CN117606801A - Cross-domain bearing fault diagnosis method based on multi-characterization self-adaptive network - Google Patents
Cross-domain bearing fault diagnosis method based on multi-characterization self-adaptive network Download PDFInfo
- Publication number
- CN117606801A CN117606801A CN202311364218.0A CN202311364218A CN117606801A CN 117606801 A CN117606801 A CN 117606801A CN 202311364218 A CN202311364218 A CN 202311364218A CN 117606801 A CN117606801 A CN 117606801A
- Authority
- CN
- China
- Prior art keywords
- domain
- characterization
- data
- fault diagnosis
- adaptive network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000012512 characterization method Methods 0.000 title claims abstract description 58
- 238000000034 method Methods 0.000 title claims abstract description 53
- 238000003745 diagnosis Methods 0.000 title claims abstract description 38
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 230000006870 function Effects 0.000 claims description 31
- 238000012549 training Methods 0.000 claims description 28
- 238000013527 convolutional neural network Methods 0.000 claims description 22
- 230000003044 adaptive effect Effects 0.000 claims description 20
- 238000004422 calculation algorithm Methods 0.000 claims description 13
- 238000012360 testing method Methods 0.000 claims description 13
- 238000000605 extraction Methods 0.000 claims description 8
- 238000005457 optimization Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 7
- 230000036541 health Effects 0.000 claims description 6
- 230000003862 health status Effects 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 4
- 210000002569 neuron Anatomy 0.000 claims description 4
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 230000006835 compression Effects 0.000 claims description 2
- 238000007906 compression Methods 0.000 claims description 2
- 238000009826 distribution Methods 0.000 abstract description 22
- 230000005012 migration Effects 0.000 description 10
- 238000013508 migration Methods 0.000 description 10
- 230000004913 activation Effects 0.000 description 9
- 230000006978 adaptation Effects 0.000 description 5
- 230000000694 effects Effects 0.000 description 5
- 238000005096 rolling process Methods 0.000 description 5
- 238000002474 experimental method Methods 0.000 description 4
- 238000000691 measurement method Methods 0.000 description 3
- 238000011176 pooling Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 238000012795 verification Methods 0.000 description 3
- 235000014653 Carica parviflora Nutrition 0.000 description 2
- 241000243321 Cnidaria Species 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000002779 inactivation Effects 0.000 description 2
- 230000003213 activating effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 230000009849 deactivation Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000007620 mathematical function Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
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
-
- 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
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- General Physics & Mathematics (AREA)
- Test And Diagnosis Of Digital Computers (AREA)
Abstract
The invention discloses a cross-domain bearing fault diagnosis method based on a multi-characterization self-adaptive network, which comprises the steps of collecting original vibration signals under different working conditions as source domain data and target domain data respectively; obtaining a frequency domain signal through preprocessing operation, and taking the frequency domain signal as the input of a model; constructing a multi-representation self-adaptive network model, extracting multiple representations of a source domain and a target domain, and minimizing the joint distribution difference between the domains based on the joint maximum mean difference and the pseudo tag, thereby extracting domain invariant features of the source domain and the target domain, and finally forming the domain joint distribution self-adaptation. Experimental results prove that the method has high classification precision and strong generalization capability, and is suitable for cross-domain bearing fault diagnosis under variable working conditions and multiple scenes.
Description
Technical Field
The invention relates to the technical field of fault diagnosis, in particular to a cross-domain bearing fault diagnosis method based on a multi-characterization self-adaptive network.
Background
Rolling bearings are widely used in various rotating machinery equipment, which generally operate in severe environments (such as high speed, high pressure, high temperature, speed change, etc.), so that there is a high failure rate of the rolling bearings, and the running and maintenance costs are increased, which is also indispensable for failure diagnosis of the rolling bearings.
In the past few years, data driving methods such as deep learning and the like are widely applied to the field of mechanical fault diagnosis, for example, the invention patent with publication number of CN116361723a discloses a bearing fault diagnosis classification method based on multi-scale features and attention, which acquires vibration signals through an acceleration sensor and performs preprocessing, and inputs preprocessed data into a pre-built multi-scale feature classification module, a transducer attention mechanism learning module and a full connection module in sequence, so that a good diagnosis effect is finally obtained. However, the training set and the test set are required to have the same probability distribution in the learning mode, and in practical application, a domain offset problem usually occurs, that is, the probability distribution of labeled training data (called a source domain) is different from that of unlabeled test data (called a target domain).
There is also a great deal of research on domain offset problem, for example, the invention patent with publication number CN115062690a discloses a "domain adaptive network-based bearing fault diagnosis method", which uses the sum of the maximum mean difference (Maximum Mean Discrepancy, MMD) of source domain data and target domain data as a loss function and uses a counter propagation algorithm to distribute the probability of Ji Yuanyu data and target domain data, thereby achieving domain adaptation and solving the domain offset problem to a certain extent.
However, these methods mainly align distributions of tokens extracted by a single structure, the single token contains only partial information, so the alignment is also focused on only partial information, and MMD focuses on only edge distributions of aligned data, ignoring joint distributions, however in practical applications, the joint distributions of data are more characteristic of data, so a method of solving the domain offset problem by aligning only edge distributions of a single token cannot always achieve satisfactory results.
Disclosure of Invention
Aiming at the problems, the invention provides a cross-domain bearing fault diagnosis method based on a Multi-characterization self-adaptive network, which is used for extracting multiple characterizations of data based on the Multi-characterization self-adaptive network (Multi-representation Adaptation Network, MRAN), and based on joint maximum mean difference (Joint Maximum Mean Discrepancy, JMMD) and Pseudo labels (Pseudo-Labeling), the joint probability distribution of the multiple characterizations of Ji Yuanyu and a target domain is realized, so that the domain offset problem is effectively solved, the bearing fault diagnosis under different working conditions can be adapted, and the good diagnosis effect is obtained.
In order to achieve the above purpose, the present invention adopts the following technical scheme:
the invention provides a multi-characterization self-adaptive network-based cross-domain bearing fault diagnosis method, which comprises the following steps:
collecting original vibration signals under different working conditions, preprocessing the collected signals to obtain a data set comprising a source domain data set and a target domain data set, and dividing the data set into training data and test data;
constructing a characteristic extraction model based on a multi-characterization adaptive network; the multi-characterization adaptive network comprises a feature extractor F and a classifier C, wherein the feature extractor F comprises a plurality of one-dimensional convolutional neural networks, can deeply extract a plurality of characterizations of data and is used for distributing the characterizations among cross-domain alignment fields; the classifier C is a fully-connected neural network and outputs the health state category of the bearing;
training the feature extraction model based on the training data;
and inputting the test data into a trained feature extraction model to obtain a bearing fault diagnosis result.
Further, preprocessing the acquired signals, including:
expanding the acquired signals in a sliding window mode;
converting the time domain signal into a frequency domain signal through fast Fourier transform;
and taking the labeled sample as source domain data, and taking unlabeled samples under different working conditions as target domain data.
Further, the feature extractor F includes: an initial shared one-dimensional convolutional neural network and a plurality of parallel one-dimensional convolutional neural networks of different-size convolutional kernels, the initial shared one-dimensional convolutional neural network being used for preliminary compression and dimension reduction of data, the parallel one-dimensional convolutional neural networks of different-size convolutional kernels comprising: the convolution layers of the convolution kernels with different sizes can capture information in different aspects, and multiple discriminant characterizations of the data are extracted in depth.
Further, the classifier C includes: the neuron number of the last full-connection layer corresponds to the health state category number of the bearing.
Further, training the feature extraction model based on the training data includes:
inputting the training data into a feature extractor F for forward propagation to obtain multiple characterizations H;
inputting the multiple characterizations H into a classifier C to obtain pseudo tags
Combining multiple characterizations H and pseudo tagsCalculating the joint maximum mean difference;
combining real label y and pseudo labelCalculating fault classification Loss and obtaining total Loss total ；
Updating model parameters based on gradient descent and back propagation algorithms;
repeating the steps until the parameters of the network model are converged.
Further, the process of obtaining multi-characterization and pseudo tags is as follows:
wherein X is an original data sample, H k Representing the extracted kth data representation, n r For the representation, g (·) is the initial shared one-dimensional convolutional neural network, j k (. Cndot.) is a multi-layer one-dimensional convolutional neural network, f (. Cndot.) is a classifier, θ g ，And theta f G (.cndot.) and h k Parameters of (-) and f (-).
Further, combining multiple characterizations H and pseudo tagsCalculating a joint maximum mean difference comprising:
wherein JMMD is the joint maximum mean difference, K - (. Cndot.) is expressed as a kernel function, s, t representing the source domain and the target domain, H s ，H t Multiple characterizations of the source domain and the target domain respectively,pseudo tag for source domain and target domain, n r For the number of features obtained by passing the sample through the feature extractor, n in the present invention r Is 3, n s ，n t For the number of samples of the source domain and the target domain, < >>Represents the kth feature of the ith sample of the source domain,/->Represents the jth sample kth feature of the target domain,/->Pseudo tag representing the ith sample of the source domain, for example>Pseudo tag, K, representing the jth sample of the target field H (·)，/>Representing kernel functions for features and pseudo tags, respectively.
Further, the total Loss total The method comprises the following steps:
wherein, alpha is a balance parameter, belongs to one of model super parameters and is used for controlling the proportion of the JMMD value in the whole loss function; j (·) is D S Is a fault classification loss of (1), wherein n is the number of bearing health status categories.
Further, the finally determined optimization targets are:
updating model parameters based on gradient descent and back propagation algorithms, comprising:
where θ is a model parameter, and μ is a learning rate.
Further, the gradient descent employs Adam optimization algorithm.
Compared with the prior art, the invention has the following beneficial effects:
the method of the invention is based on a multi-characterization self-adaptive network model and deep mines a plurality of characterizations of data, so that the model performs field self-adaptation in a plurality of representing spaces, domain invariant features of a source domain and a target domain are fully extracted, cross-field fault diagnosis is completed, and the method is a simple and effective method for solving the domain offset problem.
According to the method, based on the JMMD and the pseudo tag, the distribution difference of the source domain and the target domain is minimized by aligning the joint probability distribution of a plurality of characterizations of the data, and compared with other methods, knowledge migration and domain self-adaption can be better realized, so that the effect of cross-domain fault diagnosis is achieved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings may be obtained according to the drawings without inventive effort to a person skilled in the art.
FIG. 1 is a flow chart of a rolling bearing fault cross-domain diagnostic in an embodiment of the invention;
FIG. 2 is a schematic diagram of a multi-characterization adaptive network model according to an embodiment of the present invention;
FIG. 3 is a model of the present invention and other metrics based model at task D S →D T11 The characteristic dimension reduction distribution map of t-sne (a is the invention, b is CORAL, c is MK-MMD, and d is DANN).
Detailed Description
In order that those skilled in the art will better understand the present invention, a technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in which it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, shall fall within the scope of the present invention.
It should be noted that the terms "first," "second," and the like in the description and the claims of the present invention and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the invention described herein may be implemented in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
As shown in fig. 1, the multi-characterization adaptive network-based fault diagnosis method for the cross-domain bearing provided by the embodiment of the invention specifically includes the following steps:
step 1: collecting an original vibration signal and preprocessing:
collecting original vibration signals under different working conditions, expanding a data set in a sliding window mode, converting a time domain signal into a frequency domain signal through fast Fourier transform (Fast Fourier Transform, FFT), and taking a sample with a label as source domain data D S Label-free samples under different working conditions are used as target domain data D T The source domain data set and the target domain data set are processed according to 7:2:1 into a training set, a verification set and a test set; the expression is as follows: source domain data setTarget Domain dataset +.>Wherein (1)>Sample point for source domain, +.>Label for source field sample point, +.>For sample points of the target domain, n s N is the number of source domain samples t Is the number of target domain samples.
Step 2: constructing an MRAN model:
the multi-characterization adaptive network comprises a feature extractor F and a classifier C, wherein the feature extractor F is composed of a plurality of one-dimensional convolutional neural networks and can deeply extract a plurality of characterizations of data, and the multi-characterization adaptive network specifically comprises: the method comprises the steps of an initial shared one-dimensional convolutional neural network and a plurality of parallel one-dimensional convolutional neural networks with different sizes of convolutional kernels, wherein the former is mainly used for primarily compressing and reducing the dimension of data, the latter is formed by constructing a one-dimensional convolutional Layer (Conv 1d Layer), an Activation Layer (Activation Layer) and a Pooling Layer (Pooling Layer), the convolutional layers with different sizes of convolutional kernels can capture information in different aspects, and a plurality of discriminant representations of the data are deeply extracted and used for distributing a plurality of representations among cross-domain alignment fields. Classifier C is a fully connected neural network, outputting the health status category of the bearing, specifically including: the number of neurons of the last fully-connected Layer corresponds to the number of health status categories of the bearing as a function of the multi-Layer fully-connected Layer (Fully Connected Layer), the activation Layer, the random inactivation (Dropout Layer) Layer and the Softmax.
Step 3: training an MRAN model:
d in step 1 S And D T The training set data of (1) is input into a shared feature extractor F for forward propagation to obtain multiple representations of the data, and is input into a classifier C to obtain D S And D T Based on joint distribution of JMMD and pseudo tag alignment data, minimizing D S And D T The distribution difference between the two is extracted, so that domain invariant features of the two are extracted; at the same time combine with D S The real label in the method calculates the failure classification loss of the source domain, is used for standard supervision and learning, and ensures the accuracy of failure classification; taking the sum of the JMMD value and the fault classification loss as an objective function, carrying out iterative training through gradient descent and back propagation algorithm, and optimizing model parameters to obtain a trained multi-characterization self-adaptive network model;
the training process in step 3 specifically comprises the following steps:
(1) Will D S And D T Inputting the characteristic extractor F to obtain multiple characterizations H;
(2) Obtaining pseudo labels in the multi-characterization H classifier C；
The method comprises the following steps of firstly predicting unlabeled data by using a trained model, then retraining the model by combining the labeled data and the pseudo-label data, repeating iteration, and generating a more accurate pseudo-label in each iteration, so that the model performance is further improved;
the forward propagation process for obtaining multi-characterization and pseudo tags is specifically as follows:
wherein X is an original data sample, H k Representing the extracted kth data representation, n r For the representation number, g (·) is the initially shared one-dimensional convolutional neural network, h k (. Cndot.) is a multi-layer one-dimensional convolutional neural network, f (. Cndot.) is a classifier, θ g ，And theta f G (.cndot.) and h k Parameters of (-) and f (-).
(3) Combining multiple characterizations H and pseudo tagsCalculating JMMD;
wherein JMMD is a method for measuring distribution differences between source and target domains, which is based on the concept of MMD, by minimizing these distribution differences to achieve domain adaptation; MMD is a method for measuring the distance between two probability distributions, which judges the similarity between two probability distributions by comparing the difference of the average values of the two distributions in a certain feature space, and is used as an extension method of MMD for measuring the difference between a plurality of fields, and is more suitable for the transfer learning between the fields.
Wherein JMMD is the joint maximum mean difference, K - (. Cndot.) is expressed as a kernel function, s, t representing the source domain and the target domain, H s ，H t Multiple characterizations of the source domain and the target domain respectively,pseudo tag for source domain and target domain, n r Features obtained for the sample through the feature extractorNumber, in the present invention, n r Is 3, n s ，n t For the number of samples of the source domain and the target domain, < >>Represents the kth feature of the ith sample of the source domain,/->Represents the jth sample kth feature of the target domain,/->Pseudo tag representing the ith sample of the source domain, for example>Pseudo tag, K, representing the jth sample of the target field H (·)，/>Representing kernel functions for features and pseudo tags, respectively.
The JMMD is the joint maximum mean difference of all features of all samples, and if one batch_size JMMD is calculated, n can be calculated s And n t The value is replaced with the batch_size value.
In the formula, K (-) is expressed as a Kernel Function (Kernel Function), the Kernel Function is a mathematical Function for nonlinear data conversion, and common Kernel functions include polynomial Kernel functions, linear Kernel functions, gaussian Kernel functions and the like, the invention adopts Gaussian Kernel functions (Gaussian Kernel Function), also called radial basis Function kernels (Radial Basis Function Kernel, RBF Kernel), and the calculation formula is as follows:
where σ is the bandwidth.
(4) Combination D S True tag y and false tag of (2)Calculating fault classification Loss and obtaining total Loss total ；
The determination objective function is as follows:
wherein, alpha is a balance parameter, belongs to one of model super parameters and is used for controlling the proportion of the JMMD value in the whole loss function; j (·) is D S Since the fault classification is a classification problem, the cross entropy loss function has the following calculation formula:
wherein n is the number of bearing health status categories;
(5) Updating model parameters based on gradient descent and back propagation algorithms:
wherein θ is a model parameter, μ is a learning rate;
the gradient descent algorithm adopts an Adam optimization algorithm, and combines the advantages of two optimization algorithms, namely AdaGrad and RMSProp; the first moment estimate (First Moment Estimation, i.e., the mean of the gradient) and the second moment estimate (Second Moment Estimation, i.e., the non-centered variance of the gradient) of the gradient are taken into account together to calculate the update step size.
The finally determined optimization targets are as follows:
(6) Repeating the steps until the parameters of the network model are converged.
Step 4: testing the MRAN model:
d in step 1 T And (3) inputting the test set data into the multi-characterization self-adaptive network model obtained by training in the step (3) to obtain a classification result, analyzing the diagnosis performance and judging the quality of the model.
In order to verify the feasibility of the proposed model, the method for diagnosing the cross-domain bearing fault based on the multi-characterization adaptive network is further described in a comparative experiment mode.
The data set used for the experiment was a bearing failure data set from kesixi Chu Da (CWRU) of united states, which contains four conditions, collected by the acceleration sensor at four different loads (0, 1, 2 and 3 HP), with a sampling frequency of 12kHz, each condition containing four health states, normal (Normal), inner ring failure (IR), outer ring failure (OR) and rolling body failure (B), each failure site containing three failure dimensions (0.007 inch, 0.014 inch and 0.021 inch), and therefore one condition contains 10 health states, kesixi Chu Da (CWRU) bearing failure data set (load 0 HP) is specifically shown in table 1:
TABLE 1
The experimental steps are as follows:
s1: data preprocessing and partitioning data sets:
taking data with 0HP and no noise as a source domain data set, respectively adding the data with other loads into 10dB, 5dB, 0dB and minus 5dB Gaussian noise to obtain 12 target domain data sets with different probability distributions, wherein 12 migration tasks exist, and the diagnosis effect of the model is analyzed according to the average accuracy of the 12 migration tasks; expanding and sampling each data set in a sliding window mode, setting the window size as 1024 and the step length as 256, performing fast Fourier transform on the time domain signals obtained by sampling to obtain frequency domain signals, and performing fast Fourier transform on the time domain signals according to 7:2:1, a training set, a verification set and a test set are divided according to the proportion, and the fault diagnosis experiment data set is specifically shown in table 2:
TABLE 2
S2: constructing an MRAN model:
the MRAN model consists of a feature extractor F and a classifier C, as shown in FIG. 2;
the feature extractor F comprises an initial shared one-dimensional convolutional neural network and three branch convolutional neural networks, wherein the former comprises two convolutional layers, and the convolution kernels are respectively 8 and 16; the convolution kernel sizes are 5 and 3 respectively, and the step sizes are 2; each one-dimensional convolutional neural network of the latter is composed of 3 convolutional layers, 3 activating layers, 3 pooling layers and 1 unfolding layer, the convolution kernel size of each branch is 5, 9 and 13 respectively, the other parameters are consistent, the convolution kernel number is 32, and the step length is 2;
the activation layer is arranged behind each convolution layer, and the used activation function is a ReLU function;
the unfolding layer is arranged at the end of each branch, and the extracted features are unfolded into one-dimensional feature vectors, so that feature fusion is facilitated;
the classifier C consists of three full-connection layers, two activation layers, two random inactivation (dropout) layers and a Softmax function, wherein the neuron numbers of the full-connection layers are 256, 64 and 10 respectively; the activation layer is arranged behind each full connection layer, and the used activation function is also a ReLU function; a random deactivation layer is provided after each activation layer, which acts to prevent overfitting; the Softmax function maps the output of the last layer to a probability distribution with the sum of 1, and a fault classification result is obtained.
S3: training an MRAN model:
training the MRAN model constructed in the step S22 by combining the source domain training set and the target domain training set in the step S1, and evaluating a training result by using a verification set in the training process to prevent overfitting; the training adopts an Adam optimization algorithm, the learning rate is set to 0.001, the batch size is set to 64, and the preset iteration number is 100.
S4: testing the MRAN model:
inputting the test set of the target domain in the S1 into the MRAN model trained in the S3 to obtain a fault classification result, and generalizing and diagnosing the performance of the test model.
Experimental results:
in order to verify the superiority of the field adaptation based on the JMMD and the pseudo tag adopted by the method, a model based on other measurement methods (CORAL, MK-MMD, DANN) is selected for comparison experiments, a plurality of groups of migration tasks are tested, and comparison results of the different measurement methods in the plurality of groups of migration tasks are shown in the following table 3:
TABLE 3 Table 3
To further verify the diagnostic effect and migration ability of the MRAN model, the same metric method but a Single characterization adaptive network model (SRAN-representation Adaptation Network) was selected as a comparison method, multiple sets of migration tasks were tested, and the comparison results of different models in the multiple sets of migration tasks are shown in table 4 below:
TABLE 4 Table 4
Experimental results show that the method provided by the invention obtains 94.67% of average diagnosis accuracy in 12 groups of migration tasks of the CWRU data set, which is obviously superior to other methods and models, and shows that the method provided by the invention has good cross-domain diagnosis capability and domain generalization capability.
The invention carries out cross-domain fault diagnosis on target domain data which does not contain labels and has different working conditions based on a multi-characterization self-adaptive network, the JMMD measurement method and the pseudo-label technology, completes the migration of diagnosis knowledge, can generate excellent classification precision, has strong field self-adaptation capability, and provides an effective method for the intelligent fault diagnosis field.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some or all of the technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit of the invention.
Claims (10)
1. The multi-characterization self-adaptive network-based cross-domain bearing fault diagnosis method is characterized by comprising the following steps of:
collecting original vibration signals under different working conditions, preprocessing the collected signals to obtain a data set comprising a source domain data set and a target domain data set, and dividing the data set into training data and test data;
constructing a characteristic extraction model based on a multi-characterization adaptive network; the multi-characterization adaptive network comprises a feature extractor F and a classifier C, wherein the feature extractor F comprises a plurality of one-dimensional convolutional neural networks, can deeply extract a plurality of characterizations of data and is used for distributing the characterizations among cross-domain alignment fields; the classifier C is a fully-connected neural network and outputs the health state category of the bearing;
training the feature extraction model based on the training data;
and inputting the test data into a trained feature extraction model to obtain a bearing fault diagnosis result.
2. The multi-characterization adaptive network-based cross-domain bearing fault diagnosis method according to claim 1, wherein preprocessing the collected signals comprises:
expanding the acquired signals in a sliding window mode;
converting the time domain signal into a frequency domain signal through fast Fourier transform;
and taking the labeled sample as source domain data, and taking unlabeled samples under different working conditions as target domain data.
3. The multi-characterization adaptive network-based cross-domain bearing fault diagnosis method according to claim 1, wherein the feature extractor F comprises: an initial shared one-dimensional convolutional neural network and a plurality of parallel one-dimensional convolutional neural networks of different-size convolutional kernels, the initial shared one-dimensional convolutional neural network being used for preliminary compression and dimension reduction of data, the parallel one-dimensional convolutional neural networks of different-size convolutional kernels comprising: the convolution layers of the convolution kernels with different sizes can capture information in different aspects, and multiple discriminant characterizations of the data are extracted in depth.
4. A multi-characterization adaptive network-based cross-domain bearing fault diagnosis method according to claim 1 or 3, wherein the classifier C comprises: the neuron number of the last full-connection layer corresponds to the health state category number of the bearing.
5. The multi-characterization adaptive network-based cross-domain bearing fault diagnosis method according to claim 1, wherein training the feature extraction model based on the training data comprises:
inputting the training data into a feature extractor F for forward propagation to obtain multiple characterizations H;
inputting the multiple characterizations H into a classifier C to obtain pseudo tags
Combining multiple characterizations H and pseudo tagsCalculating the joint maximum mean difference;
combining real label y and pseudo labelCalculating fault classification Loss and obtaining total Loss total ；
Updating model parameters based on gradient descent and back propagation algorithms;
repeating the steps until the parameters of the network model are converged.
6. The multi-characterization adaptive network-based cross-domain bearing fault diagnosis method according to claim 5, wherein the process of obtaining the multi-characterization and pseudo tag is as follows:
wherein X is an original data sample, H k Representing the extracted kth data representation, n r For the representation number, g (·) is the initially shared one-dimensional convolutional neural network, h k (. Cndot.) is a multi-layer one-dimensional convolutional neural network, f (. Cndot.) is a classifier, θ g ，And theta f G (.cndot.) and h k Parameters of (-) and f (-).
7. The multi-token based adaptive network cross-domain according to claim 6A bearing fault diagnosis method is characterized by combining multiple characterization H and pseudo tagsCalculating a joint maximum mean difference comprising:
wherein JMMD is the joint maximum mean difference, K (-) is expressed as a kernel function, s, t represent the source domain and the target domain, H respectively s ，H t Multiple characterizations of the source domain and the target domain respectively,pseudo tag for source domain and target domain, n r For the number of features obtained by passing the sample through the feature extractor, n in the present invention r Is 3, n s ，n t For the number of samples of the source domain and the target domain, < >>Represents the kth feature of the ith sample of the source domain,/->Represents the jth sample kth feature of the target domain,/->Pseudo tag representing the ith sample of the source domain, for example>Pseudo tag, K, representing the jth sample of the target field H (·)，/>Representing kernel functions for features and pseudo tags, respectively.
8. The multi-characterization adaptive network-based cross-domain bearing fault diagnosis method according to claim 7, wherein the total Loss is total The method comprises the following steps:
wherein, alpha is a balance parameter, belongs to one of model super parameters and is used for controlling the proportion of the JMMD value in the whole loss function; j (·) is D S Is a fault classification loss of (1), wherein n is the number of bearing health status categories.
9. The multi-characterization adaptive network-based cross-domain bearing fault diagnosis method according to claim 8, wherein the finally determined optimization objective is:
updating model parameters based on gradient descent and back propagation algorithms, comprising:
where θ is a model parameter, and μ is a learning rate.
10. The multi-characterization adaptive network-based cross-domain bearing fault diagnosis method according to claim 9, wherein the gradient descent employs Adam optimization algorithm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311364218.0A CN117606801A (en) | 2023-10-20 | 2023-10-20 | Cross-domain bearing fault diagnosis method based on multi-characterization self-adaptive network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311364218.0A CN117606801A (en) | 2023-10-20 | 2023-10-20 | Cross-domain bearing fault diagnosis method based on multi-characterization self-adaptive network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117606801A true CN117606801A (en) | 2024-02-27 |
Family
ID=89954033
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311364218.0A Pending CN117606801A (en) | 2023-10-20 | 2023-10-20 | Cross-domain bearing fault diagnosis method based on multi-characterization self-adaptive network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117606801A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117830750A (en) * | 2024-03-04 | 2024-04-05 | 青岛大学 | Mechanical fault prediction method based on graph converter |
| CN117830750B (en) * | 2024-03-04 | 2024-06-04 | 青岛大学 | Mechanical fault prediction method based on graph converter |
-
2023
- 2023-10-20 CN CN202311364218.0A patent/CN117606801A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117830750A (en) * | 2024-03-04 | 2024-04-05 | 青岛大学 | Mechanical fault prediction method based on graph converter |
| CN117830750B (en) * | 2024-03-04 | 2024-06-04 | 青岛大学 | Mechanical fault prediction method based on graph converter |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Choi et al. | Gan-based anomaly detection and localization of multivariate time series data for power plant | |
| Karim et al. | A new generalized deep learning framework combining sparse autoencoder and Taguchi method for novel data classification and processing | |
| CN110048827B (en) | Class template attack method based on deep learning convolutional neural network | |
| Li et al. | Unsupervised feature learning for human activity recognition using smartphone sensors | |
| CN107066759B (en) | Steam turbine rotor vibration fault diagnosis method and device | |
| CN111161315B (en) | Multi-target tracking method and system based on graph neural network | |
| Gao et al. | Bearing fault diagnosis based on adaptive convolutional neural network with Nesterov momentum | |
| CN114755745B (en) | Hail weather identification and classification method based on multi-channel depth residual shrinkage network | |
| CN111562108A (en) | Rolling bearing intelligent fault diagnosis method based on CNN and FCMC | |
| Liang et al. | Multi-scale dynamic adaptive residual network for fault diagnosis | |
| CN113673346B (en) | Motor vibration data processing and state identification method based on multiscale SE-Resnet | |
| Özbılge et al. | Tomato disease recognition using a compact convolutional neural network | |
| CN111597991A (en) | Rehabilitation detection method based on channel state information and BilSTM-Attention | |
| Ye et al. | Multiscale weighted morphological network based feature learning of vibration signals for machinery fault diagnosis | |
| Peng et al. | Research on fault diagnosis method of rolling bearing based on 2DCNN | |
| CN113076920B (en) | Intelligent fault diagnosis method based on asymmetric domain confrontation self-adaptive model | |
| Wang et al. | A novel rolling bearing fault diagnosis method based on adaptive denoising convolutional neural network under noise background | |
| CN114152442A (en) | Rolling bearing cross-working condition fault detection method based on migration convolutional neural network | |
| CN117591950A (en) | Rolling bearing variable working condition fault diagnosis method, device, terminal and storage medium | |
| CN117009916A (en) | Actuator fault diagnosis method based on multi-sensor information fusion and transfer learning | |
| CN117606801A (en) | Cross-domain bearing fault diagnosis method based on multi-characterization self-adaptive network | |
| CN116361723A (en) | Bearing fault diagnosis and classification method based on multi-scale characteristics and attention | |
| CN115618933A (en) | Cross-domain fault diagnosis method based on sampling convolution and counterstudy | |
| Tian et al. | Fault diagnosis based on LMD-SVD and information-geometric support vector machine for hydraulic pumps | |
| Hao et al. | New fusion features convolutional neural network with high generalization ability on rolling bearing fault diagnosis |
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 |