance intelligent diagnosis of mechanical faults by utilizing the balanced data set.
In order to solve the technical problems, the application provides an intelligent diagnosis method for mechanical faults under an unbalanced-like data set, which comprises the following steps:
step (1), data preprocessing: converting the mechanical vibration signal to the frequency domain and normalizing the amplitude to the [0,1] range;
step (2), building a model: combining an automatic encoder with a generation countermeasure network, and building a data generation model;
step (3), model training: training the data generation model by using fault data according to a preset loss function and an optimization algorithm;
step (4), data generation: the low-dimensional characteristics of fault data learned in training by the data generation model are utilized, and fault data of corresponding classes are generated after repeated interpolation and noise addition, so that various data balances are realized;
step (5), fault diagnosis: training a preset fault diagnosis model by using the class balance data set, and performing intelligent diagnosis on mechanical faults by using the trained fault diagnosis model.
In one embodiment, in step (2), the automatic encoder is composed of an encoder and a decoder, the generation countermeasure network is composed of a generator and a discriminator, and the decoder is the generator; the automatic encoder learns the low-dimensional characteristics of the input data, namely the true data, and outputs the generated data, namely the false data, consistent with the distribution characteristics of the input data through the decoder by the low-dimensional characteristics and class labels thereof; and the discriminator in the generated countermeasure network respectively carries out true and false discrimination and category classification on the input data and the generated data.
In one embodiment, the encoder, the decoder, and the arbiter each comprise a construction by one of a deep convolutional network, a deep confidence network, and a residual network.
In one embodiment, in step (3), the preset loss function includes a mean square error loss function between the generator generated data and the encoder input data, a cross entropy classification loss function of the arbiter for true and false data, a waserstein distance or a binary cross entropy loss function of the arbiter for true and false authentication of the data, and a mean square error loss function between the encoder output feature and an intermediate implicit feature of the arbiter.
In one embodiment, in the step (3), the preset optimization algorithm includes, but is not limited to, one of a random gradient descent method (SGD), a random gradient descent with Momentum (Momentum), a Nesterov Momentum method, an adagard algorithm, and an adaptive moment estimation method (Adam).
In one embodiment, in step (4), the interpolation is performed in different low-dimensional features of the same class of fault samples, the class of labels are embedded before generating the fault data, and the added noise is low-amplitude random noise.
In one embodiment, in step (5), the preset fault diagnosis model includes one of a support vector machine, a k-nearest neighbor algorithm, a random forest, a fuzzy system, or a deep neural network.
Based on the same inventive concept, the present application also provides a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, said processor implementing the steps of any one of the methods when executing said program.
Based on the same inventive concept, the present application also provides a computer-readable storage medium, on which a computer program is stored, which program, when being executed by a processor, implements the steps of any of the methods.
Based on the same inventive concept, the present application also provides a processor for running a program, wherein the program runs to execute the method of any one of the above.
The application has the beneficial effects that:
compared with the prior art, the application discloses an intelligent diagnosis method for mechanical faults under an unbalanced-like data set. Aiming at the problem of diagnosis precision reduction caused by unbalanced data sets in mechanical fault diagnosis, the method provides a novel data generation method, the characteristic mining capability of deep learning and an countermeasure training mechanism are utilized to learn the data distribution characteristics of a small number of fault samples, interpolation and noise adding are utilized to generate novel characteristics in a low-dimensional characteristic space of data, and after labels are embedded, the novel fault samples are obtained through a generator. The influence of measurement noise in the signal can be eliminated by interpolation in a low-dimensional space, the diversity of generated samples can be improved by adding random noise, and the consistency of the generated samples and the data distribution of the same-category fault samples can be ensured by embedding the labels. Thus, the method has at least the following advantages: (1) The method can learn the low-dimensional distribution characteristics of the data and eliminate the interference of measurement noise; (2) The generated data has consistency with the same type of fault data, and has certain diversity, so that the quality of the generated data is high; and (3) the accuracy of intelligent recognition of mechanical faults is high.
Drawings
FIG. 1 is a flow chart of a method for intelligent diagnosis of mechanical faults under an unbalanced data set of the present application.
Fig. 2 is a comparison chart of generated data and real data of four fault types obtained in the mechanical fault intelligent diagnosis method under the unbalanced data set, wherein the left side is the real data under four fault states, and the right side is the generated data corresponding to the real data.
FIG. 3 is a graph showing the classification accuracy of the intelligent diagnosis method of mechanical failure under the unbalanced data set and the conventional method under five unbalanced rates.
Detailed Description
The present application will be further described with reference to the accompanying drawings and specific examples, which are not intended to be limiting, so that those skilled in the art will better understand the application and practice it.
The intelligent diagnosis method for mechanical faults under the unbalanced data set is shown in the flow chart of fig. 1, and the technology specifically comprises the following steps:
step 101: and (5) preprocessing data. The vibration signal is fourier transformed, the mechanical vibration signal is converted to the frequency domain, and the amplitude is normalized to the [0,1] range.
Step 102: and (5) building a model. Combining the automatic encoder with the generation countermeasure network to build a data generation model.
The automatic encoder consists of an encoder and a decoder, the generation countermeasure network consists of a generator and a discriminator, and the decoder is the generator. The automatic encoder learns the low-dimensional characteristics of the input data (true data) through the encoder, and then outputs the generated data (false data) consistent with the distribution characteristics of the input data through the decoder through the low-dimensional characteristics and class labels thereof. The discriminator in the generation countermeasure network performs true and false discrimination and category classification on the input data and the generated data, respectively.
The encoder, decoder, arbiter include, but are not limited to, being constructed by one of a deep convolutional network, a deep belief network, a residual network.
Step 103: and (5) model training. And training data to generate a model according to a preset loss function and an optimization algorithm by using the fault data.
The loss function of the data generation model in the training process comprises the following steps:
1) The generator generates a mean square error loss function between the data and the encoder input data. Optimizing the loss function may ensure consistency of the generated data and the input data distribution characteristics.
2) The discriminator classifies the loss function for the cross entropy of the true and false data. Optimizing the cross entropy classification loss function of the real data can improve the classification capacity of the discriminator on the real data; optimizing the cross entropy classification loss function of the false data can improve the class discrimination capability of the discriminator for the generated data and the learning capability of the generator for class features, and ensure the feature consistency between the generated data in the same class and the feature difference between the generated data in different classes.
3) The discriminator discriminates the data true and false from the Wasserstein distance or the binary cross entropy loss function. Optimizing the loss function may further improve the quality of the data generated by the generator and the discriminatory power of the arbiter.
4) A mean square error loss function between the encoder output feature and the intermediate implicit feature of the arbiter. Optimizing the loss function may improve consistency of the extracted features of the same class of data by the generator and the arbiter.
Through optimizing the loss functions, the generated data is finally close to the data distribution of the input data of the same class, the discriminator is difficult to discriminate the true and false of the generated data and the input data of the same class, and the balance is achieved between the generator and the discriminator, so that the training of the data generation model is completed.
Preset optimization algorithms include, but are not limited to, one of random gradient descent (SGD), momentum-loaded random gradient descent (Momentum), nestrov Momentum, adagard algorithm, adaptive moment estimation (Adam).
Step 104: and (5) generating data. And generating corresponding fault data by utilizing the low-dimensional characteristics of the fault data learned by the data generation model in training through repeated interpolation and noise addition, so as to realize the balance of various data.
Interpolation is performed in different low-dimensional features of the same class of fault samples, the class of labels are embedded before generating fault data, and the added noise is low-amplitude random noise.
Step 105, fault diagnosis. Training a preset fault diagnosis model by using the class balance data set, and performing intelligent diagnosis on mechanical faults by using the trained fault diagnosis model.
The preset fault diagnosis model comprises one of a support vector machine, a k nearest neighbor algorithm, a random forest, a fuzzy system and a deep neural network, but is not limited to the support vector machine, the k nearest neighbor algorithm and the fuzzy system.
In order to more clearly understand the technical scheme and the effect of the present application, the following detailed description is provided with reference to a specific embodiment.
Taking intelligent diagnosis of gear box faults as an example, building a planetary gear box fault simulation test platform, and manually setting four fault states respectively: five health states including broken teeth, missing teeth, tooth root cracks and tooth surface abrasion and normal states are added. An acceleration sensor is arranged on the planetary gear box to collect vibration signals of the gear box, and the sampling frequency is 5kHz. Each health state contained 2000 sets of signals, 1000 of which did not participate in training as test data, each set of signals being 2048 data points in length. In order to verify the effectiveness of the intelligent diagnosis method for mechanical faults under the unbalanced data set, 5 unbalanced rates are set in the example, namely the ratio of the number of healthy samples of the gear box to the number of the fault samples of each type is 5:1, 10:1, 20:1, 50:1 and 100:1, and the data size of the healthy samples under each unbalanced rate is 1000.
The technology disclosed by the application is adopted to process the 5 groups of unbalanced data sets, the steps are shown in fig. 1, and the detailed information is as follows.
And (3) preprocessing data. The vibration signal is fourier transformed, the mechanical vibration signal is converted to the frequency domain, and the amplitude is normalized to the [0,1] range. The length of the original time domain signal is 2048 data points, and 1024-length frequency domain signals are taken as input data of the model after Fourier transformation.
And (2) building a model. Combining an automatic encoder with a generation countermeasure network, and constructing a data generation model, wherein the specific embodiment is as follows:
(1) automatic encoder: including encoders and decoders, which function primarily to encode and decode input data. The encoder adopts a four-layer one-dimensional convolutional neural network structure, the dimensions of each layer are 8, 16, 32 and 64, a convolutional kernel with the length of 15 is adopted, the convolutional layers are connected with a LeakyReLU activation function layer, and a sample outputs a 64-dimensional potential feature vector after passing through the encoder. The decoder adopts a four-layer one-dimensional deconvolution neural network structure, each layer of dimension is 64, 32, 16 and 8, a deconvolution core with the length of 15 is adopted, a ReLU activation function layer is connected between deconvolution layers, the last deconvolution layer of the decoder is connected to a Sigmoid activation function, and the amplitude of generated data is limited in the range of [0,1 ].
(2) Generating an antagonizing network: comprising a generator and a discriminator. The generator is the decoder in the automatic encoder. The discriminator designs four one-dimensional convolution layers and two full-connection layers, wherein each layer has dimensions of 8, 16, 32 and 64, a convolution kernel with the length of 15 is adopted, a LeakyRelu activation function layer is connected between each convolution layer, and the convolution layers finally output feature vectors with the length of 64 dimensions. The feature is then input into two fully connected layers, respectively, the first fully connected layer reduces the 64-dimensional feature vector to 1-dimensional for calculation of the wasperstein distance between the generated data and the real data. The second fully-connected layer reduces the 64-dimensional feature vector to 4-dimensional (i.e. the number of fault categories requiring up-sampling) and connects with the Softmax active layer to judge the category of the signal.
And (3) training a model. And training data to generate a model according to a preset loss function and an optimization algorithm by using the fault data. In this embodiment there is a 4-part loss function:
(1) the generator generates a mean square error loss function between the data and the encoder input data;
(2) classifying loss functions of the cross entropy of the true and false data by the discriminator;
(3) a Wasserstein distance function for discriminating the true and false of the data by the discriminator;
(4) a mean square error loss function between the encoder output feature and the intermediate implicit feature of the arbiter.
After the loss functions of all the parts are added, the back propagation is carried out through a root mean square transfer algorithm (RmsPorp), and the discriminator and the automatic encoder are optimized in sequence. And repeatedly executing the model training, and after the model training is iterated for 2000 times, the model loss tends to be balanced, and ending the network training.
And (4) generating data. The low-dimensional characteristics of fault data learned in training by the data generation model are utilized, and fault data of corresponding classes are generated after repeated interpolation and noise addition, so that various data balances are realized;
the training samples of the same class are input into an encoder, and potential feature vectors of input data are acquired. Then selecting similar feature vectors for interpolation, in the embodiment, selecting a feature vector by adopting a K nearest neighbor method, selecting one feature vector from low-dimensional feature vectors, finding out 3 nearest neighbor vectors, and then selecting one from adjacent vectors for vector interpolation. After interpolation amplification, adding 0.02 times of standard Gaussian white noise to the newly acquired vector, and embedding a label of a sample into the noisy vector to realize amplification of the potential feature vector. And finally, inputting the processed feature vector into a decoder to generate a new sample. Fig. 2 shows a comparison of the generated signals of the four fault types and the real signals, and it can be seen that the generated signals follow the distribution rule of the real signals, and have a certain difference.
And (5) fault diagnosis. Training a preset fault diagnosis model by using the class balance data set, and performing intelligent diagnosis on mechanical faults by using the trained fault diagnosis model.
The fault diagnosis model selects a support vector machine, and the input data of the fault diagnosis model is 6 main features of each data sample extracted by using a principal component analysis method. Firstly training a support vector machine by adopting a class balance data set, and then testing the classification accuracy of the trained support vector machine by adopting test set data (each class data amount is 1000). FIG. 2 shows the classification accuracy obtained after training the support vector machine by using the class balance data set obtained by the method and the synthetic minority class upsampling technology respectively, and also shows the classification accuracy obtained without adopting the data generation method. Under different unbalance rates, the method and the synthetic minority class upsampling technology provided by the application can improve the classification accuracy of the classifier, and the method provided by the application can obtain the highest classification accuracy, so that the quality of generated data obtained by the data generation method provided by the application is high, and the performance of the classifier is improved.
In summary, the present application can learn the data distribution characteristics of a small number of failure samples by combining the automatic encoder with the generation countermeasure network, using the feature mining capability of deep learning and the countermeasure training mechanism. In addition, potential characteristics are generated by interpolation and noise addition in a low-dimensional space, and data are generated by a decoder, so that the anti-interference capability and the data quality of measured noise can be improved, and the performance of intelligent diagnosis of mechanical faults is improved.
The above-described embodiments are merely preferred embodiments for fully explaining the present application, and the scope of the present application is not limited thereto. Equivalent substitutions and modifications will occur to those skilled in the art based on the present application, and are intended to be within the scope of the present application. The protection scope of the application is subject to the claims.
Claims (8)
1. An intelligent diagnosis method for mechanical faults under an unbalanced-like data set is characterized by comprising the following steps:
step (1), data preprocessing: converting the mechanical vibration signal to the frequency domain and normalizing the amplitude to the [0,1] range;
step (2), building a model: combining an automatic encoder with a generation countermeasure network, and building a data generation model, wherein the automatic encoder learns low-dimensional characteristics of input data, namely true data, through the encoder, and then outputs generated data, namely false data, consistent with the distribution characteristics of the input data through a decoder by the aid of the low-dimensional characteristics and class labels thereof; the discriminator in the generated countermeasure network respectively carries out true and false discrimination and category classification on the input data and the generated data;
step (3), model training: training the data generation model by using fault data according to a preset loss function and an optimization algorithm, wherein the preset loss function comprises a mean square error loss function between the generated data of the generator and the input data of the encoder, a cross entropy classification loss function of a discriminator on true and false data, a Wasserstein distance or a binary cross entropy loss function of the discriminator on true and false discrimination of the data, and a mean square error loss function between an output characteristic of the encoder and an intermediate implicit characteristic of the discriminator;
step (4), data generation: the fault data low-dimensional characteristics learned by the data generation model in training are utilized, corresponding classes of fault data are generated through repeated interpolation and noise adding, so that various types of data balance is realized, the interpolation is carried out in different low-dimensional characteristics of the same class of fault samples, labels of the class are needed to be embedded before the fault data are generated, and the added noise is low-amplitude random noise, and the method comprises the following steps: inputting training samples of the same class into an encoder to obtain potential feature vectors of input data; then selecting similar feature vectors for interpolation, selecting a feature vector by adopting a K nearest neighbor method, selecting one feature vector from the low-dimensional feature vector, finding out 3 nearest neighbor vectors, and then selecting one from the nearest neighbor vectors for vector interpolation; after interpolation amplification, adding 0.02 times of standard Gaussian white noise to the newly acquired vector, and embedding a label of a sample into the noisy vector to realize amplification of potential feature vectors; finally, inputting the processed feature vector into a decoder to generate a new sample;
step (5), fault diagnosis: training a preset fault diagnosis model by using the class balance data set, and performing intelligent diagnosis on mechanical faults by using the trained fault diagnosis model.
2. The intelligent diagnosis method for mechanical failure under unbalanced-like data set of claim 1, wherein in the step (2), the automatic encoder is composed of an encoder and a decoder, the generation countermeasure network is composed of a generator and a discriminator, and the decoder is the generator.
3. The method of intelligent diagnosis of mechanical failure under an unbalanced-like data set of claim 2, wherein the encoder, the decoder and the arbiter are each constructed by one of a deep convolutional network, a deep belief network, and a residual network.
4. The intelligent diagnosis method for mechanical failure under unbalanced-like data set of claim 1, wherein in the step (3), the preset optimization algorithm comprises one of a random gradient descent method, a random gradient descent with momentum, a Nesterov momentum method, an adagard algorithm and an adaptive moment estimation method.
5. The intelligent diagnosis method for mechanical failure under unbalanced-like data set of claim 1, wherein in step (5), the preset failure diagnosis model comprises one of a support vector machine, a k-nearest neighbor algorithm, a random forest, a fuzzy system or a deep neural network.
6. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of the method according to any one of claims 1 to 5 when the program is executed by the processor.
7. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the steps of the method according to any one of claims 1 to 5.
8. A processor for running a program, wherein the program when run performs the method of any one of claims 1 to 5.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111136682.5A CN113935460B (en) | 2021-09-27 | 2021-09-27 | Intelligent diagnosis method for mechanical faults under unbalanced-like data set |
| PCT/CN2021/123198 WO2023044979A1 (en) | 2021-09-27 | 2021-10-12 | Mechanical fault intelligent diagnosis method under class unbalanced dataset |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111136682.5A CN113935460B (en) | 2021-09-27 | 2021-09-27 | Intelligent diagnosis method for mechanical faults under unbalanced-like data set |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113935460A CN113935460A (en) | 2022-01-14 |
| CN113935460B true CN113935460B (en) | 2023-08-11 |
Family
ID=79276976
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111136682.5A Active CN113935460B (en) | 2021-09-27 | 2021-09-27 | Intelligent diagnosis method for mechanical faults under unbalanced-like data set |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN113935460B (en) |
| WO (1) | WO2023044979A1 (en) |
Families Citing this family (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114057053B (en) * | 2022-01-18 | 2022-04-26 | 杭州浅水数字技术有限公司 | Method for monitoring fatigue degree of component of special machine |
| CN114611233B (en) * | 2022-03-08 | 2022-11-11 | 湖南第一师范学院 | Rotating machinery fault imbalance data generation method and computer equipment |
| CN114993677B (en) * | 2022-05-11 | 2023-05-02 | 山东大学 | Rolling bearing fault diagnosis method and system for unbalanced small sample data |
| CN116204786B (en) * | 2023-01-18 | 2023-09-15 | 北京控制工程研究所 | Method and device for generating designated fault trend data |
| CN116401596B (en) * | 2023-06-08 | 2023-08-22 | 哈尔滨工业大学（威海） | Early fault diagnosis method based on depth index excitation network |
| CN116432091B (en) * | 2023-06-15 | 2023-09-26 | 山东能源数智云科技有限公司 | Equipment fault diagnosis method based on small sample, construction method and device of model |
| CN116993319B (en) * | 2023-07-14 | 2024-01-26 | 南京先维信息技术有限公司 | Remote equipment health monitoring method and device based on Internet of things |
| CN116701948B (en) * | 2023-08-03 | 2024-01-23 | 东北石油大学三亚海洋油气研究院 | Pipeline fault diagnosis method and system, storage medium and pipeline fault diagnosis equipment |
| CN116821697B (en) * | 2023-08-30 | 2024-05-28 | 聊城莱柯智能机器人有限公司 | Mechanical equipment fault diagnosis method based on small sample learning |
| CN117056814B (en) * | 2023-10-11 | 2024-01-05 | 国网山东省电力公司日照供电公司 | Transformer voiceprint vibration fault diagnosis method |
| CN117056734B (en) * | 2023-10-12 | 2024-02-06 | 山东能源数智云科技有限公司 | Method and device for constructing equipment fault diagnosis model based on data driving |
| CN117076935B (en) * | 2023-10-16 | 2024-02-06 | 武汉理工大学 | Digital twin-assisted mechanical fault data lightweight generation method and system |
| CN117593783B (en) * | 2023-11-20 | 2024-04-05 | 广州视景医疗软件有限公司 | Visual training scheme generation method and device based on self-adaptive SMOTE |
| CN117332342B (en) * | 2023-11-29 | 2024-02-27 | 北京宝隆泓瑞科技有限公司 | Machine pump equipment operation fault classification method and device based on semi-supervised learning |
| CN117725419A (en) * | 2023-12-22 | 2024-03-19 | 兰州理工大学 | Small sample unbalanced rotor fault diagnosis method and system |
| CN117610614B (en) * | 2024-01-11 | 2024-03-22 | 四川大学 | Attention-guided generation countermeasure network zero sample nuclear power seal detection method |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109753998A (en) * | 2018-12-20 | 2019-05-14 | 山东科技大学 | The fault detection method and system, computer program of network are generated based on confrontation type |
| CN113239991A (en) * | 2021-04-28 | 2021-08-10 | 浙江工业大学 | Flame image oxygen concentration prediction method based on regression generation countermeasure network |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110428004B (en) * | 2019-07-31 | 2021-02-05 | 中南大学 | Mechanical part fault diagnosis method based on deep learning under data imbalance |
| CN112396088B (en) * | 2020-10-19 | 2023-05-12 | 西安交通大学 | Mechanical fault intelligent diagnosis method for implicit excitation countermeasure training under small sample |
| CN113298230B (en) * | 2021-05-14 | 2024-04-09 | 武汉嫦娥医学抗衰机器人股份有限公司 | Prediction method based on unbalanced data set generated against network |
| CN113255078A (en) * | 2021-05-31 | 2021-08-13 | 南京信息工程大学 | Bearing fault detection method and device under unbalanced sample condition |
-
2021
- 2021-09-27 CN CN202111136682.5A patent/CN113935460B/en active Active
- 2021-10-12 WO PCT/CN2021/123198 patent/WO2023044979A1/en unknown
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109753998A (en) * | 2018-12-20 | 2019-05-14 | 山东科技大学 | The fault detection method and system, computer program of network are generated based on confrontation type |
| CN113239991A (en) * | 2021-04-28 | 2021-08-10 | 浙江工业大学 | Flame image oxygen concentration prediction method based on regression generation countermeasure network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113935460A (en) | 2022-01-14 |
| WO2023044979A1 (en) | 2023-03-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113935460B (en) | Intelligent diagnosis method for mechanical faults under unbalanced-like data set | |
| CN109580215B (en) | Wind power transmission system fault diagnosis method based on deep generation countermeasure network | |
| Qin et al. | The optimized deep belief networks with improved logistic sigmoid units and their application in fault diagnosis for planetary gearboxes of wind turbines | |
| CN108062572B (en) | Hydroelectric generating set fault diagnosis method and system based on DdAE deep learning model | |
| CN111242071B (en) | Attention remote sensing image target detection method based on anchor frame | |
| CN111220958A (en) | Radar target Doppler image classification and identification method based on one-dimensional convolutional neural network | |
| CN103728551B (en) | A kind of analog-circuit fault diagnosis method based on cascade integrated classifier | |
| CN106124212A (en) | Based on sparse coding device and the Fault Diagnosis of Roller Bearings of support vector machine | |
| CN109000876B (en) | SNS optical fiber impact identification method based on automatic encoder deep learning | |
| CN103926526A (en) | Analog circuit fault diagnosis method based on improved RBF neural network | |
| CN114778112A (en) | Audio identification and fault diagnosis method for mechanical fault of wind turbine generator system | |
| CN110866448A (en) | Flutter signal analysis method based on convolutional neural network and short-time Fourier transform | |
| CN105335698A (en) | Gear failure diagnosis method based on adaptive genetic algorithm and SOM (Self-Organizing Map) network | |
| CN112926728B (en) | Small sample turn-to-turn short circuit fault diagnosis method for permanent magnet synchronous motor | |
| CN111783531A (en) | Water turbine set fault diagnosis method based on SDAE-IELM | |
| CN111010356A (en) | Underwater acoustic communication signal modulation mode identification method based on support vector machine | |
| CN111275108A (en) | Method for performing sample expansion on partial discharge data based on generation countermeasure network | |
| CN115114965B (en) | Wind turbine generator gearbox fault diagnosis method, device, equipment and storage medium | |
| CN113884844A (en) | Transformer partial discharge type identification method and system | |
| CN116628592A (en) | Dynamic equipment fault diagnosis method based on improved generation type countering network | |
| CN116186587A (en) | Edge end fault diagnosis method and system based on depth migration | |
| CN115409052A (en) | Fault diagnosis method and system for wind generating set bearing under data imbalance | |
| CN113657244A (en) | Fan gearbox fault diagnosis method and system based on improved EEMD and speech spectrum analysis | |
| CN115356599B (en) | Multi-mode urban power grid fault diagnosis method and system | |
| CN116541771A (en) | Unbalanced sample bearing fault diagnosis method based on multi-scale feature fusion |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |