An active control system and method for a thrust augmentor
By combining real-time status monitoring and deep learning prediction with high-frequency fuel flow control, the difficulties in hardware modification and response lag in the active control of afterburners are solved, achieving efficient combustion stability control without hardware modification.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HARBIN ENG UNIV
- Filing Date
- 2026-04-21
- Publication Date
- 2026-06-12
AI Technical Summary
Existing active control schemes for afterburners suffer from problems such as difficulty in hardware modification, large computational load, and response lag, making it difficult to effectively cope with nonlinear and transient combustion instability.
The system employs a real-time status monitoring module to collect multi-source heterogeneous data, performs deep learning predictions through a digital signal processing and prediction module, and combines this with an active control execution module to implement high-frequency fuel flow control, thereby suppressing combustion oscillations and flameout.
It achieves efficient active control without modifying the engine's core hardware, improves response margin and real-time performance, reduces computational complexity, and can identify and suppress combustion instability in advance.
Smart Images
Figure CN122191584A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to an active control system and method for an afterburner, belonging to the field of engine combustion stability control technology. Background Technology
[0002] Under extreme conditions (such as high-altitude low-speed and high-maneuver flight), the afterburner of an engine is prone to thermoacoustic oscillations or high-altitude flameout. This phenomenon is particularly pronounced in aero engines, seriously affecting flight safety and mission completion. Thermoacoustic oscillations are caused by positive feedback coupling between combustion heat release and the sound field within the combustion chamber, manifesting as in-phase pulsations of pressure and heat release rate, with amplitudes ranging from several kPa to tens of kPa, which can easily lead to fatigue damage to the engine structure. High-altitude flameout is caused by unstable flame residence and local deviations in stoichiometry from the combustible limit, resulting in loss of engine power.
[0003] Existing combustion stability control technologies are mainly divided into two categories: passive control and active control. Passive control suppresses oscillating combustion at specific frequencies by adjusting combustion organization or using acoustic resonators, but it suffers from drawbacks such as high cost, low flexibility, and reliance on specific system designs. Active control has proven to be more flexible and effective. For example, CN118088322A, an invention entitled "An Active Control Method for Combustion Instability Based on Deep Learning," discloses a method that pre-trains a deep neural network on clustered data. The trained model can achieve online prediction of combustion instability. The deep learning-based prediction model has high accuracy, fast computation speed, and stronger universality. However, when the model predicts potential combustion instability, it uses indirect feedback control via neural network PID, which lags when facing unstable combustion. The constructed ResNet+LSTM deep learning model relies excessively on flame images. The invention disclosed in CN121031282A, entitled "A Combustion Instability Boundary Prediction System Based on Complex Environmental Condition Analysis," employs a dynamic environmental interference compensation mechanism. During real-time prediction of combustion instability boundaries, spectral correction parameters are configured according to different environmental conditions to eliminate spectral distortion caused by environmental disturbances in combustion dynamic parameters. This mechanism automatically senses and corrects spectral interference deviations at the combustion measurement location, ensuring the accuracy of combustion instability sensitive feature extraction and reducing instability warning errors. Furthermore, by deploying a real-time flame pose tracking system, fuel control parameters are adaptively adjusted immediately upon detecting pose anomalies, achieving autonomous correction of the combustion instability feature identification location and ensuring accurate spatial positioning of feature capture. However, the use of Fourier transform results in severe distortion of non-stationary combustion signals and further exacerbates the problem of insufficient adaptability to strongly nonlinear and rapidly transient conditions in subsequent processes. Therefore, existing active control schemes still have the following shortcomings:
[0004] 1. Most existing deep learning-based solutions rely excessively on flame images, which leads to difficulties in placing observation windows and high computational demands for image processing in engineering applications. At the same time, some solutions require modifications to the internal hardware structure of the afterburner, which greatly limits their engineering applications. 2. The control algorithms are mostly threshold-triggered or simple phase-delayed, resulting in a severely lagging response that makes it difficult to cope with nonlinear and transient combustion instability.
[0005] Therefore, there is an urgent need to propose an active control system and method for afterburners to solve the above-mentioned technical problems. Summary of the Invention
[0006] To address the aforementioned problems, an active control system and method for an afterburner are provided. A brief overview of the invention is given below to provide a basic understanding of certain aspects of the invention. It should be understood that this overview is not an exhaustive summary of the invention. It is not intended to identify key or essential parts of the invention, nor is it intended to limit the scope of the invention.
[0007] The technical solution of the present invention: An active control system for an afterburner includes: Real-time status monitoring module: collects multi-source heterogeneous data characterizing the operating status of the afterburner, including high-frequency dynamic signals and low-frequency operating condition signals; Digital signal processing and prediction module: preprocesses the acquired data and extracts time-frequency features, and performs online prediction of combustion instability trends based on a deep learning model; Active control execution module: Controls the afterburner fuel flow based on the prediction results of the deep learning model to suppress combustion oscillations and flameout.
[0008] Preferably, the real-time status monitoring module includes: The dynamic pressure sensor has a frequency response range of 0.5Hz to 15kHz. The dynamic pressure sensor is installed at the pressure measurement port of the afterburner to capture high-frequency dynamic signals inside the combustion chamber in real time. The flame feature detection unit uses a fiber optic probe combined with a photomultiplier tube to capture ultraviolet radiation spectral signals that characterize the fluctuation of combustion heat release rate. The operating condition parameter input unit obtains low-frequency operating condition parameters in real time from the engine full authority digital electronic control system via the data bus. The low-frequency operating condition parameters include compressor outlet temperature, pressure, high-pressure rotor speed, and afterburner fuel flow.
[0009] Preferably, the digital signal processing and prediction module is connected to the real-time status monitoring module, and the digital signal processing and prediction module adopts a heterogeneous computing architecture platform based on Xilinx Zynq UltraScale+MPSoC ZU9EG; The FPGA side of the digital signal processing and prediction module is responsible for high-speed data acquisition and preprocessing, including anti-aliasing filtering based on FIR filters, noise reduction, and time alignment and synchronization processing of signals from different sensors. The CPU of the digital signal processing and prediction module is responsible for running the core algorithms, including continuous wavelet transform, Transformer deep learning prediction, and dynamic mode decomposition.
[0010] Preferably, the active control execution module is connected to the digital signal processing and prediction module. The active control execution module uses a direct-acting servo proportional valve as the actuator, which is installed at the inlet of the afterburner fuel main. The actuator's step response time is less than 2ms, and the response frequency is not less than 500Hz. The module is controlled by a PWM signal.
[0011] An active control method for an afterburner, employing the aforementioned active control system for an afterburner, includes the following steps: S01. Data Acquisition and Preprocessing: Real-time synchronous acquisition of high-frequency dynamic signals and low-frequency operating condition signals during the operation of the afterburner, followed by noise reduction and time synchronization processing. S02, Feature Extraction: Perform time-frequency analysis on high-frequency dynamic signals to extract the dominant frequency and energy characteristics that characterize the precursors of combustion instability; S03, Deep Learning Short-Term Prediction: Input historical time series data and extracted features into a pre-trained deep learning prediction model to predict the burning state within a preset time window in the future. S04, Instability Warning: Determines whether the predicted combustion state exceeds the set safety threshold; if it does, a warning is triggered. S05. Flow field structure identification: After the warning is triggered, modal analysis is performed on the current combustion field to identify the dominant mode frequency and disturbance propagation direction that cause combustion instability. S06. High-frequency fuel fluctuation decoupling control: In response to the warning signal, the control actuator superimposes high-frequency pulsation on the basic fuel flow. The high-frequency pulsation decouples combustion heat release from acoustic pressure fluctuation, thereby suppressing combustion instability.
[0012] Preferably, in S02, continuous wavelet transform is used to perform time-frequency analysis on the pressure signal;
[0013] In the formula, This is a scaling factor used to control the scaling of the wavelet to correspond to different frequencies; This is a translation factor used to reflect the time position; This indicates the pressure signal after preprocessing. The complex conjugate of the selected Morlet mother wavelet; This represents the calculated wavelet coefficients, through which the dominant frequency of the current combustion state and its local energy characteristics within a specific frequency band are extracted.
[0014] Preferably, in step S03, the multidimensional time series obtained in steps S01 and S02 over a past period are fused into an input matrix. The multidimensional time series includes dynamic pressure, flame intensity, wavelet time-frequency features, and low-frequency operating parameters. The core of the model employs a multi-head self-attention mechanism, and its single-head attention calculation principle is as follows:
[0015] In the formula, , , These represent the query matrix, key matrix, and value matrix, respectively, all obtained from the input matrix through a linear mapping. Let be the dimension of the key vector. This represents the single-head attention weight output. Deep learning models use this mechanism to effectively capture the deep coupling relationship between multi-source heterogeneous signals in long-term time series, and calculate the predicted values of the shutdown probability and oscillation amplitude in the future through model forward propagation.
[0016] Preferred step S04: The model prediction result is compared with a preset threshold. If the predicted flameout probability or oscillation amplitude exceeds the set value, a combustion instability warning is triggered, and the process proceeds to step S05; otherwise, the current state is determined to be stable, and the process returns to step S01.
[0017] Preferred: In S05, after the warning is triggered, the dynamic mode decomposition algorithm is used to process the recent pressure time series snapshots, construct the data matrix of two adjacent time points, and assume that the system evolution satisfies the linear mapping. The eigenvalues and eigenvectors of the approximate matrix are solved by singular value decomposition.
[0018] In the formula, Indicates the first The continuous-time eigenvalues of the first mode represent the first mode's continuous-time eigenvalues. The growth rate of the first mode, where the imaginary part represents the first mode. The frequency of the first mode, This represents the eigenvalues of the approximate matrix obtained by solving. The time interval between adjacent moments is represented by the algorithm. The algorithm extracts the mode with the largest real part as the target control mode and records the dominant frequency and phase information of the mode, thereby accurately identifying the main cause of combustion instability.
[0019] Preferred step S06: Based on the dominant frequency obtained in step S05, a mismatch phase control strategy is formulated; the fuel pulsation frequency is set to the target dominant frequency plus an offset; the drive unit generates a PWM signal with corresponding frequency and phase, and controls the servo proportional valve to superimpose pulsation on the basic fuel flow, so that the pulsation phase of the heat release rate is misaligned with the pressure pulsation phase; when the control system detects that the predicted risk has been eliminated and stabilized, it determines that combustion has recovered and stabilized, stops high-frequency fuel modulation, and returns to step S01.
[0020] The present invention has the following beneficial effects: (1) This invention does not require changes to the core engine hardware and has strong engineering adaptability: by reusing the existing pressure test hole and fuel line, combined with high-speed solenoid valve and sensor, active control modification can be achieved, avoiding the addition of complex hardware equipment inside the combustion chamber.
[0021] (2) This invention achieves advanced prediction and improves control response margin: the Transformer deep learning model is used to predict the risk of shutdown or oscillation in the very short time in the future, changing the traditional "post-event remedy" to active defense, which significantly increases the response time margin of the control system.
[0022] (3) This invention reduces computational complexity and improves real-time performance: It abandons the computationally intensive flame image processing scheme and uses the fusion of one-dimensional high-frequency dynamic signals and low-frequency operating parameters as the input of the deep learning model, which greatly reduces the hardware computing burden and ensures the real-time performance of online prediction. Attached Figure Description
[0023] Figure 1 This is a diagram of an active control system architecture for an afterburner.
[0024] Figure 2 This is a flowchart of an active control method for an afterburner. Detailed Implementation
[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the invention is described below with reference to specific embodiments shown in the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.
[0026] Specific implementation method one: Combining Figure 1This embodiment describes an active control system for an afterburner, comprising: Real-time status monitoring module: The real-time status monitoring module is configured to collect multi-source heterogeneous data characterizing the operating status of the afterburner, including high-frequency dynamic signals and low-frequency operating condition signals and flame heat release rate signals. Digital signal processing and prediction module: The digital signal processing and prediction module is configured to preprocess the acquired data and extract time-frequency features, and to perform online prediction of combustion instability trends based on a deep learning model; Active control execution module: The active control execution module is configured to implement high-frequency perturbation control on the afterburner fuel flow based on the prediction results of the deep learning model in order to suppress combustion oscillation and flameout.
[0027] Specific Implementation Method Two: Combining Figure 2 This embodiment describes an active control method for an afterburner, which includes the following steps: S01. Data Acquisition and Preprocessing: Real-time synchronous acquisition of high-frequency dynamic signals and low-frequency operating condition signals during the operation of the afterburner, followed by noise reduction and time synchronization processing. S02, Feature Extraction: Perform time-frequency analysis on high-frequency dynamic signals to extract the dominant frequency and energy characteristics that characterize the precursors of combustion instability; S03, Deep Learning Short-Term Prediction: Input historical time series data and extracted features into a pre-trained deep learning prediction model to predict the burning state within a preset time window in the future. S04, Instability Warning: Determines whether the predicted combustion state exceeds the set safety threshold; if it does, a warning is triggered. S05. Flow field structure identification: After the warning is triggered, modal analysis is performed on the current combustion field to identify the dominant mode frequency and disturbance propagation direction that cause combustion instability. S06, High-frequency fuel fluctuation decoupling control: In response to the warning signal, the control actuator superimposes high-frequency pulsation on the basic fuel flow, thereby decoupling combustion heat release from acoustic pressure fluctuation through the high-frequency pulsation, thereby suppressing combustion instability; This invention uses multi-source sensor fusion and deep learning prediction to identify and precisely suppress early signs of instability. It can be implemented on existing hardware structures, is easy to install and operate, and facilitates cost control and widespread application.
[0028] Specific implementation method three: Combining Figure 1-2 This embodiment describes an active control method for an afterburner, which employs an active control system for the afterburner. The real-time status monitoring module includes: The dynamic pressure sensor uses a PCB Piezotronics 112A05 high-frequency piezoelectric pressure sensor. This model of sensor has a frequency response range of 0.5Hz to 15kHz and excellent high-temperature stability. The dynamic pressure sensor is installed at the existing casing pressure measurement port of the afterburner through a water-cooled sleeve to capture high-frequency dynamic signals (high-frequency acoustic pressure pulsations) inside the combustion chamber in real time. The flame feature detection unit uses a fiber optic probe with a sapphire window combined with a Hamamatsu H10722 series photomultiplier tube. The fiber optic probe extracts the self-luminous emission of the flame in the combustion zone through the optical fiber. The center wavelength of the filter is set to 308nm (corresponding to OH groups) to accurately capture the ultraviolet radiation spectrum signal that characterizes the fluctuation of the combustion heat release rate. The operating condition parameter input unit obtains low-frequency operating condition parameters in real time from the engine full authority digital electronic control system (FADEC) via ARINC429 or CAN serial data bus. The low-frequency operating condition parameters include compressor outlet temperature, pressure, high-pressure rotor speed and afterburner fuel flow. The digital signal processing and prediction module is connected to the real-time status monitoring module. The digital signal processing and prediction module adopts a heterogeneous computing architecture platform based on Xilinx Zynq UltraScale+MPSoC ZU9EG. The FPGA side of the digital signal processing and prediction module is responsible for high-speed data acquisition and preprocessing, including anti-aliasing filtering based on FIR filters, noise reduction, and time alignment and synchronization processing of signals from different sensors. The CPU of the digital signal processing and prediction module is responsible for running the core algorithms, including continuous wavelet transform, Transformer deep learning prediction, and dynamic mode decomposition. The active control execution module is connected to the digital signal processing and prediction module. The active control execution module uses a Moog D633 series direct-acting servo proportional valve as the actuator, which is installed at the inlet of the boost fuel main. This valve actuator has extremely high dynamic response characteristics (step response time less than 2ms, response frequency not less than 500Hz). The active control execution module drives the valve by generating a high-frequency PWM signal from the drive control unit, realizing high-frequency fine-tuning and modulation control of the fuel.
[0029] The method includes the following steps: In S01, data acquisition and preprocessing are performed by synchronously acquiring high-frequency pressure pulsation and light intensity signals at a sampling rate of 20kHz, reading low-frequency operating parameters at a sampling rate of 100Hz, and performing digital low-pass filtering, normalization, and outlier removal in the FPGA, while also achieving time synchronization. Feature extraction in S02 is based on continuous wavelet transform. The continuous wavelet transform is used to perform time-frequency analysis on the preprocessed pressure signal. Compared with the traditional fast Fourier transform, the continuous wavelet transform can provide better time-frequency localization features.
[0030] In the formula, This is a scaling factor used to control the scaling of the wavelet to correspond to different frequencies; This is a translation factor used to reflect the time position; This indicates the pressure signal after preprocessing. The complex conjugate of the selected Morlet mother wavelet; This represents the calculated wavelet coefficients, through which the dominant frequency of the current combustion state and its local energy characteristics within a specific frequency band are extracted; It is a time variable; For time ,scale The wavelet mother function at that point, differential; In S03, deep learning short-term prediction is based on the Transformer architecture, fusing the multi-dimensional time series obtained in steps S01 and S02 within a past period (a period can be 10ms) into an input matrix. The multi-dimensional time series includes dynamic pressure, flame intensity, wavelet time-frequency features, and low-frequency operating condition parameters. The core of the model adopts a multi-head self-attention mechanism, and its single-head attention calculation principle is as follows:
[0031] In the formula, , , These represent the query matrix, key matrix, and value matrix, respectively, all obtained from the input matrix through a linear mapping. Let be the dimension of the key vector. This represents the single-head attention weight output. Deep learning models use this mechanism to effectively capture the deep coupling relationships of multi-source heterogeneous signals over long-range time series, and calculate the predicted values of the shutdown probability and oscillation amplitude within a future time period (which can be 5ms to 20ms) through model forward propagation. T This is the matrix transpose. In step S04, the model prediction results are compared with the preset threshold. If the predicted flameout probability exceeds 0.7 or the oscillation amplitude exceeds 5 kPa, a combustion instability warning is triggered, and the process proceeds to step S05; otherwise, the current state is determined to be stable, and the process returns to step S01. In S05, flow field structure identification is based on dynamic mode decomposition. After the warning is triggered, the dynamic mode decomposition algorithm is used to process the recent pressure time series snapshots, construct the data matrix of two adjacent time points, and assume that the system evolution satisfies the linear mapping. The eigenvalues and eigenvectors of the approximate matrix are solved by singular value decomposition.
[0032] In the formula, Indicates the first The continuous-time eigenvalues of the first mode represent the first mode's continuous-time eigenvalues. The growth rate of the first mode, where the imaginary part represents the first mode. The frequency of the first mode, This represents the eigenvalues of the approximate matrix obtained by solving. The time interval between adjacent moments is represented by the algorithm. The algorithm extracts the mode with the largest real part as the target control mode and records the dominant frequency and phase information of the mode, thereby accurately identifying the main cause of combustion instability. In step S06, a mismatch phase control strategy is formulated based on the dominant frequency obtained in step S05; the fuel pulsation frequency is set to the target dominant frequency plus a 10Hz offset; the drive unit generates a PWM signal with corresponding frequency and phase to control the high-speed servo proportional valve to superimpose a small (1% to 3% of the basic flow) pulsation on the basic fuel flow, so that the pulsation phase of the heat release rate is misaligned with the pressure pulsation phase; when the control system detects that the predicted risk has been eliminated and stabilized for 50ms, it determines that the combustion has recovered and stabilized, stops the high-frequency fuel modulation and returns to step S01.
[0033] It should be noted that in the above embodiments, as long as the technical solutions are not contradictory, they can be permuted and combined. Those skilled in the art can exhaust all possibilities based on the mathematical knowledge of permutation and combination. Therefore, the present invention will not describe the technical solutions after permutation and combination one by one, but it should be understood that the technical solutions after permutation and combination have been disclosed by the present invention.
[0034] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. An active control system for an afterburner, characterized in that: include: Real-time status monitoring module: collects multi-source heterogeneous data characterizing the operating status of the afterburner, including high-frequency dynamic signals and low-frequency operating condition signals; Digital signal processing and prediction module: preprocesses the acquired data and extracts time-frequency features, and performs online prediction of combustion instability trends based on a deep learning model; Active control execution module: Controls the afterburner fuel flow based on the prediction results of the deep learning model to suppress combustion oscillations and flameout.
2. The active control system for an afterburner according to claim 1, characterized in that: The real-time status monitoring module includes: The dynamic pressure sensor has a frequency response range of 0.5Hz to 15kHz. The dynamic pressure sensor is installed at the pressure measurement port of the afterburner to capture high-frequency dynamic signals inside the combustion chamber in real time. The flame feature detection unit uses a fiber optic probe combined with a photomultiplier tube to capture ultraviolet radiation spectral signals that characterize the fluctuation of combustion heat release rate. The operating condition parameter input unit obtains low-frequency operating condition parameters in real time from the engine full authority digital electronic control system via the data bus. The low-frequency operating condition parameters include compressor outlet temperature, pressure, high-pressure rotor speed, and afterburner fuel flow.
3. The active control system for an afterburner according to claim 2, characterized in that: The digital signal processing and prediction module is connected to the real-time status monitoring module. The digital signal processing and prediction module adopts a heterogeneous computing architecture platform based on Xilinx Zynq UltraScale+MPSoC ZU9EG. The FPGA side of the digital signal processing and prediction module is responsible for high-speed data acquisition and preprocessing, including anti-aliasing filtering based on FIR filters, noise reduction, and time alignment and synchronization processing of signals from different sensors. The CPU of the digital signal processing and prediction module is responsible for running the core algorithms, including continuous wavelet transform, Transformer deep learning prediction, and dynamic mode decomposition.
4. The active control system for an afterburner according to claim 3, characterized in that: The active control execution module is connected to the digital signal processing and prediction module. The active control execution module uses a direct-acting servo proportional valve as the actuator, which is installed at the inlet of the afterburner fuel main. The actuator's step response time is less than 2ms and the response frequency is not less than 500Hz. The module is driven by a PWM signal to achieve control.
5. A method for active control of an afterburner, characterized in that: The active control system for an afterburner as described in any one of claims 1-4 includes the following steps: S01. Data Acquisition and Preprocessing: Real-time synchronous acquisition of high-frequency dynamic signals and low-frequency operating condition signals during the operation of the afterburner, followed by noise reduction and time synchronization processing. S02, Feature Extraction: Perform time-frequency analysis on high-frequency dynamic signals to extract the dominant frequency and energy characteristics that characterize the precursors of combustion instability; S03, Deep Learning Short-Term Prediction: Input historical time series data and extracted features into a pre-trained deep learning prediction model to predict the burning state within a preset time window in the future. S04, Instability Warning: Determines whether the predicted combustion state exceeds the set safety threshold; if it does, a warning is triggered. S05. Flow field structure identification: After the warning is triggered, modal analysis is performed on the current combustion field to identify the dominant mode frequency and disturbance propagation direction that cause combustion instability. S06. High-frequency fuel fluctuation decoupling control: In response to the warning signal, the control actuator superimposes high-frequency pulsation on the basic fuel flow. The high-frequency pulsation decouples combustion heat release from acoustic pressure fluctuation, thereby suppressing combustion instability.
6. The active control method for an afterburner according to claim 5, characterized in that: In S02, continuous wavelet transform is used to perform time-frequency analysis on the pressure signal; In the formula, This is a scaling factor used to control the scaling of the wavelet to correspond to different frequencies; This is a translation factor used to reflect the time position; This indicates the pressure signal after preprocessing. The complex conjugate of the selected Morlet mother wavelet; This represents the calculated wavelet coefficients, through which the dominant frequency of the current combustion state and its local energy characteristics within a specific frequency band are extracted.
7. The active control method for an afterburner according to claim 6, characterized in that: In step S03, the multidimensional time series obtained in steps S01 and S02 over a past period are fused into an input matrix. The multidimensional time series includes dynamic pressure, flame intensity, wavelet time-frequency features, and low-frequency operating parameters. The core model employs a multi-head self-attention mechanism, and its single-head attention calculation principle is as follows: In the formula, , , These represent the query matrix, key matrix, and value matrix, respectively, all obtained from the input matrix through a linear mapping. Let be the dimension of the key vector. This represents the single-head attention weight output. Deep learning models use this mechanism to effectively capture the deep coupling relationship between multi-source heterogeneous signals in long-term time series, and calculate the predicted values of the shutdown probability and oscillation amplitude in the future through model forward propagation.
8. The active control method for an afterburner according to claim 7, characterized in that: In step S04, the model prediction results are compared with the preset threshold. If the predicted flameout probability or oscillation amplitude exceeds the set value, a combustion instability warning is triggered, and the process proceeds to step S05; otherwise, the current state is determined to be stable, and the process returns to step S01.
9. The active control method for an afterburner according to claim 8, characterized in that: In S05, after the warning is triggered, the dynamic mode decomposition algorithm is used to process the recent pressure time series snapshots, construct the data matrix of two adjacent time points, and assume that the system evolution satisfies the linear mapping. The eigenvalues and eigenvectors of the approximate matrix are solved by singular value decomposition. In the formula, Indicates the first The continuous-time eigenvalues of the first mode represent the first mode's continuous-time eigenvalues. The growth rate of the first mode, where the imaginary part represents the first mode. The frequency of the first mode, This represents the eigenvalues of the approximate matrix obtained by solving. The time interval between adjacent moments is represented by the algorithm. The algorithm extracts the mode with the largest real part as the target control mode and records the dominant frequency and phase information of the mode, thereby accurately identifying the main cause of combustion instability.
10. The active control method for an afterburner according to claim 9, characterized in that: In step S06, a mismatch phase control strategy is formulated based on the dominant frequency obtained in step S05; the fuel pulsation frequency is set to the target dominant frequency plus an offset; the drive unit generates a PWM signal with corresponding frequency and phase to control the servo proportional valve to superimpose pulsation on the basic fuel flow, so that the pulsation phase of the heat release rate is misaligned with the pressure pulsation phase; when the control system detects that the predicted risk has been eliminated and stabilized, it determines that combustion has recovered and stabilized, stops high-frequency fuel modulation and returns to step S01.