CN114528769B - Intelligent monitoring method and system for combustion mode of scramjet engine - Google Patents
Intelligent monitoring method and system for combustion mode of scramjet engine Download PDFInfo
- Publication number
- CN114528769B CN114528769B CN202210180557.2A CN202210180557A CN114528769B CN 114528769 B CN114528769 B CN 114528769B CN 202210180557 A CN202210180557 A CN 202210180557A CN 114528769 B CN114528769 B CN 114528769B
- Authority
- CN
- China
- Prior art keywords
- combustion
- self
- image
- data
- combustion mode
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- 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
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/14—Force analysis or force optimisation, e.g. static or dynamic forces
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Probability & Statistics with Applications (AREA)
- Evolutionary Biology (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Combined Controls Of Internal Combustion Engines (AREA)
- Testing Of Engines (AREA)
Abstract
The invention provides a method and a system for intelligently monitoring a combustion mode of a scramjet, which are used for measuring wall surface pressure data and image data of the scramjet obtained by a plurality of sensors; combining wall surface pressure data, an isolation section flow field structure and combustion chamber Mach number distribution to form a combustion mode judgment criterion, and dividing the combustion mode into a super-combustion mode, a sub-combustion mode and a mixed mode; a combustion field pressure image data set is established based on a combustion mode judgment criterion, a deep learning self-attention recognition network is adopted to judge the combustion mode, a data-driven high-dynamic combustion mode monitoring and analyzing method based on multi-information fusion is established, and the problem of insufficient information caused by the fact that only single pressure data is used in the prior art is avoided; the method adopts a plurality of sensors, a CPU platform and a GPU platform, builds an intelligent monitoring system for the combustion mode of the scramjet, runs an intelligent monitoring algorithm for the combustion mode of the scramjet, and detects the combustion mode in real time.
Description
Technical Field
The invention relates to the technical field of scramjet engines, in particular to an intelligent monitoring method for a scramjet engine combustion mode.
Background
The dual-mode scramjet engine can work in a wide Mach number range, and in order to obtain good propulsion performance, interaction between the combustion chamber and the isolation section is different along with continuous change of flight Mach number, so that the working state of the combustion chamber can be changed, and different combustion modes can appear. The thrust characteristics of engines in different modes are very important for controlling the hypersonic cruise flight, and the hypersonic flight vehicle working in a wide range can safely and stably fly in the atmosphere by monitoring the combustion mode and further designing a mode conversion strategy.
The combustion modes of the dual-mode scramjet engine can be divided into a sub-combustion mode, a mixed mode and a scramjet mode, and the conversion among the combustion modes can cause nonlinear strong sudden changes of pressure distribution of a combustion chamber, thrust of an aircraft, flow field characteristics in the combustion chamber and the like. Therefore, for an actual flight system or a ground test system, in order to avoid the conditions of unstable flame combustion, flameout and the like caused by the continuous change of the working state of the engine, the engine needs to be monitored in real time, the current combustion mode type of the engine is accurately identified, the time for mode conversion of the engine can be predicted, and the method has great significance for the active control of the mode of the scramjet engine.
At present, on one hand, a combustion mode identification method is based on an advanced non-contact measurement technology, and based on the obtained speed, temperature and the like of a combustion field and a shock wave structure, the combustion mode is determined, but optical measurement equipment with heavy weight cannot be carried in a hypersonic flight process. On the other hand, the judgment is carried out based on physical parameters representing combustion, for example, the value of the Mach number of a combustion flow field is inversely calculated by measuring the wall pressure data of the combustion chamber, and then the combustion mode is determined. Publication No. CN107420221A discloses an engine combustion mode identification method for distinguishing whether an engine is in a super-combustion mode or a sub-combustion mode by determining a critical pressure value of an outlet of an isolation section when a negative entropy interval occurs at the outlet of the isolation section of a scramjet engine and calculating a margin of combustion mode conversion. In addition, learners judge the combustion mode based on methods such as equivalence ratio or heat release, on one hand, the methods are limited in real-time performance, on the other hand, if combustion flow field information is not considered, the accuracy and the precision of combustion mode identification are affected, and the current method does not have the capability of early warning combustion mode conversion.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to provide an intelligent monitoring method and an intelligent monitoring system for the combustion mode of the scramjet engine, so as to solve the problems of poor instantaneity, low precision, low intelligent level and the like in the identification of the combustion mode of the scramjet engine in the prior art.
In order to achieve the purpose, the technical scheme of the invention is as follows:
an intelligent monitoring method for a combustion mode of a scramjet engine comprises the following steps:
s1, acquiring image data and wall surface pressure signal data of a flow field of an isolation section of a scramjet engine and a combustion field of a combustion chamber by adopting various measuring sensors;
s2, combining wall surface pressure data, an isolation section flow field structure and combustion chamber Mach number distribution to form a combustion mode judgment criterion, dividing the combustion mode into a super-combustion mode, a sub-combustion mode and a mixed mode, and determining combustion mode types under different equivalence ratios;
s3, simultaneously aligning multiple image data and wall pressure data of the combustion chamber under the same equivalence ratio, and cutting, brightness conversion and contrast adjustment are carried out on the image data;
s4, performing combined comparison on multiple image data acquired by the combustion chamber, and manually marking the image data according to the combustion mode types under different equivalence ratios;
s5, constructing an engine combustion field spectral image database, forming a data set for subsequent deep learning network training, and dividing the data set into a training set Train and a Test set Test;
s6, building a combustion modal self-attention recognition network suitable for the data set in the step S5; respectively acquiring a characteristic matrix of a combustion flow field image and pressure data through a convolution kernel and a multilayer perceptron by using a training set; further extracting representative semantic features using a self-attention coding unit; training a combustion mode self-attention recognition network through a deep learning reverse propagation gradient descent mechanism, finally fusing the extracted features, inputting the fused features into a full-connection layer classifier, and accurately judging the combustion mode category;
and S7, performing performance test on the trained network by using the test data set, inputting the test data set, and comparing the output value with the true value of the combustion mode to obtain the accuracy of classification judgment of the combustion mode.
As a preferred mode, the data set formed in S5 for subsequent deep learning network training is to perform fusion by combining spectral image data of non-contact measurement of multiple combustion chambers according to flow characteristics and wall pressure values of an isolation section flow field under different equivalence ratios, and perform classification of a super-combustion mode, a sub-combustion mode and a mixed mode; and mining the rule between the isolated section flow field and the spectrum image of the combustion chamber multi-information fusion under different equivalence ratios, labeling the image data, and constructing an engine combustion flow field database.
Preferably, the step S1 further includes: performing a test on an oxyhydrogen combustion wind tunnel, applying a bimodal scramjet engine, obtaining high-enthalpy polluted air by adopting a hydrogen-burning oxygen-supplementing method, and injecting normal-temperature kerosene into a combustion chamber through a circular hole; a series of pressure sensors are arranged on the upper wall surface and the lower wall surface of the isolation section and the combustion chamber and are used for acquiring wall surface pressure signal data; acquiring the wave system structure of the isolation section and the flow field structure information of the combustion chamber by adopting high-speed schlieren; the evolution process of pioneer hydrogen flame is obtained by adopting a hydroxyl free radical planar laser induced fluorescence PLIF method, and various flow field image data and wall pressure data under different equivalence ratios are obtained.
Preferably, the step S2 further includes: determining the heat release rate according to the wall surface pressure data, the isolation section flow field image data and the combustion chamber Mach number distribution condition under the condition of different equivalence ratios acquired in the step S1, and analyzing the influence of the heat release rate generated under different equivalence ratios on the combustion mode; meanwhile, whether the isolation section has a shock wave string or not is used for reflecting the occurrence of a large flow separation area in the flow field and the typical characteristics of a sub-combustion mode, the possible position of a thermal throat generated in the flow field is reflected by the position with the mass weighted Mach number equal to 1.0, and finally the mode is comprehensively judged by whether the isolation section has a shock wave string structure or not, whether the mass weighted Mach number is smaller than 1.0 or not and whether heat release is concentrated or not; when a shock wave string structure exists in the isolation section, the mass weighted Mach number is larger than 1.0 region, and the heat release rate is centralized, the combustion mode of the bimodal scramjet engine is a scramjet mode; when no shock wave string structure exists in the isolation section, no mass weighting Mach number is smaller than 1.0 region, and the heat release rate is distributed, the combustion mode of the bimodal scramjet engine is a sub-combustion mode; otherwise, the mixed mode is adopted; finally, the different equivalence ratios are divided into a scramjet engine scramjet mode, a sublagration mode and a mixed mode, so that data under the three combustion modes can be obtained, and a foundation is laid for subsequently establishing a data set.
Preferably, the step S3 further includes: firstly, extracting wall surface pressure data according to a fixed step length, and carrying out time sequence synchronization on combustion chamber schlieren, planar laser induced fluorescence PLIF and self-luminous multiple image data and pressure data obtained under each equivalence ratio; secondly, cutting, normalizing and denoising the three image data, reserving image information of a combustion chamber area, and enhancing the consistency of the same image among different individuals under different imaging conditions; secondly, in order to reduce the memory occupation and the calculated amount during training and shorten the deep learning training time, down-sampling is carried out on image data, and the image resolution is reduced; and filtering the wall pressure signal data by using Fast Fourier Transform (FFT), so as to weaken high-frequency noise and low-frequency interference.
Preferably, the step S4 further includes: firstly, selecting a plurality of scramjet engine models with different configurations, carrying out a plurality of ground wind tunnel tests on a plurality of fuel pulse injection and a plurality of steady injection and different equivalence ratios in the step S1, and observing combustion modes of a sub-combustion mode, a scramjet engine and a mixed mode to obtain data; preprocessing the acquired data by applying the step S3, determining the modes according to the step S2, and finally grouping the combustion modes according to the values of different equivalence ratios; secondly, the industry expert combines three types of combustion flow field images and pressure data at the same time to carefully analyze according to the mode discrimination criteria in the step S2, and judges the accuracy of the grouped combustion modes again to ensure the accuracy of a data set; finally, samples and labels are obtained that classify all image and pressure data into a sub-burn modality, a hyper-burn modality, and a mixed modality.
Preferably, the step S5 further includes: constructing an engine combustion field spectral image database through marking work in the step S4, and obtaining flow field images and wall surface pressure data under different configurations, different injection and different equivalence ratios in total, wherein each group of data specifically comprises 1 schlieren image, 1 plane laser induced fluorescence PLIF image, 1 self-luminous image and pressure data at the same time; the database is divided into a training set Train and a Test set Test in proportion without repetition, and the proportion is 8.
Preferably, the step S6 further includes: firstly, respectively extracting features of a schlieren image, a planar laser induced fluorescence PLIF image and a self-luminous image by adopting convolution operation to obtain an image feature matrix; secondly, extracting the characteristics of the pressure data by adopting a multilayer perceptron to obtain a pressure characteristic matrix; thirdly, transforming the image characteristic matrix and the pressure characteristic matrix into the same shape, and adding the image characteristic matrix and the pressure characteristic matrix to obtain a multi-source characteristic matrix; fourthly, constructing a self-attention encoder to further extract the features of the multi-source feature matrix, and improving the feature semantic expression capability; fifthly, the extracted features are sent into a full connection layer, and a combustion mode prediction category is obtained through softmax;
the Self-Attention encoder is completely realized by depending on a Self-Attention mechanism and consists of L identical layers, wherein each Layer mainly consists of two components, namely a Multi-Head Self-Attention Layer Multi-Head Self attribute as shown in a formula 10 and a Multi-Layer Perceptron Multi-Layer Percertron as shown in a formula 11; the multilayer sensor consists of two full-connection layers and a GeLu activation function in the middle, wherein the two components both adopt a residual error structure and use layer normalization at the front end;
z' l ＝MSA(LN(z l-1 ))+z l-1 ,l＝1,...,L (10)
z l ＝MLP(LN(z' l ))+z' l ,l＝1,...,L (11)
wherein z is l-1 Sequence of embedded images for l-1 layers, z l Sequence of embedded images of l layers, z' l For calculating the intermediate value, l is the number of layers;
the multi-head self-attention layer is a core component of the self-attention coding unit and consists of a linear layer, a self-attention head, a connecting layer and a final linear mapping layer; the self-attention head completes self-attention calculation by calculating the correlation of each element in the image embedding sequence with other elements, and the calculation method is as follows: first, the image sequence z will be embedded from the attention head 0 With three learnable self-attention weight matrices (W) for each element in (A) q ,W k ,W v ) Multiplying as in equation 12, generating three values (q, k, v), learning the self-attention weight by calculating the dot product of (q, k, v); then, calculating the dot product between an element q vector and other element k vectors in the embedded image sequence from the attention head, determining the correlation between the element and other elements, and then scaling the dot product result and sending the scaled result into softmax as formula 13, wherein the scaling factor D is k Is an attention weight matrix W k Dimension (d); finally, multiplying the v vectors of all elements embedded in the image sequence by the output of softmax from the attention head to obtain the output with the highest attention scoreSequence, complete self-attention calculation as in equation 14; the multi-head self-attention layer is formed by stacking 12 self-attention heads, the self-attention calculation processes are executed in parallel, and the results are projected to a high-dimensional space such as a formula 15 through a learnable linear mapping layer after being spliced;
SA(z)＝A·v (14)
wherein (q, k, v) is a self-attention vector, W is a self-attention calculation weight matrix, D k For the scaling factor, SA (z) is a single-headed self-attention calculation result vector.
Preferably, the step S7 further includes: on the test data set constructed in the step S5, performing detailed performance test on the combustion mode self-attention recognition network trained in the step S6, wherein the test selection accuracy Precision, recall rate and F1 score F1S are used as evaluation indexes, the mathematical formulas of the specific indexes are respectively shown as a formula 16, a formula 17 and a formula 18, and finally, the parameters of the deep learning model are further optimized according to the performance test indexes, so that the performance of the combustion mode recognition model is improved;
wherein TP is true positive by correctly predicting the number of combustion modes; FP is the number of mispredicted combustion modes, false positive; FN is the number of mispredicted combustion modes, false negative.
The invention also provides an intelligent monitoring system for the combustion mode of the scramjet engine, which comprises the following components:
the system comprises a contact pressure measurement module, a non-contact self-luminous image acquisition module, a non-contact schlieren and plane laser induced fluorescence PLIF image acquisition module, a CPU processing platform and a GPU development board;
contact pressure acquisition module: mounting pressure sensor probes on an isolation section of an engine model and the upper wall surface and the lower wall surface of a combustion chamber at equal intervals, and outputting pressure data of each point in real time, wherein the output frequency is 10KHZ;
the non-contact self-luminous image acquisition module comprises: shooting images of a combustion chamber by adopting a high-speed camera, wherein the frame rate is 10kfps;
the non-contact schlieren and plane laser induced fluorescence PLIF image acquisition module comprises: laser emitted by the continuous laser reaches the flow field test area after passing through the reflecting mirror and the schlieren mirror, and is split by the dichroic spectroscope after passing through the flow field test area; meanwhile, fluorescence excited by the pulse laser is also in the test area, OH group information is collected by the enhanced charge coupled device ICCD, and schlieren information is collected by the high-speed camera; therefore, synchronous information in the flow field is obtained through synchronous measurement of the PLIF and the schlieren;
a CPU processing platform: receiving and preprocessing pressure data, self-luminous image data, schlieren images and PLIF image data, and sending the preprocessed data to a GPU deep learning platform; receiving and outputting an identification result output by the intelligent identification deep convolutional neural network model;
GPU development board: firstly, receiving preprocessed pressure data and image data; then, sending the data into a trained intelligent recognition CNN model, and carrying out model reasoning acceleration through a TensorRT library provided by NVIDIA; and finally, outputting probability values corresponding to the three combustion modes.
Preferably, the intelligent monitoring system for the combustion mode of the scramjet engine comprises: the method comprises the steps of monitoring pressure change conditions of an engine isolation section and upper and lower wall surfaces of a combustion chamber by adopting a 10kHZ high-frequency pressure sensor, acquiring a flame hydroxyl image in the combustion chamber by adopting a 10kHZ plane laser induced fluorescence sensor, acquiring a schlieren image by adopting a high-speed camera, and acquiring flame image information in the combustion chamber by adopting self-luminescence photography, wherein the sensors are used for synchronous measurement.
As a preferred mode, the intelligent monitoring system for the combustion mode of the scramjet engine comprises:
the English WEIDA platform adopts a development board of an XCZU3CG-1SFVC784 chip to accelerate the built intelligent recognition deep neural network CNN of the scramjet combustion mode.
Compared with the prior art, the invention has the following advantages: (1) The invention applies the wall surface pressure data and the image data of the scramjet obtained by measuring with various sensors, breaks through the traditional thought of judging and analyzing the combustion mode based on the basic physical parameters representing combustion, and can establish a data-driven high-dynamic combustion mode monitoring and analyzing method based on multi-information fusion. (2) The method applies the deep learning technology to extract the characteristics of the multi-information fused image, further classifies the combustion modes of the scramjet, and avoids the defect that the combustion modes under the large-range working conditions are predicted without considering the flame shape and the pressure signal caused by the response of the flame shape to the flow disturbance and using the grasped knowledge range difficultly when the mode is judged by only applying single wall pressure data in the prior art.
Drawings
FIG. 1 is a flow chart of an intelligent monitoring method for combustion mode of a scramjet engine according to the present invention;
FIG. 2 is a block diagram of an intelligent monitoring system for combustion mode of a scramjet engine according to the present invention;
FIG. 3 is a schematic view of the contact pressure sensor of the present invention collecting pressure;
FIG. 4 shows the optical path for the non-contact schlieren and planar laser induced fluorescence PLIF system of the present invention;
FIG. 5 is a block diagram of the self-attention encoder of the present invention;
fig. 6 is a block diagram of a combustion modality self-attention recognition network of the present invention.
Detailed Description
The following embodiments of the present invention are provided by way of specific examples, and other advantages and effects of the present invention will be readily apparent to those skilled in the art from the disclosure herein. The invention is capable of other and different embodiments and of being practiced or of being carried out in various ways, and its several details are capable of modification in various respects, all without departing from the spirit and scope of the present invention.
Example 1:
the embodiment of the invention provides an intelligent monitoring method for a combustion mode of a scramjet engine,
the method comprises the following steps:
s1, acquiring image data and wall surface pressure signal data of a flow field of an isolation section of a scramjet engine and a combustion field of a combustion chamber by adopting various measuring sensors;
s2, combining wall surface pressure data, an isolation section flow field structure and combustion chamber Mach number distribution to form a combustion mode judgment criterion, dividing the combustion mode into a super-combustion mode, a sub-combustion mode and a mixed mode, and determining combustion mode types under different equivalence ratios;
s3, simultaneously aligning multiple image data and wall pressure data of the combustion chamber under the same equivalence ratio, and cutting, brightness conversion and contrast adjustment are carried out on the image data;
s4, performing combined comparison on multiple image data acquired by the combustion chamber, and manually marking the image data according to the combustion mode types under different equivalence ratios;
s5, constructing an engine combustion field spectral image database, forming a data set for subsequent deep learning network training, and dividing the data set into a training set Train and a Test set Test;
s6, building a combustion mode self-attention identification network suitable for the data set in the step S5; respectively acquiring a characteristic matrix of a combustion flow field image and pressure data through a convolution kernel and a multilayer perceptron by using a training set; further extracting representative semantic features using a self-attention coding unit; training a combustion mode self-attention recognition network through a deep learning reverse propagation gradient descent mechanism, finally fusing the extracted features, inputting the fused features into a full-link classifier, and accurately judging the combustion mode category;
and S7, performing performance test on the trained network by using the test data set, inputting the test data set, and comparing the output value with the true value of the combustion mode to obtain the accuracy of classification judgment of the combustion mode.
Example 2:
the embodiment of the invention provides an intelligent monitoring method for a combustion mode of a scramjet engine, which specifically comprises the following steps S1-S7 as shown in FIG. 1:
s1, acquiring image data and wall surface pressure signal data of a flow field of an isolation section of the scramjet engine and a combustion field of a combustion chamber by adopting various measuring sensors.
In the embodiment of the present invention, step S1 specifically includes:
the test is carried out on a hydrogen-oxygen combustion wind tunnel, a bi-modal scramjet engine is used, and a hydrogen-oxygen combustion and supplementation method is adopted to obtain high-enthalpy polluted air, wherein the total temperature is 1350K, the total pressure is 1.75Mpa, the flow rate is 2.89kg/s, and O is 2 、N 2 And H 2 The molar fractions of the O component were 20.09%,65.65% and 19.26%, respectively, and the incoming stream Mach number was 2.5, normal temperature kerosene was injected into the combustor through 15 circular holes having a hole diameter of 0.3mm, the injection position was located 15mm upstream of the exit of the separation section, and the equivalence ratios of the kerosene in the tests were 0.4, 0.5, 0.6, 0.65, 0.7, 0.75, 0.8 and 0.9, respectively. A series of 10kHZ pressure sensors are arranged on the upper wall surface and the lower wall surface of the isolation section and the combustion chamber and are used for acquiring wall surface pressure signal data. Obtaining the wave system structure of the isolation section and the flow field structure information of the combustion chamber by adopting high-speed schlieren, wherein the shooting frame frequency is 1 multiplied by 10 4 fps, camera exposure time set to 4.62 mus. And (3) acquiring the evolution process of the pioneer hydrogen flame by adopting a hydroxyl radical (OH) plane induced laser technology (PLIF). And acquiring various flow field image data and wall pressure data under different equivalence ratios.
S2, combining wall surface pressure data, an isolation section flow field structure and combustion chamber Mach number distribution to form a combustion mode judgment criterion, dividing the combustion mode into a super-combustion mode, a sub-combustion mode and a mixed mode, and determining combustion mode types under different equivalence ratios;
in the embodiment of the present invention, step S2 specifically includes:
determining the heat release rate according to the wall pressure data, the isolation section flow field image data and the combustion chamber Mach number distribution under the conditions of the equivalence ratios of 0.4, 0.5, 0.6, 0.65, 0.7, 0.75, 0.8 and 0.9, which are obtained in the step S1, and analyzing the influence of the heat release rate on the combustion mode under different equivalence ratios. Meanwhile, whether the isolation section has a shock wave string or not is used for reflecting the occurrence of a large flow separation area in the flow field and the typical characteristics of a sub-combustion mode, the possible position of a thermal throat generated in the flow field is reflected by the position with the mass weighted Mach number equal to 1.0, and finally the mode is comprehensively judged according to whether the isolation section has a shock wave string structure, whether the mass weighted Mach number is smaller than 1.0 and whether heat release is concentrated or not. When a shock wave string structure exists in the isolation section, the mass weighted Mach number is smaller than 1.0 region, and the heat release rate is centralized, the combustion mode of the dual-mode scramjet engine is a sub-combustion mode. When no shock wave string structure exists in the isolation section, no mass weighted Mach number is smaller than 1.0 region, and the heat release rate is distributed, the combustion mode of the dual-mode scramjet engine is a sub-combustion mode. Otherwise, the mixed mode is adopted. Finally, the equivalence ratios of 0.4, 0.5, 0.6, 0.65, 0.7, 0.75, 0.8 and 0.9 are divided into scramjet super-burn modes, sub-burn modes and mixed modes according to the above definitions, and data under the three combustion modes can be obtained, and a foundation is laid for subsequently establishing a data set.
S3, simultaneously aligning multiple image data and wall pressure data of the combustion chamber under the same equivalence ratio, and cutting, brightness conversion and contrast adjustment are carried out on the image data;
in the embodiment of the present invention, step S3 specifically includes:
first, wall pressure data is extracted at a fixed step length, and timing synchronization is performed on a plurality of types of image data such as combustion chamber streaks, PLIFs, and self-luminescence obtained at each equivalence ratio and the pressure data. And secondly, cutting, normalizing and denoising the three image data, reserving image information of a combustion chamber area, and enhancing the consistency of the same image among different individuals under different imaging conditions. Then, in order to reduce memory usage and calculation amount during training and shorten deep learning training time, down-sampling is performed on high-resolution image data. In addition, because the test site environment of the scramjet is complex, and the pressure sensor is influenced by factors such as electromagnetic interference and structural vibration generated by a high-temperature, high-pressure and high-frequency spark plug, the wall surface pressure signal data is filtered by Fast Fourier Transform (FFT), so that high-frequency noise and low-frequency interference are weakened.
S4, performing combined comparison on multiple image data acquired by the combustion chamber, and manually marking the image data according to the combustion mode types under different equivalence ratios;
in the embodiment of the present invention, step S4 specifically includes:
firstly, selecting 3 scramjet engine models with different configurations, carrying out ground wind tunnel tests for 50 times by 2 fuel pulse injections and 2 steady injections and equivalent ratios of 0.4, 0.5, 0.6, 0.65, 0.7, 0.75, 0.8 and 0.9 respectively, and observing combustion modes of three bimodal scramjet engines, namely a sub-combustion mode, a scramjet mode, a mixed mode and the like to obtain rich data. Preprocessing the acquired data by applying S3, determining the modes according to S2, and finally grouping the combustion modes according to the values of different equivalence ratios; secondly, the industry expert combines three types of combustion flow field images and careful analysis of pressure data at the moment according to the mode discrimination criterion in the S2, and judges the accuracy of the grouped combustion modes again to ensure the accuracy of a data set; finally, samples and labels are obtained that classify all image and pressure data into a sub-burn modality, a hyper-burn modality, and a mixed modality.
S5, constructing an engine combustion field spectral image database, forming a data set for subsequent deep learning network training, and dividing the data set into a training set Train and a Test set Test;
in the embodiment of the present invention, step S5 specifically includes:
and constructing an engine combustion field spectral image database through S4 labeling work, and obtaining flow field images and wall surface pressure data under different configurations, different injection and different equivalence ratios in total, wherein each group of data specifically comprises 1 schlieren image, 1 PLIF image, 1 self-luminous image and pressure data at the moment. The database is divided into a training set Train and a Test set Test in proportion without repetition, and the proportion is 8.
The data set formed in the S5 and used for subsequent deep learning network training is characterized in that under the condition of different equivalence ratios, the data set is fused by combining spectral image data of non-contact measurement of various combustion chambers according to the flow characteristics and wall surface pressure values of the flow field of the isolation section, and classification of a super-combustion mode, a sub-combustion mode and a mixed mode is carried out; and mining the rule between the isolated section flow field and the spectral image of the combustion chamber with multi-information fusion under different equivalence ratios, labeling the image data, and constructing an engine combustion flow field database.
S6, building a combustion mode self-attention identification network suitable for the data set in the step S5; respectively acquiring a combustion flow field image and a characteristic matrix of pressure data through a convolution kernel and a multilayer perceptron by applying the constructed training set Train; the representative semantic features are further extracted using a self-attention coding unit. Training a combustion mode self-attention recognition network through a deep learning reverse propagation gradient descent mechanism, finally fusing the extracted features, inputting the fused features into a full-link classifier, and accurately judging the category of the combustion mode;
in the embodiment of the present invention, step S6 specifically includes:
as shown in fig. 6, in a first step, performing feature extraction on the schlieren image, the PLIF image and the self-luminous image respectively by using convolution operation to obtain an image feature matrix; secondly, extracting the characteristics of the pressure data by adopting a multilayer perceptron to obtain a pressure characteristic matrix; thirdly, transforming the image characteristic matrix and the pressure characteristic matrix into the same shape, and adding the image characteristic matrix and the pressure characteristic matrix to obtain a multi-source characteristic matrix; fourthly, constructing a self-attention encoder to further extract the features of the multi-source feature matrix, and improving the feature semantic expression capability; and fifthly, sending the extracted features into a full connection layer, and obtaining a combustion mode prediction category through softmax.
The Self-Attention encoder is implemented completely by means of a Self-Attention mechanism (fig. 5), and consists of L identical layers, each Layer mainly consists of two components, namely a Multi-Head Self-Attention Layer (MSA) (formula 19) and a Multi-Layer Perceptron (MLP) (formula 20). The multilayer perceptron is composed of two full-connection layers and a GeLu activation function in the middle, the two components both adopt a residual error structure, and layer normalization is used at the front end.
z l '＝MSA(LN(z l-1 ))+z l-1 ,l＝1,,L (19)
z l ＝MLP(LN(z' l ))+z' l ,l＝1,...,L (20)
Wherein z is l-1 Sequence of embedded images for l-1 layers, z l Sequence of embedded images of l layers, z' l For calculating the intermediate value, l is the number of layers;
the multi-head self-attention layer is a core component of the self-attention coding unit and consists of a linear layer, a self-attention head, a connection layer and a final linear mapping layer. The self-attention head completes self-attention calculation by calculating the correlation of each element in the image embedding sequence with other elements, and the calculation method is as follows: first, the image sequence z will be embedded from the attention head 0 With three learnable self-attention weight matrices (W) for each element in (A) q ,W k ,W v ) Multiplying (formula 21), generating three values (q, k, v), and learning a self-attention weight by calculating a dot product of (q, k, v); then, calculating the dot product between the q vector of one element and the k vector of other elements in the embedded image sequence from the attention head, determining the correlation between the element and other elements, and then scaling the result of the dot product to send to softmax (formula 22), wherein the scaling factor D is k Is an attention weight matrix W k Dimension (d); finally, the self-attention head multiplies the v vectors embedded in all elements of the image sequence by the output of softmax, obtains the sequence with the highest attention score, and completes the self-attention calculation (formula 23). The multi-head self-attention layer adopts 12 self-attention layersThe heads are stacked, the above self-attention calculation processes are executed in parallel, and the results are projected to a high-dimensional space through a learnable linear mapping layer after being spliced (formula 24).
SA(z)＝A·v (23)
Wherein (q, k, v) is a self-attention vector, W is a self-attention calculation weight matrix, D k For the scaling factor, SA (z) is a single-headed self-attention calculation result vector.
And S7, performing performance test on the trained network by using the test data set Train, firstly inputting the test data set Train to the trained network model of S6, and comparing the predicted value of the combustion mode with the true value of the combustion mode to obtain the accuracy of classification judgment of the combustion mode.
In the embodiment of the present invention, step S7 specifically includes:
and (3) performing detailed performance test on the combustion mode self-attention recognition network trained in the step (6) on the test data set constructed in the step (5), wherein the test selection accuracy (Precision), the Recall rate (Recall) and the F1 score (F1S) are used as evaluation indexes, the mathematical formulas of the specific indexes are respectively shown as a formula 25, a formula 26 and a formula 27, and finally, the parameters of the deep learning model are further optimized according to the performance test indexes, so that the performance of the combustion mode recognition model is improved.
Wherein TP is true positive for correctly predicting the number of combustion modes; FP is the number of mispredicted combustion modes, false positive; FN is the number of mispredicted combustion modes, false negative.
Example 3:
the embodiment of the invention provides an intelligent monitoring system for a combustion mode of a scramjet engine, which comprises the following components as shown in fig. 2:
the system comprises a contact pressure acquisition module, a non-contact self-luminous image acquisition module, a non-contact schlieren and PLIF image acquisition module, a CPU processing platform and an English viand development board;
contact pressure acquisition module: installing pressure sensor probes on the isolation section of the engine model and the upper wall surface and the lower wall surface of a combustion chamber at equal intervals, wherein 28 upper wall surfaces and 28 lower wall surfaces are used for outputting 56 point pressure data in real time, and the output frequency is 10KHZ;
the non-contact self-luminous image acquisition module comprises: shooting images of the combustion chamber by using a high-speed camera, wherein the output resolution is 1024 × 256, and the frame rate is 10kfps;
a non-contact schlieren and plane laser induced fluorescence PLIF image acquisition module: laser emitted by the continuous laser reaches the flow field test area after passing through the reflecting mirror and the schlieren mirror, and is split by the dichroic spectroscope after passing through the flow field test area. Meanwhile, fluorescence excited by the pulse laser is also in the test area, OH radical information is collected by the enhanced charge coupled device ICCD, and schlieren information is collected by the high-speed camera. Therefore, synchronous information in the flow field is obtained through synchronous measurement of the PLIF and the schlieren.
A CPU processing platform: receiving and preprocessing pressure data, self-luminous image data, schlieren images and PLIF image data, and sending the preprocessed data to a deep learning platform; receiving and outputting an identification result output by the intelligent identification deep convolutional neural network model;
england development board: firstly, receiving preprocessed pressure data and image data; then, sending the data into a trained intelligent recognition deep convolutional neural network model CNN model, and carrying out model inference acceleration through a TensorRT library provided by NVIDIA; and finally, outputting probability values corresponding to the three combustion modes.
The system adopts a 10kHZ high-frequency pressure sensor to monitor the pressure change conditions of the upper wall surface and the lower wall surface of an engine isolation section and a combustion chamber, adopts a 10kHZ plane laser induced fluorescence sensor to acquire a flame hydroxyl image in the combustion chamber, adopts a high-speed camera to acquire a schlieren image, adopts self-luminous photography to acquire flame image information in the combustion chamber, and carries out synchronous measurement by the sensors.
The English WEIDA platform adopts a development board of an XCZU3CG-1SFVC784 chip to accelerate the built intelligent recognition deep neural network CNN of the scramjet combustion mode.
The working method of the system comprises the following steps S1-S2:
s1, performing multi-sensor data acquisition by using contact and non-contact optical measurement sensors, and realizing data transmission between a CPU platform and a GPU development board platform by using a high-speed data transmission protocol;
and S2, the CPU platform receives measured multi-sensor data information, processes the data, converts the processed sensor data into a tensor form, transmits the tensor form to the GPU development board platform, completes the intelligent combustion mode monitoring method based on deep learning according to the built deep learning model, and finally accurately outputs the combustion mode discrimination types of a sub-combustion mode, a super-combustion mode and a mixed mode and transmits the combustion mode discrimination types to the PC end upper computer.
In the embodiment of the present invention, step S2 specifically includes:
the data acquired by the various types of sensors are converted into tensor forms and transmitted into a trained intelligent recognition model, the model is subjected to accelerated operation by using a development board of an Invitta platform XCZU3CG-1SFVC784 chip, the model outputs probability values corresponding to three combustion modes through operation, and finally the data are processed by a CPU (Central processing Unit) to output the combustion mode corresponding to the maximum probability value. The system is shown in figure 2.
The foregoing embodiments are merely illustrative of the principles and utilities of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or change the above-mentioned embodiments without departing from the spirit and scope of the present invention. Accordingly, it is intended that all equivalent modifications or changes which can be made by those skilled in the art without departing from the spirit and technical spirit of the present invention be covered by the claims of the present invention.
Claims (6)
1. An intelligent monitoring method for a combustion mode of a scramjet engine is characterized by comprising the following steps:
s1, acquiring image data and wall surface pressure signal data of a flow field of an isolation section of a scramjet engine and a combustion field of a combustion chamber by adopting various measuring sensors; the step S1 further comprises: performing a test on an oxyhydrogen combustion wind tunnel, applying a bimodal scramjet engine, obtaining high-enthalpy polluted air by adopting a hydrogen-burning oxygen-supplementing method, and injecting normal-temperature kerosene into a combustion chamber through a circular hole; a series of pressure sensors are arranged on the upper wall surface and the lower wall surface of the isolation section and the combustion chamber and are used for acquiring wall surface pressure signal data; acquiring the wave system structure of the isolation section and the flow field structure information of the combustion chamber by adopting high-speed schlieren; adopting a hydroxyl free radical planar laser induced fluorescence PLIF method to obtain the evolution process of pioneer hydrogen flame, and obtaining various flow field image data and wall surface pressure data under different equivalence ratios;
s2, combining wall surface pressure data, an isolation section flow field structure and combustion chamber Mach number distribution to form a combustion mode judgment criterion, dividing the combustion mode into a super-combustion mode, a sub-combustion mode and a mixed mode, and determining combustion mode types under different equivalence ratios; the step S2 further comprises: determining the heat release rate according to the wall surface pressure data, the isolation section flow field image data and the combustion chamber Mach number distribution condition under the condition of different equivalence ratios acquired in the step S1, and analyzing the influence of the heat release rate generated under different equivalence ratios on the combustion mode; meanwhile, whether the isolation section has a shock wave string or not is used for reflecting the occurrence of a large flow separation area in the flow field and the typical characteristics of a sub-combustion mode, the possible position of a thermal throat generated in the flow field is reflected by the position with the mass weighted Mach number equal to 1.0, and finally the mode is comprehensively judged according to whether the isolation section has a shock wave string structure, whether the mass weighted Mach number is smaller than 1.0 and whether heat release is concentrated or not; when a shock wave string structure exists in the isolation section, the mass weighted Mach number is larger than 1.0 region, and the heat release rate is centralized, the combustion mode of the bimodal scramjet engine is a scramjet mode; when no shock wave string structure exists in the isolation section, no mass weighting Mach number is smaller than 1.0 region, and the heat release rate is distributed, the combustion mode of the bimodal scramjet engine is a sub-combustion mode; otherwise, the mixed mode is adopted; finally, the different equivalence ratios are divided into a scramjet engine scramjet mode, a sublagration mode and a mixed mode, so that data under the three combustion modes can be obtained, and a foundation is laid for subsequently establishing a data set;
s3, simultaneously aligning multiple image data and wall pressure data of the combustion chamber under the same equivalence ratio, and cutting, brightness conversion and contrast adjustment are carried out on the image data; the step S3 further comprises: firstly, extracting wall surface pressure data according to a fixed step length, and carrying out time sequence synchronization on combustion chamber schlieren, planar laser induced fluorescence PLIF and self-luminous multiple image data and pressure data obtained under each equivalence ratio; secondly, cutting, normalizing and denoising the three image data, reserving image information of a combustion chamber area, and enhancing the consistency of the same image among different individuals under different imaging conditions; then, in order to reduce the memory occupation and the calculated amount during training and shorten the deep learning training time, down-sampling is carried out on the image data, and the image resolution is reduced; filtering the wall surface pressure signal data by using Fast Fourier Transform (FFT), and weakening high-frequency noise and low-frequency interference;
s4, performing combined comparison on multiple image data acquired by the combustion chamber, and manually marking the image data according to the combustion mode types under different equivalence ratios; the step S4 further includes: firstly, selecting a plurality of scramjet engine models with different configurations, carrying out a plurality of ground wind tunnel tests by a plurality of fuel pulse injection and a plurality of steady injection and different equivalence ratios in the step S1, and observing combustion modes of three bimodal scramjet engines, namely a sub-combustion mode, a scramjet mode and a mixed mode to obtain data; preprocessing the acquired data by applying the step S3, determining the modes according to the step S2, and finally grouping the combustion modes according to the values of different equivalence ratios; secondly, the industry expert combines three types of combustion flow field images and pressure data at the same time to carefully analyze according to the mode discrimination criteria in the step S2, and judges the accuracy of the grouped combustion modes again to ensure the accuracy of a data set; finally, samples and labels are obtained for all image and pressure data classifications for sub-ignition, hyper-ignition, and mixed modalities;
s5, constructing an engine combustion field spectral image database, forming a data set for subsequent deep learning network training, and dividing the data set into a training set Train and a Test set Test; the step S5 is further as follows: constructing an engine combustion field spectral image database through marking work in the step S4, and obtaining flow field images and wall surface pressure data under different configurations, different injection and different equivalence ratios in total, wherein each group of data specifically comprises 1 schlieren image, 1 plane laser induced fluorescence PLIF image, 1 self-luminous image and pressure data at the same time; randomly dividing the database into a training set Train and a Test set Test in a non-repetitive manner according to a ratio of 8;
s6, building a combustion modal self-attention recognition network suitable for the data set in the step S5; respectively acquiring a characteristic matrix of a combustion flow field image and pressure data through a convolution kernel and a multilayer perceptron by using a training set; further extracting representative semantic features using a self-attention coding unit; training a combustion mode self-attention recognition network through a deep learning reverse propagation gradient descent mechanism, finally fusing the extracted features, inputting the fused features into a full-link classifier, and accurately judging the combustion mode category; the step S6 further includes: firstly, respectively extracting features of a schlieren image, a planar laser induced fluorescence PLIF image and a self-luminous image by adopting convolution operation to obtain an image feature matrix; secondly, extracting the characteristics of the pressure data by adopting a multilayer perceptron to obtain a pressure characteristic matrix; thirdly, transforming the image characteristic matrix and the pressure characteristic matrix into the same shape, and adding the image characteristic matrix and the pressure characteristic matrix to obtain a multi-source characteristic matrix; fourthly, constructing a self-attention encoder to further extract the features of the multi-source feature matrix, and improving the feature semantic expression capability; fifthly, the extracted features are sent into a full connection layer, and a combustion mode prediction category is obtained through softmax;
the Self-Attention encoder is completely realized by depending on a Self-Attention mechanism and consists of L identical layers, wherein each Layer mainly consists of two components, namely a Multi-Head Self-Attention Layer Multi-Head Self attribute as formula 1 and a Multi-Layer Perceptron Multi-Layer Percertron as formula 2; the multilayer sensor consists of two full-connection layers and a GeLu activation function in the middle, wherein the two components both adopt a residual error structure and use layer normalization at the front end;
z′ l ＝MSA(LN(z l-1 ))+z l-1 ,l＝1,...,L (1)
z l ＝MLP(LN(z′ l ))+z′ l ,l＝1,...,L (2)
wherein z is l-1 Sequence of embedded images for l-1 layers, z l Sequence of embedded images of l layers, z' l For calculating the intermediate value, l is the number of layers;
the multi-head self-attention layer is a core component of the self-attention coding unit and consists of a linear layer, a self-attention head, a connecting layer and a final linear mapping layer; the self-attention head completes self-attention calculation by calculating the correlation of each element in the image embedding sequence with other elements, and the calculation method is as follows: first, the image sequence z will be embedded from the attention head 0 With three learnable self-attention weight matrices (W) for each element in (A) q ,W k ,W v ) Multiplying as in equation 3, generating three values (q, k, v), learning the self attention weight by calculating the dot product of (q, k, v); then, calculating dot product between q vector of one element and k vector of other elements in the embedded image sequence by self attention head, determining correlation between the element and other elements, and then scaling the dot product result and sending the scaled dot product result into softmax as formula 4, wherein scaling factor D is k Is an attention weight matrix W k Dimension (d); finally, multiplying the v vectors of all elements of the embedded image sequence by output of softmax by the self-attention head to obtain a sequence with the highest attention score, and finishing self-attention calculation as formula 5; the multi-head self-attention layer is formed by stacking 12 self-attention heads, the self-attention calculation processes are executed in parallel, and the results are projected to a high-dimensional space through a learnable linear mapping layer after being spliced as shown in a formula 6;
SA(z)＝A·v (5)
wherein (q, k, v) is a self-attention vector, W is a self-attention calculation weight matrix, D k SA (z) is a single-headed self-attention calculation result vector as a scaling factor;
and S7, performing performance test on the trained network by using the test data set, inputting the test data set, and comparing the output value with the true value of the combustion mode to obtain the accuracy of classification judgment of the combustion mode.
2. The intelligent monitoring method for the combustion mode of the scramjet engine according to claim 1, characterized in that: the data set formed in the S5 and used for subsequent deep learning network training is characterized in that under the condition of different equivalence ratios, the flow characteristics and wall pressure values of the flow field of the isolation section are combined with the spectral image data of non-contact measurement of various combustion chambers for fusion, and classification of a super-combustion mode, a sub-combustion mode and a mixed mode is performed; and mining the rule between the isolated section flow field and the spectrum image of the combustion chamber multi-information fusion under different equivalence ratios, labeling the image data, and constructing an engine combustion flow field database.
3. The intelligent monitoring method for the combustion mode of the scramjet engine according to claim 1, characterized in that: the step S7 is further: on the test data set constructed in the step S5, carrying out detailed performance test on the combustion modal self-attention recognition network trained in the step S6, wherein the test selects accuracy Precision, recall and F1 score F1S as evaluation indexes, mathematical formulas of specific indexes are respectively shown as a formula 7, a formula 8 and a formula 9, and finally, the parameters of the deep learning model are further optimized according to the performance test indexes, so that the performance of the combustion modal recognition model is improved;
wherein TP is true positive for correctly predicting the number of combustion modes; FP is the number of mispredicted combustion modes, false positive; FN is the number of mispredicted combustion modes, false negative.
4. An intelligent monitoring system for combustion modes of a scramjet engine, which uses the intelligent monitoring method for combustion modes of the scramjet engine as claimed in any one of claims 1 to 3, and is characterized by comprising the following steps:
the system comprises a contact pressure measurement module, a non-contact self-luminous image acquisition module, a non-contact schlieren and plane laser induced fluorescence PLIF image acquisition module, a CPU processing platform and a GPU development board;
contact pressure acquisition module: mounting pressure sensor probes on an isolation section of an engine model and the upper wall surface and the lower wall surface of a combustion chamber at equal intervals, and outputting pressure data of each point in real time, wherein the output frequency is 10KHZ;
the non-contact self-luminous image acquisition module comprises: shooting images of a combustion chamber by adopting a high-speed camera, wherein the frame rate is 10kfps;
the non-contact schlieren and plane laser induced fluorescence PLIF image acquisition module comprises: laser emitted by the continuous laser passes through the reflecting mirror and the schlieren mirror and then reaches the flow field test area, and the laser passes through the flow field test area and is split by the dichroic spectroscope; meanwhile, fluorescence excited by the pulse laser is also in the test area, OH group information is collected by the enhanced charge coupled device ICCD, and schlieren information is collected by the high-speed camera; therefore, synchronous information in the flow field is obtained through synchronous measurement of the PLIF and the schlieren;
a CPU processing platform: receiving and preprocessing pressure data, self-luminous image data, schlieren images and PLIF image data, and sending the preprocessed data to a GPU deep learning platform; receiving and outputting an identification result output by the intelligent identification deep convolution neural network model;
GPU development board: firstly, receiving preprocessed pressure data and image data; then, sending the data into a trained intelligent recognition CNN model, and carrying out model reasoning acceleration through a TensorRT library provided by NVIDIA; and finally, outputting probability values corresponding to the three combustion modes.
5. The intelligent monitoring system for the combustion mode of the scramjet engine according to claim 4, wherein:
the method comprises the steps of monitoring pressure change conditions of an engine isolation section and upper and lower wall surfaces of a combustion chamber by adopting a 10kHZ high-frequency pressure sensor, acquiring a flame hydroxyl image in the combustion chamber by adopting a 10kHZ Planar laser induced fluorescence sensor, acquiring a schlieren image by adopting a high-speed camera, and acquiring flame image information in the combustion chamber by adopting self-luminous photography, wherein the sensors are used for synchronous measurement.
6. The intelligent monitoring system for the combustion mode of the scramjet engine according to claim 4, wherein:
the English viada platform adopts a development board of an XCZU3CG-1SFVC784 chip to accelerate the built scramjet combustion mode intelligent recognition deep neural network CNN.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210180557.2A CN114528769B (en) | 2022-02-25 | 2022-02-25 | Intelligent monitoring method and system for combustion mode of scramjet engine |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210180557.2A CN114528769B (en) | 2022-02-25 | 2022-02-25 | Intelligent monitoring method and system for combustion mode of scramjet engine |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114528769A CN114528769A (en) | 2022-05-24 |
| CN114528769B true CN114528769B (en) | 2023-04-07 |
Family
ID=81623898
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210180557.2A Active CN114528769B (en) | 2022-02-25 | 2022-02-25 | Intelligent monitoring method and system for combustion mode of scramjet engine |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114528769B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114923695B (en) * | 2022-07-18 | 2022-10-28 | 中国空气动力研究与发展中心空天技术研究所 | Intelligent detection method and system for combustion instability of scramjet engine |
| CN115311535B (en) * | 2022-09-29 | 2022-12-30 | 中国空气动力研究与发展中心低速空气动力研究所 | Icing wind tunnel cloud and mist field parameter acquisition method and storage medium |
| CN116771519B (en) * | 2023-06-20 | 2024-02-23 | 哈尔滨工业大学 | PLIF-based fuel engine equivalence ratio identification and regulation device and method |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107420221A (en) * | 2016-05-23 | 2017-12-01 | 杨庆春 | A kind of engine combustion modal identification method |
| CN111579248A (en) * | 2020-05-14 | 2020-08-25 | 中国空气动力研究与发展中心 | Optical synchronous measurement method for combustion flow field of scramjet engine |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110378431A (en) * | 2019-07-24 | 2019-10-25 | 中国人民解放军国防科技大学 | Convolutional neural network-based supersonic combustion chamber combustion mode detection method |
| CN113137634B (en) * | 2021-06-02 | 2022-04-26 | 厦门大学 | Variable-structure bimodal stamping combustion chamber |
| CN113837147B (en) * | 2021-10-29 | 2022-08-05 | 山东省人工智能研究院 | Transform-based false video detection method |
-
2022
- 2022-02-25 CN CN202210180557.2A patent/CN114528769B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107420221A (en) * | 2016-05-23 | 2017-12-01 | 杨庆春 | A kind of engine combustion modal identification method |
| CN111579248A (en) * | 2020-05-14 | 2020-08-25 | 中国空气动力研究与发展中心 | Optical synchronous measurement method for combustion flow field of scramjet engine |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114528769A (en) | 2022-05-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114528769B (en) | Intelligent monitoring method and system for combustion mode of scramjet engine | |
| Han et al. | An adaptive spatiotemporal feature learning approach for fault diagnosis in complex systems | |
| US11454567B2 (en) | Fault diagnosis method of reciprocating machinery based on keyphasor-free complete-cycle signal | |
| Ravikumar et al. | Gearbox fault diagnosis based on Multi-Scale deep residual learning and stacked LSTM model | |
| CN114923695B (en) | Intelligent detection method and system for combustion instability of scramjet engine | |
| CN113158445B (en) | Prediction algorithm for residual service life of aero-engine with convolution memory residual error self-attention mechanism | |
| CN108363382A (en) | A kind of complex equipment fault diagnosis method and system | |
| JP2007124880A (en) | Method of forming input vector of nerve network for automatic inference of partial discharge causes | |
| Kayaalp et al. | Developing a model for prediction of the combustion performance and emissions of a turboprop engine using the long short-term memory method | |
| Atasoy et al. | A comparative analysis of exhaust gas temperature based on machine learning models for aviation applications | |
| Ghazaly et al. | Prediction of misfire location for SI engine by unsupervised vibration algorithm | |
| CN110826600B (en) | Engine surge prediction method based on adaptive resonance network online incremental learning | |
| Li et al. | Structural health monitoring data anomaly detection by transformer enhanced densely connected neural networks | |
| CN114880767B (en) | Aero-engine residual service life prediction method based on attention mechanism Dense-GRU network | |
| US20230359863A1 (en) | Virtual-sensing integrated for predictive reliability | |
| Deng et al. | An image understanding based model with ion current signals for predicting combustion information | |
| Kozlenko et al. | Software implemented fault diagnosis of natural gas pumping unit based on feedforward neural network | |
| CN115017984A (en) | Early warning method and system for failure risk of aircraft engine | |
| Aramburu et al. | Knock detection in spark ignited heavy duty engines: An application of machine learning techniques with various knock sensor locations | |
| Tabaszewski et al. | Vibration-based identification of engine valve clearance using a convolutional neural network | |
| CN113723199A (en) | Airport low visibility detection method, device and system | |
| Gao et al. | Flight Data Adaptive Segmentation and Classification for Fleet-Level Anomaly Detection | |
| CN112037215B (en) | Insulator defect detection method and system based on zero sample learning | |
| KR102372124B1 (en) | Method of Fault Classification of Solenoid Pumps based on Machine Learning | |
| Khattak et al. | Multi-modal LSTM network for anomaly prediction in piston engine aircraft |
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 |