CN110378431A - Convolutional neural network-based supersonic combustion chamber combustion mode detection method - Google Patents
Convolutional neural network-based supersonic combustion chamber combustion mode detection method Download PDFInfo
- Publication number
- CN110378431A CN110378431A CN201910669920.5A CN201910669920A CN110378431A CN 110378431 A CN110378431 A CN 110378431A CN 201910669920 A CN201910669920 A CN 201910669920A CN 110378431 A CN110378431 A CN 110378431A
- Authority
- CN
- China
- Prior art keywords
- convolutional neural
- neural networks
- combustion mode
- combustion chamber
- data
- 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
- G06F18/21—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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Combined Controls Of Internal Combustion Engines (AREA)
Abstract
The invention discloses a combustion chamber combustion mode detection method based on a convolutional neural network, which comprises the steps of S1, collecting original data, dividing the original data into a plurality of data samples and forming a data sample set; s2, randomly dividing the data sample set into a training set, a verification set and a test set, and preprocessing the data in the training set, the verification set and the test set; s3, building a convolutional neural network model capable of mapping the features in the data samples to the corresponding combustion modes; s4, training a model, verifying the model and evaluating the effect to obtain a combustion mode detection model; s5, detecting the sample to be detected with unknown combustion mode by using the combustion mode detection model. The combustion mode detection method based on the convolutional neural network does not need manual design and extraction of data characteristics, and can effectively realize combustion mode detection by directly utilizing the original data acquired in the working process of the supersonic combustion chamber.
Description
Technical field
The present invention relates to supersonic combustion ramjet engine combustion chamber conditions monitoring technical fields, especially a kind of based on volume
The supersonic speed combustion chamber combustion mode detection method of product neural network.
Background technique
Supersonic combustion technology is the key technology of scramjet engine, and realizes the important of hypersonic aircraft
Basis.Guarantee that the indoor supersonic combustion process of burning is stable, controllable, is always the hot research problem in the field, and it is real-time
Grasping indoor combustion mode of burning is to realize combustion process stabilization, controllable precondition.
As the relevant technologies gradually mature, hypersonic aircraft also gradually walks out laboratory, various military and civilians
It has been suggested, or even has tried out using concept.In scramjet engine actual moving process, its operation conditions is monitored
It is very important.Since the distribution of combustion process fuel and the control of other parameters depend greatly on combustion mode
Judgement, therefore combustion mode detection is the important component of course of work monitoring.
The judgement that current ultrasonic quick burning burns Indoor Combustion mode relies primarily on the variation of the special parameter of some specific positions,
It is realized in conjunction with statistics and the method for conventional machines study, indoor equivalent proportion of such as burning, the wall surface of distance piece feature locations
Pressure ratio etc..Although these methods can provide some with reference to still for supersonic speed combustion chamber course of work Design of Monitoring and Control System
There are many problems: first is that carrying out combustion mode inspection by the variation and trend of the certain selected parameters of observation or dimensionless index
It is usually time-consuming and insecure for surveying, especially when data are by noise effect；Second is that such methods are for expertise
It is very strong with the dependence of engineering experience, when related physical and chemical process is especially complex, or even can not be understood completely, disadvantage
End is especially apparent；Third is that such methods are very strong generally directed to property, it is only applicable to the combustion chamber of a certain configuration, if combustion chamber structure
Type is different or even structural parameters change, and requires that parameter and feature are chosen and designed again, versatility is bad.
Summary of the invention
The present invention provides a kind of combustion chambers burn mode detection method based on convolutional neural networks, for overcoming existing skill
Less reliable in art, it is strong to expertise and engineering experience dependence, versatility is bad the defects of, this method is directly using super
The initial data progress combustion mode detection acquired during velocity of sound combustion chamber operational, the testing result high reliablity of this method,
Weak to expertise and engineering experience dependence, versatility is good.
To achieve the above object, the present invention proposes a kind of combustion chambers burn mode detection side based on convolutional neural networks
Method, comprising the following steps:
S1: during collecting work in supersonic combustion room each different sensors initial data, initial data is corresponding
Combustion mode it is known that initial data is divided into several data samples, all data sample composition data sample sets；
S2: by set of data samples random division be training set, verifying collection and test set, and respectively to the training set, test
Data in card collection and test set are pre-processed；
S3: the convolutional neural networks mould that can be realized by Feature Mapping potential in data sample to corresponding combustion mode is built
Type；
S4: convolutional neural networks model is trained using training set, to obtain the weight of convolutional neural networks model
Matrix parameter is verified the convolutional neural networks model after training using verifying collection, to convolutional neural networks model
Hyper parameter optimizes, and obtains supersonic speed combustion chamber combustion mode detection model, later using test set to the ultrasonic quick burning
It burns room combustion mode detection model and carries out recruitment evaluation；
S5: it is carried out using data to be tested sample of the supersonic speed combustion chamber combustion mode detection model to unknown combustion mode
Detection obtains supersonic speed combustion chamber combustion mode instantly.
Combustion chambers burn mode detection method provided by the invention based on convolutional neural networks is not necessarily to initial data spy
Sign carries out engineer and extraction, directly can realize burning mould using the initial data acquired in the supersonic speed combustion chamber course of work
Formula detection, and method versatility is good, testing result accuracy is high.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
The structure shown according to these attached drawings obtains other attached drawings.
Fig. 1 is supersonic speed combustion chamber combustion mode detection model schematic diagram provided by the invention；
Fig. 2 is the combustion chambers burn mode detection method flow chart provided by the invention based on convolutional neural networks；
Fig. 3 is supersonic speed combustion chamber structural schematic diagram of the present invention；
Fig. 4 is the schematic diagram that initial data is divided into several data samples in detection method provided by the invention；
Fig. 5 is the convolutional neural networks schematic diagram established in the embodiment of the present invention.
The embodiments will be further described with reference to the accompanying drawings for the realization, the function and the advantages of the object of the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiment is only a part of the embodiments of the present invention, instead of all the embodiments.Base
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts it is all its
His embodiment, shall fall within the protection scope of the present invention.
It in addition, the technical solution between each embodiment of the present invention can be combined with each other, but must be general with this field
Based on logical technical staff can be realized, it will be understood that when the combination of technical solution appearance is conflicting or cannot achieve this
The combination of technical solution is not present, also not the present invention claims protection scope within.
Fig. 1 show supersonic speed combustion chamber combustion mode detection model schematic diagram, Fig. 2 in the present invention and show in the present invention
Combustion chambers burn mode detection method flow chart based on convolutional neural networks, including following specific embodiments:
S1: during collecting work in supersonic combustion room each different sensors initial data, initial data is corresponding
Combustion mode it is known that initial data is divided into several data samples, all data sample composition data sample sets；The original
Beginning data include the corresponding parameter of each different sensors in supersonic speed combustion chamber, such as chamber wall surface pressure, propellant spray pressure
Power, chamber temperature, fuel flow rate etc..
S2: by set of data samples random division be training set, verifying collection and test set, and respectively to the training set, test
Data in card collection and test set are pre-processed；
S3: the convolutional neural networks mould that can be realized by Feature Mapping potential in data sample to corresponding combustion mode is built
Type；
S4: convolutional neural networks model is trained using training set, to obtain the weight of convolutional neural networks model
Matrix parameter is verified the convolutional neural networks model after training using verifying collection, to convolutional neural networks model
Hyper parameter optimizes, and obtains supersonic speed combustion chamber combustion mode detection model, later using test set to the ultrasonic quick burning
It burns room combustion mode detection model and carries out recruitment evaluation；
S5: it is carried out using data to be tested sample of the supersonic speed combustion chamber combustion mode detection model to unknown combustion mode
Detection obtains supersonic speed combustion chamber combustion mode instantly.
Supersonic speed combustion chamber includes a variety of configurations, such as cavity formula combustion chamber, the board-like combustion chamber of branch, involved in the present embodiment
It and is cavity formula combustion chamber.It, can be by being based on convolution using method provided by the invention for the combustion chamber of other configurations
The supersonic speed combustion chamber combustion mode detection method of neural network realizes detection to burning Indoor Combustion mode.
It is auxiliary that supersonic speed combustion chamber combustion mode generally comprises unburned mode, cavity shear layer flame stabilization mode, cavity
Help jet stream tail flame stabilization mode, joint cavity shear layer recirculating zone flame stabilization mode, mode of being jammed etc..To combustion mode
The each research institute of definition, seminar, colleges and universities etc. may be all different, thus will appear a large amount of miscellaneous present invention not
The combustion mode mentioned, but the combustion chambers burn based on convolutional neural networks that these combustion modes can provide through the invention
Mode detection method is detected.
In S1, the initial data in the supersonic speed combustion chamber course of work in some period is acquired, is wrapped in initial data
Include chamber wall surface pressure, propellant spray pressure, fuel flow rate, chamber temperature etc., the quantity root of initial data collected
Depending on the actual conditions such as number of sensors, combustion chamber operational duration, general acquisition data more as far as possible.
Initial data is divided into several data samples, as shown in figure 4, the data sample is the matrix of size m × n,
Wherein m is the sampling number for including in individual data sample, and n is number of probes, the parameter for as including in data sample
Number, parameter include chamber wall surface pressure, propellant spray pressure, fuel flow rate, chamber temperature etc..For convenience of convolutional Neural
Network processes, general m and n are 2 power, or are the product of 2 power and other numbers.
In general, data sample quantity is The more the better, in the supersonic speed combustion chamber course of work that training set is included
Operating condition is more comprehensive, and the convolutional neural networks model performance that training obtains is better, more accurate to the judgement of combustion mode.Therefore,
The method of overlapped partitioning can be used in the case where initial data limited amount to expand the quantity of data sample, as shown in Figure 4.
In S2, needs to be training set, verifying collection and test set by set of data samples random division obtained in S1, draw
During point, to guarantee that the data sample of various known combustion modes is evenly distributed on training set, verifying collection and test set as far as possible
In.Wherein, the training for the network weight that training set is automatically updated for convolutional neural networks model itself, verifying collection is for need
The selection for the hyper parameter manually to set optimizes, and test set is used to carry out the supersonic speed combustion chamber combustion mode detection model
Recruitment evaluation, generally using the accuracy rate of model inspection result as the evaluation index of modelling effect.Training set, verifying collection and test
The ratio of collection can be adjusted according to the actual situation, generally 3:1:1, if data sample is especially more, number shared by test set
Can suitably reduce according to the ratio of sample (such as: if a data set only has 100 data samples, the ratio that test set accounts for
Unsuitable too small, not so a sample mistake point will have a huge impact accuracy rate；In turn, if there is 100,000 data samples
This, then even also there are 1000 samples in test set and verifying collection only 1%, the inside, the several samples of mistake point are to whole accuracy rate
Influencing will not be very big).
It, before training, be respectively to the data in training set, verifying collection and test set after the completion of set of data samples divides
It is pre-processed.Due to including a plurality of types of data such as pressure, temperature in initial data, different types of data have different
Dimension and the order of magnitude, and the order of magnitude differs greatly.To avoid order of magnitude difference from causing different parameters can not in feature extraction
It is impartial to, guarantees the reliability of gained feature, need to be standardized the data in training set, verifying collection and test set
Processing.Common standardized method generally has min-max standardization (deviation is also made to standardize), z-score standardized method etc.,
It is common statistical method, can be selected according to the actual situation.What is selected in this example is the standardization side z-score
Method, i.e.,
Wherein, x ' is pretreated data sample, and x is the data sample before pretreatment, and μ is the sample standard deviation of training set
Value, s are the standard deviation of training set.It is worth noting that, being located in advance to the data sample in training set, verifying collection and training set
Reason, the mean value and variance of use are all from training set, to guarantee only to obtain information from training data during model training.
In S3, the convolutional neural networks model built mainly include input layer, convolutional layer, pond layer, full articulamentum and
Softmax output layer.Wherein, input layer is inputted for data sample；Convolutional layer is the core component of convolutional neural networks,
It is mainly used for automatically extracting the feature in data sample；Pond layer is usually after convolutional layer, it is therefore an objective to reduce the sky of the feature
Between dimension；Convolutional layer and pond layer are usually alternately arranged；Full articulamentum plays the work of classifier in entire convolutional neural networks
With by the label space of the Feature Mapping extracted to data sample；Finally by softmax output layer will that treated be one-dimensional
The data sample that feature vector is converted into input belongs to the probability of all kinds of combustion modes, and highest probability is that data sample is corresponding
Combustion mode.It is specific as follows:
Have for a certain data sample x ' by pretreated training set by first convolutional layer
h1=σ1(W1*x′+b1) (2)
Wherein, h1For the eigenmatrix exported after first convolutional layer, σ1、W1、b1Respectively first convolutional layer
Activation primitive, convolution kernel and biasing, * are convolution operation.
It is later first pond layer, has
h2=pooling (h1) (3)
Wherein, pooling is pondization operation, and common pondization operation includes maximum pond and average pond etc.；h2For warp
Cross the eigenmatrix exported after first pond layer；h1For the eigenmatrix exported after first convolutional layer.
So after several convolutional layers and pond layer, eigenmatrix h is obtainedl, behind usually connect several full articulamentums.
By eigenmatrix hlExpand into one-dimensional characteristic vector h 'l, for adjacent full articulamentum, have
hl+1=σl+1(Wl+1h′l+bl+1) (4)
Wherein, hl+1For the feature vector exported after first full articulamentum；h′lIt is characterized matrix hlIt is obtained after expansion
One-dimensional characteristic vector,；σl+1、Wl+1、bl+1The respectively activation primitive of this layer, weight matrix and biasing.
By the full articulamentum of several layers, the one-dimensional vector for representing the length of classification results as k is finally obtainedIt is as follows:
Wherein, the value of k is equal to the species number for participating in all combustion modes of classification；For in vector
The specific value of each element.
Finally, converting the one-dimensional vector that length is k to by softmax function (normalization exponential function) number of input
Belong to the probability of all kinds of combustion modes according to sampleSpecially
Wherein,For vectorIn each element specific value；qiBelong to all kinds of burnings for the data sample of input
The probability of mode；
Thus obtained input data sample belongs to the probability i.e. q=(q of each combustion mode1,q2,…,qk), probability value is most
High is last testing result.
In S4, training set is for the weight matrix in training convolutional neural networks；Verifying collection is in the training process
The training effect of evaluation model, to carry out the optimization of guidance model hyper parameter；It finally selects to collect the best mould of upper effect in verifying
Type is as final supersonic speed combustion chamber combustion mode detection model.Test set is inputted into the trained supersonic speed combustion chamber combustion
Mode detection model is burnt, using this testing result as the standard for evaluating the detection model effect.
It is general to lose letter by construction loss function, and by minimizing in the training process of convolutional neural networks model
Number is as final optimization aim.For combustion mode detects (belonging to classification problem), common loss function is to intersect
Entropy loss function.
As described in S3, the output of convolutional neural networks model is q=(q1,q2,…,qk), desired probability distribution is p
=(p1,p2,…,pk), then the intersection entropy loss of convolutional neural networks model is
Wherein, loss is loss function；piBelong to the desired probability of all kinds of combustion modes for the data sample of input；qiFor
The data sample of input belongs to the probability of all kinds of combustion modes.
Training process is trained model using back-propagation algorithm, is carried out more using following formula to network weight
Newly, to realize the training of convolutional neural networks itself, i.e.,
Wherein,Loss function is indicated to the gradient of weight matrix, η is learning rate, and W is weight to be updated, WnewFor
Updated weight.
When loss function tends towards stability and no longer declines, terminate this training set to the instruction of convolutional neural networks model
Practice.
After convolutional neural networks model training, program is run on verifying collection, checks verifying collection in the convolutional Neural
Result on network model.If verifying collection accuracy rate on convolutional neural networks model instantly not up to requires, super ginseng is adjusted
Number, then training is re-started to convolutional neural networks model adjusted.It repeats the above steps, until model reaches on verifying collection
To requirement accuracy rate or find one group of highest hyper parameter of accuracy rate.Whole parameters of fixed model at this time complete training.
After the completion of model training, the model is run on test set, accuracy rate at this time can represent the detection method and exist
Performance in supersonic speed combustion chamber combustion mode.
In S5, for supersonic speed combustion chamber combustion mode to be detected, the initial data under its working condition is acquired, it will
Supersonic speed combustion chamber combustion mode detection model after initial data pretreatment, as the input S4 acquisition of data to be tested sample
It is detected, the combustion mode of supersonic speed combustion chamber at this time can be directly obtained.
Embodiment 1
Involved supersonic speed combustion chamber is cavity formula combustion chamber, involved supersonic speed combustion chamber combustion mode in the present embodiment
Including unburned, cavity shear layer flame stabilization mode, cavity assisted jet tail flame stabilization mode, mode of being jammed, totally four
Kind combustion mode.
Fig. 1 show supersonic speed combustion chamber combustion mode detection model schematic diagram, Fig. 2 in the present invention and show in the present invention
Combustion chambers burn mode detection method flow chart based on convolutional neural networks, including following specific embodiments:
S1: during collecting work in supersonic combustion room each different sensors initial data, initial data is corresponding
Combustion mode it is known that initial data is divided into several data samples, all data sample composition data sample sets；
In S1, cavity in 61 supersonic speed combustion chamber combustion processes is collected altogether by 24 static pressure sensors
Wall pressure data in combustion chamber in the axial direction, frequency acquisition 50Hz, combustion mode using one-hot it is known that and compiled
Code encodes 4 kinds of combustion modes, as described in Table 1.
Method divides initial data as shown in Figure 4, will be former using the method for overlapped partitioning to increase sample size
Beginning data are divided into 1388 data samples, and it includes that 24 sensors are continuous in that is, each data sample that size, which is 40 × 24,
40 pressure datas of acquisition.
S2: by set of data samples random division be training set, verifying collection and test set, and respectively to the training set, test
Data in card collection and test set are pre-processed；
In S2,1388 data sample random divisions obtained in S1 are training set, verify collection and test set, specifically
Division result is as shown in table 2.Since sample number is less, the division proportion used here is about 3:1:1.Recycle z-score mark
Quasi-ization method is standardized sample, i.e., first calculates the mean value and variance of each data point of sample in training set, and recycling should
Mean value and variance are standardized training set, verifying collection and test set.
S3: building the convolutional neural networks model that can be realized by Feature Mapping in data sample to corresponding combustion mode,
As shown in Figure 5；
In S3, build convolutional neural networks model as shown in Figure 5, altogether include 1 input layer, 2 convolutional layers, 2
Pond layer, 3 full articulamentums and 1 softmax output layer.
Data sample inputs convolutional neural networks model by input layer and obtains by 2 convolutional layers and 2 pond layers
10 × 12 × 20 eigenmatrix.The one-dimensional vector that eigenmatrix length of run is 2400, obtains length by 3 full articulamentums
It is converted into the vector that length is 4 finally by softmax output layer for 100 feature vector, four values difference in vector
The probability that the data sample belongs to 4 kinds of combustion modes is represented, maximum probability is last testing result.
S4: convolutional neural networks model is trained using training set to obtain the weight square of convolutional neural networks model
Battle array parameter verifies with the super ginseng to convolutional neural networks model the convolutional neural networks model after training using verifying collection
Number, which optimizes, obtains supersonic speed combustion chamber combustion mode detection model, is fired later using test set to the supersonic speed combustion chamber
It burns mode detection model and carries out recruitment evaluation；
In S4, using the weight matrix in training set training convolutional neural networks model in the present embodiment, when in training
When loss function on collection occurs fluctuation and no longer declines, deconditioning, convolutional neural networks model weight matrix parameter training
It finishes；Using verifying collection the hyper parameter of convolutional neural networks model is adjusted, the hyper parameter refer in the training process without
Method automatically updates, and needs the parameter of the manual setting before training starts, be Optimum learning rate in the present embodiment, regularization coefficient and
Batch size；It finally selects to collect the best convolutional neural networks model of upper effect as final supersonic speed combustion chamber combustion in verifying
Mode detection model is burnt, the accuracy rate in the present embodiment on verifying collection has been up to 97.9%, selects hyper parameter at this time
The hyper parameter final as convolutional neural networks model.So far, convolutional neural networks model parameter is all fixed, convolutional Neural net
Network model training is completed, and convolutional neural networks model at this time is supersonic speed combustion chamber combustion mode detection model.
Test set is inputted the supersonic speed combustion chamber combustion mode detection model to detect, with the accurate of the testing result
Rate is as the standard for evaluating the supersonic speed combustion chamber combustion mode detection model effect.Supersonic speed combustion chamber burns in the present embodiment
Mode detection model is 93.22% to the accuracy rate of testing result of test set.
S5: it is examined using to be detected sample of the supersonic speed combustion chamber combustion mode detection model to unknown combustion mode
It surveys, obtains supersonic speed combustion chamber combustion mode instantly.
In S5, for sample to be detected, the initial data under its working condition is acquired, which is pre-processed
Afterwards, it is detected as the data to be tested sample input S4 supersonic speed combustion chamber combustion mode detection model obtained, it can be direct
Obtain the combustion mode of supersonic speed combustion chamber at this time.
1 combustion mode coding schematic diagram of table
| Code name | One-hot coding | Number | |
| It is unburned | 0 | (1,0,0,0) | 5 |
| Cavity shear layer flame stabilization mode | 1 | (0,1,0,0) | 41 |
| Cavity assisted jet tail flame stabilization mode | 2 | (0,0,1,0) | 13 |
| It is jammed mode | 3 | (0,0,0,1) | 2 |
2 data set of table divides
| Training set | Verifying collection | Test set | |
| It is unburned | 77 | 29 | 28 |
| Cavity shear layer flame stabilization mode | 578 | 192 | 192 |
| Cavity assisted jet tail flame stabilization mode | 155 | 51 | 51 |
| It is jammed mode | 15 | 5 | 5 |
| It is total | 845 | 277 | 276 |
The above description is only a preferred embodiment of the present invention, is not intended to limit the scope of the invention, all at this
Under the inventive concept of invention, using equivalent structure transformation made by description of the invention and accompanying drawing content, or directly/use indirectly
It is included in other related technical areas in scope of patent protection of the invention.
Claims (10)
1. a kind of supersonic speed combustion chamber combustion mode detection method based on convolutional neural networks, which is characterized in that including following
Step:
S1: during collecting work in supersonic combustion room each different sensors initial data, the corresponding combustion of initial data
Burning mode is it is known that be divided into several data samples, all data sample composition data sample sets for initial data；
S2: it is training set, verifying collection and test set by set of data samples random division, and the training set, verifying are collected respectively
It is pre-processed with the data in test set；
S3: the convolutional neural networks model that can be realized by Feature Mapping potential in data sample to corresponding combustion mode is built；
S4: convolutional neural networks model is trained using training set, to obtain the weight matrix of convolutional neural networks model
Parameter is verified the convolutional neural networks model after training using verifying collection, with the super ginseng to convolutional neural networks model
Number optimizes, and obtains supersonic speed combustion chamber combustion mode detection model, later using test set to the supersonic speed combustion chamber
Combustion mode detection model carries out recruitment evaluation；
S5: it is examined using data to be tested sample of the supersonic speed combustion chamber combustion mode detection model to unknown combustion mode
It surveys, obtains supersonic speed combustion chamber combustion mode instantly.
2. a kind of supersonic speed combustion chamber combustion mode detection method based on convolutional neural networks as described in claim 1,
It is characterized in that, the combustion mode includes: unburned mode, cavity shear layer flame stabilization mode, cavity assisted jet tail
Flame stabilization mode, joint cavity shear layer recirculating zone flame stabilization mode, mode of being jammed.
3. a kind of supersonic speed combustion chamber combustion mode detection method based on convolutional neural networks as described in claim 1,
Be characterized in that, in the step S1, the data sample be size m × n matrix, wherein m be individual data sample in include
Sampling number, n is number of probes.
4. a kind of supersonic speed combustion chamber combustion mode detection method based on convolutional neural networks as described in claim 1,
It is characterized in that, in the step S2, the pretreatment is to be standardized to data.
5. a kind of supersonic speed combustion chamber combustion mode detection method based on convolutional neural networks as claimed in claim 4,
It is characterized in that, the method for the standardization is z-score standardized method, i.e.,
Wherein, x ' is pretreated data sample, and x is the data sample before pretreatment, and μ is the sample average of training set, and s is
The standard deviation of training set.
6. a kind of supersonic speed combustion chamber combustion mode detection method based on convolutional neural networks as described in claim 1,
It is characterized in that, in the step S2, the training set, verifying integrates and test set accounts for the ratio of data sample as 3:1:1.
7. a kind of supersonic speed combustion chamber combustion mode detection method based on convolutional neural networks as described in claim 1,
It is characterized in that, in the step S3, the convolutional neural networks model is specifically included that
Input layer is inputted for data sample；
Convolutional layer, for automatically extracting the feature in data sample；
Pond layer, for reducing the space dimensionality of feature；
Full articulamentum, for two dimensional character to be mapped as one-dimensional characteristic vector, and to extracted feature further progress feature
Fusion and dimensionality reduction；
Softmax output layer, the data sample for converting input for one-dimensional characteristic vector belong to the general of all kinds of combustion modes
Rate, highest probability is the corresponding combustion mode of data sample.
8. a kind of supersonic speed combustion chamber combustion mode detection method based on convolutional neural networks as described in claim 1,
It is characterized in that, in the step S4, the training specifically:
It is trained using network weight of the training set to convolutional neural networks, the network weight refers to convolutional neural networks model
In it is all can automatically update in the training process, the parameter without manually adjusting；
The hyper parameter of convolutional neural networks model is adjusted using verifying collection, the hyper parameter, which refers to, in the training process can not
It automatically updates, needs the parameter of the manual setting before training starts；
Convolutional neural networks model after test set input training is detected, the convolutional neural networks model after training is to survey
The accuracy rate of examination collection combustion mode detection is the performance for representing the model in the detection of supersonic speed combustion chamber combustion mode.
9. a kind of supersonic speed combustion chamber combustion mode detection method based on convolutional neural networks as described in claim 1,
Be characterized in that, in the step S4, when being trained using training set to convolutional neural networks model by loss function come
The effect for assessing training, when loss function value starts fluctuation and no longer declines, this training terminates.
10. a kind of supersonic speed combustion chamber combustion mode detection method based on convolutional neural networks as claimed in claim 9,
It being characterized in that, the loss function is cross entropy loss function,
Wherein, loss is loss function；piBelong to the desired probability of all kinds of combustion modes for the data sample of input；qiFor input
Data sample belong to the probability of all kinds of combustion modes.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910669920.5A CN110378431A (en) | 2019-07-24 | 2019-07-24 | Convolutional neural network-based supersonic combustion chamber combustion mode detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910669920.5A CN110378431A (en) | 2019-07-24 | 2019-07-24 | Convolutional neural network-based supersonic combustion chamber combustion mode detection method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110378431A true CN110378431A (en) | 2019-10-25 |
Family
ID=68255337
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910669920.5A Pending CN110378431A (en) | 2019-07-24 | 2019-07-24 | Convolutional neural network-based supersonic combustion chamber combustion mode detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110378431A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111298735A (en) * | 2020-02-12 | 2020-06-19 | 苏州垣瑞环境科技有限公司 | Self-circulation intelligent heat supply system for oil sludge resource utilization |
| CN111796180A (en) * | 2020-06-23 | 2020-10-20 | 广西电网有限责任公司电力科学研究院 | Automatic identification method and device for mechanical fault of high-voltage switch |
| CN112782976A (en) * | 2020-12-14 | 2021-05-11 | 中国大唐集团科学技术研究院有限公司火力发电技术研究院 | CNN algorithm-based intelligent combustion optimization closed-loop control method |
| CN113008742A (en) * | 2021-02-23 | 2021-06-22 | 中国农业大学 | Method and system for detecting deposition amount of fog drops |
| CN114492187A (en) * | 2022-01-25 | 2022-05-13 | 中国空气动力研究与发展中心空天技术研究所 | Supersonic combustion chamber pulse injection control method and system based on humanoid active disturbance rejection |
| CN114528769A (en) * | 2022-02-25 | 2022-05-24 | 中国空气动力研究与发展中心空天技术研究所 | Intelligent monitoring method and system for combustion mode of scramjet engine |
| CN116522803A (en) * | 2023-06-29 | 2023-08-01 | 西南科技大学 | Supersonic combustor flow field reconstruction method capable of explaining deep learning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108182452A (en) * | 2017-12-29 | 2018-06-19 | 哈尔滨工业大学（威海） | Aero-engine fault detection method and system based on grouping convolution self-encoding encoder |
| CN108229404A (en) * | 2018-01-09 | 2018-06-29 | 东南大学 | A kind of radar echo signal target identification method based on deep learning |
| CN108416429A (en) * | 2018-02-24 | 2018-08-17 | 华北水利水电大学 | Hydrogen-fuel engine fault diagnosis system and its method based on SOM neural networks |
| CN109918752A (en) * | 2019-02-26 | 2019-06-21 | 华南理工大学 | Mechanical failure diagnostic method, equipment and medium based on migration convolutional neural networks |
-
2019
- 2019-07-24 CN CN201910669920.5A patent/CN110378431A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108182452A (en) * | 2017-12-29 | 2018-06-19 | 哈尔滨工业大学（威海） | Aero-engine fault detection method and system based on grouping convolution self-encoding encoder |
| CN108229404A (en) * | 2018-01-09 | 2018-06-29 | 东南大学 | A kind of radar echo signal target identification method based on deep learning |
| CN108416429A (en) * | 2018-02-24 | 2018-08-17 | 华北水利水电大学 | Hydrogen-fuel engine fault diagnosis system and its method based on SOM neural networks |
| CN109918752A (en) * | 2019-02-26 | 2019-06-21 | 华南理工大学 | Mechanical failure diagnostic method, equipment and medium based on migration convolutional neural networks |
Non-Patent Citations (1)
| Title |
|---|
| 胡茑庆: "基于经验模态分解和深度卷积神经网络的行星齿轮箱故障诊断方法", 《机械工程学报》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111298735A (en) * | 2020-02-12 | 2020-06-19 | 苏州垣瑞环境科技有限公司 | Self-circulation intelligent heat supply system for oil sludge resource utilization |
| CN111796180A (en) * | 2020-06-23 | 2020-10-20 | 广西电网有限责任公司电力科学研究院 | Automatic identification method and device for mechanical fault of high-voltage switch |
| CN112782976A (en) * | 2020-12-14 | 2021-05-11 | 中国大唐集团科学技术研究院有限公司火力发电技术研究院 | CNN algorithm-based intelligent combustion optimization closed-loop control method |
| CN112782976B (en) * | 2020-12-14 | 2023-03-07 | 中国大唐集团科学技术研究院有限公司火力发电技术研究院 | CNN algorithm-based intelligent combustion optimization closed-loop control method |
| CN113008742A (en) * | 2021-02-23 | 2021-06-22 | 中国农业大学 | Method and system for detecting deposition amount of fog drops |
| CN114492187A (en) * | 2022-01-25 | 2022-05-13 | 中国空气动力研究与发展中心空天技术研究所 | Supersonic combustion chamber pulse injection control method and system based on humanoid active disturbance rejection |
| CN114492187B (en) * | 2022-01-25 | 2022-12-02 | 中国空气动力研究与发展中心空天技术研究所 | Supersonic combustor pulse injection control method and system based on humanoid active disturbance rejection |
| CN114528769A (en) * | 2022-02-25 | 2022-05-24 | 中国空气动力研究与发展中心空天技术研究所 | Intelligent monitoring method and system for combustion mode of scramjet engine |
| CN116522803A (en) * | 2023-06-29 | 2023-08-01 | 西南科技大学 | Supersonic combustor flow field reconstruction method capable of explaining deep learning |
| CN116522803B (en) * | 2023-06-29 | 2023-09-05 | 西南科技大学 | Supersonic combustor flow field reconstruction method capable of explaining deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110378431A (en) | Convolutional neural network-based supersonic combustion chamber combustion mode detection method | |
| CN112100865B (en) | Method for predicting remaining life of aircraft engine based on parallel CNN model | |
| CN108717474B (en) | It is a kind of to compose preparation method to using relevant aero-engine comprehensive task | |
| CN104061960B (en) | Pressure altitude parameter determination method on a kind of subsonic flight body | |
| CN108573116B (en) | Aero-engine transition state thrust estimation method based on long-time and short-time memory network | |
| CN108256173A (en) | A kind of Gas path fault diagnosis method and system of aero-engine dynamic process | |
| Duer | Artificial Neural Network-based Technique for Operation Process Control of a Technical Object. | |
| CN113241132B (en) | Method and system for establishing machine learning model for predicting material irradiation damage | |
| CN114462319A (en) | Active regulation and control method for combustion performance of aero-engine and intelligent prediction model | |
| CN114923695B (en) | Intelligent detection method and system for combustion instability of scramjet engine | |
| CN112597702B (en) | Pneumatic modeling generation type confrontation network model training method based on radial basis function | |
| CN110334449A (en) | A kind of aerofoil profile Fast design method based on online agent model algorithm | |
| Liao et al. | Uncertainties in gas-path diagnosis of gas turbines: Representation and impact analysis | |
| CN113255432A (en) | Turbine vibration fault diagnosis method based on deep neural network and manifold alignment | |
| CN115456460A (en) | Multi-quality index output prediction method and system for tobacco shred loosening and conditioning process | |
| CN115163424A (en) | Wind turbine generator gearbox oil temperature fault detection method and system based on neural network | |
| CN112947080B (en) | Scene parameter transformation-based intelligent decision model performance evaluation system | |
| CN117370766A (en) | Satellite mission planning scheme evaluation method based on deep learning | |
| CN105203327A (en) | Gas channel measurement parameter selecting method applied to engine gas channel analysis | |
| CN111474905B (en) | Parameter drift fault diagnosis method in manufacturing process of electromechanical product | |
| CN114996863A (en) | Turbofan engine T-S fuzzy modeling method based on feature extraction | |
| CN114398769A (en) | Automatic score obtaining method for unmanned helicopter flight control system | |
| CN114136622B (en) | DBN-DTL-based aeroengine gas circuit fault diagnosis method | |
| CN112036403A (en) | Intelligent detection method for missing of bolt pin of power transmission tower based on attention mechanism | |
| Ma et al. | Modeling method of small sample data and optimization in ultrasonic extraction process of botanicals |
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 |
Application publication date: 20191025 |
|
| RJ01 | Rejection of invention patent application after publication |