A hydrogen combustion tempering monitoring system and method based on multi-modal data fusion

Abstract

The application discloses a hydrogen combustion tempering monitoring system and method based on multi-modal data fusion, and relates to the technical field of combustion monitoring and diagnosis. A data acquisition module acquires physical parameters and image data of a hydrogen combustion process; a data preprocessing module pre-processes the physical parameters and the image data; a feature extraction module extracts features from the pre-processed physical parameters and the image data; a multi-modal feature fusion module splices the extracted features to form a joint feature vector and performs dimension reduction; a tempering state discrimination module constructs a tempering state discrimination model, outputs a probability value of a tempering state occurrence, and judges a combustion state according to a set safety threshold; and an audible and visual alarm module alarms based on a result selection. Multi-modal features are extracted based on physical parameters and image data, and are fused to form a joint feature vector; a tempering state discrimination model is constructed according to a random forest algorithm; and in combination with a set safety threshold, the audible and visual alarm module can be triggered in real time, so that real-time and accurate discrimination of a tempering phenomenon is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of combustion monitoring and diagnosis, and particularly relates to a hydrogen combustion backfire monitoring system and method based on multi-modal data fusion. BACKGROUND

[0002] With the increasing concern about global climate change, low-carbon development has become an important and urgent requirement in the energy field. Hydrogen fuel is considered one of the cleanest and most prominent carbon-neutral fuels in gas turbine operation due to its non-polluting, high-efficiency, and recyclable characteristics. Therefore, low-carbon fuels containing hydrogen, hydrogen-rich, or pure hydrogen have received widespread attention.

[0003] Due to the active chemical properties and fast diffusion rate of hydrogen, hydrogen combustion technology faces great challenges, with the most prominent problem being backfire. Backfire refers to the movement of the flame from the reaction zone of the combustion chamber to the upstream gas premixing zone, which is the result of the coupling of factors such as combustion speed, local flow field conditions, and combustion instability. Simply put, backfire occurs when the flame speed is higher than the gas flow speed, but in the premixing section of the burner, the flame propagation speed is slow, and backfire can still occur by re-entering the combustion chamber through the boundary layer. Backfire not only affects combustion efficiency but also can cause serious safety problems. Therefore, monitoring backfire is of great significance to improving combustion safety and optimizing the combustion process.

[0004] Traditional backfire monitoring methods generally use temperature sensors or pressure sensors, which have certain limitations. They can only measure the local temperature or pressure at the installation location and cannot reflect the overall state of the combustion chamber. If backfire occurs in the sensor blind area, monitoring will fail. In addition, the measurement results of the sensors are easily disturbed by environmental factors, which may cause measurement errors and reduce the monitoring accuracy.

[0005] High-frequency planar laser-induced fluorescence (PLIF) technology is a cutting-edge optical measurement technology in the field of combustion diagnosis, with high temporal and spatial resolution and high sensitivity. The characteristics of the flame change at the initial stage of backfire, accompanied by changes in flame structure, flame position, and dynamic characteristics, which provides a new opportunity for backfire monitoring.

[0006] In summary, monitoring OH radicals, a product of hydrogen combustion, using PLIF technology can directly capture flame characteristics and provide more comprehensive information on the combustion state. Physical parameters (such as temperature and pressure) and OH-PLIF images respectively reflect different aspects of the combustion process; temperature and pressure data reflect the macroscopic state of combustion, while image data reflects the microscopic structure and local characteristics. Therefore, combining traditional monitoring methods with high-frequency optical diagnostic techniques can help provide high-frequency and multi-dimensional information, which is of great significance for achieving accurate monitoring of combustion chamber backfire. Summary of the Invention

[0007] To address the shortcomings of the prior art, this invention provides a hydrogen combustion flashback monitoring system and method based on multimodal data fusion. It extracts multimodal features from physical parameters and image data and fuses them to form a joint feature vector. It constructs a flashback state discrimination model based on the random forest algorithm and, combined with a set safety threshold, can trigger an audible and visual alarm module in real time, thereby achieving real-time and accurate discrimination of flashback phenomena.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] A hydrogen combustion flashback monitoring system based on multimodal data fusion includes, in sequence, a data acquisition module, a data preprocessing module, a feature extraction module, a multimodal feature fusion module, a flashback state discrimination module, and an audible and visual alarm module, wherein:

[0010] The data acquisition module consists of a sensor system and an OH-PLIF system, which acquire physical parameters and image data of the hydrogen combustion process, respectively. The physical parameters include temperature and pressure data, and the image data is an OH-PLIF image of the distribution of OH free radicals.

[0011] The data preprocessing module preprocesses the acquired physical parameters and image data respectively. For the physical parameters, discrete wavelet transform is used to remove noise and normalize them. For the image data, median filtering is used to reduce noise and then the maximum inter-class variance method is combined to segment the background region and flame region of the target location in the image.

[0012] The feature extraction module performs feature extraction on the preprocessed physical parameters and image data respectively. For physical parameters, the mean, variance, slope and peak value of temperature and pressure data are calculated as extracted features. For image data, the shape features, color features and texture features of the flame are obtained based on the flame region as extracted features.

[0013] The multimodal feature fusion module is composed of feature splicing and feature dimension reduction. The feature splicing part splices the extracted features of the physical parameters and the image data to form a joint feature vector. The feature dimension reduction part uses principal component analysis to reduce the dimension of the joint feature vector.

[0014] The temper state discrimination module adopts a random forest algorithm to construct a temper state discrimination model. Based on the reduced joint feature vector, the data of normal combustion and temper state are labeled and input as a data set for training and testing the model. The probability value of the occurrence of the temper state is output. The real-time collected joint feature vector is input into the temper state discrimination model. According to the set safety threshold, it is determined whether it belongs to normal combustion or temper state.

[0015] The sound-light alarm module is based on the output result of the temper state discrimination module. If it is determined to be a temper state, an alarm is given.

[0016] Further, the sensor system is synchronized with the data acquisition frequency of the OH-PLIF system to ensure the time consistency of the physical parameters and the image data.

[0017] Further, the sensor system is installed with a set of temperature sensors and pressure sensors on the inner wall of the fuel nozzle and the outlet of the fuel nozzle to monitor and collect the temperature and pressure data in real time during the hydrogen combustion process.

[0018] Further, the OH-PLIF system includes a laser, an optical system, a high-speed camera, and a timing control system. The timing control system ensures that the laser and the high-speed camera are synchronized through a digital signal generator, so that the laser beam emitted by the laser is vertically incident on the combustion chamber through the optical system, and the OH-PLIF image of the OH free radical distribution during the hydrogen combustion process is captured in real time.

[0019] A hydrogen combustion temper monitoring method based on multi-modal data fusion, comprising the following steps:

[0020] Step 1: Data acquisition

[0021] The sensor system and the OH-PLIF system are started simultaneously to collect physical parameters and image data during ignition.

[0022] Step 2: Data preprocessing

[0023] For the preprocessing of physical parameters, first, the discrete wavelet transform is selected to decompose the data into approximation coefficients and detail coefficients. The detail coefficients of each layer are threshold processed to remove noise components. The processed detail coefficients and approximation coefficients are inverse wavelet transformed to reconstruct the denoised temperature and pressure data. Then, the denoised temperature and pressure data are normalized respectively.

[0024] For the preprocessing of image data, first, the median filtering method is used for noise reduction processing, a fixed window of 3*3 size is selected to traverse each pixel of the image, the pixel values in the window are sorted, and the median value is taken to replace the current pixel value, and the operation is repeated until all pixels are processed, then the maximum between-class variance method is used for image segmentation on the image after noise reduction processing, the number of pixels n of each gray level of the image is counted i , the probability of each gray level is calculated , wherein N is the total number of pixels of each gray level, and the gray level k is taken as the threshold to divide the image into background area and flame area, and the probability mean of the background area and the flame area is calculated according to and , wherein L is the gray level range of the image, and the mean of the background area and the flame area is calculated according to , then the inter-class variance is calculated according to σ 2 (k) = ω0(k)·ω1(k)·(μ0(k)-μ1(k)) 2 , and the threshold k * that makes the inter-class variance maximum is found by traversing all possible thresholds k * , and the image is binarized using the threshold k * , the pixels with a gray level less than or equal to the threshold k * are taken as the background area, and the pixels with a gray level greater than the threshold k 1 are taken as the flame area.

[0025] Step three: feature extraction

[0026] For the feature extraction of physical parameters, the mean, variance, slope and peak value of temperature and pressure data are calculated as extraction features;

[0027] For the feature extraction of image data, the shape feature, color feature and texture feature of the flame in the hydrogen combustion process are extracted from the segmented flame area as extraction features, wherein:

[0028] The shape feature includes flame area, flame perimeter and flame circularity, the flame area calculation formula is , wherein I(i,j) represents the value of any pixel (i,j) inside the flame area, row 1 and column 1 represent the number of rows and columns inside the flame area respectively, the flame perimeter calculation formula is , wherein B(i,j) represents the value of any pixel (i,j) on the boundary of the flame area, row 2 and column 2 represent the number of rows and columns on the boundary of the flame area respectively, and the flame circularity calculation formula is , the value range is [0,1];

[0029] The color feature includes a mean gray value and a gray variance, and the mean gray value is calculated according to the formula The gray variance is calculated according to the formula In the formula, I(i) represents the gray value of the i-th pixel, and K represents the total number of pixels of the flame region in the image.

[0030] The texture feature describes the flame texture by extracting the gray value distribution of pixel pairs in the statistical image through a gray level co-occurrence matrix, selects the directions of 0°, 45°, 90° and 135° pixel pairs, and statistics the gray value distribution of pixel pairs satisfying the direction and distance conditions by traversing the flame region with a distance of 1 pixel, constructs the gray level co-occurrence matrix and normalizes the matrix, so that each element represents a probability, and the normalized gray level co-occurrence matrix The texture feature is calculated, and P(i,j) represents the number of times of occurrence of pixel pairs with gray values of i and j.

[0031] Step four: multi-modal feature fusion

[0032] For the feature splicing part, the feature vector of the physical parameter is represented as F temp-pressure =[f1,f2,...,f M ], the feature vector of the image data is represented as F OH-PLIF =[g1,g2,...,g K ], and the joint feature vector is F joint =[f1,f2,...,f M ,g1,g2,...,g K ], each feature in the joint feature vector is standardized so that the mean is 0 and the variance is 1, then the covariance matrix is calculated and eigenvalue decomposition is performed to obtain eigenvalues and eigenvectors, the eigenvalues are sorted in descending order, the corresponding eigenvectors are principal components, the principal components with a cumulative contribution rate of 95% are selected according to the size of the eigenvalues, and the standardized feature matrix is projected onto the selected principal components for dimension reduction.

[0033] Step five: temper state discrimination and acousto-optic alarm

[0034] The tempering state discrimination module adopts a random forest algorithm to construct a tempering state discrimination model, labels the combustion state of each sample as 1 representing a tempering state and 0 representing normal combustion based on the result of multi-modal feature fusion, divides the data set into a training set and a test set, sets the parameters of the random forest, uses the training set to train the tempering state discrimination model, uses the test set to evaluate the tempering state discrimination model, loads the trained tempering state discrimination model in the system, inputs the real-time collected joint feature vector, outputs the probability value of the occurrence of the tempering state, and determines whether it belongs to normal combustion or a tempering state according to the set safety threshold value, if the output probability value is greater than the safety threshold value, it is determined as a tempering state, and the sound and light alarm module alarms at the same time, otherwise it is determined as normal combustion.

[0035] Further, the step three of calculating the texture feature based on the normalized gray level co-occurrence matrix comprises:

[0036] The contrast, the calculation formula is

[0037] The energy, the calculation formula is

[0038] The homogeneity, the calculation formula is

[0039] The correlation, the calculation formula is In the formula, μ i And μ j Respectively represent the mean of the gray value of the pixel pair i and j, σ i And σ j Respectively represent the standard deviation of the gray value of the pixel pair i and j.

[0040] Compared with the prior art, the beneficial effects of the present application are: the present application uses a sensor system to collect the temperature and pressure of the combustion chamber as physical parameters, simultaneously captures the OH free radical distribution image in the hydrogen combustion process by an OH-PLIF system as image data, respectively extracts a plurality of features of the two kinds of physical parameters and image data, and fuses the multi-modal features to form a joint feature vector, and according to the tempering state discrimination model constructed by the random forest algorithm, the sound and light alarm module is triggered in real time in combination with the set safety threshold value, the real-time and accurate discrimination of the tempering phenomenon is realized, the accuracy of the hydrogen combustion tempering monitoring is improved, and reference is provided for the combustion parameter adjustment. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 It is a schematic diagram of the hydrogen combustion tempering monitoring system of the present application;

[0042] Figure 2 It is a schematic diagram of the data acquisition module in the present application. DETAILED DESCRIPTION

[0043] The technical solutions in the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort belong to the scope of protection of the present application.

[0044] A hydrogen combustion tempering monitoring system based on multi-modal data fusion, combining Figure 1 As shown in the figure, it comprises data acquisition module, data preprocessing module, feature extraction module, multi-modal feature fusion module, tempering state discrimination module and sound-light alarm module connected in sequence.

[0045] The data acquisition module is composed of a sensor system and an OH-PLIF system, combining Figure 2 As shown in the figure, the physical parameters and image data of the hydrogen combustion process are obtained by the two respectively. The sensor system is installed with a set of temperature sensors and pressure sensors at the inner wall of the fuel nozzle and the outlet of the fuel nozzle, for real-time monitoring and collecting the temperature and pressure data in the hydrogen combustion process as physical parameters; the OH-PLIF system includes a laser, an optical system, a high-speed camera and a timing control system, the timing control system ensures the synchronization of the laser and the high-speed camera through a digital signal generator, so that the laser beam emitted by the laser is vertically incident to the combustion chamber through the optical system, and the OH-PLIF image of the OH free radical distribution in the hydrogen combustion process is captured in real time as image data. It is worth noting that the data acquisition frequency of the sensor system and the OH-PLIF system should also be synchronized to ensure the time consistency of the physical parameters and the image data.

[0046] The data preprocessing module sets different preprocessing methods for the collected physical parameters and image data, and stores the preprocessed physical parameters and image data. For the physical parameters, discrete wavelet transform is used to remove noise and normalize the temperature and pressure data after removing noise to the range of [0, 1] to eliminate the dimensional gap; for the image data, median filtering is used for noise reduction, and then the maximum inter-class variance method is used to segment the image data according to the distribution characteristics of OH free radicals, and the background area and flame area of the image target position are segmented out.

[0047] The feature extraction module sets different extraction methods for the preprocessed physical parameters and image data. For the physical parameters, the mean, variance, slope and peak value of the temperature and pressure data are calculated respectively as the extracted features; for the image data, the shape feature, color feature and texture feature of the flame in the hydrogen combustion process are obtained as the extracted features according to the segmented flame area.

[0048] The multimodal feature fusion module is composed of feature splicing and feature dimension reduction, and the purpose is to obtain more abundant multimodal features to provide input data for the hydrogen combustion tempering state discrimination module. The feature splicing part splices the extracted features of the physical parameters and image data to form a joint feature vector; the feature dimension reduction part reduces the dimension of the joint feature vector by principal component analysis to reduce redundant information and improve the calculation efficiency of the model.

[0049] The tempering state discrimination module adopts a random forest algorithm to construct a tempering state discrimination model, labels the data of normal combustion and tempering state based on the joint feature vector after dimension reduction, and inputs the data set to train and test the model, and outputs the probability value of the tempering state occurrence. The joint feature vector collected in real time is input into the tempering state discrimination model, and whether it belongs to normal combustion or tempering state is determined according to the set safety threshold.

[0050] The sound-light alarm module generates a high-level signal based on the output result of the tempering state discrimination module, and transmits the signal to the sound-light alarm through a wired mode. The sound-light alarm generates an alarm after detecting the high-level signal.

[0051] It should be noted that the temperature and pressure data and OH-PLIF image collected by the data acquisition module in the system are transmitted to the computer by a transmission line, and the tempering state discrimination module and the sound-light alarm module are connected by a common cable. The data transmission between the remaining modules is completed in the computer.

[0052] As shown in Figures 1-2 A hydrogen combustion tempering monitoring method based on multimodal data fusion, comprising the following steps:

[0053] Step 1: Data acquisition

[0054] A set of temperature sensors and pressure sensors are installed on the inner wall of the fuel nozzle and the outlet of the fuel nozzle. The optical system is adjusted so that the laser beam emitted by the laser is vertically incident into the combustion chamber. When igniting, the sensor system and the high-speed camera are started to collect physical parameters and image data during hydrogen combustion, and the data is stored in the computer.

[0055] Step 2: Data preprocessing

[0056] For the preprocessing of physical parameters, the wavelet transform is used to remove noise for temperature and pressure data. The discrete wavelet transform is selected to decompose the data into approximation coefficients and detail coefficients. The detail coefficients of each layer are threshold processed to remove noise components. The processed detail coefficients and approximation coefficients are inverse wavelet transformed to reconstruct the denoised temperature and pressure data. Then, the denoised temperature and pressure data are respectively normalized according to Normalization is performed, where X is the original temperature or pressure data, X max is the maximum value in the temperature or pressure data, X min is the minimum value in the temperature or pressure data, and X norm is the normalized temperature or pressure data.

[0057] For the preprocessing of image data, first, a median filter method is used for noise reduction processing. A fixed window of size 3x3 is selected to traverse each pixel of the image, and the pixel values in the window are sorted, and the median value is used to replace the current pixel value. The operation is repeated until all pixels are processed. Then, the maximum inter-class variance method is used for image segmentation on the noise-reduced image. The number of pixels n i of each gray level of the image is counted, and the probability of each gray level is calculated. where N is the total number of pixels of each gray level, and the gray level k is set as the threshold to divide the image into a background region and a foreground region (flame region). The probability mean of the background region and the foreground region is calculated according to and respectively, where L is the gray level range of the image. The mean of the background region and the foreground region is then calculated according to and respectively. Then, the inter-class variance is calculated according to σ 2 (k) = ω0(k)·ω1(k)·(μ0(k)-μ1(k)) 2 , and the threshold k * that maximizes the inter-class variance is found by traversing all possible thresholds k. The image is binarized using the threshold k * , and the pixels with a gray level less than or equal to the threshold k * are set as the background region, and the pixels with a gray level greater than the threshold k * are set as the foreground region.

[0058] Step three: feature extraction

[0059] For the feature extraction of physical parameters, the mean, variance, slope, and peak value of the temperature and pressure data are calculated as extraction features, where:

[0060] The mean is calculated according to , which represents the central tendency of the data, the variance is calculated according to , which represents the volatility of the data, the slope is calculated according to , which represents the rising or falling speed of the data, and the peak value is calculated according to P = max(x i ), which represents the extreme case of the data, where x i represents the i-th data point, and M represents the total number of data points.

[0061] For feature extraction of image data, shape feature, color feature and texture feature of flame in hydrogen combustion process are extracted as extraction features according to the segmented flame region, wherein:

[0062] The shape feature mainly includes flame area, flame perimeter and flame circularity, and the calculation formula of the flame area is wherein, I(i,j) represents the value of any pixel (i,j) in the flame region, row 1 and column 1 represent the number of rows and columns in the flame region respectively. The calculation formula of the flame perimeter is wherein, B(i,j) represents the value of any pixel (i,j) on the boundary of the flame region, row 2 and column 2 represent the number of rows and columns on the boundary of the flame region respectively. The calculation formula of the flame circularity is wherein, the value of the flame circularity is an index of the degree of the flame region close to a circle in the image, and the value range is [0,1], and the value closer to 1 represents the closer to a circle;

[0063] The color feature mainly includes the mean gray value and the gray variance, and the statistics will be extracted from the color information of the image. The calculation formula of the mean gray value is wherein, the mean gray value is the average value of the color intensity of all pixels in the image. The calculation formula of the gray variance is wherein, the gray variance is the variance of the color intensity of all pixels in the flame region and reflects the fluctuation degree of the color, wherein I(i) represents the gray value of the i-th pixel, and K represents the total number of pixels in the flame region in the image;

[0064] The texture feature mainly describes the flame texture by extracting the gray value distribution of the pixel pairs in the image through the gray level co-occurrence matrix. The directions of the pixel pairs of 0°, 45°, 90° and 135° are selected, the distance is 1 pixel, the flame region is traversed, the gray value distribution of the pixel pairs satisfying the direction and distance conditions is counted, the gray level co-occurrence matrix is constructed, and the matrix is normalized through so that each element represents a probability, wherein P(i,j) represents the number of times of the pixel pairs with the gray values of i and j. The texture feature is calculated based on the normalized gray level co-occurrence matrix, including:

[0065] Contrast, the calculation formula is which reflects the local change degree of the image;

[0066] Energy, the calculation formula is which reflects the uniformity of the image;

[0067] Homogeneity, the calculation formula is which reflects the local smoothness of the image;

[0068] Correlation, calculated using the following formula: Reflecting the linear correlation of images, where μ i and μ j Let σ represent the mean gray values ​​of pixel pairs i and j, respectively. i and σ j These represent the standard deviations of the gray values ​​for pixel pairs i and j, respectively.

[0069] All extracted features from the above physical parameters and image data are stored as feature vectors for subsequent analysis.

[0070] Step 4: Multimodal Feature Fusion

[0071] The feature concatenation part refers to joining feature vectors from different data sources in a certain order to form a higher-dimensional joint feature vector. The feature vector of the physical parameters is represented as F. temp-pressure =[f1,f2,...,f M The feature vector of the image data is represented as F. OH-PLIF =[g1,g2,...,g K If the joint eigenvector is F, then the joint eigenvector is F. joint =[f1,f2,...,f M ,g1,g2,...,g K Based on the feature concatenation results, according to Each feature in the joint feature vector is standardized so that the mean is 0 and the variance is 1. Then, through... Calculate the covariance matrix, where Z represents the standardized eigenvalue matrix. Perform eigenvalue decomposition Zυ on the covariance matrix Σ. i =λ i υ i The eigenvalue λ is obtained. i and eigenvector υ i , eigenvalue λ i Sort by size from largest to smallest, the corresponding feature vector υ i Principal components are selected based on the eigenvalues, with a cumulative contribution rate of 95%. The standardized feature matrix Z is then projected onto the selected principal components for dimensionality reduction.

[0072] Step 5: Retardation Status Identification and Audible / Visual Alarm

[0073] The backfire state discrimination module mainly adopts a random forest algorithm to construct a backfire state discrimination model, labels the combustion state of each sample based on the above multi-modal feature fusion result, 1 represents the backfire state, 0 represents normal combustion, and divides the data set into a training set and a test set. The parameters of the random forest are set, including the number of trees, the maximum depth, etc., the training set is used to train the backfire state discrimination model, the backfire state discrimination model will construct multiple decision trees, and will be trained through randomly selected features and samples. After the backfire state discrimination model is constructed, the test set is used to evaluate the performance of the model. In order to perform real-time discrimination of the backfire state, the trained backfire state discrimination model is loaded in the system, and the joint feature vector collected in real time is input into the backfire state discrimination model. The backfire state discrimination model will output the probability value of the backfire state. According to the set safety threshold, it is judged whether it belongs to normal combustion or backfire state. If the output probability value is greater than the safety threshold, it is judged as backfire state, otherwise it is judged as normal combustion. If it is judged as backfire state, the controller of the sound and light alarm module outputs a high level signal to trigger the sound and light alarm, and the sound and light alarm emits a high frequency buzzing sound and warning light.

[0074] The present application innovatively combines physical parameters with image data, overcomes the limitations of a single data source, significantly improves the accuracy and robustness of backfire state discrimination, is suitable for real-time monitoring and early warning in complex combustion environments, and has important engineering application value.

[0075] It should be apparent to those skilled in the art that the application is not limited to the details of the above-described exemplary embodiments, but can be implemented in other embodiments without departing from the spirit or essential characteristics of the application. Therefore, the embodiments should be considered exemplary and non-limiting, and the scope of the application is defined by the appended claims, not the above description, and all changes falling within the meaning and scope of the equivalent conditions of the claims are intended to be included in the application. Any reference signs in the claims should not be considered as limiting the claims involved.

[0076] In addition, it should be understood that although the present specification is described in terms of embodiments, not every embodiment contains only one independent technical solution, and the description of the specification is only for the sake of clarity, and those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can be appropriately combined to form other embodiments that those skilled in the art can understand.

Claims

1. A hydrogen combustion tempering monitoring system based on multi-modal data fusion, characterized by: The system comprises sequentially connected data acquisition module, data preprocessing module, feature extraction module, multi-modal feature fusion module, tempering state discrimination module and acousto-optic alarm module. The data acquisition module is composed of a sensor system and an OH-PLIF system, which respectively acquire physical parameters and image data of the hydrogen combustion process, wherein the physical parameters include temperature and pressure data, and the image data is OH-PLIF image of OH radical distribution. The data preprocessing module respectively preprocesses the acquired physical parameters and image data, wherein for the physical parameters, discrete wavelet transform is used to remove noise and normalize the data, and for the image data, median filtering is used to remove noise and combine the maximum inter-class variance method to segment the background area and flame area of the image target position. The feature extraction module respectively extracts features from the preprocessed physical parameters and image data, wherein for the physical parameters, the mean, variance, slope and peak value of the temperature and pressure data are calculated as the extracted features, and for the image data, the shape feature, color feature and texture feature of the flame are obtained as the extracted features. The multi-modal feature fusion module is composed of feature concatenation and feature dimension reduction, wherein the feature concatenation part concatenates the extracted features of the physical parameters and image data to form a joint feature vector, and the feature dimension reduction part uses principal component analysis to reduce the dimension of the joint feature vector. The tempering state discrimination module uses a random forest algorithm to construct a tempering state discrimination model, based on the reduced joint feature vector, labels the data of normal combustion and tempering state as a data set and inputs them, trains and tests the model, and outputs the probability value of tempering state occurrence, inputs the real-time collected joint feature vector into the tempering state discrimination model, and determines whether it belongs to normal combustion or tempering state according to the set safety threshold. The acousto-optic alarm module alarms based on the output result of the tempering state discrimination module.

2. The hydrogen combustion tempering monitoring system based on multi-modal data fusion according to claim 1, characterized in that: The data acquisition frequency of the sensor system and the OH-PLIF system is synchronized to ensure the time consistency of the physical parameters and image data.

3. The hydrogen combustion tempering monitoring system based on multi-modal data fusion according to claim 2, characterized in that: The sensor system is installed with a set of temperature sensors and pressure sensors on the inner wall of the fuel nozzle and the outlet of the fuel nozzle to monitor and collect the temperature and pressure data in real time during the hydrogen combustion process.

4. The hydrogen combustion tempering monitoring system based on multi-modal data fusion of claim 2, wherein: The OH-PLIF system includes a laser, an optical system, a high-speed camera and a timing control system, the timing control system ensures synchronization of the laser and the high-speed camera through a digital signal generator, makes the laser beam emitted by the laser vertically incident to the combustion chamber through the optical system, and captures the OH-PLIF image of OH radical distribution in real time during the hydrogen combustion process.

5. A hydrogen combustion tempering monitoring method based on multi-modal data fusion, characterized by: The hydrogen combustion tempering monitoring system according to claim 1 has a monitoring method comprising the following steps: Step one: data acquisition Start the sensor system and the OH-PLIF system to collect physical parameters and image data at the same time when ignition; Step two: data preprocessing For the preprocessing of physical parameters, firstly, the discrete wavelet transform is selected to decompose the data into approximation coefficients and detail coefficients, the detail coefficients of each layer are thresholded to remove noise components, the processed detail coefficients and approximation coefficients are used for inverse wavelet transform to reconstruct the denoised temperature and pressure data, and then the denoised temperature and pressure data are normalized respectively; For the preprocessing of image data, first, the median filtering method is used for noise reduction processing, a fixed window of 3x3 size is selected to traverse each pixel of the image, the pixel values in the window are sorted, and the median value is replaced with the current pixel value. Repeat the operation until all pixels are processed, then use the maximum inter-class variance method for image segmentation on the noise-reduced image, count the number of pixels n of each gray level of the image i , calculate the probability of each gray level where N is the total number of pixels of each gray level, and the gray level k is the threshold value that divides the image into background and flame regions, respectively and calculate the mean probability of the background and flame regions, where L is the gray level range of the image, and then calculate the mean of the background and flame regions according to and respectively, then calculate the inter-class variance according to 2 (k) = ω0(k)·ω1(k)·(μ0(k)-μ1(k)) 2 , traverse all possible thresholds k to find the threshold k that maximizes the inter-class variance * , use the threshold k * to binarize the image, set the pixels with gray level less than or equal to the threshold k * as the background region, and set the pixels with gray level greater than the threshold k * as the flame region; Step three: feature extraction For the feature extraction of physical parameters, the mean, variance, slope and peak value of temperature and pressure data are calculated respectively as the extracted features; For the feature extraction of image data, the shape feature, color feature and texture feature of the flame in the hydrogen combustion process are extracted according to the segmented flame region as the extracted features, wherein: The shape features include a flame area, a flame perimeter and a flame circularity, the flame area is calculated by where I(i,j) represents a value of any pixel (i,j) inside the flame region, row 1 and column 1 respectively represent a row number and a column number inside the flame region, the flame perimeter is calculated by where B(i,j) represents a value of any pixel (i,j) on the boundary of the flame region, row 2 and column 2 respectively represent a row number and a column number on the boundary of the flame region, the flame circularity is calculated by with a value range of [0,1]. The color features include a gray mean value and a gray variance, and the gray mean value is calculated according to the formula The gray variance is calculated according to the formula In the formula, I(i) represents the gray value of the i-th pixel, and K represents the total number of pixels of the flame region in the image. The texture feature describes the flame texture by extracting the gray value distribution of pixel pairs in the statistical image through a gray level co-occurrence matrix, selects the directions of 0°, 45°, 90° and 135° pixel pairs, and counts the gray value distribution of pixel pairs satisfying the direction and distance conditions by traversing the flame region, constructs the gray level co-occurrence matrix and normalizes the matrix, so that each element represents a probability, and the normalized gray level co-occurrence matrix is used as the texture feature of the flame image The texture feature is calculated, and P(i,j) represents the number of times of occurrence of pixel pairs with gray values of i and j. Step four: multi-modal feature fusion For the feature splicing part, the feature vector of the physical parameter is represented as F temp-pressure = [f1, f2,..., f M ] and the feature vector of the image data is represented as F OH-PLIF = [g1, g2,..., g K ], then the joint feature vector is F joint = [f1, f2,..., f M , g1, g2,..., g K ], each feature in the joint feature vector is standardized so that the mean is 0 and the variance is 1, then the covariance matrix is calculated and eigenvalue decomposition is performed to obtain eigenvalues and eigenvectors, the eigenvalues are sorted in descending order, the corresponding eigenvectors are principal components, the principal components with cumulative contribution rate of 95% are selected according to the size of the eigenvalues, and the standardized feature matrix is projected onto the selected principal components for dimension reduction; Step five: tempering state discrimination and sound-light alarm The tempering state discrimination module adopts the random forest algorithm to construct the tempering state discrimination model, labels the combustion state of each sample based on the results of multi-modal feature fusion, 1 represents the tempering state and 0 represents the normal combustion, divides the data set into training set and test set, sets the parameters of random forest, trains the tempering state discrimination model using the training set, evaluates the tempering state discrimination model using the test set, loads the trained tempering state discrimination model in the system, inputs the real-time collected joint feature vector, outputs the probability value of tempering state occurrence, and judges whether it belongs to normal combustion or tempering state according to the set safety threshold, if the output probability value is greater than the safety threshold, it is judged as tempering state, and the sound-light alarm module alarms at the same time, otherwise it is judged as normal combustion.

6. The hydrogen combustion tempering monitoring method based on multi-modal data fusion according to claim 5, characterized in that: The texture feature calculation based on normalized gray level co-occurrence matrix in step three comprises: Contrast, calculated as Energy, calculated as Homogeneity, calculated as Correlation, the calculation formula is In the formula, μ i and μ j respectively represent the mean of the gray value of the pixel pair i and j, σ i and σ j respectively represent the standard deviation of the gray value of the pixel pair i and j.

Images