CN108596261A - Based on the gear parameter oversampler method for generating confrontation network model - Google Patents
Based on the gear parameter oversampler method for generating confrontation network model Download PDFInfo
- Publication number
- CN108596261A CN108596261A CN201810399022.8A CN201810399022A CN108596261A CN 108596261 A CN108596261 A CN 108596261A CN 201810399022 A CN201810399022 A CN 201810399022A CN 108596261 A CN108596261 A CN 108596261A
- Authority
- CN
- China
- Prior art keywords
- data
- generator
- classification
- gear parameter
- arbiter
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Biophysics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Neurology (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention provides a kind of gear parameter oversampler method based on generation confrontation network model, carries out according to the following steps:S1:It determines and needs the gear parameter sampled and acquire corresponding data;S2:A generator and an arbiter are built by neural network;S3:In input noise sequence signal to generator, obtain generating data;S4:The data of generation and original sampling data input arbiter are subjected to classification judgement;S5:Using Softmax layers of progress linear transformation, final classification result is obtained;S6:Classification results and setting value are relatively obtained into error in classification, when the weight for more than goal-selling, then updating each neuron in generator generates new data;Less than goal-selling, then the weight for updating each neuron in arbiter reclassifies judgement;S7:Data and the original sampling data fusion that generator is generated.Its effect is:The data of generation have similar distribution with original sampling data, and enough data resources are provided for gear performance evaluation.
Description
Technical field
The present invention relates to the data sampling techniques in big data field, and in particular to one kind is based on generation confrontation network model
Gear parameter oversampler method.
Background technology
Chief component of the speed changer as automobile drastically influences stability and the safety of automobile.And gear is
The main reason for critical component in speed changer, gear distress is typically TRANS PROGRAM.Gear train is complicated, non-linear, no
It can directly reflect the relationship between mechanical breakdown and physical parameter.Many researchs are dedicated to improving the safety of gear, including change
Into its physical arrangement, the relationship etc. between its each physical parameter and safety coefficient is found.
In the prior art, automaker designs transmission gear generally according to artificial 6336 standard of experience and ISO, and
And most of research sampling gear size parameters evaluate the performance of gear, it is usually mutual with full-scale parameter when practical operation
Coordinate to carry out the assessment of safety coefficient, relatively cumbersome due to obtaining each item data, one side assessment cycle is longer, another party
Face causes gear safe condition precision of prediction relatively low because sampled data is rare.
Invention content
To solve the above-mentioned problems, the present invention proposes a kind of based on the gear parameter over-sampling side for generating confrontation network model
Method creates reliable data by over-sampling, reduces the trouble of data artificial sample, while promoting its gear safe condition
Precision of prediction.
To achieve the goals above, the technology used in the present invention is as follows:
It is a kind of based on generate confrontation network model gear parameter oversampler method, key be according to the following steps into
Row:
S1:It determines and needs the gear parameter sampled and acquire corresponding data, definition original sampling data is D;
S2:A generator and an arbiter are built by neural network;
S3:In input noise sequence signal Z to the generator, obtain generating data;
S4:The data that generator is generated are inputted into the arbiter simultaneously with the original sampling data D obtained by step S1
Carry out classification judgement;
S5:Linear transformation is carried out using the Softmax layers of output result to arbiter, obtains final classification result;
S6:Classification results and setting value are relatively obtained into error in classification, it is when error in classification is more than goal-selling, then more newborn
The weight of each neuron in growing up to be a useful person is then back to S3 and generates new data;When error in classification is less than goal-selling, then update is sentenced
The weight of each neuron in other device is then back to S4 and re-starts classification judgement;
S7:The data and original sampling data that generator is generated are that D is merged collectively as needed for gear parameter analysis
Sampled data.
Optionally, the gear parameter sampled 85 parameters and 2 safety coefficients as needs in step S1, described 85
Parameter includes tooth load, application factor KA, internal dynamic factor Kv, facewidth load factor KHβ,KFβ, end load factor KHα,
KFα, engagement loading coefficient Kγ, 2 safety coefficients are bending safety coefficient SFWith touch-safe coefficient SH。
Optionally, the generator is made of 4 layers of neural network, the number of neuron is followed successively by 128,1024,256,
87 neurons of the 87, the 4th layer of neural network correspond to output for 85 parameters and 2 safety coefficients, in each layer of neural network
Activation primitive be Relu functions.
Optionally, the arbiter is made of 3 layers of neural network, and the number of neuron is followed successively by 256,1024,128,
Activation primitive in each layer of neural network is leaky_Relu functions.
Optionally, input noise sequence signal Z obediences are uniformly distributed or Gaussian Profile in step S3.
Optionally, the data obtained by over-sampling are admitted to reconstructed error pca model extraction feature, are then calculated by PSO-BP
Method obtains the estimated value of 2 safety coefficients.
Beneficial effects of the present invention:
It can meet data sampling by generating new data in the case where sampled data is on the low side based on the above method
The data of the requirement of quantity, generation have similar distribution with original sampling data, can solve gear data sampling difficulty
Problem provides reliable guarantee for gear safe prediction.
Description of the drawings
Fig. 1 is the system architecture diagram of the present invention;
Fig. 2 is gear safety coefficient forecasting system Organization Chart;
Fig. 3 is the influence relational graph of initial data and generation data to bending safety coefficient distribution;
Fig. 4 is the influence relational graph of initial data and generation data to contact safety coefficient distribution.
Specific implementation mode
The embodiment of technical solution of the present invention will be described in detail below.Following embodiment is only used for clearer
Ground illustrates technical scheme of the present invention, therefore is only used as example, and not intended to limit the protection scope of the present invention.
It should be noted that unless otherwise indicated, technical term or scientific terminology used in this application should be this hair
The ordinary meaning that bright one of ordinary skill in the art are understood.
As shown in Figure 1, the present embodiment discloses a kind of gear parameter oversampler method based on generation confrontation network model,
Key is to carry out according to the following steps:
S1:It determines and needs the gear parameter sampled and acquire corresponding data, definition original sampling data is D;
Existing automaker is generally according to 6336 standard design vehicle gearbox gear of artificial experience and ISO, the mark
Arrive the basic principle of design of gears involved in standard, General Influence factor, the intensity and quality of surface durability technology and material,
Common parameter index includes tooth load, application factor KA, internal dynamic factor Kv, facewidth load factor KHβ,KFβ, end load
Factor KHα,KFα, engagement loading coefficient Kγ, 2 safety coefficients are bending safety coefficient SFWith touch-safe coefficient SH。
One basic demand of design of gears is to keep gear safety coefficient in a certain spy on the basis of full-scale parameter
In fixed range.Existing gear data have 214 projects, each is related to 90 parameters and 2 safety coefficients, join at this 90
In number, 5 parameters are maintained in fixed value due to manufacturing process, therefore it may only be necessary to discuss remaining 85 parameter.
In this step and using in existing 214 projects 85 parameters and 2 safety coefficients as goal in research, i.e.,
Original sampling data is that D contains 214 projects, and each project includes 85 parameters and 2 safety coefficients.
S2:A generator and an arbiter are built by neural network;
As seen in Figure 1, generator is made of 4 layers of neural network, the number of wherein neuron is followed successively by 128,
1024,256,87, it is 85 parameters and 2 safety coefficients, each layer of god that 87 neurons of the 4th layer of neural network, which correspond to output,
It is Relu functions through the activation primitive in network, and arbiter is made of 3 layers of neural network, the number of neuron is followed successively by
256,1024,128, the activation primitive in each layer of neural network is leaky_Relu functions.
S3:In input noise sequence signal Z to the generator, obtain generating data, here input noise sequence signal Z
Obedience is uniformly distributed or Gaussian Profile;
S4:The data that generator is generated are inputted into the arbiter simultaneously with the original sampling data D obtained by step S1
Carry out classification judgement;
S5:Linear transformation is carried out using the Softmax layers of output result to arbiter, obtains final classification result;
S6:Classification results and setting value are relatively obtained into error in classification, it is when error in classification is more than goal-selling, then more newborn
The weight of each neuron in growing up to be a useful person is then back to S3 and generates new data;When error in classification is less than goal-selling, then update is sentenced
The weight of each neuron in other device is then back to S4 and re-starts classification judgement;
In above-mentioned training process, when in two network models one be fixed when, another network weight is then
It is updated.Both sides' successive optimization during alternating iteration, to form competition, until reaching dynamic equilibrium (Na Shijun
Weighing apparatus).Based on above-mentioned thought, the error in classification of grader will be helpful to generator and update its weight and training, and new data will again
Secondary testing classification device, such iteration carry out always, eventually arrive at until required iterations or balance.
S7:The data and original sampling data that generator is generated are that D is merged collectively as needed for gear parameter analysis
Sampled data, mixed data can be solved the few problem of data sampling amount, realize data oversampling.
As shown in Fig. 2, in the specific implementation, the data obtained by over-sampling are admitted to reconstructed error pca model extraction feature,
Then the estimated value of 2 safety coefficients is obtained by PSO-BP algorithms.
As can be seen that data D in Fig. 2bFor original sampling data, data DaIt is adopted for what is generated using the method for the invention
Sample data, by data DbWith data DaIt is sent into reconstructed error pca model extracting parameter feature simultaneously, is finally based on these parameters spy
Sign is predicted using 2 safety coefficients of PSO-BP algorithms pair.
Star point in figure, which is can be seen that, in conjunction with Fig. 3 and Fig. 4 indicates that original sampling data, circle points indicate the data generated,
The data space of the safety coefficient of gear is extended on all reliability steps, rather than focuses only on existing data
Region, based on the advantages of fighting e-learning is generated, these data remain the build-in attribute of initial data simultaneously, can from Fig. 3
To find out, the bending safety coefficient S that is obtained from original sampling dataFAverage value be from generating the obtained bending of data safely
Number SFAverage value it is very close, similarly, from Fig. 4 it can also be seen that the touch-safe coefficient S obtained from original sampling dataH's
Average value and the touch-safe coefficient S obtained from generation dataHAverage value it is also very close.Unquestionably, the mistake proposed
The method of sampling has studied data pattern, and enough data are produced in whole region.
Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention., rather than its limitations;
Although present invention has been described in detail with reference to the aforementioned embodiments, it will be understood by those of ordinary skill in the art that:Its
It still can be with technical scheme described in the above embodiments is modified, either to which part or all technical features
Carry out equivalent replacement;And these modifications or replacements, various embodiments of the present invention skill that it does not separate the essence of the corresponding technical solution
The range of art scheme should all cover in the claim of the present invention and the range of specification.
Claims (6)
1. a kind of based on the gear parameter oversampler method for generating confrontation network model, it is characterised in that carry out according to the following steps:
S1:It determines and needs the gear parameter sampled and acquire corresponding data, definition original sampling data is D;
S2:A generator and an arbiter are built by neural network;
S3:In input noise sequence signal Z to the generator, obtain generating data;
S4:The data that generator is generated are inputted the arbiter simultaneously with the original sampling data D obtained by step S1 to carry out
Classification judges;
S5:Linear transformation is carried out using the Softmax layers of output result to arbiter, obtains final classification result;
S6:Classification results and setting value are relatively obtained into error in classification, when error in classification is more than goal-selling, then update generator
In each neuron weight, be then back to S3 and generate new data;When error in classification be less than goal-selling, then update arbiter
In each neuron weight, be then back to S4 re-start classification judge;
S7:The data and original sampling data that generator is generated are that D is merged collectively as the sampling needed for gear parameter analysis
Data.
2. according to claim 1 based on the gear parameter oversampler method for generating confrontation network model, which is characterized in that
The gear parameter for sampling 85 parameters and 2 safety coefficients as needs in step S1,85 parameters include that the gear teeth are negative
It carries, application factor KA, internal dynamic factor Kv, facewidth load factor KHβ,KFβ, end load factor KHα,KFα, engage loading coefficient
Kγ, 2 safety coefficients are bending safety coefficient SFWith touch-safe coefficient SH。
3. according to claim 2 based on the gear parameter oversampler method for generating confrontation network model, which is characterized in that
The generator is made of 4 layers of neural network, and the number of neuron is followed successively by the 128,1024,256,87, the 4th layer of neural network
87 neurons to correspond to output be 85 parameters and 2 safety coefficients, the activation primitive in each layer of neural network is Relu
Function.
4. according to claim 1 or 3 based on the gear parameter oversampler method for generating confrontation network model, feature exists
In the arbiter is made of 3 layers of neural network, and the number of neuron is followed successively by 256,1024,128, each layer of nerve net
Activation primitive in network is leaky_Relu functions.
5. according to claim 1 or 3 based on the gear parameter oversampler method for generating confrontation network model, feature exists
In input noise sequence signal Z obediences are uniformly distributed or Gaussian Profile in step S3.
6. according to claim 1 based on the gear parameter oversampler method for generating confrontation network model, which is characterized in that
Data obtained by over-sampling are admitted to reconstructed error pca model extraction feature, then obtain 2 safety systems by PSO-BP algorithms
Several estimated values.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810399022.8A CN108596261A (en) | 2018-04-28 | 2018-04-28 | Based on the gear parameter oversampler method for generating confrontation network model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810399022.8A CN108596261A (en) | 2018-04-28 | 2018-04-28 | Based on the gear parameter oversampler method for generating confrontation network model |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108596261A true CN108596261A (en) | 2018-09-28 |
Family
ID=63610625
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810399022.8A Pending CN108596261A (en) | 2018-04-28 | 2018-04-28 | Based on the gear parameter oversampler method for generating confrontation network model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108596261A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110020683A (en) * | 2019-04-02 | 2019-07-16 | 重庆青山工业有限责任公司 | Gear sample data base construction method |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20040064426A1 (en) * | 2002-09-30 | 2004-04-01 | Depold Hans R. | Physics based neural network for validating data |
US20120316835A1 (en) * | 2010-01-14 | 2012-12-13 | Shunji Maeda | Anomaly detection method and anomaly detection system |
CN107180392A (en) * | 2017-05-18 | 2017-09-19 | 北京科技大学 | A kind of electric power enterprise tariff recovery digital simulation method |
CN107480788A (en) * | 2017-08-11 | 2017-12-15 | 广东工业大学 | A kind of training method and training system of depth convolution confrontation generation network |
CN107491606A (en) * | 2017-08-17 | 2017-12-19 | 安徽工业大学 | Variable working condition epicyclic gearbox sun gear method for diagnosing faults based on more attribute convolutional neural networks |
CN107563428A (en) * | 2017-08-25 | 2018-01-09 | 西安电子科技大学 | Classification of Polarimetric SAR Image method based on generation confrontation network |
-
2018
- 2018-04-28 CN CN201810399022.8A patent/CN108596261A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20040064426A1 (en) * | 2002-09-30 | 2004-04-01 | Depold Hans R. | Physics based neural network for validating data |
US20120316835A1 (en) * | 2010-01-14 | 2012-12-13 | Shunji Maeda | Anomaly detection method and anomaly detection system |
CN107180392A (en) * | 2017-05-18 | 2017-09-19 | 北京科技大学 | A kind of electric power enterprise tariff recovery digital simulation method |
CN107480788A (en) * | 2017-08-11 | 2017-12-15 | 广东工业大学 | A kind of training method and training system of depth convolution confrontation generation network |
CN107491606A (en) * | 2017-08-17 | 2017-12-19 | 安徽工业大学 | Variable working condition epicyclic gearbox sun gear method for diagnosing faults based on more attribute convolutional neural networks |
CN107563428A (en) * | 2017-08-25 | 2018-01-09 | 西安电子科技大学 | Classification of Polarimetric SAR Image method based on generation confrontation network |
Non-Patent Citations (3)
Title |
---|
刘玉杰等: "基于条件生成对抗网络的手绘图像检索", 《计算机辅助设计与图形学学报》 * |
张营营: "生成对抗网络模型综述", 《电子设计工程》 * |
裔隽等著: "《Python机器学习实战》", 28 February 2018, 科学技术文献出版社 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110020683A (en) * | 2019-04-02 | 2019-07-16 | 重庆青山工业有限责任公司 | Gear sample data base construction method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112149316B (en) | Aero-engine residual life prediction method based on improved CNN model | |
CN110147450B (en) | Knowledge complementing method and device for knowledge graph | |
WO2023020388A1 (en) | Gearbox fault diagnosis method and apparatus, gearbox signal collection method and apparatus, and electronic device | |
CN112685961B (en) | Method and system for predicting remaining service life of analog circuit | |
CN110212528B (en) | Power distribution network measurement data missing reconstruction method | |
CN106951512A (en) | A kind of end-to-end session control method based on hybrid coding network | |
CN110223785A (en) | A kind of infectious disease transmission network reconstruction method based on deep learning | |
CN105654117A (en) | Hyperspectral image spectral-spatial cooperative classification method based on SAE depth network | |
CN116011125B (en) | Response prediction method for uncertain axle coupling system | |
CN108168577A (en) | MEMS gyro random error compensation method based on BP neural network | |
CN109840639A (en) | A kind of late time forecasting methods of high speed rail train operation | |
CN116610092A (en) | Method and system for vehicle analysis | |
CN114898121A (en) | Concrete dam defect image description automatic generation method based on graph attention network | |
CN102034111A (en) | Method for identifying and detecting aircraft structural damage conditions in diversified way | |
CN111126794A (en) | Data enhancement and neural network confrontation training system based on small samples | |
CN113015983A (en) | Autonomous system including continuous learning world model and related methods | |
CN114266201B (en) | Self-attention elevator trapping prediction method based on deep learning | |
CN108596261A (en) | Based on the gear parameter oversampler method for generating confrontation network model | |
CN103354073B (en) | A kind of LCD color deviation correction method | |
Kim et al. | A study toward appropriate architecture of system-level prognostics: physics-based and data-driven approaches | |
CN115808627A (en) | Lithium battery SOH prediction method and device | |
CN116843544A (en) | Method, system and equipment for super-resolution reconstruction by introducing hypersonic flow field into convolutional neural network | |
CN113722308B (en) | Acceleration response data complement method and device based on EEMD-MultiCNN-LSTM | |
CN112819143B (en) | Working memory computing system and method based on graph neural network | |
CN109490793A (en) | A kind of X-ray high voltage power supply failure prediction method based on wavelet decomposition and LSTM |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180928 |