A method for regulating temperature field of combustion chamber outlet and multi-path fuel injection rod

Abstract

The embodiment of the application provides a kind of combustion chamber outlet temperature field's regulation and control method and multi-path fuel injection rod, it is related to gas turbine aero-engine technical field, wherein, including the following steps: obtaining engine operating condition state parameter and the oil supply flow instruction of multi-path fuel, engine operating condition state parameter and the flow instruction of multi-path fuel are normalized, model input feature is generated;The deep learning prediction model of layered-fusion network architecture is constructed, the predicted combustion chamber outlet temperature field performance parameter is obtained by model, and the objective function for evaluating temperature field quality is constructed;The flow distribution ratio of multi-path fuel is used as decision variable, the fuel flow ratio of each path is determined by objective function, and the optimal fuel distribution instruction set is generated, to drive actuator to independently adjust the fuel flow of precombustion stage nozzle region and each main combustion stage nozzle sub-region oil supply oil line respectively.The quality of combustion chamber outlet temperature field is improved by the regulation and control method of the application.

Description

Technical Field

[0001] This invention relates to the field of gas turbine aero-engine technology, and in particular to a method for controlling the temperature field at the combustion chamber outlet and a multi-path fuel injector. Background Technology

[0002] High-temperature combustion chamber is one of the important technical features of the new generation of engines. It has a high total fuel-air ratio and heat release rate. Due to the heat resistance of turbine materials, higher requirements are placed on the temperature distribution coefficient of the combustion chamber outlet.

[0003] like Figure 2 As shown, traditional combustion chambers organize combustion through the head and main combustion orifice jets, with the mixed gas mainly playing a role in regulating the outlet temperature field. However, compared with traditional combustion chambers, high-temperature rise combustion chambers have a higher fuel-air ratio, resulting in a larger amount of combustion air and a smaller amount of available mixed gas. Regulating the outlet temperature distribution of the combustion chamber through mixed gas presents a significant technical challenge.

[0004] Currently, common high-temperature combustion chambers typically employ a dual fuel supply system, with a main fuel line supplying multiple main combustion stage nozzles simultaneously, while the secondary fuel line supplies a pre-combustion stage nozzle located in the center. The main and pre-combustion stage nozzles are arranged in a concentric circle pattern. (See...) Figure 3 As shown, although the total fuel supply to the combustion chamber varies at different states such as ground takeoff point and cruise, the fuel mist distribution characteristics in the downstream annular region of the combustion chamber are similar due to the influence of the combustion chamber flow characteristics and the similar fuel supply characteristics of each nozzle in the main combustion stage. The outlet temperature field is mainly regulated by the mixed airflow determined by the geometric flow path of the combustion chamber, which is a passive regulation. Summary of the Invention

[0005] In view of this, embodiments of this specification provide a method for controlling the temperature field at the combustion chamber outlet and a multi-channel fuel injection boom.

[0006] The embodiments in this specification provide the following technical solutions:

[0007] A method for controlling the temperature field at the combustion chamber outlet, comprising:

[0008] The engine operating condition parameters and the fuel supply flow commands of multiple fuel lines are obtained. The engine operating condition parameters and the fuel supply flow commands of multiple fuel lines are normalized to generate model input features. The fuel supply flow commands of multiple fuel lines include the fuel supply flow commands of the pre-combustion stage nozzle area and the fuel supply flow commands of the main combustion stage nozzle sub-areas of multiple main combustion stage nozzle sub-areas.

[0009] A deep learning prediction model with a hierarchical-fusion network architecture is constructed. The input features of the model are input into the trained deep learning prediction model to obtain the predicted combustion chamber outlet temperature field performance parameters. Based on the predicted combustion chamber outlet temperature field performance parameters, an objective function for evaluating the quality of the temperature field is constructed. The combustion chamber outlet temperature field performance parameters include the outlet temperature distribution coefficient and the radial temperature distribution coefficient.

[0010] The flow distribution ratio of multiple fuel lines is used as a decision variable. The Bayesian optimization algorithm is used for iterative search. The flow ratio of each fuel line is determined by the objective function, generating the optimal fuel distribution instruction set. The optimal fuel distribution instruction set is then converted into control signals for the metering valves of the multiple fuel lines, driving the actuators to independently adjust the fuel flow of the fuel supply lines in the pre-combustion stage nozzle area and the fuel supply lines in each main combustion stage nozzle sub-area.

[0011] Furthermore, engine operating condition parameters and multi-path fuel flow commands are acquired, and these parameters and commands are normalized to construct model input features, including:

[0012] A fuel supply tensor is constructed based on the pre-combustion stage fuel supply flow command and the main combustion stage fuel supply flow command. The first dimension of the fuel supply tensor is used to store the path number of all fuel supply lines, and the second dimension is used to store the normalized fuel supply flow command corresponding to each path number.

[0013] The engine operating condition parameters are divided into combustion chamber inlet flow field parameters that affect the combustion chamber inlet flow field and engine power setting parameters that characterize the engine power setting state. The combustion chamber inlet flow field parameters and engine power setting parameters are normalized respectively.

[0014] The normalized combustion chamber inlet flow field parameters, normalized engine power setting parameters, and fuel supply tensor are combined to generate model input features.

[0015] Furthermore, a deep learning prediction model with a hierarchical-fusion network architecture is constructed, including:

[0016] The deep learning prediction model consists of interconnected partition feature extraction layers, radial spatial interaction layers, and a global combustion process layer;

[0017] The partition feature extraction layer contains multiple sub-networks that share parameters. Each sub-network receives the fuel supply tensor of each fuel supply path and each sub-network independently extracts and outputs the state feature vector of the fuel for each fuel supply path. The number of sub-networks is the same as the number of fuel supply paths.

[0018] The relative radial positions of each main combustion stage nozzle sub-region in the pre-combustion stage nozzle region and the main combustion stage nozzle region on the fuel injector are obtained. The multiple input state feature vectors are arranged according to their relative positions through a radial spatial interaction layer to generate a feature sequence. The feature sequence is then processed by a convolutional neural network to output a spatial fusion feature that integrates the radial mixing pattern.

[0019] By fusing spatial fusion features, normalized combustion chamber inlet flow field parameters, and normalized engine power setting parameters through a global combustion process layer, and mapping through a fully connected network, a two-dimensional temperature field matrix is ​​output to characterize the radial-circumferential temperature field distribution of the predicted combustion chamber outlet section.

[0020] Furthermore, the feature sequences are processed using a convolutional neural network to output spatial fusion features that incorporate radial mixing patterns, including:

[0021] The feature sequence is input into a convolutional neural network, and the feature sequence is subjected to sliding window convolution operation through the convolutional layer of the convolutional neural network to extract the spatial correlation and interaction between adjacent and cross-regional fuel state features.

[0022] The output after the convolution operation is nonlinearly activated and feature compressed to generate a spatial fusion feature vector. The spatial fusion feature vector is used to characterize the overall state of the multi-path fuel mixture in the radial space of the combustion chamber head.

[0023] Furthermore, the spatial fusion features, normalized combustion chamber inlet flow field parameters, and normalized engine power setting parameters are fused through a global combustion process layer. This is then mapped via a fully connected network to output a two-dimensional temperature field matrix characterizing the predicted radial-circumferential temperature field distribution at the combustion chamber outlet cross-section, including:

[0024] The spatial fusion features, normalized combustion chamber inlet flow field parameters, and normalized engine power setting parameters are spliced ​​together to generate a comprehensive input vector;

[0025] The integrated input vector is input into a subnetwork containing at least two fully connected layers, and the output is a reconstructed two-dimensional matrix. The length of the reconstructed two-dimensional matrix is ​​the number of grid points in the radial direction of the predicted temperature field, and the width of the reconstructed two-dimensional matrix is ​​the number of grid points in the circumferential direction of the predicted temperature field.

[0026] The reconstructed two-dimensional matrix is ​​normalized to generate a two-dimensional temperature field matrix that characterizes the radial-circumferential temperature field distribution of the combustion chamber outlet section.

[0027] Furthermore, it also includes:

[0028] The composite loss function L for constructing a deep learning prediction model includes:

[0029] The predicted outlet temperature distribution coefficient OTDF_pred and the predicted radial temperature distribution coefficient RTDF_pred are calculated using the two-dimensional temperature field matrix.

[0030] The mean square error of the outlet temperature distribution coefficient is calculated using the predicted outlet temperature distribution coefficient OTDF_pred.

[0031] The mean square error of the radial temperature distribution coefficient is calculated using the predicted radial temperature distribution coefficient RTDF_pred.

[0032] The sum of the mean square error of the outlet temperature distribution coefficient and the mean square error of the radial temperature distribution coefficient is used as the coefficient loss term L_coeff.

[0033] The deep learning prediction model is trained using a composite loss function L, where L = α × L_dist + β × L_coeff, and L_dist is the error between the two-dimensional temperature field matrix predicted by the deep learning prediction model and the actual temperature field distribution matrix. α and β are both preset weight coefficients.

[0034] Furthermore, based on the predicted combustion chamber outlet temperature field performance parameters, an objective function for evaluating the quality of the temperature field is constructed, including:

[0035] The objective function is J = -[λ1×OTDF_pred+λ2×|RTDF_pred-RTDF_target|], where OTDF_pred is the predicted outlet temperature distribution coefficient, RTDF_pred is the predicted radial temperature distribution coefficient, RTDF_target is the preset target value of the ideal radial temperature distribution coefficient, and λ1 and λ2 are preset weighting coefficients.

[0036] Furthermore, the flow distribution ratio of multiple fuel lines is used as a decision variable. An iterative search is performed using a Bayesian optimization algorithm to determine the flow ratio of each fuel line through an objective function, generating an optimal fuel distribution instruction set, including:

[0037] Within the feasible region that satisfies the total fuel flow constraint, an initial fuel allocation scheme is generated. The trained deep learning prediction model is used to obtain the prediction performance parameters corresponding to the initial fuel allocation scheme, and the initial objective function value is calculated to generate an initial sample set.

[0038] Based on the fuel allocation schemes in the sample set and the corresponding initial objective function values, a Gaussian process regression model is constructed.

[0039] Using a Gaussian process regression model, the initial sample set is iterated to generate the final sample set, and the parameters of the Gaussian process regression model are updated.

[0040] Repeat the iteration until the preset iteration termination condition is met, and select the fuel allocation ratio scheme with the optimal objective function value from the final sample set as the optimal fuel allocation instruction set.

[0041] Furthermore, it also includes:

[0042] After executing the optimal fuel distribution command set to adjust the fuel supply circuit, the actual temperature field data at the combustion chamber outlet is collected. Actual performance parameters are calculated from this data, and the deep learning prediction model is updated iteratively using these parameters, including:

[0043] After executing the optimal fuel distribution command set, the actual temperature field distribution data at the combustion chamber outlet is collected;

[0044] Based on the actual temperature field distribution data, the actual outlet temperature distribution coefficient and the actual radial temperature distribution coefficient are calculated.

[0045] The engine operating condition parameters, the set of optimal fuel distribution commands executed, the actual outlet temperature distribution coefficient, and the actual radial temperature distribution coefficient are combined to generate new training samples with labels.

[0046] Add new training samples to the historical training dataset of the deep learning prediction model;

[0047] The deep learning prediction model is incrementally trained using a historical training dataset containing new training samples, and the parameters of the deep learning prediction model are adjusted.

[0048] A multi-path fuel injector, based on which a method for controlling the combustion chamber outlet temperature field of an aero-engine is used to perform zoned control of the combustion chamber outlet temperature field, the multi-path fuel injector comprising:

[0049] The pre-combustion nozzle area is located at the center of the fuel injector rod, and the main combustion nozzle area is located on both sides of the pre-combustion nozzle area.

[0050] The pre-combustion stage nozzle area includes multiple arrayed pre-combustion stage nozzles, and the main combustion stage nozzle area is divided into multiple main combustion stage nozzle sub-areas along the radial direction of the injection rod. Each main combustion stage nozzle sub-area includes multiple arrayed main combustion stage nozzles.

[0051] The pre-combustion stage nozzle area and each main combustion stage nozzle sub-area are supplied with fuel through independent fuel supply lines.

[0052] Compared with the prior art, the beneficial effects that at least one technical solution adopted in the embodiments of this specification can achieve include at least:

[0053] The combustion chamber outlet temperature field control method of this invention improves the quality of the combustion chamber outlet temperature field. Attached Figure Description

[0054] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 This is a flowchart of a method for controlling the temperature field at the combustion chamber outlet provided in an embodiment of the present invention;

[0056] Figure 2 This is a schematic diagram of the working principle of a traditional combustion chamber;

[0057] Figure 3 This is a schematic diagram of the nozzle for fuel supply to the main pre-combustion stage of the existing high-temperature combustion chamber;

[0058] Figure 4 This is a three-dimensional schematic diagram of the overall structure of the multi-channel fuel injector according to an embodiment of the present invention;

[0059] Figure 5 This is a schematic diagram of the pre-combustion stage nozzle area and the main combustion stage nozzle area according to an embodiment of the present invention;

[0060] Figure 6 This is a schematic diagram illustrating the principle of the combustion chamber outlet temperature field control method based on a multi-path fuel injection boom according to an embodiment of the present invention.

[0061] The attached diagram is labeled as follows: 1. Pre-combustion stage nozzle area; 11. Pre-combustion stage nozzle; 2. Main combustion stage nozzle area; 21. Main combustion stage nozzle sub-area; 211. Main combustion stage nozzle. Detailed Implementation

[0062] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0063] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. This application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0064] This invention proposes an active control scheme for the outlet temperature field based on multi-path fuel zone control at the head. Through the multi-path fuel zone structure, each fuel zone independently controls the fuel supply, forming a relatively independent and actively adjustable array-type combustion zone, improving the distribution of local high-temperature hot spots in combustion, and realizing active control of the combustion chamber outlet temperature field.

[0065] like Figure 1 and Figure 6 As shown, the method includes:

[0066] Step S101: Obtain engine operating condition parameters and multi-path fuel supply flow instructions, normalize the engine operating condition parameters and multi-path fuel flow instructions, and generate model input features. The multi-path fuel supply flow instructions include the pre-combustion stage fuel supply flow instructions of the fuel supply path of the pre-combustion stage nozzle region 1 and the main combustion stage fuel supply flow instructions of the fuel supply paths of multiple main combustion stage nozzle sub-regions 21 of the main combustion stage nozzle region 2.

[0067] Step S102: Construct a deep learning prediction model with a hierarchical-fusion network architecture, input the model input features into the trained deep learning prediction model, obtain the predicted combustion chamber outlet temperature field performance parameters, and construct an objective function to evaluate the quality of the temperature field through the predicted combustion chamber outlet temperature field performance parameters. The combustion chamber outlet temperature field performance parameters include the outlet temperature distribution coefficient and the radial temperature distribution coefficient.

[0068] Step S103: Using the flow distribution ratio of the multi-path fuel as a decision variable, perform iterative search through the Bayesian optimization algorithm, determine the flow ratio of each path fuel through the objective function, generate the optimal fuel distribution instruction set, and convert the optimal fuel distribution instruction set into control signals for the metering valves of the multi-path fuel, driving the actuator to independently adjust the fuel flow of the fuel supply path of the pre-combustion stage nozzle area 1 and the fuel supply path of each main combustion stage nozzle sub-area 21.

[0069] Based on the radially similar distribution of the downstream flow field in the jet-stabilized flame combustion configuration, multi-path fuel zoning and independent control of fuel supply in each path can form a relatively independent and actively adjustable array-type combustion zone. By adjusting the local oil mist distribution at different radial positions, the distribution of high-temperature hot spots in combustion can be optimized, thereby achieving active control of the combustion chamber outlet temperature field.

[0070] Based on deep learning and other methods, this approach aims to improve the quality of the combustion chamber outlet temperature field. It collaboratively optimizes the fuel distribution patterns in each zone, establishing a correlation between fuel distribution parameters and combustion chamber outlet temperature field performance parameters under different engine conditions. This yields the optimal fuel distribution ratio between different zones, allowing for fine-tuning of fuel mist distribution and improved combustion chamber outlet temperature field quality. Through active regulation of the outlet temperature field based on multi-path fuel distribution control at the head, high outlet temperature field quality is achieved under various engine conditions.

[0071] In practice, the number of the full-ring combustion chamber head can be adjusted according to the number of high-pressure turbine guide vanes to achieve a mapping match between one combustion chamber head and several high-pressure turbine guide vanes, thereby improving the quality of the combustion chamber outlet temperature field, increasing the reliability of turbine components, and improving the overall performance of the engine.

[0072] The first dimension of the fuel supply tensor stores the path numbers of all fuel supply circuits. For example, the pre-combustion stage fuel circuit is numbered 1, and the fuel circuits of each main combustion stage zone are numbered 2, 3, 4, etc. The second dimension stores the normalized fuel flow command corresponding to each path number. The normalization process specifically involves dividing the real-time flow command of each fuel circuit by its respective design point reference flow value.

[0073] The engine operating parameters are divided into combustion chamber inlet flow field parameters affecting the combustion chamber inlet flow field and engine power setting parameters characterizing the engine power setting state. The combustion chamber inlet flow field parameters include at least the total pressure and total temperature at the combustion chamber inlet. The engine power setting parameters include at least flight altitude, Mach number, and engine core speed. Both the combustion chamber inlet flow field parameters and the engine power setting parameters are then normalized by dividing each parameter by its design point reference value.

[0074] The normalized combustion chamber inlet flow field parameters, normalized engine power setting parameters, and fuel supply tensor are combined to generate the model input features. These features are a unified data structure that integrates the current engine state and the fuel distribution scheme.

[0075] Specifically, the engine operating condition parameters and multi-path fuel flow commands are acquired, and these parameters and commands are normalized to construct model input features, including:

[0076] A fuel supply tensor is constructed based on the pre-combustion stage fuel supply flow command and the main combustion stage fuel supply flow command. The first dimension of the fuel supply tensor is used to store the path number of all fuel supply lines, and the second dimension is used to store the normalized fuel supply flow command corresponding to each path number.

[0077] The engine operating condition parameters are divided into combustion chamber inlet flow field parameters that affect the combustion chamber inlet flow field and engine power setting parameters that characterize the engine power setting state. The combustion chamber inlet flow field parameters and engine power setting parameters are normalized respectively.

[0078] The normalized combustion chamber inlet flow field parameters, normalized engine power setting parameters, and fuel supply tensor are combined to generate model input features.

[0079] The partitioned feature extraction layer contains multiple sub-networks that share parameters, with the number of sub-networks being the same as the number of fuel supply paths. Each sub-network receives data corresponding to a path number from the fuel supply tensor and performs feature extraction operations independently, outputting a state feature vector that represents the initial state of the fuel in the corresponding fuel supply path (such as atomization and local equivalence ratio trends).

[0080] The relative radial positions (e.g., normalized coordinates with the center of the injection rod) of each main combustion stage nozzle sub-region 21 in the pre-combustion stage nozzle region 1 and the main combustion stage nozzle region 2 are obtained on the injection rod. A radial spatial interaction layer arranges the multiple input state feature vectors according to their radial position order in their corresponding fuel path regions (e.g., from the center of the pre-combustion stage upwards to the main combustion stage, the middle main combustion stage, and the lower main combustion stage), generating an ordered feature sequence. This sequence is processed by a one-dimensional convolutional neural network, which extracts the spatial correlation and interaction patterns between adjacent and cross-region features through a sliding window operation of the convolutional kernel, outputting a spatially fused feature vector that incorporates the mixing patterns of multiple fuel paths in the radial space.

[0081] The spatially fused feature vector, the normalized combustor inlet flow field parameter vector, and the normalized engine power setting parameter vector are concatenated and fused through a global combustion process layer to form a comprehensive feature. This comprehensive feature is then nonlinearly transformed and mapped through a subnetwork consisting of at least two fully connected layers, ultimately outputting a two-dimensional matrix. The number of rows and columns of this two-dimensional matrix corresponds to the number of discrete grid points in the radial and circumferential directions of the predicted combustor outlet section, respectively. Each element value in the matrix represents the predicted relative temperature at the corresponding grid position, thus characterizing the predicted radial-circumferential temperature field distribution of the combustor outlet section.

[0082] Specifically, constructing a deep learning prediction model with a layered-fusion network architecture includes:

[0083] The deep learning prediction model consists of interconnected partition feature extraction layers, radial spatial interaction layers, and a global combustion process layer;

[0084] The partition feature extraction layer contains multiple sub-networks that share parameters. Each sub-network receives the fuel supply tensor of each fuel supply path and each sub-network independently extracts and outputs the state feature vector of the fuel for each fuel supply path. The number of sub-networks is the same as the number of fuel supply paths.

[0085] The relative positions of each main combustion stage nozzle sub-region 21 in the pre-combustion stage nozzle region 1 and the main combustion stage nozzle region 2 in the radial direction of the fuel injection rod are obtained. Multiple input state feature vectors are arranged according to their relative positions through a radial spatial interaction layer to generate a feature sequence. The feature sequence is then processed by a convolutional neural network to output a spatial fusion feature that integrates the radial mixing pattern.

[0086] By fusing spatial fusion features, normalized combustion chamber inlet flow field parameters, and normalized engine power setting parameters through a global combustion process layer, and mapping through a fully connected network, a two-dimensional temperature field matrix is ​​output to characterize the radial-circumferential temperature field distribution of the predicted combustion chamber outlet section.

[0087] A feature sequence arranged radially is fed into a one-dimensional convolutional neural network. The network contains at least one convolutional layer, each configured with multiple one-dimensional convolutional kernels. These kernels slide along the feature sequence, performing weighted summations and non-linear activations (such as ReLU) on features within local regions to extract local spatial relationships between adjacent fuel zone state features, and to capture cross-region interactions by stacking convolutional layers or using larger kernels.

[0088] The output of the convolutional layer is further subjected to nonlinear activation and feature compression operations (e.g., through pooling layers or additional fully connected layers) to generate a spatially fused feature vector with reduced dimensionality and concentrated information. This vector no longer retains the independent information of each partition, but comprehensively represents the overall state of multiple fuel streams after mixing and interaction in the radial space of the combustion chamber head, serving as the input for subsequent global combustion process layers.

[0089] Specifically, the feature sequences are processed using a convolutional neural network to output spatial fusion features that incorporate radial mixing patterns, including:

[0090] The feature sequence is input into a convolutional neural network, and the feature sequence is subjected to sliding window convolution operation through the convolutional layer of the convolutional neural network to extract the spatial correlation and interaction between adjacent and cross-regional fuel state features.

[0091] The output after the convolution operation is nonlinearly activated and feature compressed to generate a spatial fusion feature vector. The spatial fusion feature vector is used to characterize the overall state of the multi-path fuel mixture in the radial space of the combustion chamber head.

[0092] The spatial fusion feature vector, the normalized combustion chamber inlet flow field parameter vector (such as inlet pressure and temperature), and the normalized engine power setting parameter vector (such as flight Mach number and speed) are spliced ​​together to form a higher-dimensional comprehensive input vector.

[0093] The synthesized input vector is fed into a subnetwork consisting of at least two fully connected layers. Each fully connected layer performs a linear transformation on its input and is typically connected to a nonlinear activation function (e.g., ReLU) to introduce nonlinear modeling capabilities.

[0094] The number of neurons in the last fully connected layer is set to be equal to the total number of elements in the desired output temperature field 2D matrix, i.e., H (radial grid number) × W (circumferential grid number). The output vector of this layer is then transformed into a 2D matrix of H rows and W columns through a reshaping operation.

[0095] The reshaped two-dimensional matrix is ​​then normalized (by using the Sigmoid function to map its element values ​​to the [0,1] interval, or by performing min-max normalization) so that its numerical range corresponds to the predicted relative temperature distribution, thereby generating the final two-dimensional temperature field matrix used to characterize the predicted radial-circumferential temperature field distribution at the combustion chamber outlet section.

[0096] Specifically, the spatial fusion features, normalized combustion chamber inlet flow field parameters, and normalized engine power setting parameters are fused through a global combustion process layer. This is then mapped via a fully connected network to output a two-dimensional temperature field matrix characterizing the predicted radial-circumferential temperature field distribution at the combustion chamber outlet cross-section, including:

[0097] The spatial fusion features, normalized combustion chamber inlet flow field parameters, and normalized engine power setting parameters are spliced ​​together to generate a comprehensive input vector;

[0098] The integrated input vector is input into a subnetwork containing at least two fully connected layers, and the output is a reconstructed two-dimensional matrix. The length of the reconstructed two-dimensional matrix is ​​the number of grid points in the radial direction of the predicted temperature field, and the width of the reconstructed two-dimensional matrix is ​​the number of grid points in the circumferential direction of the predicted temperature field.

[0099] The reconstructed two-dimensional matrix is ​​normalized to generate a two-dimensional temperature field matrix that characterizes the radial-circumferential temperature field distribution at the combustion chamber outlet section.

[0100] Calculate the average value of all elements in the two-dimensional moment of the temperature field to obtain T_avg_pred;

[0101] Find the maximum value of all elements in the matrix to obtain T_max_pred;

[0102] Calculate the average value of the matrix row by row (radial) to obtain the radial average temperature vector, and find the maximum value of the vector to obtain T_radial_max_pred.

[0103] Using the known combustion chamber inlet temperature T_inlet, calculations are performed according to the formulas OTDF_pred = (T_max_pred - T_avg_pred) / (T_avg_pred - T_inlet) and RTDF_pred = (T_radial_max_pred - T_avg_pred) / (T_avg_pred - T_inlet). This calculation process is implemented through a differentiable operation module to ensure that the gradient can propagate backward.

[0104] The mean square error of the outlet temperature distribution coefficient, MSE_OTDF, is calculated by matching the predicted outlet temperature distribution coefficient OTDF_pred with its corresponding true label value OTDF_true. MSE_OTDF = (OTDF_pred - OTDF_true) 2 .

[0105] The mean square error of the radial temperature distribution coefficient, MSE_RTDF, is calculated by comparing the predicted radial temperature distribution coefficient RTDF_pred with its corresponding true label value RTDF_true. The result is: MSE_RTDF = (RTDF_pred - RTDF_true). 2 .

[0106] Adding the mean square error of the outlet temperature distribution coefficient to the mean square error of the radial temperature distribution coefficient yields the coefficient loss term, i.e., L_coeff = MSE_OTDF + MSE_RTDF.

[0107] The deep learning prediction model is trained using a composite loss function L, where L = α × L_dist + β × L_coeff. L_dist represents the error (e.g., mean square error) between the two-dimensional temperature field matrix predicted by the deep learning prediction model and the actual temperature field distribution matrix (from simulation or experiment). α and β are preset weighting coefficients used to balance the importance of distribution accuracy and index accuracy during training.

[0108] Specifically, it also includes:

[0109] The composite loss function L for constructing a deep learning prediction model includes:

[0110] The predicted outlet temperature distribution coefficient OTDF_pred and the predicted radial temperature distribution coefficient RTDF_pred are calculated using the two-dimensional temperature field matrix.

[0111] The mean square error of the outlet temperature distribution coefficient is calculated using the predicted outlet temperature distribution coefficient OTDF_pred. The average value of all elements in the two-dimensional moment of the temperature field is then calculated to obtain T_avg_pred.

[0112] The mean square error of the radial temperature distribution coefficient is calculated using the predicted radial temperature distribution coefficient RTDF_pred.

[0113] The sum of the mean square error of the outlet temperature distribution coefficient and the mean square error of the radial temperature distribution coefficient is used as the coefficient loss term L_coeff.

[0114] The deep learning prediction model is trained using a composite loss function L, where L = α × L_dist + β × L_coeff, and L_dist is the error between the two-dimensional temperature field matrix predicted by the deep learning prediction model and the actual temperature field distribution matrix. α and β are both preset weight coefficients.

[0115] Specifically, by predicting the performance parameters of the combustion chamber outlet temperature field, an objective function for evaluating the quality of the temperature field is constructed, including:

[0116] The objective function is J = -[λ1×OTDF_pred+λ2×|RTDF_pred-RTDF_target|], where OTDF_pred is the predicted outlet temperature distribution coefficient, RTDF_pred is the predicted radial temperature distribution coefficient, RTDF_target is the preset target value of the ideal radial temperature distribution coefficient, and λ1 and λ2 are preset weighting coefficients.

[0117] Specifically, the flow distribution ratio of multiple fuel lines is used as a decision variable. A Bayesian optimization algorithm is used for iterative search to determine the flow ratio of each fuel line through an objective function, generating an optimal fuel distribution instruction set, including:

[0118] Within the feasible region that satisfies the total fuel flow constraint, an initial fuel allocation scheme is generated. The trained deep learning prediction model is used to obtain the prediction performance parameters corresponding to the initial fuel allocation scheme, and the initial objective function value is calculated to generate an initial sample set.

[0119] Based on the fuel allocation schemes in the sample set and the corresponding initial objective function values, a Gaussian process regression model is constructed.

[0120] Using a Gaussian process regression model, the initial sample set is iterated to generate the final sample set, and the parameters of the Gaussian process regression model are updated.

[0121] Repeat the iteration until the preset iteration termination condition is met, and select the fuel allocation ratio scheme with the optimal objective function value from the final sample set as the optimal fuel allocation instruction set.

[0122] Specifically, it also includes:

[0123] After executing the optimal fuel distribution command set to adjust the fuel supply circuit, the actual temperature field data at the combustion chamber outlet is collected. Actual performance parameters are calculated from this data, and the deep learning prediction model is updated iteratively using these parameters, including:

[0124] After executing the optimal fuel distribution command set, the actual temperature field distribution data at the combustion chamber outlet is collected;

[0125] Based on the actual temperature field distribution data, the actual outlet temperature distribution coefficient and the actual radial temperature distribution coefficient are calculated.

[0126] The engine operating condition parameters, the set of optimal fuel distribution commands executed, the actual outlet temperature distribution coefficient, and the actual radial temperature distribution coefficient are combined to generate new training samples with labels.

[0127] Add new training samples to the historical training dataset of the deep learning prediction model;

[0128] The deep learning prediction model is incrementally trained using a historical training dataset containing new training samples, and the parameters of the deep learning prediction model are adjusted.

[0129] After executing the optimal fuel distribution command set and waiting for the combustion chamber to stabilize, the actual temperature field distribution data is collected by an array of temperature sensors (such as thermocouple rakes) arranged at the combustion chamber outlet section.

[0130] Based on actual temperature field distribution data, the actual outlet temperature distribution coefficient OTDF_act and the actual radial temperature distribution coefficient RTDF_act are calculated according to the definitions of OTDF and RTDF. The engine operating condition parameters that triggered this regulation, the actual executed optimal fuel distribution command set, and the calculated OTDF_act and RTDF_act are combined to generate a new training sample containing input features and output labels. This new training sample is then added to the historical training dataset of the deep learning prediction model.

[0131] Periodically (after a certain number of tuning cycles) or when a certain number of new samples have been accumulated, an incremental training or fine-tuning round is performed on the deep learning prediction model using an updated historical training dataset containing new training samples. This is to adjust the network parameters of the deep learning prediction model so that the model's predictive ability can adapt to the slow changes in engine performance or the drift of measurement system characteristics.

[0132] like Figure 4 and Figure 5 As shown, the multi-channel fuel injector includes:

[0133] The pre-combustion stage nozzle area 1 and the main combustion stage nozzle area 2 are located at the center of the fuel injection rod, and the main combustion stage nozzle area 2 is located on both sides of the pre-combustion stage nozzle area 1.

[0134] The pre-combustion stage nozzle region 1 includes multiple arrayed pre-combustion stage nozzles 11, and the main combustion stage nozzle region 2 is divided into multiple main combustion stage nozzle sub-regions 21 along the radial direction of the fuel injection rod. Each main combustion stage nozzle sub-region 21 includes multiple arrayed main combustion stage nozzles 211.

[0135] The pre-combustion stage nozzle area 1 and each main combustion stage nozzle sub-area 21 are supplied with oil through independent oil circuits.

[0136] In one embodiment of the present invention, the present invention is based on a jet-stabilized flame combustion head configuration, with main and pre-combustion stages of fuel, and the main and pre-combustion stage injection rods having a rectangular structure. The pre-combustion stage nozzles are supplied with fuel from a single fuel manifold, referred to as fuel line 1, and their nozzles are located at the center of the injection rod, arranged in a multi-point radial array. The end faces of the main combustion stage nozzles are symmetrically located on both sides of the injection rod at a certain angle to the end faces of the pre-combustion stage nozzles. The main combustion stage fuel injection, under the strong shearing action of the jet air, improves fuel atomization performance. The main combustion stage nozzles are divided into multiple radial regions, each region's nozzles are supplied with fuel from a separate fuel manifold, and the fuel flow rate of each manifold is independently adjustable, also arranged in a multi-point radial array. The number of main and pre-combustion stage nozzles is determined by comprehensively considering factors such as the design flow rate of a single nozzle unit, the nozzle supply pressure difference, and the sensitivity of the local fuel flow rate at the combustion chamber head to the outlet temperature, to determine the specific number of radially divided main combustion stage nozzles, such as... Figure 5 As shown, in one embodiment of the present invention, the main combustion stage nozzle is supplied with fuel by three fuel lines, which are referred to as fuel line 2, fuel line 3, and fuel line 4 respectively from top to bottom radially.

[0137] The embodiments of the present invention achieve the following technical effects:

[0138] According to the embodiments of the present invention, the fuel supply flow rate of each region can be actively adjusted under different conditions based on the correlation between fuel supply distribution parameters and performance parameters such as combustion chamber outlet temperature field, so as to obtain the optimal fuel distribution ratio between different zones, refine the control of fuel mist distribution, and improve the quality of combustion chamber outlet temperature field; the multi-channel fuel injector of the multi-channel fuel zone structure can arrange nozzles with different atomization performance according to the fuel distribution requirements of different radial positions, optimize the atomization performance of the head fuel nozzle in the small state, and improve combustion efficiency.

[0139] Obviously, those skilled in the art should understand that the modules or steps of the above-described embodiments of the present invention can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the embodiments of the present invention are not limited to any particular hardware and software combination.

[0140] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations can be made to the embodiments of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for controlling the temperature field at the combustion chamber outlet, characterized in that, include: The engine operating condition parameters and the fuel supply flow instructions of multiple fuel lines are obtained. The engine operating condition parameters and the fuel supply flow instructions of multiple fuel lines are normalized to generate model input features. The fuel supply flow instructions of multiple fuel lines include the fuel supply flow instructions of the pre-combustion stage fuel in the fuel supply path of the pre-combustion stage nozzle region (1) and the fuel supply flow instructions of the main combustion stage fuel in the fuel supply path of multiple main combustion stage nozzle sub-regions (21) of the main combustion stage nozzle region (2). A deep learning prediction model with a hierarchical-fusion network architecture is constructed. The input features of the model are input into the trained deep learning prediction model to obtain the predicted combustion chamber outlet temperature field performance parameters. Based on the predicted combustion chamber outlet temperature field performance parameters, an objective function for evaluating the quality of the temperature field is constructed. The combustion chamber outlet temperature field performance parameters include the outlet temperature distribution coefficient and the radial temperature distribution coefficient. The flow distribution ratio of the multi-path fuel is used as a decision variable. The Bayesian optimization algorithm is used for iterative search. The flow ratio of each fuel is determined by the objective function to generate the optimal fuel distribution instruction set. The optimal fuel distribution instruction set is converted into the control signal of the metering valve of the multi-path fuel, and the actuator is driven to independently adjust the fuel flow of the fuel supply line of the pre-combustion stage nozzle area (1) and the fuel supply line of each of the main combustion stage nozzle sub-areas (21).

2. The method for controlling the combustion chamber outlet temperature field as described in claim 1, characterized in that, Obtain engine operating condition parameters and multi-path fuel flow commands, normalize the engine operating condition parameters and multi-path fuel flow commands, and construct model input features, including: A fuel supply tensor is constructed based on the pre-combustion stage fuel supply flow command and the main combustion stage fuel supply flow command. The first dimension of the fuel supply tensor is used to store the path number of all fuel supply lines, and the second dimension is used to store the normalized fuel supply flow command corresponding to each path number. The engine operating condition parameters are divided into combustion chamber inlet flow field parameters that affect the combustion chamber inlet flow field and engine power setting parameters that characterize the engine power setting state, and the combustion chamber inlet flow field parameters and the engine power setting parameters are normalized respectively. The normalized combustion chamber inlet flow field parameters, the normalized engine power setting parameters, and the fuel supply tensor are combined to generate the model input features.

3. The method for controlling the combustion chamber outlet temperature field as described in claim 1, characterized in that, Constructing a deep learning prediction model with a layered-fusion network architecture includes: The deep learning prediction model includes interconnected partition feature extraction layers, radial spatial interaction layers, and a global combustion process layer; The partition feature extraction layer includes multiple sub-networks with shared parameters. Each sub-network receives the fuel supply tensor of each fuel supply circuit. Each sub-network independently extracts and outputs the state feature vector of the fuel for each fuel supply circuit. The number of sub-networks is the same as the number of fuel supply circuits. The relative positions of each main combustion stage nozzle sub-region (21) of the pre-combustion stage nozzle region (1) and the main combustion stage nozzle region (2) in the radial direction of the fuel injector are obtained. The multiple input state feature vectors are arranged according to the relative positions through the radial spatial interaction layer to generate a feature sequence. The feature sequence is then processed by a convolutional neural network to output a spatial fusion feature that integrates the radial mixing pattern. The global combustion process layer fuses the spatial fusion features, normalized combustion chamber inlet flow field parameters, and normalized engine power setting parameters. Through fully connected network mapping, a two-dimensional temperature field matrix is ​​output to characterize the predicted radial-circumferential temperature field distribution at the combustion chamber outlet section.

4. The method for controlling the combustion chamber outlet temperature field as described in claim 3, characterized in that, The feature sequence is processed by a convolutional neural network to output spatial fusion features that fuse radial mixing patterns, including: The feature sequence is input into a convolutional neural network, and the feature sequence is subjected to a sliding window convolution operation through the convolutional layer of the convolutional neural network to extract the spatial correlation and interaction between adjacent and cross-regional fuel state features. The output after the convolution operation is nonlinearly activated and feature compressed to generate a spatial fusion feature vector, which is used to characterize the overall state of the multi-fuel mixture in the radial space of the combustion chamber head.

5. The method for controlling the combustion chamber outlet temperature field as described in claim 3, characterized in that, The global combustion process layer fuses the spatial fusion features, normalized combustion chamber inlet flow field parameters, and normalized engine power setting parameters. Through fully connected network mapping, it outputs a two-dimensional temperature field matrix characterizing the predicted radial-circumferential temperature field distribution at the combustion chamber outlet cross-section, including: The spatial fusion features, the normalized combustion chamber inlet flow field parameters, and the normalized engine power setting parameters are spliced ​​together to generate a comprehensive input vector; The integrated input vector is input into a subnetwork containing at least two fully connected layers, and the output is a reconstructed two-dimensional matrix, wherein the length of the reconstructed two-dimensional matrix is ​​the number of grid points in the radial direction of the predicted temperature field, and the width of the reconstructed two-dimensional matrix is ​​the number of grid points in the circumferential direction of the predicted temperature field. The reconstructed two-dimensional matrix is ​​normalized to generate a two-dimensional temperature field matrix that characterizes the predicted radial-circumferential temperature field distribution at the combustion chamber outlet section.

6. The method for controlling the combustion chamber outlet temperature field as described in claim 3, characterized in that, Also includes: The composite loss function L of the deep learning prediction model is constructed as follows: The predicted outlet temperature distribution coefficient OTDF_pred and the predicted radial temperature distribution coefficient RTDF_pred are calculated using the two-dimensional temperature field matrix. The mean square error of the outlet temperature distribution coefficient is calculated using the predicted outlet temperature distribution coefficient OTDF_pred. The mean square error of the radial temperature distribution coefficient is calculated using the predicted radial temperature distribution coefficient RTDF_pred. The sum of the mean square error of the outlet temperature distribution coefficient and the mean square error of the radial temperature distribution coefficient is used as the coefficient loss term L_coeff; The deep learning prediction model is trained using a composite loss function L, where L = α × L_dist + β × L_coeff, and L_dist is the error between the two-dimensional temperature field matrix predicted by the deep learning prediction model and the actual temperature field distribution matrix, and α and β are preset weight coefficients.

7. The method for controlling the combustion chamber outlet temperature field as described in claim 1, characterized in that, Based on the predicted combustion chamber outlet temperature field performance parameters, an objective function for evaluating the quality of the temperature field is constructed, including: The objective function J = -[λ1×OTDF_pred+λ2×|RTDF_pred-RTDF_target|], where OTDF_pred is the predicted outlet temperature distribution coefficient, RTDF_pred is the predicted radial temperature distribution coefficient, RTDF_target is the preset target value of the ideal radial temperature distribution coefficient, and λ1 and λ2 are preset weighting coefficients.

8. The method for controlling the combustion chamber outlet temperature field as described in claim 1, characterized in that, Using the multi-path fuel flow allocation ratio as a decision variable, an iterative search is performed using a Bayesian optimization algorithm. The objective function is used to determine the fuel flow ratio for each path, generating an optimal fuel allocation instruction set, including: Within the feasible region that satisfies the total fuel flow constraint, an initial fuel allocation ratio scheme is generated. The trained deep learning prediction model is used to obtain the prediction performance parameters corresponding to the initial fuel allocation ratio scheme, and the initial objective function value is calculated to generate an initial sample set. Based on the fuel allocation scheme in the sample set and the corresponding initial objective function value, a Gaussian process regression model is constructed. Using the Gaussian process regression model, the initial sample set is iterated to generate the final sample set, and the parameters of the Gaussian process regression model are updated. The process is repeated until a preset iteration termination condition is met, and the fuel allocation ratio scheme with the optimal objective function value is selected from the final sample set as the optimal fuel allocation instruction set.

9. The method for controlling the combustion chamber outlet temperature field as described in any one of claims 1 to 7, characterized in that, Also includes: After executing the optimal fuel distribution command set to adjust the fuel supply circuit, actual temperature field data at the combustion chamber outlet is collected. Actual performance parameters are calculated using this data, and the deep learning prediction model is updated iteratively using these actual performance parameters, including: After executing the optimal fuel distribution command set, the actual temperature field distribution data at the combustion chamber outlet is collected; Based on the actual temperature field distribution data, the actual outlet temperature distribution coefficient and the actual radial temperature distribution coefficient are calculated. The engine operating condition parameters, the executed optimal fuel distribution command set, the actual outlet temperature distribution coefficient, and the actual radial temperature distribution coefficient are combined to generate a new training sample with labels. The new training samples are added to the historical training dataset of the deep learning prediction model; The deep learning prediction model is incrementally trained using the historical training dataset containing the new training samples, and the parameters of the deep learning prediction model are adjusted.

10. A multi-channel fuel injection boom, based on which the combustion chamber outlet temperature field of an aero-engine is controlled in zones using the combustion chamber outlet temperature field control method according to any one of claims 1 to 9, characterized in that, The multi-channel fuel injector includes: The pre-combustion stage nozzle area (1) and the main combustion stage nozzle area (2) are provided. The pre-combustion stage nozzle area (1) is located at the center of the fuel injection rod, and the main combustion stage nozzle area (2) is located on both sides of the pre-combustion stage nozzle area (1). The pre-combustion stage nozzle region (1) includes multiple arrayed pre-combustion stage nozzles (11), and the main combustion stage nozzle region (2) is divided into multiple main combustion stage nozzle sub-regions (21) along the radial direction of the fuel injector rod. Each main combustion stage nozzle sub-region (21) includes multiple arrayed main combustion stage nozzles (211). The pre-combustion stage nozzle region (1) and each of the main combustion stage nozzle sub-regions (21) are supplied with oil through independent oil supply lines.

Images