CN111721536A - Rolling bearing fault diagnosis method for improving model migration strategy - Google Patents
Rolling bearing fault diagnosis method for improving model migration strategy Download PDFInfo
- Publication number
- CN111721536A CN111721536A CN202010636625.2A CN202010636625A CN111721536A CN 111721536 A CN111721536 A CN 111721536A CN 202010636625 A CN202010636625 A CN 202010636625A CN 111721536 A CN111721536 A CN 111721536A
- Authority
- CN
- China
- Prior art keywords
- model
- network
- learning
- domain
- target
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/04—Bearings
- G01M13/045—Acoustic or vibration analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
- G06F2218/06—Denoising by applying a scale-space analysis, e.g. using wavelet analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Acoustics & Sound (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
A rolling bearing fault diagnosis method for improving a model migration strategy belongs to the technical field of rolling bearing fault diagnosis. The method is provided for solving the problem that the distribution of the same state data in a source domain and a target domain is large in difference. Acquiring time-frequency spectrums of vibration signals of bearings of different models by utilizing wavelet transformation and constructing an image data set; selecting data of a certain model as a source domain, and selecting data of other models as a target domain; training a ResNet-34 deep convolution network by using source domain data to obtain a source domain data classification model; the implicit gradual change element learning self-adaption is utilized to determine the migration knowledge level and the knowledge content to realize the improvement of the model migration strategy, and the phenomenon that the gradient in the heterogeneous system structure is not easy to converge is avoided; introducing the migration knowledge into the data training process of the ResNet-152 convolutional neural network of the target domain to realize model migration through parameter transmission; and optimizing network parameters by adopting a random gradient descent algorithm when training the source domain network and the target domain network, and establishing fault diagnosis models of rolling bearings of different models.
Description
Technical Field
The invention relates to a rolling bearing fault diagnosis method for improving a model migration strategy, and belongs to the technical field of rolling bearing fault diagnosis.
Background
Rolling bearing is used as key component of rotary machine, and is extensively used in industrial production, and can be used for making effective fault diagnosis so as to prevent serious accident[1]. The rolling bearings are various in types, so that training data of a certain type with labels are lacked or cannot be acquired in actual work[2]. The method has important practical significance for carrying out state recognition on the vibration signals of unknown state information under other models according to the vibration signals of the rolling bearing with known certain state information[3]。
Fault diagnosis of different positions and different damage levels of rolling bearings essentially involves the identification of the operating state of the rolling bearing[4]. The traditional state identification method needs to manually extract features firstly, has rich signal processing experience and feature extraction knowledge, and selects proper features according to different fault types[5-6]. In recent years, with the research of deep learning by researchers, the fault diagnosis method begins to use the deep learning to automatically extract the required features, and overcomes the defect of manually extracting the features in the traditional method[7-8]. Document [9 ]]The rolling bearing fault diagnosis method based on the stacking automatic encoder is provided, the characteristics do not need to be extracted manually, and the influence of manual experience on a diagnosis result is greatly reduced. Document [10 ]]And the bearing fault diagnosis is realized by using the map and the sequence model of deep learning, and better fault classification precision is obtained. Document [11]The method provides the method for enhancing the input classification information by utilizing the multi-scale cascade convolution neural network, and the method is used for bearing failure under the non-steady working conditionThe barrier diagnosis has better results. Document [12 ]]Extracting features from time domain and frequency domain data of the bearing, fusing a plurality of deep features into inherent low-dimensional features through local and global principal component analysis, and realizing bearing fault diagnosis by utilizing an integrated kernel extreme learning machine. Document [13 ]]The vibration signal is converted into a spectrogram and is input into the full convolution neural network, so that the fault position and the damage degree of the bearing can be accurately identified, and the bearing has better generalization.
In actual operation, the operating load of the rolling bearing varies, and the vibration characteristics of the rolling bearing are different under different loads, which is more complicated than under a constant load. When vibration data under different loads are respectively used as training data and test data, the data distribution of the training data and the test data is different, and the diagnosis effect of the fault diagnosis method based on deep learning is reduced.
In recent years, transfer learning has received a wide attention from trainees, which can solve problems in a target domain using knowledge already existing in a source domain. Domain adaptation is a common method of transfer learning, aiming at reducing the distribution difference between the source domain and the target domain[14]. Document [15 ]]A deep Adaptation network (JAN) is provided, and distribution difference among different domain data is reduced by combining a maximum average difference criterion and aligning the Joint distribution of a plurality of cross-domain specific field layers. Document [16 ]]And providing multi-layer adaptation and multi-core maximum mean difference, and reducing mean difference among different domain data. Document [17 ]]A Joint Distribution Adaptation (JDA) method is provided, and differences between the domains are reduced by reducing Joint probability distribution distances between a source domain and a target domain. Document [18]By applying a Geodesic Flow Kernel (GFK), source domain data is aligned to target domain data step by step through a large number of intermediate subspaces, and the method has a good effect in the field of visual identification. Document [19 ]]A multi-layer multi-core variable framework based on the maximum average difference is provided, a core method is introduced to replace a high-dimensional graph of the maximum average error, the difference between data distribution of different domains is reduced to a great extent, and bearing fault diagnosis under different working conditions is realized with higher accuracy. Document [20 ]]Provides a neural network based on depth generationThe bearing fault diagnosis method can generate a forged sample for domain adaptation, solve the problem of less sample data and improve the diagnosis accuracy. Document [21 ]]The cross-domain stacking noise reduction automatic encoder with the new adaptive training strategy is provided, the advantages of domain adaptation and manifold learning are combined, fault diagnosis is carried out on gearbox and engine rolling bearing data sets under different working conditions, and high accuracy is obtained. Document [22 ]]The knowledge learned by a plurality of hidden layers of the deep convolutional network is utilized to provide a representation clustering algorithm, the intra-class distance is minimized, the inter-class distance is maximized, the maximum average difference between training data and testing data is minimized by adopting a domain adaptation method, and the robustness of the rolling bearing fault diagnosis method under different working conditions can be obviously improved. Document [23]]The method comprises the steps of constructing a domain-shared residual error network, extracting migration fault characteristics, applying domain adaptation regular term constraint in a training process, forming a deep migration diagnosis model, and performing effective fault diagnosis on mechanical equipment.
The domain adaptation method requires that some cross features exist in the source domain data and the target domain data, the source domain features and the target domain features are put into the same space through mapping, distribution differences are reduced in the mapping space, but feature mapping is difficult, and an overfitting phenomenon is easy to occur. The model migration method is to find the shared parameter information from the source domain and the target domain, and realize migration through parameter fine tuning, so that the over-fitting phenomenon can be effectively avoided. Document [24] selects output-sensitive parameters in a source domain model, freezes only the sensitive parameters during training, reduces trainable parameters and protects informational parameters, effectively prevents overfitting, adapts to a target domain through a target data volume and an optimal freezing degree, and effectively diagnoses faults of bearings of different loads. Document [25] proposes a deep migration non-negative constraint sparse automatic encoder, which combines the advantages of deep learning and migration learning, solves the problem of fault diagnosis of a rolling bearing with scarce tag data under different working conditions by using parameter transfer, and has a good effect. Document [26] proposes that the fault diagnosis of the rolling bearing is realized by utilizing AlexNet model migration, only the last full connection layer of the network model needs to be replaced to realize parameter transmission, the modeling time can be reduced, and the fault diagnosis of the rolling bearing has good effect under different working conditions.
If the same network is used for extracting the vibration characteristics of the rolling bearings with different loads, some characteristic information may be omitted, and the diagnosis effect is influenced. Document [27] maps the source domain and target domain features to the same space through different automatic encoder networks, introduces a central distance in the domain to evaluate the similarity of distribution between different domains, and realizes bearing fault diagnosis through a support vector machine, thereby having a better effect. However, in actual work, rolling bearings are various in model and a large amount of labeled vibration data under different models are difficult to obtain comprehensively, so that training data is lacked, objective difference exists among data in the same state in different domains, and the diagnosis result is not ideal or even modeling cannot be performed. Meanwhile, when the traditional model migration strategy is applied between heterogeneous architectures and tasks, the phenomenon of non-convergence is easily caused.
Disclosure of Invention
The technical problem to be solved by the invention is as follows:
the invention provides a rolling bearing fault diagnosis method for improving a model migration strategy, aiming at the problems that vibration data of a rolling bearing of a certain model is difficult to obtain comprehensively in actual work, the distribution difference of data of the same state in a source domain and a target domain is large, and the identification accuracy of different fault positions and different damage degrees is low when vibration data of rolling bearings of different models are selected in different domains.
The technical scheme adopted by the invention for solving the technical problems is as follows:
a rolling bearing fault diagnosis method for improving a model migration strategy is realized by the following steps:
(1) data selection and processing
Acquiring all state (normal state and different fault degrees of an inner ring, an outer ring and a rolling body) vibration signals of a rolling bearing with a known model, and taking the vibration signals as source domain data; using vibration signals of rolling bearings of other models with unknown specific states as target domain data; performing wavelet transformation on source domain data and target domain data and constructing a two-dimensional image data set as input of a source domain network and a target domain network;
(2) source domain network model training
Inputting the processed source domain data into a ResNet-34 network, and obtaining a source domain classification model through multiple iterative training;
(3) model migration parameter transfer
Deriving a source domain classification model, and realizing model migration and improving model migration strategies by using implicit gradual change element learning to transfer parameters
The method comprises the following steps of achieving the purpose of model migration by using implicit gradual change element learning transfer parameters, determining which layer of knowledge is migrated and what knowledge is migrated to a target domain through the weight of each layer of the network, and assisting the target domain network in training;
(4) target domain network model training and multi-state recognition
And putting the processed target domain data into a ResNet-152 network, establishing a multi-state recognition model after parameter migration by continuously and alternately updating target model parameters and meta-network parameters, obtaining a label finally labeled by the target domain, and comparing the label with a target domain data real label to obtain the accuracy of multi-state recognition so as to measure the recognition performance of the model.
Further, the process of model migration by using the implicit gradient learning transfer parameter is realized based on a model migration strategy of implicit gradient learning, wherein the model migration strategy of implicit gradient learning is as follows: learning the delivery rules, automatically considering the differences of architecture and tasks between the source domain and the target domain, without manually adjusting the delivery configuration, the meta-network will generate corresponding weights for each feature and between each pair of source network layer and target network layer together with the target network, with the main purpose of: (1) automatically deciding which feature mappings of the source domain model are useful for learning the target domain task; (2) the internal circulation process is accelerated.
Further, the weighted feature matching process of the model migration strategy of the implicit gradual change learning:
1) what is migrated
Using weighted feature matching penalties, emphasize what to migrate according to its effect on the target task:
in the formula, Sm(x) Is a feature map of the mth layer of the pre-trained network,is a profile of the nth layer of the target network, rθIs the linear transformation of the parameter theta, H × W represents the size of the characteristic diagram of the m and n layers of the source domain network and the target domain network,is a non-negative weight of channel c andthe channel weights are set to:
in the formula (I), the compound is shown in the specification,a parameter representing a meta network;
2) where to migrate
When knowledge is migrated from a source domain model to a target domain model, determining that (m, n) corresponding to a network layer in the source domain model and the target domain model is important to the effectiveness of the source domain model and the target domain model, and introducing a learnable parameter for each pair of (m, n): lambda [ alpha ]m,nThe transfer quantity between the mth layer and the nth layer of the source domain model and the target domain model can be respectively determined; for each pair of (m, n) is setAs the output of the meta-network, automatically determining the important layer pair of the learning target task;
combining the weight omega of the channel and the weight of the matching pair lambda to obtain the weight transmission loss defined as:
wherein C represents a candidate pair. The final penalty for training the target model is therefore:
in the formula, LorgIs the initial loss, β is a hyperparameter and β > 0.
The rolling bearing fault diagnosis method can be used for multi-state identification of the rolling bearing.
The rolling bearing multi-state identification comprises identification of normal states, faults of inner rings, outer rings and rolling bodies and degradation degrees of different performances of rolling bearings of different models.
All states of the rolling bearing with a certain type comprise a normal state and different fault degrees of an inner ring, an outer ring and a rolling body.
The invention has the following beneficial technical effects: the fault diagnosis method provided by the invention improves the strategy for realizing model migration, and avoids the phenomenon of gradient non-convergence in a heterogeneous system structure. The method comprises the steps of performing wavelet transformation processing on an original time domain vibration signal to construct a two-dimensional frequency spectrum data set; training a ResNet-34 deep convolution network by using a source domain data set, introducing implicit gradual change element learning, and adaptively determining which layer of knowledge and what knowledge are transferred; putting the migrated knowledge into the process of training the ResNet-152 deep convolution network by the target domain data to realize model migration; and optimizing network parameters by adopting a random gradient descent optimization method in the training process, and realizing multi-state identification of different fault positions and damage degrees of rolling bearings of different models.
The model migration is realized by transferring parameters by implicit gradual change element learning, the migration means in the rolling bearing fault diagnosis method is changed, and the migration means in the rolling bearing fault diagnosis method is changed, so that the rolling bearing fault diagnosis method improves the improved model migration strategy.
The method solves the problems that vibration data of a rolling bearing with a certain type is difficult to acquire comprehensively in actual work, and the distribution difference of data in the same state in a source domain and a target domain is large. The method comprises the steps of obtaining time frequency spectrums of vibration signals of bearings of different models by utilizing wavelet transformation and constructing an image data set; selecting data of a certain model as a source domain, and selecting data of other models as a target domain; training a ResNet-34 deep convolution network by using source domain data to obtain a source domain data classification model; the method utilizes implicit gradual change element learning to adaptively determine the knowledge hierarchy and knowledge content to be migrated, realizes the improvement of a model migration strategy, and avoids the phenomenon that the gradient in a heterogeneous system structure is not easy to converge; introducing the migration knowledge into the data training process of the target domain ResNet-152 convolutional neural network to achieve the purpose of realizing model migration through parameter transfer; and optimizing network parameters by adopting a random gradient descent algorithm when training the source domain network and the target domain network, and establishing fault diagnosis models of rolling bearings of different models. Experiments prove that the method can realize the multi-state classification of rolling bearings of different models, and has better stability and higher accuracy. Experiments show that compared with other methods compared by the invention, the identification accuracy of the method is improved by at least 8%, and the average identification accuracy is up to 99%.
Drawings
FIG. 1 is a schematic diagram of "what is migrated";
FIG. 2 is a schematic of "where migrated";
FIG. 3 is a block diagram of a multi-state identification process for rolling bearings of different models;
FIG. 4 is a schematic of a laboratory bench;
FIG. 5 is a graph of the variation of the loss value with the number of iterations (a → B), the variation of the training accuracy with the number of iterations (B), and the variation of the verification, testing, and optimal accuracy with the number of iterations (c);
FIG. 6 is a training target domain network (A → B), in which: (a) the loss value is a graph of the change of the loss value along with the number of iterations, (b) is a graph of the change of the training accuracy along with the number of iterations, and (c) is a graph of the change of the verification, the test and the optimal accuracy along with the number of iterations;
fig. 7 shows the training target domain network (a → C), (a) shows the variation of the loss value with the number of iterations, (b) shows the variation of the training accuracy with the number of iterations, and (C) shows the variation of the verification, testing, and optimal accuracy with the number of iterations.
FIG. 8 is a graph showing the comparison between the method of the present invention and other methods.
Detailed Description
The implementation process of the rolling bearing fault diagnosis method for improving the model migration strategy is described in detail as follows with reference to the accompanying drawings:
1 Yuan learning
The meta-learning algorithm and the related theory when applied to the field of supervised learning are explained.
1.1 Yuan learning theory
Meta-learning, or learning, is a discipline of the door system that looks at how other learning methods perform over a wide range of learning tasks, learning from acquired experience or metadata, learning new tasks more quickly than other learning methods[28]. Meta-learning was first developed in cognitive psychology and is becoming a formal concept in machine learning in recent years.
Meta-learning focuses more on how to improve the learning ability of the neural network, and is widely applied to the fields of memory enhancement, gradient prediction, parameter prediction, loss prediction, attention-based task commonalities searching and the like. Roughly classified into 3 types according to the core idea:
(1) model-based meta learning
The core research idea is as follows:
pθ(y|x,S)=fθ(x,S) (1)
in the formula, S stands for meta learning, pθRepresentative probability, fθIs the probability density function, x is the input and y is the output.
(2) Metric-based meta-learning
The core research idea is as follows:
in the formula, kθIs a kernel function for measuring x and xiBetweenThe similarity of (c).
(3) Meta learning based on optimization
The core research idea is as follows:
pθ(y|x,S)=fθ(S)(x,S) (3)
1.2 Meta-learning Algorithm in the field of supervised learning
The small sample learning is widely concerned and researched in the field of supervised learning, and aims to perform meta-learning by means of prior knowledge of similar learning tasks and then learn new parameters from a few inputs and outputs of the tasks to be learned. For example, a learning objective may classify images according to only a few examples, but the model has learned other similar knowledge. In meta-learning, if the task obeys TiP (T), function f with model as parameter thetaθWhen learning a new task, the model parameters are changed from θ to θ'i. Updated parameter vector θ'iIn the learning task TiUp is updated by one or more gradient dips. It is vigilant that the algorithm is meta-optimized on the model parameters θ, and the meta-objective is calculated from the updated model parameters θ'. Meta-object:
wherein α is the learning rate when learning a new task,the representative element learning function, ▽, is a gradient operator.
And when the cross-task element is updated, the updated model parameter theta' is used for calculation. A loss function is obtained:
if a random gradient descent algorithm is adopted, updating the model parameter theta:
where β is the learning rate in the meta update.
On the classification problem of supervised learning, a cross entropy loss function is often used. Cross entropy loss functional form in discrete classification task:
in the formula (I), the compound is shown in the specification,is a model parameter, x(j)、y(j)Is from a learning task TiThe input and output of the intermediate sample.
The meta-learning algorithm flow for classifying problems in supervised learning is as follows:
model migration
The method is used for explaining the implicit gradual change meta-learning related theory and how to use meta-learning to transfer parameters so as to achieve the purpose of migration, and a new model migration method is provided.
2.1 implicit gradual metamorphic learning
Implicit gradual change meta learning is the popularization and deepening of meta learning under the background of supervised learning and is an algorithm for calculating the meta gradient[29]. Hypothesis slave learning taskTwo disjoint sets are sampled:andeach data is composed of K inputs and outputs, thenSimilar thereto. The loss function is defined as a function of the parameter vector and the data set:learning task TiIs aimed at usingLearning parameters for a particular taskSo as to minimize the test penalty function of the learning task
In general two-level meta-learning, an algorithm space of learning task specific parameters, such as task T, is typically computed using a set of parameters θ and a training data set of the learning taskiThe algorithm space of (a):the main objective is to generate the meta-parameters of the specific parameters after the learning parameters are updated:
will generallyViewed as an explicit or implicit two-layer optimization problem, when performing meta-testing, use with a new task TjWhen corresponding data set is represented, parameters can be learned through the pair elementsAn adaptive procedure is used to achieve better generalization performance. For meta-learning, Alg (θ, D) corresponds toA single or multiple gradient descent step of theta initialization, if corresponding to the single step of gradient descent, then
Where α is a scalar hyperparameter and can also be a learning rateThere is a learned induction bias that is suitable for fine tuning the learning task using K samples.
To avoid overfitting, Alg needs to incorporate some form of regularization, since meta-learning uses fewer optimal gradient steps, so the regularization algorithm:
where λ is the regularization strength. For convenience of writing, the definition is:
according to equations (8), (10), (11), (12) and (13), the two-layer optimization problem is defined as:
the gradient descent update is extended in meta-learning using a chain rule, which expresses:
in the formula (I), the compound is shown in the specification,is a jacobian matrix of d × d dimensions the gradient descent update is extended using the chain rule as:
if it is notAnd (3) through an iterative algorithm, calculating a forward or backward propagation derivative in an iterative process:but it is very easy to cause the optimized path to go beyond memory. Therefore, the implicit Jacobian matrix is used for solving, paths do not need to be considered, only results need to be considered, and the Jacobian matrix solving process comprises the following steps:
for large neural networks, it is difficult to solve the solution of the jacobian matrix in equation (17), and therefore, it is considered to solve an approximation thereof. And (3) solving:
where, representing a constant in the approximation algorithm,' is the jacobian vector product,girepresents the approximation of the element gradient of the implicit gradient element learning, and omega is the parameter of the conjugate gradient algorithm.
2.2 model migration strategy based on implicit gradual change element learning
When the difference between the source domain data and the target domain data is large and the network architecture is different, if the traditional model migration parameter fine tuning method is directly used, the gradient convergence is difficult. Documents [30] and [31] propose parameter transfer using knowledge distillation and jacobian matrix, respectively, but there is a phenomenon in which useless knowledge is transferred. Therefore, a novel parameter transfer strategy based on implicit gradual change meta learning is provided, mainly a transfer rule is learned, the difference of the architecture and the task between a source domain and a target domain is automatically considered, the transfer configuration does not need to be adjusted manually, and a meta network and a target network can generate corresponding weight for each feature and between each pair of a source network layer and a target network layer. The main purpose is:
(1) automatically deciding which feature mappings of the source domain model are useful for learning the target domain task.
(2) The internal circulation process is accelerated.
2.2.1 weighted feature matching
1) What is migrated
In migration learning, not all features of the source model are applicable to learning the target task, so consider using weighted feature matching penalties to emphasize what migrates according to their effect on the target task:
in the formula, Sm(x) Is a feature map of the mth layer of the pre-trained network,is a profile of the nth layer of the target network, rθIs the linear transformation of the parameter theta, H × W represents the size of the characteristic diagram of the m and n layers of the source domain network and the target domain network,is a non-negative weight of channel c andsince the important channels to be transmitted may be different from input image to input image, the channel weights are set as:
in the formula (I), the compound is shown in the specification,parameters representing a meta-network.
The "what to migrate" operation is shown in FIG. 1.
2) Where to migrate
When migrating knowledge from a source domain model to a target domain model, it is important to determine that the source domain model corresponds to (m, n) the network layer in the target domain model for its effectiveness. A learnable parameter is thus introduced for each pair (m, n): lambda [ alpha ]m,nAnd the transfer quantity between the mth layer and the nth layer of the source domain model and the target domain model can be respectively determined to be more than or equal to 0. Further, for each pair of (m, n) is setAs an output of the meta-network, an importance layer pair of the learning target task is automatically determined.
Combining the weight omega of the channel and the weight of the matching pair lambda to obtain the weight transmission loss defined as:
wherein C represents a candidate pair. The final penalty for training the target model is therefore:
in the formula,LorgIs the initial loss, β is a hyperparameter and β > 0.
The "where migrated" operation is shown in FIG. 2.
3 rolling bearing multi-state identification method
The flow of the multi-state identification method for normal rolling bearings, faults of inner rings, outer rings and rolling bodies and different performance degradation degrees of rolling bearings of different models is shown in fig. 3.
The specific process steps are as follows:
(1) data selection and processing
All state (normal state and different fault degrees of an inner ring, an outer ring and a rolling body) vibration signals of a rolling bearing of a known type are obtained and are used as source domain data. And taking the vibration signals of rolling bearings of other models with unknown specific states as target domain data. And performing wavelet transformation on the source domain data and the target domain data and constructing a two-dimensional image data set as the input of a source domain network and a target domain network.
(2) Source domain network model training
The method comprises the steps of inputting processed source domain data into a ResNet-34 network by utilizing the characteristics that a residual network has typical local connection, translation, scaling and distortion of input signals have high invariance, residual blocks are not easy to be subjected to gradient disappearance and the like, and the residual network has the capability of mining high-dimensional data characteristics, and obtaining a source domain classification model through multiple iterative training.
(3) Parameter passing for model migration
And (3) deriving a source domain classification model, utilizing implicit gradual change element learning transfer parameters to achieve the purpose of model migration, determining knowledge of which layer is migrated and what knowledge is migrated to a target domain through the weight of each layer of the network, and assisting the target domain network to train.
(4) Target domain network model training and multi-state recognition
And putting the processed target domain data into a ResNet-152 network, continuously and alternately updating target model parameters and meta-network parameters, establishing a multi-state recognition model after parameter migration, and obtaining a label finally labeled by the target domain. And comparing the real label with the target domain data to obtain the accuracy of multi-state identification so as to measure the identification performance of the model.
The specific process of the implicit gradual change learning transfer parameter assisted target domain training is as follows:
4 application and analysis
4.1 Experimental data
The experimental data set adopts a bearing data set of the university of Kaiser West reservoir of America[32]. As shown in fig. 4, the experiment table is provided with a motor-driving-end deep groove ball bearing of the type SKF6205 and a fan-end deep groove ball bearing of the type SKF6203, and a 16-channel data recorder is used for collecting vibration signals of a rolling bearing, wherein the sampling frequency is 12 kHz.
The experiment is that under 0hp, 1hp, 2hp, 3hp different loads, gather different model antifriction bearing vibration signal, including normal condition, inner circle, rolling element, outer lane trouble, and every damage state contains 7 mils, 14 mils and 21 mils three kinds of damage degree, totally 40 kinds of states. For convenience, the failure states of the damaged diameters of the inner ring of the rolling bearing at 0hp of 7mil, 14mil and 21mil are respectively represented by 0-IR07, 0-IR14 and 0-IR21, the different failure states of the rolling elements are respectively represented by 0-B07, 0-B14 and 0-B21, the different failure states of the outer ring are respectively represented by 0-OR07, 0-OR14 and 0-OR21, the normal state is represented by 0-N, and the failure states under the other load conditions are similar to the failure states. The specific status information is shown in table 1.
TABLE 1 Experimental data State information
The present invention shares 3 sets of data sets as shown in table 2. Data set a is model SKF6205 rolling bearing vibration data, 300 samples for each of the 40 states, for a total of 12000 samples. Data set B is model SKF6203 rolling bearing vibration data, 300 samples for each of the 40 states, for a total of 12000 samples.
Data set C is a model 6307E rolling bearing, rotating at 680rps and sampling frequency 8.192 kHz. And 3 kinds of state data of inner ring faults, outer ring faults and normal states are obtained through experiments. 300 samples per state for a total of 900 samples.
TABLE 2 bearing test data set
According to the rolling bearing fault diagnosis problem provided by the invention, the effectiveness of the provided method is verified by adopting migration tasks such as B → A, C → A, C → B, A → B, A → C and B → C, wherein B → A represents that the knowledge of the source domain data set B is migrated to the target domain data set A (containing different damage positions and degrees). In the experiment, 5/6 in the source domain and target domain data sets were used for training, and 1/6 was used for testing. Because the state types in the data set C are few, if the data set C is used as a source domain in the migration process, the health state type of A or B needs to be set to be the same as C, and the data set C does not need to be used as a target domain.
4.2 Fault diagnosis experiment for rolling bearings of different models
The method for model migration based on the implicit gradient element learning improved model migration strategy is used for achieving fault diagnosis of rolling bearings of different models, a ResNet-34 convolution network is used as a source domain network model, ResNet-152 is used as a target domain network model, and ' knowledge of where and what ' is migrated ' is achieved through implicit gradient element learning. According to multiple experimental results and experiences, the learning rate of the convolution network is set to be 0.05, the iteration times are 50, and the initial weight is 0.9; the meta learning rate was set to 0.0001. In order to reduce the influence of random initialization to-be-trained parameters and experimental contingency on the fault diagnosis result of the rolling bearing, each group of experiments are verified repeatedly for 5 times. The experimental data set adopts A, B, C types of rolling bearing data in the table 2, and the experimental results of different migration tasks are shown in the table 3.
Table 3 experimental results of different migration tasks
From the results of the migration task a → B alone, the knowledge of the data set a can be migrated to the data set B with better results, i.e. the fault status of the rolling bearing model 6203 can be identified using the fault signature of the rolling bearing model 6205. From the results of all migration tasks, the method can effectively solve the problem of fault diagnosis of rolling bearings of different models, and the average test accuracy rate reaches more than 99%.
In order to better prove the effectiveness of the method, taking the data set a as the source domain and the data set B or C as the target domain as an example, the loss function, the training accuracy, the testing accuracy, the verification accuracy and the optimal accuracy in each iteration sample of the source domain and the target domain network are obtained by using the method, as shown in fig. 5 to fig. 7.
As can be seen from fig. 5 to 7, the source domain has a high accuracy and a stable loss function in the training process, the target domain data has a stable loss in the training and testing processes, and the testing and verifying accuracy is substantially consistent, so that the method can effectively implement fault diagnosis of rolling bearings of different models, has high accuracy and good stability, and further illustrates that the method can avoid the gradient non-convergence phenomenon in the heterogeneous system structure.
4.3 comparative experiments with other algorithms
In order to further prove that the method has great advantages in the fault diagnosis of rolling bearings of different models, a model migration method provided by JAN, DAN, JDA, GFK and a document [26] is selected for comparative experimental research. The same source domain and target domain datasets were used during the experiment. The comparative results are shown in FIG. 8.
As can be seen from fig. 8, in different migration task experiments, when the rolling bearings of different models are subjected to state classification, the average accuracy of the proposed model migration method is as high as 98%, and the accuracy is improved by at least 8% compared with the depth domain adaptation method and other model migration methods. Therefore, the model migration method based on implicit gradual change element learning can effectively solve the problem of fault diagnosis of rolling bearings of different models.
Conclusion of the invention
(1) Based on the characteristics that a residual error network has typical local connection, translation, scaling and distortion of input signals have high invariance, residual error blocks are not easy to disappear in a gradient mode and the like, and meanwhile, by means of the advantage that wavelet transformation has obvious information local characteristics, deep features of rolling bearings of different models are extracted, and a network classification model after source domain and parameter migration is trained and generated.
(2) Based on the characteristics of memory, improvement of neural network learning ability and easiness in convergence of implicit gradient element learning, the model migration method for improving transfer parameters by utilizing implicit gradient element learning is provided, and the self-adaption determination of the knowledge level and the knowledge content is achieved through training of a source domain network and a meta-learning network. And migrating the knowledge based on the parameters to a target domain network with a structure different from that of the source domain network, so that heterogeneous model migration is realized, and the phenomena of omission of characteristic information and gradient non-convergence are avoided.
(3) By utilizing a new model migration strategy, the distribution difference between the source domain characteristics and the target domain characteristics is reduced, and the problems that labeled data is scarce and even cannot be acquired and the fault states of rolling bearings of different models are classified are solved. Experiments show that compared with other methods compared by the invention, the identification accuracy of the method is improved by at least 8%, and the average identification accuracy is up to 99%.
The experimental study on the vibration data of the rolling bearing is already carried out in the experiment, and the experiment on the vibration tested on site is not carried out, which is the important point of the future work.
The references cited in the present invention are as follows:
[1] leizheng, zhang, ming ampere wave, etc. rolling bearing fault diagnosis method based on IEWT and MCKD [ J ]. mechanical engineering report, 2019,55 (23): 136-146.
Li Zheng,Zhang Wei,Ming Anbo,et al.Fault diagnosis method of rollingbearings based on IEWT and MCKD[J].Journal of Mechanical Engineering,2019,55(23):136-146.
[2]SHAO H,JIANG H,ZHANG H,et al.Rolling bearing fault featurelearning using improved convolutional deep belief network with compressedsensing[J].Mechanical Systems and Signal Processing,2018,100:743-765.
[3]HAINING L,FANGZHEN S,RENJIE D,et al.Domain adaptive faultdiagnosis based on the geodesic flow kernel under small data condition[J].Journal of Vibration and Shock,2018, 37(18):36-42.
[4]VAN M,KANGH.Bearing-fault diagnosis using non-local meansalgorithm and empirical mode decomposition-based feature extraction and two-stage feature selection[J].IET Science Measurement&Technology,2015,9(6):671-680.
[5]JIA F,LEI Y,GUO L,et al.A neural network constructed by deeplearning technique and its application to intelligent fault diagnosis ofmachines[J].Neurocomputing,2018,272: 619-928.
[6] Yaodechen, yang jianwei, cheng qing, etc. bearing fault diagnosis research based on multi-scale eigenmode arrangement entropy and SA-SVM [ J ] mechanical engineering report, 2018,54 (9): 168-176.
Yao Dechen,Yang Jianwei,Cheng Xiaoqing,et al.Research on bearingfault diagnosis based on multi-scale eigenmode permutation entropy and SA-SVM[J].Journal of Mechanical Engineering,2018,54(9):168-176.
[7]DING S,GUO L,HOU Y.Extreme learning machine with kernel modelbased on deep learning[J].Neural Computing&Applications,2017,28(8):1975-1984.
[8]SHEN C,QI Y,WANG J,et al.An automatic and robust features learningmethod for rotating machinery fault diagnosis based on contractiveautoencoder[J].Engineering Applications of Artificial Intelligence,2018,76:170-184.
[9]WANG X,QIN Y,WANG Y,et al.ReLTanh:An activation function withvanishing gradient resistance for SAE-based DNNs and its application torotating machinery fault diagnosis[J].Neurocomputing,2019,363:88-98.
[10]UDMALE S S,SINGH S K,BHIRUD S G,et al.A bearing data analysisbased on kurtogram and deep learning sequence models[J].Measurement,2019,145:665-677.
[11]HUANG W,CHENG J,YANG Y,et al.An improved deep convolutionalneural network with multi-scale information for bearing fault diagnosis[J].Neurocomputing,2019,359:77-92.
[12]PANG S,YANG X,ZHANG X,et al.Fault diagnosis of rotating machinerywith ensemble kernel extreme learning machine based on fused multi-domainfeatures[J].ISA transactions,2019.
[13]ZHANG W,ZHANG F,CHEN W,et al.Fault state recognition of rollingbearing based fully convolutional network[J].Computing in Science&Engineering,2018,21:55-63.
[14]Yan R,Shen F,Sun C,et al.Knowledge transfer for rotary machinefault diagnosis[J].IEEE Sensors Journal,2019.
[15]LONG M,ZHU H,WANG J,et al.Deep transfer learning with jointadaptation networks[C]//International Conference on Machine Learning,2017:2208-2217.
[16]LONG M,ZHU H,WANG J,et al.Deep transfer learning with jointadaptation networks[C]//International Conference on Machine Learning,2017:2208-2217.
[17]LONG M,WANG J,DING G,et al.Transfer Feature Learning with JointDistribution Adaptation[C]//International Conference on Computer Vision,2013:2200-2207.
[18]GONG B,SHI Y,SHA F,et al.Geodesic flow kernel for unsuperviseddomain adaptation[C]//IEEE Conference on Computer Vision and PatternRecognition,Providence, America,June 2012:2066-2073.
[19]AN Z,LI S,WANG J,et al.Generalization of deep neural network forbearing fault diagnosis under different working conditions using multiplekernel method[J].Neurocomputing,2019, 352:42-53.
[20]LI X,ZHANG W,DING Q.Cross-domain fault diagnosis of rollingelement bearings using deep generative neural networks[J].IEEE Transactionson Industrial Electronics,2018, 66(7):5525-5534.
[21]PANG S,YANG X.A Cross-Domain Stacked Denoising Autoencoders forRotating Machinery Fault Diagnosis under Different Working Conditions[J].IEEEAccess,2019,7: 77277-77292.
[22]LI X,ZHANG W,DING Q.A robust intelligent fault diagnosis methodfor rolling element bearings based on deep distance metric learning[J].Neurocomputing,2018,310:77-95.
[23] Depth migration diagnosis method of mechanical equipment fault under big data [ J ] mechanical engineering report, 2019,55 (7): 1-8.
Lei Yaguo,Yang Bin,Du Zhaojun,et al.Deep migration diagnosis methodof mechanical equipment faults under big data[J].Journal of MechanicalEngineering,2019,55(7):1-8.
[24]KIM H,YOUN B D.A New Parameter Repurposing Method for ParameterTransfer With Small Dataset and Its Application in Fault Diagnosis of RollingElement Bearings[J].IEEE Access,2019,7:46917-46930.
[25]LI X,JIANG H,ZHAO K,et al.A Deep Transfer Nonnegativity-Constraint Sparse Autoencoder for Rolling Bearing Fault Diagnosis With FewLabeled Data[J].IEEE Access, 2019,7:91216-91224.
[26]WANG J,MO Z,ZHANG H,et al.A Deep Learning Method for BearingFault Diagnosis Based on Time-Frequency Image[J].IEEE Access,2019,7:42373-42383.
[27]CHUNFENG W,ZHENG L,JUN Z,et al.Heterogeneous Transfer LearningBased on Stack Sparse Auto-Encoders for Fault Diagnosis[C]//ChineseAutomation Congress,2018: 4277-4281.
[28]REN Y,CHI C,JINTAO Z.A Survey of Personalized RecommendationAlgorithm Selection Based on Meta-learning[C]//The International Conferenceon Cyber Security Intelligence and Analytics.Springer,Cham,2019:1383-1388.
[29]RAJESWARAN A,FINN C,KAKADE S M,et al.Meta-Learning with ImplicitGradients[C]//Neural Information Processing Systems,2019:113-124.
[30]LI Z,HOIEM D.Learning without Forgetting[J].european conferenceon computer vision, 2018,40(12):614-629.
[31]Srinivas S,Fleuret F.Knowledge Transfer with Jacobian Matching[C]//International Conference on Machine Learning,2018:4723-4731.
[32]LOPARO,K A.Bearing data center[M/OL].Cleveland:Case WesternReserve University, 2013.http://csegroups.
case.edu/bearingdatacenter/home.
Claims (6)
1. A rolling bearing fault diagnosis method for improving a model migration strategy is characterized by comprising the following implementation processes:
(1) data selection and processing
Acquiring all state vibration signals of a rolling bearing with a known model, and taking the vibration signals as source domain data; using vibration signals of rolling bearings of other models with unknown specific states as target domain data; performing wavelet transformation on source domain data and target domain data and constructing a two-dimensional image data set as input of a source domain network and a target domain network;
(2) source domain network model training
Inputting the processed source domain data into a ResNet-34 network, and obtaining a source domain classification model through multiple iterative training;
(3) model migration parameter transfer
Deriving a source domain classification model, and transferring parameters by using implicit gradual change element learning to realize model migration and improve a model migration strategy;
the method comprises the following steps of achieving the purpose of model migration by using implicit gradual change element learning transfer parameters, determining which layer of knowledge is migrated and what knowledge is migrated to a target domain through the weight of each layer of the network, and assisting the target domain network in training;
(4) target domain network model training and multi-state recognition
And putting the processed target domain data into a ResNet-152 network, establishing a multi-state recognition model after parameter migration by continuously and alternately updating target model parameters and meta-network parameters, obtaining a label finally labeled by the target domain, and comparing the label with a target domain data real label to obtain the accuracy of multi-state recognition so as to measure the recognition performance of the model.
2. The method for diagnosing the fault of the rolling bearing with the improved model migration strategy according to claim 1, wherein the process of model migration by using the implicit gradient learning transfer parameter is realized based on the model migration strategy of the implicit gradient learning, and the model migration strategy of the implicit gradient learning is as follows: learning the delivery rules, automatically considering the differences of architecture and tasks between the source domain and the target domain, without manually adjusting the delivery configuration, the meta-network will generate corresponding weights for each feature and between each pair of source network layer and target network layer together with the target network, with the main purpose of: (1) automatically deciding which feature mappings of the source domain model are useful for learning the target domain task; (2) the internal circulation process is accelerated.
3. The method for diagnosing the fault of the rolling bearing with the improved model migration strategy according to claim 2, wherein the weighting feature matching process of the model migration strategy with the implicit gradual change element learning comprises the following steps:
1) what is migrated
Using weighted feature matching penalties, emphasize what to migrate according to its effect on the target task:
in the formula, Sm(x) Is a feature map of the mth layer of the pre-trained network,is a profile of the nth layer of the target network, rθIs the linear transformation of the parameter theta, H × W represents the size of the characteristic diagram of the m and n layers of the source domain network and the target domain network,is a non-negative weight of channel c andthe channel weights are set to:
in the formula (I), the compound is shown in the specification,a parameter representing a meta network;
2) where to migrate
When knowledge is migrated from a source domain model to a target domain model, determining that (m, n) corresponding to a network layer in the source domain model and the target domain model is important to the effectiveness of the source domain model and the target domain model, and introducing a learnable parameter for each pair of (m, n): lambda [ alpha ]m,nThe transfer quantity between the mth layer and the nth layer of the source domain model and the target domain model can be respectively determined; for each pair of (m, n) is setAs the output of the meta-network, automatically determining the important layer pair of the learning target task;
combining the weight omega of the channel and the weight of the matching pair lambda to obtain the weight transmission loss defined as:
wherein C represents a candidate pair. The final penalty for training the target model is therefore:
in the formula, LorgIs initial loss, β is hyperparametric and β>0。
4. The method for diagnosing the fault of the rolling bearing with the improved model migration strategy as claimed in claim 1, 2 or 3, wherein the method for diagnosing the fault of the rolling bearing can perform multi-state identification of the rolling bearing.
5. The method for diagnosing the fault of the rolling bearing with the improved model migration strategy according to claim 4, wherein the multi-state identification of the rolling bearing comprises identification of faults and degradation degrees of different performances of normal states, inner rings, outer rings and rolling bodies of rolling bearings of different models.
6. The method for diagnosing the fault of the rolling bearing with the improved model migration strategy as claimed in claim 1 or 5, wherein all the states of the rolling bearing with a certain model comprise a normal state and different fault degrees of an inner ring, an outer ring and a rolling body.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010636625.2A CN111721536B (en) | 2020-07-20 | 2020-07-20 | Rolling bearing fault diagnosis method for improving model migration strategy |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010636625.2A CN111721536B (en) | 2020-07-20 | 2020-07-20 | Rolling bearing fault diagnosis method for improving model migration strategy |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111721536A true CN111721536A (en) | 2020-09-29 |
CN111721536B CN111721536B (en) | 2022-05-27 |
Family
ID=72571574
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010636625.2A Active CN111721536B (en) | 2020-07-20 | 2020-07-20 | Rolling bearing fault diagnosis method for improving model migration strategy |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111721536B (en) |
Cited By (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112364836A (en) * | 2020-12-07 | 2021-02-12 | 无锡科晟光子科技有限公司 | Vibration optical fiber signal classification method based on full convolution neural network |
CN112465030A (en) * | 2020-11-28 | 2021-03-09 | 河南大学 | Multi-source heterogeneous information fusion fault diagnosis method based on two-stage transfer learning |
CN112613481A (en) * | 2021-01-04 | 2021-04-06 | 上海明略人工智能(集团)有限公司 | Bearing abrasion early warning method and system based on frequency spectrum |
CN112611563A (en) * | 2020-12-01 | 2021-04-06 | 上海明略人工智能(集团)有限公司 | Method and device for determining target fault information |
CN112633269A (en) * | 2020-12-28 | 2021-04-09 | 上海明略人工智能(集团)有限公司 | Logo recognition method and system |
CN112685857A (en) * | 2020-12-31 | 2021-04-20 | 西安交通大学 | Bearing residual life prediction method based on transfer migration |
CN112926547A (en) * | 2021-04-13 | 2021-06-08 | 北京航空航天大学 | Small sample transfer learning method for classifying and identifying aircraft electric signals |
CN112964469A (en) * | 2021-02-28 | 2021-06-15 | 哈尔滨理工大学 | Online fault diagnosis method for rolling bearing under variable load of transfer learning |
CN113111820A (en) * | 2021-04-21 | 2021-07-13 | 北京科技大学 | Rotary component fault diagnosis method and device based on improved CNN and relation module |
CN113269734A (en) * | 2021-05-14 | 2021-08-17 | 成都市第三人民医院 | Tumor image detection method and device based on meta-learning feature fusion strategy |
CN113269266A (en) * | 2021-06-08 | 2021-08-17 | 清华大学 | Knowledge distillation-based wafer manufacturing process fault detection method under multi-fault mode |
CN113378725A (en) * | 2021-06-15 | 2021-09-10 | 山东大学 | Cutter fault diagnosis method, equipment and storage medium based on multi-scale-channel attention network |
CN113420387A (en) * | 2021-06-17 | 2021-09-21 | 长安大学 | Migration diagnosis method and system for rolling bearing of compacting machine |
CN113469219A (en) * | 2021-06-08 | 2021-10-01 | 中国地质大学(武汉) | Rotary machine fault diagnosis method under complex working condition based on element transfer learning |
CN113470025A (en) * | 2021-09-02 | 2021-10-01 | 北京字节跳动网络技术有限公司 | Polyp detection method, training method and related device |
CN113538353A (en) * | 2021-07-05 | 2021-10-22 | 华北电力大学(保定) | Five-phase asynchronous motor rolling bearing fault diagnosis method based on single-channel diagram data enhancement and migration training residual error network |
CN113834656A (en) * | 2021-08-27 | 2021-12-24 | 西安电子科技大学 | Bearing fault diagnosis method, system, equipment and terminal |
CN113988126A (en) * | 2021-10-26 | 2022-01-28 | 哈尔滨理工大学 | Rolling bearing fault diagnosis method based on less-label data feature migration |
CN114550315A (en) * | 2022-01-24 | 2022-05-27 | 云南联合视觉科技有限公司 | Identity comparison and identification method and device and terminal equipment |
CN114660418A (en) * | 2022-05-24 | 2022-06-24 | 国网山西省电力公司大同供电公司 | Power distribution network single-phase earth fault line selection method based on feature fusion and clustering |
CN116541713A (en) * | 2023-06-27 | 2023-08-04 | 清华大学 | Bearing fault diagnosis model training method based on local time-frequency characteristic transfer learning |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
AU2001253522B2 (en) * | 2000-04-13 | 2007-04-05 | Accenture Global Services Limited | Method for a health care solution framework |
CN101605346A (en) * | 2008-06-10 | 2009-12-16 | 中兴通讯股份有限公司 | The fault restoration method and apparatus |
CN108414226A (en) * | 2017-12-25 | 2018-08-17 | 哈尔滨理工大学 | Fault Diagnosis of Roller Bearings under the variable working condition of feature based transfer learning |
CN109902393A (en) * | 2019-03-01 | 2019-06-18 | 哈尔滨理工大学 | Fault Diagnosis of Roller Bearings under a kind of variable working condition based on further feature and transfer learning |
CN110427875A (en) * | 2019-07-31 | 2019-11-08 | 天津大学 | Infrared image object detection method based on depth migration study and extreme learning machine |
CN110490249A (en) * | 2019-08-16 | 2019-11-22 | 哈尔滨工业大学 | Structural Damage Identification based on attribute class relations and few sample meta learning |
CN111260660A (en) * | 2020-01-14 | 2020-06-09 | 电子科技大学 | 3D point cloud semantic segmentation migration method based on meta-learning |
-
2020
- 2020-07-20 CN CN202010636625.2A patent/CN111721536B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
AU2001253522B2 (en) * | 2000-04-13 | 2007-04-05 | Accenture Global Services Limited | Method for a health care solution framework |
CN101605346A (en) * | 2008-06-10 | 2009-12-16 | 中兴通讯股份有限公司 | The fault restoration method and apparatus |
CN108414226A (en) * | 2017-12-25 | 2018-08-17 | 哈尔滨理工大学 | Fault Diagnosis of Roller Bearings under the variable working condition of feature based transfer learning |
CN109902393A (en) * | 2019-03-01 | 2019-06-18 | 哈尔滨理工大学 | Fault Diagnosis of Roller Bearings under a kind of variable working condition based on further feature and transfer learning |
CN110427875A (en) * | 2019-07-31 | 2019-11-08 | 天津大学 | Infrared image object detection method based on depth migration study and extreme learning machine |
CN110490249A (en) * | 2019-08-16 | 2019-11-22 | 哈尔滨工业大学 | Structural Damage Identification based on attribute class relations and few sample meta learning |
CN111260660A (en) * | 2020-01-14 | 2020-06-09 | 电子科技大学 | 3D point cloud semantic segmentation migration method based on meta-learning |
Non-Patent Citations (1)
Title |
---|
康守强 等: "基于特征迁移学习的变工况下滚动轴承故障诊断方法", 《中国电机工程学报》 * |
Cited By (30)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112465030A (en) * | 2020-11-28 | 2021-03-09 | 河南大学 | Multi-source heterogeneous information fusion fault diagnosis method based on two-stage transfer learning |
CN112611563A (en) * | 2020-12-01 | 2021-04-06 | 上海明略人工智能(集团)有限公司 | Method and device for determining target fault information |
CN112364836A (en) * | 2020-12-07 | 2021-02-12 | 无锡科晟光子科技有限公司 | Vibration optical fiber signal classification method based on full convolution neural network |
CN112633269B (en) * | 2020-12-28 | 2024-03-15 | 上海明略人工智能(集团)有限公司 | Logo recognition method and system |
CN112633269A (en) * | 2020-12-28 | 2021-04-09 | 上海明略人工智能(集团)有限公司 | Logo recognition method and system |
CN112685857A (en) * | 2020-12-31 | 2021-04-20 | 西安交通大学 | Bearing residual life prediction method based on transfer migration |
CN112613481A (en) * | 2021-01-04 | 2021-04-06 | 上海明略人工智能(集团)有限公司 | Bearing abrasion early warning method and system based on frequency spectrum |
CN112964469A (en) * | 2021-02-28 | 2021-06-15 | 哈尔滨理工大学 | Online fault diagnosis method for rolling bearing under variable load of transfer learning |
CN112964469B (en) * | 2021-02-28 | 2022-05-27 | 哈尔滨理工大学 | Online fault diagnosis method for rolling bearing under variable load of transfer learning |
CN112926547A (en) * | 2021-04-13 | 2021-06-08 | 北京航空航天大学 | Small sample transfer learning method for classifying and identifying aircraft electric signals |
CN112926547B (en) * | 2021-04-13 | 2022-03-29 | 北京航空航天大学 | Small sample transfer learning method for classifying and identifying aircraft electric signals |
CN113111820A (en) * | 2021-04-21 | 2021-07-13 | 北京科技大学 | Rotary component fault diagnosis method and device based on improved CNN and relation module |
CN113111820B (en) * | 2021-04-21 | 2022-05-13 | 北京科技大学 | Rotary part fault diagnosis method and device based on improved CNN and relation module |
CN113269734A (en) * | 2021-05-14 | 2021-08-17 | 成都市第三人民医院 | Tumor image detection method and device based on meta-learning feature fusion strategy |
CN113269266A (en) * | 2021-06-08 | 2021-08-17 | 清华大学 | Knowledge distillation-based wafer manufacturing process fault detection method under multi-fault mode |
CN113469219A (en) * | 2021-06-08 | 2021-10-01 | 中国地质大学(武汉) | Rotary machine fault diagnosis method under complex working condition based on element transfer learning |
CN113269266B (en) * | 2021-06-08 | 2023-09-19 | 清华大学 | Wafer manufacturing process fault detection method based on knowledge distillation under multiple fault modes |
CN113378725A (en) * | 2021-06-15 | 2021-09-10 | 山东大学 | Cutter fault diagnosis method, equipment and storage medium based on multi-scale-channel attention network |
CN113420387A (en) * | 2021-06-17 | 2021-09-21 | 长安大学 | Migration diagnosis method and system for rolling bearing of compacting machine |
CN113538353A (en) * | 2021-07-05 | 2021-10-22 | 华北电力大学(保定) | Five-phase asynchronous motor rolling bearing fault diagnosis method based on single-channel diagram data enhancement and migration training residual error network |
CN113538353B (en) * | 2021-07-05 | 2023-09-01 | 华北电力大学(保定) | Five-phase asynchronous motor rolling bearing fault diagnosis method based on single-channel diagram data enhancement and migration training residual error network |
CN113834656A (en) * | 2021-08-27 | 2021-12-24 | 西安电子科技大学 | Bearing fault diagnosis method, system, equipment and terminal |
CN113834656B (en) * | 2021-08-27 | 2024-04-30 | 西安电子科技大学 | Bearing fault diagnosis method, system, equipment and terminal |
CN113470025A (en) * | 2021-09-02 | 2021-10-01 | 北京字节跳动网络技术有限公司 | Polyp detection method, training method and related device |
CN113988126B (en) * | 2021-10-26 | 2024-04-16 | 哈尔滨理工大学 | Rolling bearing fault diagnosis method based on few-label data feature migration |
CN113988126A (en) * | 2021-10-26 | 2022-01-28 | 哈尔滨理工大学 | Rolling bearing fault diagnosis method based on less-label data feature migration |
CN114550315A (en) * | 2022-01-24 | 2022-05-27 | 云南联合视觉科技有限公司 | Identity comparison and identification method and device and terminal equipment |
CN114660418A (en) * | 2022-05-24 | 2022-06-24 | 国网山西省电力公司大同供电公司 | Power distribution network single-phase earth fault line selection method based on feature fusion and clustering |
CN116541713A (en) * | 2023-06-27 | 2023-08-04 | 清华大学 | Bearing fault diagnosis model training method based on local time-frequency characteristic transfer learning |
CN116541713B (en) * | 2023-06-27 | 2023-12-12 | 清华大学 | Bearing fault diagnosis model training method based on local time-frequency characteristic transfer learning |
Also Published As
Publication number | Publication date |
---|---|
CN111721536B (en) | 2022-05-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111721536B (en) | Rolling bearing fault diagnosis method for improving model migration strategy | |
Zhao et al. | Deep multi-scale convolutional transfer learning network: A novel method for intelligent fault diagnosis of rolling bearings under variable working conditions and domains | |
Li et al. | Intelligent fault diagnosis of rolling bearings under imbalanced data conditions using attention-based deep learning method | |
Zhou et al. | A new tool wear condition monitoring method based on deep learning under small samples | |
CN110059601B (en) | Intelligent fault diagnosis method for multi-feature extraction and fusion | |
Li et al. | A deep transfer nonnegativity-constraint sparse autoencoder for rolling bearing fault diagnosis with few labeled data | |
CN112964469B (en) | Online fault diagnosis method for rolling bearing under variable load of transfer learning | |
CN111680446B (en) | Rolling bearing residual life prediction method based on improved multi-granularity cascade forest | |
CN113988126B (en) | Rolling bearing fault diagnosis method based on few-label data feature migration | |
Su et al. | A novel method based on deep transfer unsupervised learning network for bearing fault diagnosis under variable working condition of unequal quantity | |
Li et al. | WavCapsNet: An interpretable intelligent compound fault diagnosis method by backward tracking | |
Liao et al. | Wavelet transform based convolutional neural network for gearbox fault classification | |
CN110941928A (en) | Rolling bearing residual life prediction method based on dropout-SAE and Bi-LSTM | |
CN114048769A (en) | Multi-source multi-domain information entropy fusion and model self-optimization method for bearing fault diagnosis | |
Wang et al. | A combination of residual and long–short-term memory networks for bearing fault diagnosis based on time-series model analysis | |
CN113865868B (en) | Rolling bearing fault diagnosis method based on time-frequency domain expression | |
CN113822139B (en) | Equipment fault diagnosis method based on improved 1DCNN-BiLSTM | |
Zhang et al. | Triplet metric driven multi-head GNN augmented with decoupling adversarial learning for intelligent fault diagnosis of machines under varying working condition | |
CN114429150A (en) | Rolling bearing fault diagnosis method and system under variable working conditions based on improved depth subdomain adaptive network | |
CN112132102B (en) | Intelligent fault diagnosis method combining deep neural network with artificial bee colony optimization | |
CN116465628A (en) | Rolling bearing fault diagnosis method based on improved multi-source domain heterogeneous model parameter transmission | |
Wang et al. | Bearing fault diagnosis algorithm based on granular computing | |
Ge et al. | A deep condition feature learning approach for rotating machinery based on MMSDE and optimized SAEs | |
Zhao et al. | Unbalanced fault diagnosis of rolling bearings using transfer adaptive boosting with squeeze-and-excitation attention convolutional neural network | |
Ma et al. | A collaborative central domain adaptation approach with multi-order graph embedding for bearing fault diagnosis under few-shot samples |
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 |