CN115392315A - Gearbox fault detection method based on transferable features - Google Patents
Gearbox fault detection method based on transferable features Download PDFInfo
- Publication number
- CN115392315A CN115392315A CN202211055927.6A CN202211055927A CN115392315A CN 115392315 A CN115392315 A CN 115392315A CN 202211055927 A CN202211055927 A CN 202211055927A CN 115392315 A CN115392315 A CN 115392315A
- Authority
- CN
- China
- Prior art keywords
- time sequence
- data set
- gearbox
- training
- model
- 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
- 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
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention relates to a gearbox fault detection method based on transferable characteristics, which comprises the steps of obtaining a gearbox fault information data set, dividing the data set into a candidate domain data set and a target domain data set, and obtaining a first representative time sequence and a second representative time sequence through preprocessing; obtaining a data set with the most similar mode characteristics in the time domain through calculation based on the correlation between the first representative time sequence and the second representative time sequence; pre-training an LSTM model by using the most similar data set, extracting common features, and then performing time sequence prediction on the target domain data set by using the pre-trained model to obtain time domain features of the gearbox; and training a CNN model according to the spectrogram corresponding to the second representative time sequence, acquiring the frequency domain characteristics of the gearbox spectrogram, and combining the time domain characteristics and the frequency spectrum characteristics to obtain a prediction result. The method improves the accuracy of prediction.
Description
Technical Field
The invention relates to the technical field of fault detection, in particular to a gearbox fault detection method based on transferable characteristics.
Background
At present, for the prediction analysis of abundant time sequence data in the same field, a large number of mature technical means exist: the prediction can be completed by the aid of currently observed characteristic values and historical time sequence information through technologies such as RNN/LSTM/GRU and the like; multi-step prediction can be carried out by the techniques of Seq2Seq and the like; different Attention weights can be added to the time sequence data through an Attention mechanism, and prediction based on multiple characteristics is strengthened; characteristics such as space and the like can be added through a GeoMAN model, so that the prediction accuracy is improved; through a Transformer and a related variant model, a multidimensional long input sequence can be conveniently processed, and the accuracy of a predicted long sequence is improved; historical information of a wider field of view can be captured for predictive analysis through relevant models such as CNN; the accuracy of prediction can be improved by combining a plurality of models and innovating the models. Most of the above technologies need to satisfy the feature of abundant data volume in the same field.
The existing technology generally comprises the steps of using a model trained in another related field before by methods such as model parameter reuse, source domain and target domain characteristic conversion and the like through transfer learning so as to achieve the aim of predicting a target domain; and by a data enhancement correlation method, more useful information is extracted, the prediction precision is improved, and the like. In the above methods, it is necessary to find the source domain most suitable for the methods for the destination domain, and deeply mine the association relationship between the source domain and the destination domain.
In the mechanical field, a large number of rich theoretical bases are available to aid predictive analysis: for example, it is known that the life of a rolling bearing is predicted to be in positive correlation with a basic dynamic load and in negative correlation with an equivalent dynamic load, and a basic calculation formula and a look-up table are provided. But how to select a data source which can help the target domain to predict from the commonly used device data; how to enhance the prediction of data with small data quantity by using a data source with larger data quantity; and how to mine hidden information of time series data with small data volume is a problem which is relatively troublesome.
The industry has abundant theoretical mechanism models, basic mechanism knowledge is required to be met in the process of analyzing data, and the problem of how to enhance the accuracy of the data analysis model by utilizing the mechanism knowledge is the research requirement.
Disclosure of Invention
The invention aims to provide a gearbox fault detection method based on transferable characteristics, which aims to solve the problems that the prior art is applicable to a target domain, requires a scene with rich training data, and has little equipment fault data in an industrial environment.
In order to achieve the purpose, the invention provides the following scheme:
a gearbox fault detection method based on transferable features includes:
acquiring a gearbox fault information data set, dividing the data set into a candidate domain data set and a target domain data set, and preprocessing to obtain a first representative time sequence and a second representative time sequence;
obtaining a data set with the most similar mode characteristics in the time domain through calculation based on the correlation between the first representative time sequence and the second representative time sequence;
pre-training an LSTM model by using the most similar data set, extracting common features, and then performing time sequence prediction on the target domain data set by using the pre-trained model to obtain time domain features of the gearbox;
and training a CNN model according to the spectrogram corresponding to the second representative time sequence, acquiring the frequency domain characteristics of the gearbox spectrogram, and combining the time domain characteristics and the frequency spectrum characteristics to obtain a prediction result.
Preferably, the acquiring the gearbox fault information dataset comprises: acquiring a time sequence containing hidden information based on an equipment simulator, and constructing a gearbox fault information data set based on the time sequence containing hidden information, wherein the gearbox fault information data set comprises a plurality of time sequences of hidden information.
Preferably, obtaining the first representative time series and the second representative time series by preprocessing comprises:
and selecting the first representative time sequence and the second representative time sequence by disturbing the partial value of any dimension time sequence, sequentially removing any dimension time sequence and measuring the importance of different dimension time sequences on the final prediction of the data set, wherein the first representative time sequence is a representative time sequence of the candidate domain data set, and the second representative time sequence is a representative time sequence of the target domain data set.
Preferably, the step of scrambling the partial value of the time series of any dimension and the step of sequentially removing the time series of any dimension comprises the steps of:
putting multidimensional data into the LSTM model to perform multivariate prediction to obtain a first prediction result;
the method comprises the steps of disturbing the partial value of any dimension time sequence, determining the dimension, traversing each time sequence on the same dimension by using a preset threshold value, wherein the front and back sequence of a measuring point is exchanged in each cycle; carrying out classification prediction based on the LSTM model structure to obtain a second prediction result;
sequentially removing any dimension time sequence, removing one time sequence of measuring dimension each time, reducing the input dimension of the LSTM model, and predicting through the LSTM model structure to obtain a third prediction result;
and performing calculation based on the first prediction result, the second prediction result and the third prediction result, and if the calculation result value is larger, representing that the selected dimension contains more fault information, so as to select the first representative time sequence and the second representative time sequence.
Preferably, the obtaining of the most similar data set by calculation includes:
normalizing the first representative time sequence and the second representative time sequence, summing the normalized first representative time sequence and the normalized second representative time sequence, and performing mean value removing calculation to obtain a calculation result, wherein the calculation result represents the similarity between the candidate domain data set and the target domain data set, and the first k most similar data sets are selected from the candidate domain data sets as the most similar data sets based on the calculation result; wherein smaller calculations represent more similarity.
Preferably, the obtaining the time-domain features comprises:
and taking the most similar data set as the LSTM model pre-training data set, sequentially training the LSTM models, and performing multivariate time sequence prediction on the target domain data set through the trained LSTM models to obtain the time domain characteristics.
Preferably, pre-training the LSTM model comprises:
pre-training the LSTM model based on the most similar data set, modifying the structures of a full connection layer and a softmax layer according to the classification characteristics of different candidate domain data sets, inheriting model parameters before the full connection layer obtained by the previous training before pre-training the model each time, then continuing training, and finally obtaining the pre-trained LSTM model.
Preferably, the multivariate time series prediction of the target domain data set by the pre-trained LSTM model comprises:
transferring the pre-trained LSTM model to the target domain data set for training, modifying the structures of a full connection layer and a SoftMax layer in the LSTM model, introducing a CNN model for merging and training, inheriting model parameters before the full connection layer before the model is trained, training by using the target domain data set, and finally obtaining the time domain characteristics of the gearbox.
Preferably, training a CNN model according to the spectrogram corresponding to the second representative time sequence to obtain the frequency domain characteristics of the spectrogram of the gearbox, including:
and generating a corresponding spectrogram according to the second representative time sequence, and extracting the frequency domain characteristics of the gearbox spectrogram by using the information related to the meshing frequency and the side frequency band in the CNN capture spectrogram.
Preferably, obtaining the prediction result comprises:
and combining the result output by the LSTM model and the result output by the CNN model, namely combining the time domain characteristics and the frequency domain characteristics, and outputting the result to the finally modified full-link layer and the softmax layer to obtain the prediction result.
The invention has the beneficial effects that:
(1) The invention provides a transferable feature-based method for a research object with less research data volume, such as a gear box, and the model is pre-trained by using data sets in other similar fields, so that the commonality features are extracted, the robustness of the prediction model is enhanced, and the prediction precision is improved;
(2) For judging the similarity of the fields, the invention provides the method for utilizing the similarity between DTW (dynamic time warping) scale data sets, thereby avoiding the problem of negative migration caused by the migration of unrelated fields;
(3) Aiming at the problems that a plurality of industrial data sets usually have multi-dimensional observation data, but the specific dimension data is difficult to judge and contains more fault information, the invention provides a method for disordering a certain one-dimensional characteristic value and sequentially removing a certain one-dimensional observation characteristic based on a deep learning framework so as to judge the importance of each dimension on the final judgment state of the data set;
(4) In order to improve the prediction accuracy, the method learns the expert knowledge in the related gearbox field, so that the final prediction model structure is modified. And generating a corresponding spectrogram according to the selected original time domain data of the first representative time sequence of the gearbox, capturing information related to the meshing frequency and the side frequency band in the spectrogram by using the CNN, extracting frequency domain features in the time sequence data of the gearbox, and combining the frequency domain features with the time domain features for classification and prediction.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a flow chart of a gearbox fault detection method based on transferable features in accordance with an embodiment of the present invention;
FIG. 2 is a schematic diagram of the structure of a representative sequence of selected data sets in accordance with an embodiment of the present invention;
FIG. 3 is a schematic structural diagram of a method for selecting top k best candidate domains according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a structure of k best candidate domains for model pre-training according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of a prediction structure for gearbox data according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In order to make the aforementioned objects, features and advantages of the present invention more comprehensible, the present invention is described in detail with reference to the accompanying drawings and the detailed description thereof.
The invention provides a gearbox fault detection method based on transferable characteristics. The technical scheme provided by the invention is as follows (as shown in figure 1):
1) First, the data needs to be preprocessed. Because the device may collect time series of multiple dimensions through multiple sensors, some may contain much failure information to assist in analysis, and some may not be as helpful for analysis. An algorithm is needed from which the dimension containing the most hidden information (see fig. 2) is selected as the representative time series for the second step.
The invention provides that for each data set, the importance of different dimension time sequences on the final prediction of the data set is measured by disturbing the partial value of a dimension time sequence and sequentially removing the dimension time sequence, so that the representative time sequence of each data set is selected. The concrete measures are as follows:
for each data set, firstly putting multidimensional data into an LSTM model to perform multivariate prediction to obtain a prediction result 1;
and then shuffle the value of the time series part of a certain dimension. I.e. for each dimension, each time series in this dimension is traversed with step =2, where the front-to-back order of two measurement points is swapped in each cycle (say 1234- > 2143). Then, carrying out classification prediction by using the models with the same structure in the previous step to obtain a prediction result 2;
then one dimension time series is removed in turn. Namely, each time, a time sequence of measuring dimensionality is removed, the input dimensionality of the LSTM model is reduced, and the model with the same structure as the previous step is also used for prediction to obtain a prediction result 3;
it is known that the more the model is lost after human intervention on a certain dimension time series, the more important the corresponding dimension time series (characteristic) is, and more fault information can be contained. So abs ((result 2+ result 3) -result 1 x 2) for each dimension is calculated, and the larger the value, the more fault information this dimension contains, thus selecting a representative time series for each data set.
For each data set, a representative time series is selected, and the correlation calculation of the second step is carried out.
2) For a large amount of industrial source domain data, a method is needed to measure the correlation between all source domain data and a target domain, because the smaller the data distribution difference is, the more useful knowledge for migration is, the greater the help for the target domain is, and conversely if the correlation between the two is low, but migration learning is still performed, the performance may be worse than that of unutilizing the source domain knowledge, thereby bringing about "negative migration". This is a troublesome problem.
For industrial fault equipment, the equipment often undergoes three stages of fatigue, namely a fatigue crack initiation stage, a fatigue crack growth stage and a fatigue fracture stage. The points in time at which these phases occur may be different in different devices, but if they have similar pattern sub-sequences, they indicate a high degree of similarity between the two.
The invention provides a DTW-based method for measuring the similarity of time sequence data of all candidate domains and target domains, so that the candidate domain most related to the target domain of a gearbox is selected as a candidate domain for subsequent pre-training.
Similarity between the two data sets is calculated based on DTW between all the candidate domain representative time series and the target domain representative time series respectively. And then selecting the most similar first k data sets from the multiple candidate domain data sets as the designated candidate domains of the subsequently trained prediction model (as shown in FIG. 3). In this embodiment, the first representative time sequence represents the time sequence for the candidate domain, and the second representative time sequence represents the time sequence for the target domain.
3) Due to the fact that the similarity of the selected k candidate domains and the target domain time sequence data is high, the data volume of the candidate domains is larger, and the difficulty that the existing depth model is difficult to use for prediction analysis due to the fact that the data volume of the target domain is small can be solved.
The invention pre-trains the model of the same LSTM time sequence prediction classification by using k candidate domains selected in the second step in sequence, as shown in FIG. 4, extracts some common characteristics and helps the target domain to perform prediction analysis. The specific method comprises the following steps:
structurally, only the structures of the full connection layer and the softmax layer are modified according to the classification characteristics of different candidate domain data sets (because the number of fault types of different data sets may be different), and the previous model structure is not modified; in terms of parameters, before the model is pre-trained for each time, model parameters before the full connection layer obtained by the last training are inherited, and then training is carried out.
4) And transferring the model obtained by the training in the third step to a target domain for predictive analysis, so that the convergence speed of the model under the target domain can be increased, and the problem of small data volume of the target domain is solved. The specific method is as above:
migrating the pre-trained LSTM model to the target domain data set for training, and structurally positioning 5 computing nodes of a full connection layer and a SoftMax layer (because the label types of the target domain (gearbox) data set are 5 and need to be merged with the introduced CNN model), wherein the previously designed structure is not modified; parametrically, model parameters before a full connection layer are inherited before a model is trained, and then the model is trained by using a target domain (gearbox) data set.
And according to the deep analysis of the mechanism knowledge of the gearbox fault, the transmission error forms the main excitation source of the gear vibration and noise. The transmission error is large, and the collision is intensified when the gear enters and leaves the engagement in the running process, so that a high vibration peak value is generated, and the amplitude change and the phase change of a short time are formed. The invention mainly researches the gear tooth damage error, namely the gear tooth surface damage caused by various faults in the running of the gear, and the drive error excitation of the gear can be generated in the rotation of the gear.
The process of fault analysis of vibration time series data in industry mainly focuses on the phenotype characteristics of meshing frequency and side frequency bands. There are three main failure analysis methods in industry: power spectrum analysis, side band analysis, cepstrum analysis, which are all based on the transformed spectrogram or power spectrum, etc., and the prediction judgment is performed by mining the characteristics of the meshing frequency and the side band, as shown in fig. 5. The method comprises the following steps:
and generating a corresponding spectrogram according to the first representative time sequence of the gearbox data set, and extracting the frequency domain characteristics of the gearbox spectrogram by using the information related to the meshing frequency and the side frequency band in the CNN capture spectrogram. And finally, combining the result output by the LSTM model and the result output by the CNN model, namely combining the time domain characteristic and the frequency domain characteristic, and outputting the result to the modified full-link layer and the softmax layer to obtain a prediction result.
The above-described embodiments are only intended to describe the preferred embodiments of the present invention, and not to limit the scope of the present invention, and various modifications and improvements made to the technical solution of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.
Claims (10)
1. A gearbox fault detection method based on transferable features, comprising:
acquiring a gearbox fault information data set, dividing the data set into a candidate domain data set and a target domain data set, and preprocessing to obtain a first representative time sequence and a second representative time sequence;
obtaining a data set with the most similar mode characteristics in a time domain through calculation based on the correlation between the first representative time sequence and the second representative time sequence;
pre-training an LSTM model by using the most similar data set, extracting common features, and then performing time sequence prediction on the target domain data set by using the pre-trained model to obtain time domain features of the gearbox;
and training a CNN model according to the spectrogram corresponding to the second representative time sequence, acquiring the frequency domain characteristics of the gearbox spectrogram, and combining the time domain characteristics and the frequency spectrum characteristics to obtain a prediction result.
2. The transferable feature-based gearbox fault detection method of claim 1 wherein obtaining the gearbox fault information dataset comprises: acquiring a time sequence containing hidden information based on an equipment simulator, and constructing a gearbox fault information data set based on the time sequence containing the hidden information, wherein the gearbox fault information data set comprises a plurality of time sequences of the hidden information.
3. The transferable feature-based gearbox fault detection method of claim 1 wherein deriving the first and second representative time-series by preprocessing comprises:
and selecting the first representative time sequence and the second representative time sequence by disturbing the partial value of any dimension time sequence, sequentially removing any dimension time sequence and measuring the importance of different dimension time sequences on the final prediction of the data set, wherein the first representative time sequence is a representative time sequence of the candidate domain data set, and the second representative time sequence is a representative time sequence of the target domain data set.
4. A gearbox fault detection method based on transferable features according to claim 3, wherein the scrambling of any dimension time series partial values and the sequential elimination of any dimension time series comprises:
putting multidimensional data into the LSTM model to perform multivariate prediction to obtain a first prediction result;
the method comprises the steps of disturbing partial values of time sequences of any dimension, determining the dimension, traversing each time sequence on the same dimension by using a preset threshold value, and exchanging the front and back sequence of measuring points in each cycle; carrying out classification prediction based on the LSTM model structure to obtain a second prediction result;
sequentially removing any dimension time sequence, removing one time sequence of measuring dimension each time, reducing the input dimension of the LSTM model, and predicting through the LSTM model structure to obtain a third prediction result;
and performing calculation based on the first prediction result, the second prediction result and the third prediction result, and if the calculation result value is larger, representing that the selected dimension contains more fault information, so as to select the first representative time sequence and the second representative time sequence.
5. The transferable feature-based gearbox fault detection method of claim 4 wherein the calculating the most similar data set includes:
normalizing the first representative time sequence and the second representative time sequence, summing the normalized first representative time sequence and the normalized second representative time sequence, and performing mean value removing calculation to obtain a calculation result, wherein the calculation result represents the similarity between the candidate domain data set and the target domain data set, and the first k most similar data sets are selected from the candidate domain data sets as the most similar data sets based on the calculation result; wherein smaller calculation results represent more similarity.
6. The transferable feature-based gearbox fault detection method of claim 1 wherein obtaining the time-domain feature comprises:
and taking the most similar data set as an LSTM model pre-training data set, sequentially training the LSTM models, and performing multivariate time sequence prediction on the target domain data set through the trained LSTM models to obtain the time domain characteristics.
7. The transferable feature-based gearbox fault detection method of claim 6, wherein pre-training the LSTM model comprises:
pre-training the LSTM model based on the most similar data set, modifying the structures of a full connection layer and a softmax layer according to the classification characteristics of different candidate domain data sets, inheriting model parameters before the full connection layer obtained by previous training before pre-training the model each time, then continuing training, and finally obtaining the pre-trained LSTM model.
8. The transferable feature-based gearbox fault detection method of claim 7 wherein multivariate time series prediction is performed on the target domain data set by a pre-trained LSTM model comprising:
transferring the pre-trained LSTM model to the target domain data set for training, modifying the structures of a full connection layer and a SoftMax layer in the LSTM model, introducing a CNN model for merging and training, inheriting model parameters before the full connection layer before the model is trained, training by using the target domain data set, and finally obtaining the time domain characteristics of the gearbox.
9. The gearbox fault detection method based on transferable features of claim 1, wherein training a CNN model according to the spectrograms corresponding to the second representative time series to obtain the frequency domain features of the spectrograms of the gearbox comprises:
and generating a corresponding spectrogram according to the second representative time sequence, and extracting the frequency domain characteristics of the gearbox spectrogram by using the information related to the meshing frequency and the side frequency band in the CNN capture spectrogram.
10. The transferable feature-based gearbox fault detection method of claim 1 wherein deriving the predicted outcome comprises:
and combining the result output by the LSTM model and the result output by the CNN model, namely combining the time domain characteristics and the frequency domain characteristics, and outputting the result to the finally modified full-link layer and the softmax layer to obtain the prediction result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211055927.6A CN115392315B (en) | 2022-08-31 | 2022-08-31 | Gearbox fault detection method based on transferable characteristics |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211055927.6A CN115392315B (en) | 2022-08-31 | 2022-08-31 | Gearbox fault detection method based on transferable characteristics |
Publications (2)
Publication Number | Publication Date |
---|---|
CN115392315A true CN115392315A (en) | 2022-11-25 |
CN115392315B CN115392315B (en) | 2023-04-18 |
Family
ID=84124893
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202211055927.6A Active CN115392315B (en) | 2022-08-31 | 2022-08-31 | Gearbox fault detection method based on transferable characteristics |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115392315B (en) |
Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110730046A (en) * | 2019-10-18 | 2020-01-24 | 中国人民解放军陆军工程大学 | Cross-frequency-band spectrum prediction method based on deep migration learning |
CN112380773A (en) * | 2020-11-19 | 2021-02-19 | 华中科技大学鄂州工业技术研究院 | Fuel cell degradation prediction method, device and equipment based on transfer learning |
CN112507785A (en) * | 2020-11-02 | 2021-03-16 | 北京工商大学 | Rolling bearing fault analysis based on CNN and LSTM |
CN113344016A (en) * | 2020-02-18 | 2021-09-03 | 深圳云天励飞技术有限公司 | Deep migration learning method and device, electronic equipment and storage medium |
WO2021208799A1 (en) * | 2020-04-18 | 2021-10-21 | 华为技术有限公司 | Transfer model training method and apparatus and fault detection method and apparatus |
CN113591553A (en) * | 2021-06-18 | 2021-11-02 | 北京航天动力研究所 | Turbo pump migration learning fault intelligent judgment method based on small sample weight optimization |
CN113851131A (en) * | 2021-08-17 | 2021-12-28 | 西安电子科技大学广州研究院 | Cross-modal lip language identification method |
CN113869286A (en) * | 2021-12-01 | 2021-12-31 | 中国工程物理研究院电子工程研究所 | Self-adaptive multi-task intelligent fault diagnosis model and fault diagnosis method |
CN114034486A (en) * | 2021-10-11 | 2022-02-11 | 中国人民解放军92578部队 | Unsupervised transfer learning-based bearing fault diagnosis method for pump mechanical equipment |
CN114372491A (en) * | 2021-12-31 | 2022-04-19 | 山东科讯信息科技有限公司 | Fault identification method of rotating equipment and related equipment |
WO2022134766A1 (en) * | 2020-12-24 | 2022-06-30 | 华为技术有限公司 | Scene migration method, apparatus and electronic device |
-
2022
- 2022-08-31 CN CN202211055927.6A patent/CN115392315B/en active Active
Patent Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110730046A (en) * | 2019-10-18 | 2020-01-24 | 中国人民解放军陆军工程大学 | Cross-frequency-band spectrum prediction method based on deep migration learning |
CN113344016A (en) * | 2020-02-18 | 2021-09-03 | 深圳云天励飞技术有限公司 | Deep migration learning method and device, electronic equipment and storage medium |
WO2021208799A1 (en) * | 2020-04-18 | 2021-10-21 | 华为技术有限公司 | Transfer model training method and apparatus and fault detection method and apparatus |
CN112507785A (en) * | 2020-11-02 | 2021-03-16 | 北京工商大学 | Rolling bearing fault analysis based on CNN and LSTM |
CN112380773A (en) * | 2020-11-19 | 2021-02-19 | 华中科技大学鄂州工业技术研究院 | Fuel cell degradation prediction method, device and equipment based on transfer learning |
WO2022134766A1 (en) * | 2020-12-24 | 2022-06-30 | 华为技术有限公司 | Scene migration method, apparatus and electronic device |
CN113591553A (en) * | 2021-06-18 | 2021-11-02 | 北京航天动力研究所 | Turbo pump migration learning fault intelligent judgment method based on small sample weight optimization |
CN113851131A (en) * | 2021-08-17 | 2021-12-28 | 西安电子科技大学广州研究院 | Cross-modal lip language identification method |
CN114034486A (en) * | 2021-10-11 | 2022-02-11 | 中国人民解放军92578部队 | Unsupervised transfer learning-based bearing fault diagnosis method for pump mechanical equipment |
CN113869286A (en) * | 2021-12-01 | 2021-12-31 | 中国工程物理研究院电子工程研究所 | Self-adaptive multi-task intelligent fault diagnosis model and fault diagnosis method |
CN114372491A (en) * | 2021-12-31 | 2022-04-19 | 山东科讯信息科技有限公司 | Fault identification method of rotating equipment and related equipment |
Non-Patent Citations (4)
Title |
---|
WENTAO ZHANG 等: "Intelligent Machine Fault Diagnosis Using Convolutional Neural Networks and Transfer Learning" * |
XIAORUI SHAO 等: "Fault Diagnosis of Bearing Based on Convolutional Neural Network Using Multi-Domain Features" * |
杨格: "统计金融模型及金融市场波动持续时间序列的研究" * |
鞠永慧: "基于机器学习的加密网络数据分类识别关键技术研究与实现" * |
Also Published As
Publication number | Publication date |
---|---|
CN115392315B (en) | 2023-04-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104819846B (en) | Rolling bearing sound signal fault diagnosis method based on short-time Fourier transform and sparse laminated automatic encoder | |
Chen et al. | A novel fault diagnosis model for gearbox based on wavelet support vector machine with immune genetic algorithm | |
CN103268519B (en) | Based on the power-system short-term load forecasting method and the device that improve Lyapunov index | |
CN105981025A (en) | Actually-measured marine environment data assimilation method based on sequence recursive filtering three-dimensional variation | |
CN112528564A (en) | Bridge structure damage identification method based on sparse Bayesian learning | |
US8718803B2 (en) | Method for calculating measures of similarity between time signals | |
Han et al. | Fault diagnosis of rolling element bearings based on multiscale dynamic time warping | |
Shu et al. | A multi-task learning-based automatic blind identification procedure for operational modal analysis | |
Xu et al. | Deep coupled visual perceptual networks for motor fault diagnosis under nonstationary conditions | |
CN103852525B (en) | Acoustic emission signal recognition methods based on AR-HMM | |
CN103310113A (en) | Universal blood glucose prediction method based on frequency band separation and data modeling | |
Shang et al. | An intelligent fault diagnosis system for newly assembled transmission | |
CN111122162A (en) | Industrial system fault detection method based on Euclidean distance multi-scale fuzzy sample entropy | |
Lin et al. | Adaptive stochastic resonance quantified by a novel evaluation index for rotating machinery fault diagnosis | |
CN114330834A (en) | Charging pile power consumption prediction method based on self-updating cubic exponential smoothing method | |
CN113505664A (en) | Fault diagnosis method for planetary gear box of wind turbine generator | |
CN114894468A (en) | Chaos detection-based early weak fault diagnosis method for planetary gear box | |
CN111078853B (en) | Question-answering model optimization method, device, computer equipment and storage medium | |
Parvin et al. | A comprehensive interturn fault severity diagnosis method for permanent magnet synchronous motors based on transformer neural networks | |
CN115392315B (en) | Gearbox fault detection method based on transferable characteristics | |
Yao et al. | Multiband weights-induced periodic sparse representation for bearing incipient fault diagnosis | |
Techane et al. | Rotating machinery prognostics and application of machine learning algorithms: Use of deep learning with similarity index measure for health status prediction | |
Li et al. | EMD-TEO based speech emotion recognition | |
CN113469408A (en) | Running state trend prediction method and system for phase modulator | |
CN110044619B (en) | Multi-fault feature identification method based on sparse multi-cycle group lasso |
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 |