A Virtual Throttle Recognition Method for Aircraft Engines Based on Multi-Source Parameter Fusion
By using multi-source parameter fusion and a dynamically weighted physical information neural network model, the problem of accurate identification of aircraft engine thrust management under sensor failure was solved, achieving safe control under complex operating conditions and reducing surge risk and software maintenance costs.
Patent Information
- Application Number
- CN202611391900.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-09
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2046-09-09
AI Technical Summary
Existing aircraft engine control systems struggle to accurately identify pilot throttle inputs when sensors malfunction, especially when physical throttle signals are lost. Furthermore, they are prone to surge risks during thrust management. Existing analytical redundancy technology suffers from computational divergence under complex operating conditions, failing to guarantee flight safety.
A virtual throttle recognition method for aircraft engines based on multi-source parameter fusion is adopted. By collecting multi-source feature parameters in real time, the parameters are aligned and corrected using time delay cross-correlation matrix and thermodynamic degradation baseline offset factor. The method is then combined with a dynamic weighted physical information neural network model for inversion and an asymmetric smooth switching strategy to ensure a smooth transition of control.
It improves the inversion accuracy under extreme conditions, eliminates the risk of model inversion crashes or divergence, avoids surge risk, enhances flight safety redundancy, and reduces software maintenance costs.
Smart Images

Figure CN122880630B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of aero-engine control technology, specifically relating to a method for identifying the virtual throttle of an aircraft engine based on multi-source parameter fusion. Background Technology
[0002] In modern integrated flight control systems, the thrust management system generates thrust commands through pilot manipulation of the throttle lever (TLA) or the automatic throttle system. These commands are then transmitted to the full authority digital engine control system (FADEC) via a fly-by-wire system. Although modern civil aircraft commonly employ dual-redundant or triple-redundant linear / rotating variable differential transformers (LVDT / RVDT) as throttle position sensors to improve reliability, extreme situations such as common-mode failures, mechanical linkage jamming, and circuit breaks still occur frequently due to the sensors being exposed to harsh environments of strong mechanical vibration, electromagnetic interference, and alternating high and low temperatures over long periods.
[0003] In the event of an extreme failure where the physical throttle signal is completely lost, traditional FADEC systems typically employ a simple fault-locking strategy: locking the thrust state at the moment before the failure occurred, or directly reverting to idle. During high-dynamic flight phases such as takeoff, climb, or approach and landing, thrust lock-in or loss of thrust poses a serious threat to flight safety. Existing analytical redundancy techniques (such as conventional Kalman filtering or traditional data fitting models) often fail to consider the inherent physical thermal inertia hysteresis between aero-engine sensor signals. Furthermore, under complex nonlinear conditions such as large-scale engine envelope maneuvers and long-term degradation of thermodynamic components, they are prone to computational divergence, making it difficult to accurately reflect the pilot's true throttle input intentions. Even more critically, when identifying a virtual throttle and transferring control, if the engine's current compressor surge margin boundary is not considered, a sudden thrust step can easily induce compressor stall or even in-flight engine failure. Summary of the Invention
[0004] The present invention aims to at least partially solve the aforementioned technical problems. Therefore, the objective of the present invention is to provide a method for identifying the virtual throttle of an aircraft engine based on multi-source parameter fusion.
[0005] The technical solution adopted in this invention is as follows:
[0006] A method for identifying the virtual throttle of an aircraft engine based on multi-source parameter fusion includes the following steps:
[0007] Step 1: Real-time acquisition of multi-source characteristic parameters of the aircraft engine, which consist of low-pressure rotor speed N1, high-pressure rotor speed N2, exhaust temperature EGT, fuel flow rate Wf, flight Mach number Ma, and flight altitude Alt.
[0008] Step 2: Extract time series of the multi-source feature parameters, calculate the phase hysteresis of the low-pressure rotor speed N1, high-pressure rotor speed N2 and exhaust temperature EGT relative to the fuel flow rate Wf using the time delay cross-correlation matrix, and perform time delay phase alignment processing on the data in the historical sliding window to obtain the aligned multi-source feature parameters.
[0009] Step 3: Calculate the exhaust temperature margin decay over the entire life cycle based on the engine's historical operating cycle number, generate a thermodynamic degradation baseline offset factor, and use this thermodynamic degradation baseline offset factor to normalize and correct the aligned multi-source characteristic parameters to obtain the corrected multi-source characteristic parameters.
[0010] Step 4: Input the corrected multi-source feature parameters into the pre-trained dynamic weighted physical information neural network model, and inversely calculate the current virtual throttle command; wherein, the weights of the loss function of the dynamic weighted physical information neural network model are dynamically adjusted in real time according to the flight envelope region where the current flight Mach number Ma and the flight altitude Alt are located;
[0011] Step 5: Perform cross-arbitration comparison between the virtual throttle command and the physical throttle command output by the physical throttle sensor. When the difference exceeds the set fault tolerance threshold, trigger a physical sensor fault alarm and start the asymmetric smooth switching strategy to switch the engine control command source from the physical throttle command to the virtual throttle command.
[0012] In a preferred embodiment of the present invention, the time delay cross-correlation matrix construction process in step 2 is as follows: using the fuel flow rate Wf as the reference signal, the cross-correlation functions of the low-pressure rotor speed N1, the high-pressure rotor speed N2, and the exhaust temperature EGT with the fuel flow rate Wf within the sliding time window are calculated respectively. The time delay corresponding to the maximum value of the cross-correlation function is found, and each parameter is forward-shifted and aligned according to the time delay. The principle is to eliminate the asynchronous error between the fuel flow rate (fastest response) and the rotor mechanical inertia and temperature thermal inertia, ensuring that the physical state slices of the input model belong to the same causal moment.
[0013] In a preferred embodiment of the present invention, the thermodynamic degradation baseline offset factor in step 3 is determined by calculating the increment of the steady-state value of the exhaust temperature under the current cruise condition, with the factory-set baseline exhaust temperature as zero. This thermodynamic degradation baseline offset factor is superimposed on the exhaust temperature feature nodes of the input layer of the dynamically weighted physical information neural network model in the form of a bias constant. Based on this principle, the parameter drift of the aging engine can be corrected mathematically, thus eliminating the need for frequent model retraining as the engine ages.
[0014] In a preferred embodiment of the present invention, the dynamic weighted physical information neural network model in step 4 consists of a data-driven multilayer perceptron network and a physical constraint layer embedded with the engine aerodynamic thermodynamic equations; the physical constraint layer includes residual terms of the rotor dynamics equations and residual terms of the combustion chamber energy conservation equations. By utilizing strong aerodynamic thermodynamic constraints, it is ensured that the output of the inversion model conforms to objective physical laws under any attitude.
[0015] In a preferred embodiment of the present invention, the total loss function of the dynamically weighted physical information neural network model is composed of two weighted components: data prediction error and physical constraint residual. The dynamically weighted physical information neural network model incorporates a flight envelope scheduler. When the flight envelope scheduler determines that the aircraft is in the transonic transition region, it increases the weight coefficient of the data prediction error and decreases the weight coefficient of the physical constraint residual. The principle is to reduce the weight of simplified physical mechanism equations at the edge of the extremely complex flow field envelope, thus preventing the model derivation from diverging.
[0016] In a preferred embodiment of the present invention, the cross-arbitration comparison process in step 5 includes: setting a residual evaluation function based on first-order hysteresis filtering; when the absolute difference between the physical throttle command and the virtual throttle command is continuously detected to be greater than the fault tolerance threshold for a period exceeding the confirmation time window, it is determined that the physical throttle sensor has malfunctioned; wherein, the fault tolerance threshold is dynamically relaxed or tightened according to the flight Mach number Ma. This prevents false triggering caused by occasional electromagnetic noise.
[0017] In a preferred embodiment of the present invention, the asymmetric smooth switching strategy in step 5 specifically involves: based on the engine compressor surge margin protection logic, determining the thrust change trend during control command source switching; if the virtual throttle command is greater than the physical throttle command, an acceleration trend is identified, and a first exponential smoothing function is used for slow switching; if the virtual throttle command is less than or equal to the physical throttle command, a deceleration trend is identified, and a second exponential smoothing function is used for fast switching; wherein the time constant of the first exponential smoothing function is greater than the time constant of the second exponential smoothing function. This principle maximizes system safety during control handover while ensuring safety.
[0018] As a preferred embodiment of the present invention, after obtaining the virtual throttle command in step 4 and before executing step 5, a boundary limitation step is also included: using the current flight Mach number Ma and flight altitude Alt to look up the table to obtain the minimum idle throttle limit value and the maximum continuous thrust throttle limit value of the engine in the current flight state, and performing upper and lower limit processing on the virtual throttle command obtained by inversion to form a double safety protection.
[0019] In a preferred embodiment of the present invention, the length of the sliding time window is adaptively adjusted according to the absolute rate of change of fuel flow rate Wf; when the absolute rate of change of fuel flow rate Wf exceeds the rate of change threshold, the length of the sliding time window is reduced. This aims to capture high-frequency transient characteristics under high-mobility "sudden acceleration" conditions.
[0020] As a preferred embodiment of the present invention, the output control quantity calculation formula of the asymmetric smooth switching strategy is as follows: the output control quantity is equal to the physical throttle command at the fault trigger time plus the difference between the virtual throttle command and the physical throttle command at the fault trigger time, multiplied by a natural exponential decay correction term including the asymmetric switching coefficient; when there is an acceleration trend, the asymmetric switching coefficient takes a first preset value, and when there is a deceleration trend, the asymmetric switching coefficient takes a second preset value, and the first preset value is less than the second preset value.
[0021] The beneficial effects of this invention are as follows:
[0022] First, traditional physical information neural networks often suffer from inversion divergence when operating at highly nonlinear high-altitude low-speed or transonic maneuvering boundaries due to conflicts between simplified physical equations and the actual complex flow fields. This invention introduces a physical information neural network based on dynamic weighting of the flight envelope. By dynamically scheduling weights, it achieves an unexpectedly significant improvement in inversion convergence speed at extreme envelope transitions, completely eliminating the risk of system crashes or divergence.
[0023] Secondly, the physical response times of different engine sensors vary greatly (e.g., the thermal inertia of exhaust temperature is extremely long). Without alignment, this can lead to incorrect virtual throttle spikes in the model during transient high-thrust maneuvers. This invention designs a parameter phase alignment mechanism based on the time-delay cross-correlation matrix. This alignment mechanism eliminates the inversion distortion caused by this temporal misalignment, ensuring high-fidelity recognition of the transition state.
[0024] Third, in the instant a sudden failure leads to system takeover, a conventional linear transition can easily trigger engine surge (especially when suddenly increasing thrust from low to high). This invention incorporates the characteristics of compressor surge boundaries, employing a slow transition during acceleration and a fast transition during deceleration. This improvement, highly coordinated with the high-frequency recognition of the front-end neural network, ensures a smooth transition of control while perfectly avoiding the fatal risk of compressor stall induced by the moment of accelerated takeover, greatly enhancing the underlying flight safety redundancy.
[0025] Fourth, as engines age, their parameters drift significantly under the same thrust. This invention introduces a thermodynamic degradation adaptive baseline offset based on exhaust temperature margin, explicitly extracting aging characteristics as a baseline offset factor to compensate for in the model. This ensures that the same set of fixed-weight neural networks maintains consistently high inversion accuracy on both older aircraft about to undergo major overhauls and newly manufactured aircraft, reducing software maintenance costs for airlines. Attached Figure Description
[0026] Figure 1 This is a flowchart of a virtual throttle recognition method for aircraft engines based on multi-source parameter fusion. Detailed Implementation
[0027] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0028] It should be understood that, and also noted, in the embodiments, the functions / actions may appear in a different order than those shown in the figures. For example, depending on the functions / actions involved, they may actually be performed substantially concurrently, or sometimes the two figures shown consecutively may be performed in reverse order.
[0029] The full authority digital electronic control system (FADEC) of an aero-engine acts as its brain, and the thrust throttle (TLA) signal is its most important external command input. This embodiment takes a certain type of high-bypass commercial turbofan engine as an example to illustrate in detail the entire process from sensor data acquisition to the final realization of virtual throttle safe takeover.
[0030] Step 1: Real-time acquisition of multi-source feature parameters (including N1, N2, EGT, Wf, Ma, Alt).
[0031] like Figure 1 As shown in step 1, within each FADEC operation cycle (e.g., 20 milliseconds), the system acquires the engine status in real time via the onboard ARINC429 or AFDX bus, and collects multi-source characteristic parameters (including N1, N2, EGT, Wf, Ma, and Alt) in real time. The selected parameters are not blindly piled up, but are strictly composed of strongly correlated aerodynamic and thermodynamic variables.
[0032] Low-pressure rotor speed N1: directly characterizes the total thrust level of the engine;
[0033] High-pressure rotor speed N2: characterizes the operating status and energy generation of the core engine (compressor, combustion chamber, high-pressure turbine);
[0034] Exhaust temperature EGT: characterizes the thermal load state of turbine blades;
[0035] Fuel flow rate Wf: the direct physical cause and energy input source of thrust variation;
[0036] Flight Mach number Ma and flight altitude Alt: Define the intake ram air boundary and the ambient air density boundary where the engine is located.
[0037] Step 2: Dynamically adjust the length of the sliding time window based on the absolute rate of change of Wf.
[0038] like Figure 1 As shown in step 2, the system not only dynamically adjusts the length of the sliding time window based on the absolute rate of change of Wf, but also calculates the cross-correlation matrix to extract the time delay and perform forward phase shift alignment.
[0039] Due to differences in sensor measurement principles and physical medium properties, signals exhibit significant asynchrony. Fuel flow rate Wf is controlled by a hydraulic-mechanical servo valve, with a response delay typically only 30-50 milliseconds; rotor speeds N1 and N2, due to their large metal rotational inertia, have a response delay of approximately 300-800 milliseconds; exhaust temperature EGT, constrained by thermocouple thermal capacity characteristics and heat conduction from high-temperature airflow to the turbine's metal structure, can exhibit a response hysteresis of up to 1.5 to 2 seconds.
[0040] If all currently read parameters are directly packaged and input into the neural network at time t, the network will be confused due to the causal misalignment that the current rotational speed is actually caused by the fuel flow rate one second ago, leading to violent output oscillations. Therefore, this invention designs a time-delay cross-correlation matrix processing mechanism. The system maintains a cross-correlation matrix of length T. w A sliding time window (e.g., a historical data array within 5 seconds) is used. Using the undelayed fuel flow rate Wf as a baseline, the cross-correlation function R between the N1, N2, EGT sequences and the Wf sequence is calculated. xy (τ):
[0041]
[0042] By finding R xy (τ) The time delay τ corresponding to the absolute maximum value i The system learns how many milliseconds N1 lags behind Wf and how many milliseconds EGT lags behind Wf under the current operating condition. Then, the system extracts the N1 data from time (t-τ1) and the EGT data from time (t-τ2) in the historical cache, aligns and assembles them with the Wf data at the current time t. This ensures that the multi-source parameters of the input model are at the same absolute cross-section in terms of physical causality.
[0043] Meanwhile, considering that pilots may suddenly increase or decrease the throttle during emergency avoidance, the sliding time window T...w The length is set to adaptive adjustment. When the absolute rate of change of fuel flow Wf |dWf / dt| exceeds the preset rate of change threshold (e.g., exceeding 6% per second), the system actively reduces T. w To enhance the system's sensitivity to capturing transient high-frequency change characteristics, up to 1 second.
[0044] Step 3: Extract the current cruise steady-state EGT and compare it with the factory reference to calculate the attenuation.
[0045] like Figure 1 As shown in step 3, the system first extracts the current cruise steady-state EGT and compares it with the factory reference to calculate the attenuation. Then, it generates a thermodynamic degradation baseline offset factor to perform normalization correction on the input parameters.
[0046] Aero engines exhibit significant life-cycle performance degradation. As on-wing flight cycles increase, compressor blades wear and accumulate deposits, leading to a decrease in high-pressure turbine aerodynamic efficiency. To maintain the same thrust (same N1 speed) as a new aircraft during cruise, the controller must instruct the fuel pumps to inject more fuel, resulting in a significantly higher exhaust gas temperature (EGT) compared to a new aircraft (i.e., reduced EGT margin).
[0047] If N1 and EGT with aging drift are directly input into the trained model, the model will incorrectly identify the current state as high throttle, leading to inversion errors. Therefore, this system automatically records the current EGT value during the steady-state phase of each long-haul cruise and compares it with the factory baseline EGT stored in the FADEC non-volatile memory to calculate the exhaust temperature margin decay ΔEGT over the entire lifecycle. margin .
[0048] This attenuation is converted into a thermodynamic degradation baseline offset factor. During the data preprocessing normalization stage, the system subtracts this offset factor from the real-time acquired EGT, mathematically restoring the data of a late-stage aging engine to the form of a healthy new engine. This compensation mechanism ensures that the neural network deployed on the onboard computer does not require any retraining during the engine's ten years of service, maintaining extremely high throttle command inversion accuracy at all times.
[0049] Step 4: Obtain the current Ma and Alt parsing flight envelope region features.
[0050] like Figure 1 As shown in step 4, the system first obtains the current Ma and Alt parsing flight envelope region features, then the flight envelope scheduler dynamically adjusts the PINN data prediction and physical constraint loss weights, and finally dynamically weights the PINN forward inference inversion and performs hard boundary limiting to obtain the virtual throttle.
[0051] The processed parameters are then fed into the Physical Information Neural Network (PINN). PINN is a cutting-edge AI architecture that combines the flexibility of data fitting with the rigor of physical laws. The network ontology uses a multilayer perceptron, with the input layer containing six nodes: N1, N2, EGT, Wf, Ma, and Alt (after correction), and the output layer containing one node representing the virtual throttle command percentage.
[0052] In conventional machine learning, the loss function only includes the mean squared error of the data prediction (Loss). data This invention embeds a physical constraint layer (Loss) characterizing the engine mechanism into the feedback correction of network training and real-time inference. phy The physical constraint layer focuses on calculating two residuals: one is the rotor dynamics equation residual, which is the polar moment of inertia multiplied by the angular acceleration, and must approximate the difference between the turbine output torque and the compressor consumed torque; the other is the combustion chamber energy conservation residual, which is the fuel calorific value must match the enthalpy increase.
[0053] The total loss function is defined as: Loss = w d Loss data +w p Loss phy .
[0054] This invention introduces a flight envelope scheduler. In conventional subsonic cruise or low-altitude, low-speed regions, the engine flow field is stable, and the mechanistic equations are extremely accurate. The scheduler improves the efficiency of the flight envelope. p Weights are used to constrain the network using strict physical laws to prevent overfitting of the output. When the aircraft enters high angle-of-attack maneuvers, the transonic transition zone, or the extremely high-altitude low-density region above 30,000 feet, the air intake is distorted, and the complex shock waves cause the traditional simplified thermodynamic equations to fail. In this case, forcibly constraining the PINN model will lead to non-convergence. At this time, the envelope scheduler automatically reduces w based on the Mach number Ma and altitude Alt. p The coefficient (reduces physical trust) while increasing w. d Coefficient (enhancing data reliability). This dynamic weighting mechanism, which switches the brain's decision-making logic in real time based on airspeed and altitude, solves the industry pain point of AI easily diverging at the edge of a harsh flight envelope.
[0055] Before proceeding to the next step, the network output must pass through the envelope hard boundary limiter. The system looks up the minimum idle protection value and maximum continuous thrust protection value under the current flight attitude and atmospheric conditions, and forcibly clamps the inversion results within this safe range.
[0056] Step 5: Filter and calculate the difference between the physical throttle and the virtual throttle to assess whether it is consistently greater than the fault tolerance threshold.
[0057] like Figure 1As shown in step 5 and subsequent fault-tolerant decision branches, during normal flight missions, the physical throttle sensor functions normally, and the generated virtual throttle command runs silently in the background as a shadow system. The system calculates the absolute difference between the physical throttle command and the virtual throttle command in real time. To avoid false alarms triggered by transient data glitches caused by lightning or strong electromagnetic radiation, a first-order hysteresis filter is connected in series with the residual evaluation function.
[0058] The system has a defined confirmation time window (e.g., 250 milliseconds). If the filtered absolute difference consistently exceeds the fault tolerance threshold and the confirmation time window, the physical sensor is deemed to have suffered irreversible mechanical breakage or electrical short-circuit failure. This fault tolerance threshold is also dynamic, tightening to 3% in stable airflow regions and widening to 6% in transonic turbulence regions to prevent false triggering.
[0059] Once the physical signal loss is confirmed, engine control must be smoothly transferred to the virtual throttle immediately. However, sudden changes in thrust command are extremely dangerous. If the current virtual throttle is calculated to be 85%, while the physical throttle, which was stuck before the failure, was at 40%, instantly raising the command to 85% will cause a sudden surge in fuel flow. Because the rotor speed cannot keep up, the excess fuel leads to a sharp increase in combustion chamber pressure, which can easily exceed the compressor's exhaust capacity, causing severe compressor surge or even engine stall.
[0060] Therefore, such as Figure 1 As shown in the trend prediction and branch switching, this invention designs an asymmetric smooth switching strategy. When takeover is triggered, the system first analyzes the thrust change trend:
[0061] If the virtual throttle command > the physical throttle command (indicating an acceleration trend requiring increased thrust), the system calls the first exponential smoothing function, setting the asymmetric switching coefficient k1 = 0.4 (small slope, large time constant). The command is executed according to the formula TLA. cmd (t)=TLA p0 +(TLA v -TLA p0 )⋅(1-e -0.4(t-t0) The compressor accelerates slowly. This allows sufficient acceleration time for the compressor to increase the surge margin and prevent surge.
[0062] If the virtual throttle command is less than or equal to the physical throttle command (indicating a deceleration trend, such as when the aircraft is approaching and needs to immediately reduce throttle), reducing fuel will not cause surge. The system calls the second exponential smoothing function, setting the asymmetric switching coefficient k2=3.0 (large slope, small time constant). The command quickly drops to the virtual calculated value. This ensures that the aircraft can obtain a deceleration response without delay during critical moments such as approach and landing, achieving a fast response.
[0063] This invention encompasses a complete closed loop, from bottom-level physical signal alignment, mid-level degradation mechanism compensation, core-level dynamic AI inference, to top-level asymmetric safety fault-tolerant control, forming a highly robust virtual redundancy safety barrier for aero-engines.
[0064] This invention is not limited to the above-described optional embodiments. Anyone can derive other various forms of products under the guidance of this invention. However, regardless of any changes made in their shape or structure, any technical solution that falls within the scope of the claims of this invention shall be protected by this invention.
Claims
1. A method for identifying the virtual throttle of an aircraft engine based on multi-source parameter fusion, characterized in that, Includes the following steps: Step 1: Real-time acquisition of multi-source characteristic parameters of the aircraft engine, which consist of low-pressure rotor speed N1, high-pressure rotor speed N2, exhaust temperature EGT, fuel flow rate Wf, flight Mach number Ma, and flight altitude Alt. Step 2: Extract time series of the multi-source feature parameters, calculate the phase hysteresis of the low-pressure rotor speed N1, high-pressure rotor speed N2 and exhaust temperature EGT relative to the fuel flow rate Wf using the time delay cross-correlation matrix, and perform time delay phase alignment processing on the data in the historical sliding window to obtain the aligned multi-source feature parameters. Step 3: Calculate the exhaust temperature margin decay over the entire life cycle based on the engine's historical operating cycle number, generate a thermodynamic degradation baseline offset factor, and use this thermodynamic degradation baseline offset factor to normalize and correct the aligned multi-source characteristic parameters to obtain the corrected multi-source characteristic parameters. Step 4: Input the corrected multi-source feature parameters into the pre-trained dynamic weighted physical information neural network model, and inversely calculate the current virtual throttle command; wherein, the weights of the loss function of the dynamic weighted physical information neural network model are dynamically adjusted in real time according to the flight envelope region where the current flight Mach number Ma and the flight altitude Alt are located; Step 5: Perform cross-arbitration comparison between the virtual throttle command and the physical throttle command output by the physical throttle sensor. When the difference exceeds the set fault tolerance threshold, trigger a physical sensor fault alarm and start the asymmetric smooth switching strategy to switch the engine control command source from the physical throttle command to the virtual throttle command.
2. The method for virtual throttle recognition of aircraft engines based on multi-source parameter fusion according to claim 1, characterized in that, The time delay cross-correlation matrix construction process in step 2 is as follows: taking the fuel flow rate Wf as the reference signal, calculate the cross-correlation function between the low-pressure rotor speed N1, the high-pressure rotor speed N2, and the exhaust temperature EGT and the fuel flow rate Wf within the sliding time window, find the time delay corresponding to the maximum value of the cross-correlation function, and forward shift and align each parameter according to the time delay.
3. The method for virtual throttle recognition of aircraft engines based on multi-source parameter fusion according to claim 1, characterized in that, The thermodynamic degradation baseline offset factor in step 3 is determined by calculating the increment of the steady-state value of the exhaust temperature under the current cruise state, with the factory reference exhaust temperature baseline as the zero point. This thermodynamic degradation baseline offset factor is superimposed on the exhaust temperature feature node of the input layer of the dynamic weighted physical information neural network model in the form of a bias constant.
4. The method for virtual throttle recognition of aircraft engines based on multi-source parameter fusion according to claim 1, characterized in that, The dynamic weighted physical information neural network model in step 4 consists of a data-driven multilayer perceptron network and a physical constraint layer embedded with the engine aerodynamic thermodynamic equations; the physical constraint layer includes residual terms of the rotor dynamics equations and residual terms of the combustion chamber energy conservation equations.
5. The method for virtual throttle recognition of aircraft engines based on multi-source parameter fusion according to claim 4, characterized in that, The total loss function of the dynamic weighted physical information neural network model is composed of two weighted parts: data prediction error and physical constraint residual. The dynamic weighted physical information neural network model has a built-in flight envelope scheduler. When the flight envelope scheduler determines that the aircraft is in the transonic transition region, it increases the weight coefficient of the data prediction error and decreases the weight coefficient of the physical constraint residual.
6. The method for virtual throttle recognition of aircraft engines based on multi-source parameter fusion according to claim 1, characterized in that, The cross-arbitration comparison process in step 5 includes: setting a residual evaluation function based on first-order hysteresis filtering; when the absolute difference between the physical throttle command and the virtual throttle command is continuously detected to be greater than the fault tolerance threshold for a period of time exceeding the confirmation time window, it is determined that the physical throttle sensor has failed; wherein, the fault tolerance threshold is dynamically relaxed or tightened according to the flight Mach number Ma.
7. The method for virtual throttle recognition of aircraft engines based on multi-source parameter fusion according to claim 1, characterized in that, The asymmetric smooth switching strategy in step 5 is as follows: based on the engine compressor surge margin protection logic, the thrust change trend during control command source switching is determined; if the virtual throttle command is greater than the physical throttle command, it is determined that there is an acceleration trend, and a first exponential smoothing function is used for slow switching; if the virtual throttle command is less than or equal to the physical throttle command, it is determined that there is a deceleration trend, and a second exponential smoothing function is used for fast switching; wherein the time constant of the first exponential smoothing function is greater than the time constant of the second exponential smoothing function.
8. The method for virtual throttle recognition of aircraft engines based on multi-source parameter fusion according to claim 1, characterized in that, After obtaining the virtual throttle command in step 4 and before executing step 5, a boundary limitation step is also included: using the current flight Mach number Ma and flight altitude Alt to look up the table to obtain the minimum idle throttle limit value and the maximum continuous thrust throttle limit value of the engine in the current flight state, and perform upper and lower limit processing on the virtual throttle command obtained by inversion.
9. The method for virtual throttle recognition of aircraft engines based on multi-source parameter fusion according to claim 2, characterized in that, The length of the sliding time window is adaptively adjusted according to the absolute rate of change of fuel flow Wf; when the absolute rate of change of fuel flow Wf is detected to exceed the rate of change threshold, the length of the sliding time window is reduced.
10. The method for identifying virtual throttle of an aircraft engine based on multi-source parameter fusion according to claim 7, characterized in that, The formula for calculating the output control quantity of the asymmetric smooth switching strategy is: the output control quantity equals the physical throttle command at the fault trigger time plus the difference between the virtual throttle command and the physical throttle command at the fault trigger time, multiplied by the natural exponential decay correction term that includes the asymmetric switching coefficient. When there is an acceleration trend, the asymmetric switching coefficient takes a first preset value; when there is a deceleration trend, the asymmetric switching coefficient takes a second preset value, and the first preset value is less than the second preset value.
Citation Information
Patent Citations
Gas control apparatus, in particular for turbine engine test bed
CA2492157A1
Driving path planning method for unmanned vehicle in airport
CN122531242A
