CN115165364A - Wind turbine generator bearing fault diagnosis model construction method based on transfer learning - Google Patents
Wind turbine generator bearing fault diagnosis model construction method based on transfer learning Download PDFInfo
- Publication number
- CN115165364A CN115165364A CN202210785595.0A CN202210785595A CN115165364A CN 115165364 A CN115165364 A CN 115165364A CN 202210785595 A CN202210785595 A CN 202210785595A CN 115165364 A CN115165364 A CN 115165364A
- Authority
- CN
- China
- Prior art keywords
- model
- bearing
- transmission end
- training
- state
- 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
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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Acoustics & Sound (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention relates to a method for detecting the state and diagnosing the fault of wind driven generator equipment, in particular to a method for constructing a wind driven generator bearing fault diagnosis model based on transfer learning, which comprises the following steps: acquiring training original data; (2) selecting a pre-training model; (3) Training the pre-training model by using a model migration method; (4) And performing weighted integration on the model to determine a final algorithm network model. The technical scheme of the invention has the following beneficial effects: (1) The model with excellent performance is adopted for transfer learning, so that a large amount of model parameter setting and super-parameter self-learning are omitted, and the development efficiency of the wind driven generator fault diagnosis model is improved; (2) Accurately diagnosing early faults of a transmission end and a non-transmission end of a bearing; (3) Under the condition of bearing failure, maintenance personnel can carry out state repair and preventive repair according to the failure state of the bearing, so that planned repair and excessive repair are reduced, the maintenance cost and the maintenance time are saved, and the maintenance efficiency is improved.
Description
Technical Field
The invention relates to a method for detecting the state of wind driven generator equipment and diagnosing faults, in particular to a method for constructing a wind driven generator bearing fault diagnosis model based on transfer learning.
Background
Because the wind driven generator equipment and spare parts have larger fault interval discreteness, regular maintenance can cause higher maintenance cost and more shutdown maintenance time, so that the state maintenance and the preventive maintenance can effectively reduce the maintenance cost, reduce the accident shutdown rate and have higher investment-to-income ratio. Condition monitoring is a technique for sensing the health of a device, enabling potential problems to be detected and diagnosed at an early stage of their development and corrected by appropriate recovery measures before the problem becomes serious.
The existing fault diagnosis and identification scheme based on feature analysis and machine learning is characterized in that a bearing vibration signal in operation is collected, time domain analysis is carried out on the collected data, an identification model is built by using methods such as threshold judgment and logical reasoning, expert knowledge is needed to carry out manual feature extraction and selection, and a shallow classifier is additionally arranged. However, the identification scheme only carries out data analysis on the time domain signals of vibration or audio, and has the problems that incomplete feature extraction affects the performance of the model and the identification rate of bearing faults is low.
The method is characterized in that a vibration signal of a motor is obtained through a monitoring system, a deep learning network model is built and trained to perform fault location and classification, however, a large number of model parameters such as a convolutional layer and a pooling layer need to be designed for building the model, and due to knowledge difference of professionals, the conditions that a large amount of time is needed for model training and a resource configuration result is not ideal exist.
In an actual industrial system, when the working state of the wind driven generator changes, the structure of a sound signal is changed, different sound signal characteristics appear, and the running state of equipment can be judged according to the change of the sound signal characteristics of the equipment. The transfer learning is a method for solving the domain adaptation problem, a source domain with a large amount of label data and available parameter setting is processed in the transfer learning by transferring the learned knowledge, and the target domain with a small amount of labels can be applied to different works only by finely adjusting the trained model and using a small amount of label data to train the model. Therefore, the model training time can be saved and the development efficiency can be improved by using the existing model with excellent performance to perform transfer learning.
Disclosure of Invention
The invention relates to a wind power generator bearing fault diagnosis model construction method based on transfer learning, which solves the following problems: (1) Constructing a bearing fault diagnosis model by adopting an audio signal to realize fault identification of the bearing; (2) The existing network model is applied to a wind driven generator bearing fault identification system by using a transfer learning method, so that the development efficiency of the model is improved; (3) Aiming at the problem that audio signals are difficult to identify, the audio signals are converted into two-dimensional information, and a weighted integration method is adopted to carry out weighted integration on the migration learning model, so that the bearing fault identification accuracy is improved.
The invention is realized by adopting the following technical scheme: the method for constructing the wind turbine generator bearing fault diagnosis model based on the transfer learning comprises the following steps:
(1) Acquiring a training original data set;
(2) Selecting an AlexNet network, an Inception V3 network and a VGG-16 network as a pre-training model;
(3) Training the pre-training model by applying a model migration method to obtain an AlexNet migration model, an Inception V3 migration model and a VGG-16 migration model which are respectively marked as a model 1, a model 2 and a model 3;
(4) Carrying out weighted integration on the models to determine a final integration algorithm model, wherein aiming at the recognition accuracy rate and the overall recognition accuracy rate of the models 1, 2 and 3 on different faults, a weight distribution formula is as follows:
wherein Z ij Representing the identification weight of the ith model to the jth fault, A ij Representing the identification accuracy of the ith model to the jth fault, A i Identifying an accuracy for the ith model as a whole; integration algorithmThe model weights the fault positioning and classification information of the model 1, the model 2 and the model 3 to obtain a final fault diagnosis result.
According to the method for constructing the wind driven generator bearing fault diagnosis model based on the transfer learning, the bearings are respectively arranged at the transmission end and the non-transmission end of the wind driven generator according to two fault conditions of bearing outer ring stripping and bearing inner ring stripping, and 5 operation schemes are designed: the first scheme comprises the following steps: the state of the transmission end bearing is normal, and the state of the non-transmission end bearing is normal; scheme II: the state of the bearing at the transmission end is that the outer ring is peeled off, and the state of the bearing at the non-transmission end is normal; the third scheme is as follows: the state of the bearing at the transmission end is normal, and the state of the bearing at the non-transmission end is outer ring stripping; and the scheme is as follows: the bearing state of the transmission end is that the inner ring is stripped, and the bearing state of the non-transmission end is normal; and a fifth scheme: the bearing state of the transmission end is normal, and the bearing state of the non-transmission end is inner ring stripping; each operating scenario is considered a type of fault.
According to the wind turbine generator bearing fault diagnosis model construction method based on transfer learning, the acquisition process of an original data set is as follows: the wind driven generator is installed on an experiment table in an inclined theta degree mode to operate with load, audio signal data are collected for 5 minutes under M rotating speeds in each operation scheme, and 5 groups of test data are counted;
dividing the M groups of audio original data of each scheme into N time series with the length of L, totaling 5 × M × N, respectively performing Fourier transform on the time series with the length of L, so that the 1 × L original data series become 2 × L two-dimensional data series, adding a label to each two-dimensional data series, and finally forming 5 × M N groups of original data sets with the length of 2 × L and labels.
According to the wind turbine generator bearing fault diagnosis model construction method based on transfer learning, an original data set is divided into a training set, a verification set and a test set according to a certain proportion, and a pre-training model is trained.
According to the wind turbine generator bearing fault diagnosis model construction method based on transfer learning, the AlexNet network training process comprises the following steps: and (3) retaining the structures and parameters of the convolution layer and the pooling layer, randomly initializing the last full connection layer and the output layer by adopting new parameters, namely cutting off the last softmax layer, connecting the softmax layer after the random initialization of the new parameters, training the newly constructed network by using a training set, and finally testing the network by using a test set sample.
The technical scheme of the invention has the following beneficial effects:
(1) The model with excellent performance is adopted for transfer learning, so that a large amount of model parameter setting and super-parameter self-learning are omitted, and the development efficiency of the wind driven generator fault diagnosis model is improved;
(2) The early-stage faults of the transmission end and the non-transmission end of the bearing are accurately diagnosed, and the serious faults of the bearing are avoided through maintenance modes such as generator maintenance, waste oil treatment, carbon deposition treatment, filter cotton replacement and the like during daily maintenance, so that the wind turbine generator is shut down and the generated energy is influenced;
(3) Under the condition of bearing failure, maintenance personnel can carry out state repair and preventive repair according to the failure state of the bearing, so that planned repair and excessive repair are reduced, the maintenance cost and the maintenance time are saved, and the maintenance efficiency is improved.
Drawings
Fig. 1 is a waveform diagram of an acquired signal.
FIG. 2 is a model migration flow diagram.
Fig. 3 is a confusion matrix diagram of an AlexNet migration model.
Detailed Description
The method for constructing the wind turbine generator bearing fault diagnosis model based on the transfer learning comprises the following steps: acquiring training original data; (2) selecting a pre-training model; (3) Training the pre-training model by applying a model migration method; (4) And performing weighted integration on the model to determine a final algorithm network model.
Obtaining training raw data
(1) Prefabricating bearing faults:
prefabricating common faults for the bearing, wherein the fault types comprise: bearing inner race is peeled off, bearing inner race peels off, installs the bearing respectively at aerogenerator transmission end, non-transmission end, 5 kinds of operation schemes in total:
the first scheme comprises the following steps: the state of the transmission end bearing is normal, and the state of the non-transmission end bearing is normal;
scheme II: the state of the bearing at the transmission end is that the outer ring is peeled off, and the state of the bearing at the non-transmission end is normal;
and a third scheme is as follows: the state of the bearing at the transmission end is normal, and the state of the bearing at the non-transmission end is outer ring stripping;
the scheme four is as follows: the bearing state of the transmission end is that the inner ring is stripped, and the bearing state of the non-transmission end is normal;
and a fifth scheme: the state of the bearing at the transmission end is normal, and the state of the bearing at the non-transmission end is that the inner ring is peeled off;
(2) Test data acquisition processing
The wind driven generator is installed on a test bench to run with load by inclining theta degrees, audio signal data are collected for 5 minutes under M rotating speeds in each running scheme, 5 groups of test data are calculated, and part of data are shown in figure 1.
Dividing the M groups of audio original data of each scheme into N time series with the length of L, totaling 5 × M × N, respectively performing Fourier transform on the time series with the length of L, so that the 1 × L original data series become 2 × L two-dimensional data series, adding a label to each two-dimensional data series, and finally forming 5 × M N groups of original data sets with the length of 2 × L and labels.
Training a pre-training model by applying a model migration method
(1) Model training
The method comprises the steps of dividing an original signal data set obtained by an experiment into a training set, a verification set and a test set according to a certain proportion, firstly selecting an AlexNet network, training more than 100 million images on ImageNet by each parameter in the AlexNet network, learning abundant feature representation, and proving the effectiveness of extracted features on classification. Therefore, the method can adapt to a new image only by marginally adjusting the parameters, the structures and parameters of the convolution layer and the pooling layer are reserved, the last full-connection layer and the last output layer are initialized randomly by adopting new parameters, namely the last softmax layer is cut off, then the softmax layer after the new parameter initialization is connected, a newly constructed network is trained by using a training set, and finally a network model is tested by using a test set sample, wherein the whole flow is shown in fig. 2.
An Inception V3 network and a VGG-16 network are selected, the Inception V3 is used as a third generation model of the GooLeNet series, the parameter quantity can be well controlled, the image recognition performance is good, and the VGG-16 has good classification and positioning performance. And respectively carrying out training test on the two network models by adopting the same method. Finally, 3 network models with different parameters are obtained, namely an AlexNet migration model, an Inception V3 migration model and a VGG-16 migration model, which are respectively marked as a model 1, a model 2 and a model 3.
Weighting integration is carried out on the model to determine the final algorithm network model
According to the obtained models 1, 2 and 3, different fault types are weighted according to the recognition accuracy and the integral recognition accuracy of the models to different faults, and the weight distribution formula is as follows:
wherein Z ij Representing the identification weight of the ith model to the jth fault, A ij Representing the identification accuracy of the ith model to the jth fault, A i Overall recognition accuracy for the fault for the ith model; a. The ij And A i Calculating the result of the confusion matrix tested by the algorithm model; as shown in fig. 3, the test set verification confusion matrix of the model 1 is shown in fig. 3, and it can be seen that the overall recognition rate is 91.4%, so as to avoid the frequent occurrence of fault misjudgment due to low accuracy of a single model, and improve the judgment accuracy by adopting different models to perform an integrated fusion method.
The operation scheme two fault types are taken as an example for explanation: the recognition accuracy of the model 1, the model 2 and the model 3 to the fault type of the scheme two is 90.8%, 93.5% and 92.8%, so that the weight distribution is as follows: 0.316, 0.349 and 0.335, and the weight distribution of each model to different faults is calculated as follows:
the fault type one: the weight assignment for models 1, 2, 3 is: 0.347, 0.335, 0.318;
and (2) fault type II: the weight assignment for models 1, 2, 3 is: 0.316, 0.349, 0.335;
and (3) fault type three: the weight assignment for models 1, 2, 3 is: 0.346, 0.343, 0.311;
and (4) fault type four: the weight assignment for models 1, 2, 3 is: 0.352, 0.323, 0.325;
and (5) fault type five: the weight assignment for models 1, 2, 3 is: 0.328, 0.354, 0.318;
the integration algorithm model performs weighted integration on the fault location and classification information of the input audio signals of the model 1, the model 2 and the model 3 to obtain a final fault diagnosis result.
The output result of the weighted integration model is described in the following output situations that may occur in the model:
(1) Models 1, 2, 3 output the same results
The judgment results of the models 1, 2 and 3 are respectively a fault type two, a fault type two and a fault type two, the result of the integrated algorithm model for the fault type two is 0.316+0.349+0.335=1, the results of the integrated algorithm model calculated as the fault types one, three, four and five are all 0, and the algorithm identification result is the fault type two.
(2) Model 1, 2, 3 outputs two identical results
The judgment results of the models 1, 2 and 3 are respectively the two, two and five fault types, the result of the integrated algorithm model is 0.316+0.349=0.665 for the two fault type, the result of the integrated algorithm model is 0.318 for the five fault type, and the results of the first, third and fourth fault types are all 0, so the final result is the two fault type.
(3) Model 1, 2, 3 output three different results
The judgment results of the models 1, 2 and 3 are respectively a fault type one, a fault type three and a fault type five, the result of the integrated algorithm model is 0.347, the result of the integrated algorithm model is 0.343, the result of the integrated algorithm model is 0.318, the result of the integrated algorithm model is 0, and the results of the integrated algorithm model are both 0, so that the final result is the fault type one.
The judgment results of the models 1, 2 and 3 are respectively a fault type five, a fault type one and a fault type two, the result of the integrated algorithm model is 0.335, the result of the fault type one is considered to be 0.335, the result of the fault type five is considered to be 0.328, the results of the fault types three and four are 0, the comprehensive result has the condition of the same value, the result is not credible, and data needs to be collected and judged again.
Claims (5)
1. The method for constructing the wind turbine generator bearing fault diagnosis model based on the transfer learning is characterized by comprising the following steps of: the method comprises the following steps:
(1) Acquiring a training original data set;
(2) Selecting an AlexNet network, an Inception V3 network and a VGG-16 network as a pre-training model;
(3) Training the pre-training model by applying a model migration method to obtain an AlexNet migration model, an Inception V3 migration model and a VGG-16 migration model which are respectively marked as a model 1, a model 2 and a model 3;
(4) Carrying out weighted integration on the models to determine a final integration algorithm model, wherein aiming at the recognition accuracy rate and the overall recognition accuracy rate of the models 1, 2 and 3 on different faults, a weight distribution formula is as follows:
wherein Z ij Representing the identification weight of the ith model to the jth fault, A ij Representing the identification accuracy of the ith model to the jth fault, A i Identifying an accuracy for the ith model as a whole; the integration algorithm model performs weighted integration on the fault location and classification information of the model 1, the model 2 and the model 3 to obtain a final fault diagnosis result.
2. The wind turbine generator bearing fault diagnosis model building method based on transfer learning of claim 1, characterized in that: according to bearing inner race strip, two kinds of fault conditions are peeled off to the bearing inner race, install the bearing respectively at aerogenerator transmission end, non-transmission end, design 5 kinds of operation schemes: the first scheme is as follows: the state of the transmission end bearing is normal, and the state of the non-transmission end bearing is normal; scheme II: the state of the bearing at the transmission end is that the outer ring is peeled off, and the state of the bearing at the non-transmission end is normal; and a third scheme is as follows: the state of the bearing at the transmission end is normal, and the state of the bearing at the non-transmission end is outer ring stripping; and the scheme is as follows: the state of the bearing at the transmission end is that the inner ring is stripped, and the state of the bearing at the non-transmission end is normal; and a fifth scheme: the state of the bearing at the transmission end is normal, and the state of the bearing at the non-transmission end is that the inner ring is peeled off; each operating scenario is considered a type of fault.
3. The method for building the wind turbine generator bearing fault diagnosis model based on the transfer learning of claim 2, wherein the method comprises the following steps: the acquisition process of the original data set comprises the following steps: the wind driven generator is installed on an experiment table in an inclined theta degree mode to operate with load, audio signal data are collected for 5 minutes under M rotating speeds in each operation scheme, and 5 groups of test data are counted;
dividing the M groups of audio original data of each scheme into N time series with the length of L, totaling 5 × M × N, respectively performing Fourier transform on the time series with the length of L, so that the 1 × L original data series become 2 × L two-dimensional data series, adding a label to each two-dimensional data series, and finally forming 5 × M N groups of original data sets with the length of 2 × L and labels.
4. The method for building the wind turbine generator bearing fault diagnosis model based on the transfer learning of claim 3, wherein the method comprises the following steps: the original data set is divided into a training set, a verification set and a test set according to a certain proportion, and a pre-training model is trained.
5. The method for building the wind turbine generator bearing fault diagnosis model based on the transfer learning of claim 4, wherein the method comprises the following steps: the process of training the AlexNet network comprises the following steps: and (3) reserving structures and parameters of the convolution layer and the pooling layer, randomly initializing the last full-connection layer and the output layer by adopting new parameters, namely cutting off the last softmax layer, connecting the softmax layer after the random initialization of the new parameters, training the newly constructed network by using a training set, and finally testing the network by using a test set sample.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210785595.0A CN115165364A (en) | 2022-07-04 | 2022-07-04 | Wind turbine generator bearing fault diagnosis model construction method based on transfer learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210785595.0A CN115165364A (en) | 2022-07-04 | 2022-07-04 | Wind turbine generator bearing fault diagnosis model construction method based on transfer learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115165364A true CN115165364A (en) | 2022-10-11 |
Family
ID=83491122
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210785595.0A Pending CN115165364A (en) | 2022-07-04 | 2022-07-04 | Wind turbine generator bearing fault diagnosis model construction method based on transfer learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115165364A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116628564A (en) * | 2023-04-20 | 2023-08-22 | 上海宇佑船舶科技有限公司 | Model training method and system for detecting generator state |
-
2022
- 2022-07-04 CN CN202210785595.0A patent/CN115165364A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116628564A (en) * | 2023-04-20 | 2023-08-22 | 上海宇佑船舶科技有限公司 | Model training method and system for detecting generator state |
| CN116628564B (en) * | 2023-04-20 | 2024-03-12 | 上海宇佑船舶科技有限公司 | Model training method and system for detecting generator state |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107220469B (en) | Method and system for estimating state of fan | |
| CN106769052B (en) | A kind of mechanical system rolling bearing intelligent failure diagnosis method based on clustering | |
| CN110110768B (en) | Rolling bearing fault diagnosis method based on parallel feature learning and multiple classifiers | |
| CN112036547B (en) | Rolling bearing residual life prediction method combining automatic feature extraction with LSTM | |
| CN108647786B (en) | Rotary machine on-line fault monitoring method based on deep convolution countermeasure neural network | |
| CN109255333B (en) | Fault hybrid diagnosis method for rolling bearing of large wind turbine generator | |
| CN111046945A (en) | Fault type and damage degree diagnosis method based on combined convolutional neural network | |
| Zhong et al. | An intelligent fault diagnosis method based on STFT and convolutional neural network for bearings under variable working conditions | |
| CN106762452B (en) | Fan master control system fault diagnosis and on-line monitoring method based on data-driven | |
| CN113339204B (en) | Wind driven generator fault identification method based on hybrid neural network | |
| CN112665852B (en) | Variable working condition planetary gearbox fault diagnosis method and device based on deep learning | |
| CN113240022A (en) | Wind power gear box fault detection method of multi-scale single-classification convolutional network | |
| CN113505664A (en) | Fault diagnosis method for planetary gear box of wind turbine generator | |
| CN115791174B (en) | Rolling bearing abnormality diagnosis method, system, electronic equipment and storage medium | |
| CN116502072B (en) | Robust fault diagnosis method for key components of wind generating set under complex variable working conditions | |
| CN103852255B (en) | Based on neutral net wind power generating set typical case's drive failures intelligent diagnosing method | |
| CN112766200A (en) | Planetary gear box small sample fault diagnosis method based on one-dimensional convolutional neural network | |
| CN112508242A (en) | Method for constructing bearing fault location and classification model of wind power generator | |
| CN115165364A (en) | Wind turbine generator bearing fault diagnosis model construction method based on transfer learning | |
| CN115060491A (en) | Fan bearing operation health degree assessment method and system based on multi-source data | |
| CN113673442B (en) | Variable working condition fault detection method based on semi-supervised single classification network | |
| Zare et al. | Convolutional neural networks for wind turbine gearbox health monitoring | |
| CN113887586A (en) | FFCNN-SVM transfer learning fault diagnosis method based on feature fusion under small sample | |
| CN115163424A (en) | Wind turbine generator gearbox oil temperature fault detection method and system based on neural network | |
| Yang et al. | Self-attention parallel fusion network for wind turbine gearboxes 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 |