Airborne fusion reconstruction method and system for aero-engine fault temperature and pressure sensor signals

Abstract

The invention discloses an aero-engine fault temperature and pressure sensor signal airborne fusion reconstruction method and system, and the method comprises the steps: introducing an actual measurement sensor parameter, simplifying a complete part-level model, building a model-based temperature and pressure airborne reconstruction sub-module, and achieving the model-based sensor signal reconstruction; a multi-dimensional training set is constructed according to flight conditions and actual measurement sensor data, a temperature and pressure airborne reconstruction sub-module under a Transform architecture is trained and established, and synchronous deep learning reconstruction of temperature and pressure signals is realized; by analyzing the two-parameter precision and time consumption of the two reconstruction methods, joint evaluation switching logic based on the two-channel error change rate and the maximum absolute error in the sliding window is designed, and smooth fusion of multi-parameter reconstruction signals is realized by adopting a two-channel independent weighting coefficient calculation strategy. According to the method, a high-precision temperature and pressure reconstruction signal can be obtained by fusing mechanism modeling and a deep learning method, and the fault tolerance performance of an aero-engine to a temperature and pressure sensor fault is effectively improved.

Description

Technical Field

[0001] The present invention belongs to the field of aero-engine modeling and simulation, and particularly relates to a method and system for airborne fusion reconstruction of signals of aero-engine fault temperature and pressure sensors. Background Art

[0002] Sensors are an important part of the aero-engine control system and the main channel for obtaining the operating state information of the aero-engine. The temperature and pressure of each section of the engine, as key parameters characterizing the gas path state of the engine, the accuracy and stability of their sensor signals play a decisive role in engine control, surge protection, and the safe operation of the engine. However, during the actual operation process, the sensors are long-term in a harsh environment of high-temperature gas scouring, extreme pressure fluctuations, and strong vibrations, and the failure rate is significantly higher than that of other airborne equipment. When the temperature or pressure sensor fails, the control system cannot accurately obtain the key parameters of each section, seriously threatening the safe operation of the engine. Therefore, realizing the signal reconstruction of airborne fault temperature and pressure sensors is the core requirement for improving the fault tolerance control ability of the engine. Compared with the hardware redundancy technology, the analytical redundancy technology does not need to increase the number of hardware, can avoid the problems of increased engine weight and installation space limitations, and is a better solution for realizing the collaborative reconstruction of dual-parameter temperature and pressure signals.

[0003] Traditional analytical redundancy technologies mainly use component-level models or shallow neural networks to realize signal reconstruction. However, the complete component-level model involves the coupled calculations of multiple components such as the combustion chamber and turbine, with high complexity and poor real-time performance, and it is difficult to meet the requirements of airborne real-time reconstruction; while the shallow neural network has the defect of insufficient fitting ability for systems with a high degree of non-linearity. Aiming at the above problems, the present invention designs a temperature and pressure airborne reconstruction sub-module based on the model and a deep reconstruction sub-module based on the Transformer architecture. Through the dual-channel error joint evaluation and matrix fusion strategy, while ensuring the reconstruction accuracy, it meets the requirements of airborne real-time performance, and forms a signal fusion reconstruction method for temperature and pressure sensors. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to establish a temperature and pressure airborne reconstruction sub-module based on the model and a temperature and pressure airborne reconstruction sub-module under the Transformer architecture respectively in view of the defects of the background art, and at the same time design the module switching logic, and propose a method and system for airborne fusion reconstruction of signals of aero-engine fault temperature and pressure sensors. The present invention adopts the following technical solutions to solve the above technical problems:

[0005] A method for airborne fusion reconstruction of signals of aero-engine fault temperature and pressure sensors includes the following steps:

[0006] (1) Establish a first reconstruction module based on a physical model, and obtain a first temperature reconstruction value and a first pressure reconstruction value through the calculation of aero-engine aerodynamic parameters;

[0007] (2)Construct a second reconstruction module based on deep learning, process the sensor time-series data through a trained neural network model, and output the second temperature reconstruction value and the second pressure reconstruction value;

[0008] (3)Perform real-time error evaluation on the outputs of the first reconstruction module and the second reconstruction module, and dynamically adjust the fusion weights of the two modules according to the evaluation results;

[0009] (4)Based on the adjusted fusion weights, perform weighted fusion on the reconstruction values of the two modules, and output the final temperature reconstruction signal and pressure reconstruction signal;

[0010] Among them, the first reconstruction value obtained in step (1) serves as the reference input for the error evaluation in step (3), the second reconstruction value obtained in step (2) serves as the comparison input for the error evaluation in step (3), and the output result of step (3) directly determines the weight allocation in step (4).

[0011] Preferably, step (1) specifically includes:

[0012] (1.1)Collect the parameters of the fan outlet temperature, fan outlet pressure, compressor outlet pressure, turbine outlet pressure, and rotor speed;

[0013] (1.2)Input the collected parameters into a pre-established pneumatic thermodynamics model, and sequentially calculate the fan flow rate and the core flow rate;

[0014] (1.3)Based on the calculated flow parameters, calculate the first temperature reconstruction value and the first pressure reconstruction value through the compressor characteristic curve.

[0015] Preferably, step (2) specifically includes:

[0016] (2.1)Construct a training data set including engine operating parameters and sensor time-series data;

[0017] (2.2)Design a neural network architecture with dual output branches, where the first branch is used for temperature reconstruction and the second branch is used for pressure reconstruction;

[0018] (2.3)Use the training data set to train the neural network to obtain a deep learning model that can synchronously output the second temperature reconstruction value and the second pressure reconstruction value.

[0019] Preferably, step (3) specifically includes:

[0020] (3.1)Set a sliding time window, and calculate the temperature error and pressure error output by the first reconstruction module and the second reconstruction module in real time;

[0021] (3.2) Calculate the error change rate between adjacent time steps and analyze the error change trend;

[0022] (3.3) When the error of any parameter exceeds the preset threshold or the error change rate continues to increase, generate a reconstruction module switching instruction;

[0023] (3.4) According to the error evaluation result, calculate the real-time fusion weights of the first reconstruction module and the second reconstruction module.

[0024] Preferably, step (4) specifically includes:

[0025] (4.1) Receive the fusion weights calculated in step (3);

[0026] (4.2) Perform weighted averaging on the first temperature reconstruction value and the second temperature reconstruction value according to the weights to obtain the final temperature reconstruction signal;

[0027] (4.3) Perform weighted averaging on the first pressure reconstruction value and the second pressure reconstruction value according to the weights to obtain the final pressure reconstruction signal;

[0028] (4.4) Output the fused temperature reconstruction signal and pressure reconstruction signal to the engine control system.

[0029] The present invention also provides an airborne reconstruction system for aeroengine fault temperature and pressure sensors, including:

[0030] A data acquisition unit for real-time acquisition of engine operation parameters and sensor data;

[0031] A first reconstruction unit connected to the data acquisition unit for performing the functions of the first reconstruction module;

[0032] A second reconstruction unit connected to the data acquisition unit for performing the functions of the second reconstruction module;

[0033] A fusion control unit respectively connected to the first reconstruction unit and the second reconstruction unit for performing the error evaluation and signal fusion functions;

[0034] Wherein, the output of the data acquisition unit is simultaneously input into the first reconstruction unit and the second reconstruction unit, the outputs of the first reconstruction unit and the second reconstruction unit are jointly input into the fusion control unit, and the output of the fusion control unit is used as the final output of the system.

[0035] Preferably, the fusion control unit includes:

[0036] An error calculation module for calculating the error index of the outputs of the two reconstruction units;

[0037] A weight decision module connected to the error calculation module for determining the fusion weights according to the error index;

[0038] A signal fusion module, connected to a weight decision module, for performing weighted fusion calculation;

[0039] Among them, the error calculation module receives the outputs of two reconstruction units, the weight decision module generates weight parameters according to the error calculation results, and the signal fusion module completes signal fusion according to the weight parameters.

[0040] The present invention also provides an aeroengine control system, including:

[0041] A sensor group, used for monitoring the temperature and pressure of various parts of the engine;

[0042] A control unit, used for adjusting the engine operation parameters according to the monitoring data;

[0043] The described airborne reconstruction system, connected to the sensor group and the control unit;

[0044] Among them, when the sensor group fails, the reconstruction signal output by the airborne reconstruction system automatically replaces the output of the faulty sensor to ensure that the control unit continuously obtains reliable monitoring data.

[0045] The present invention also provides a computer-readable storage medium, storing a computer program, and when the program is executed by a processor, the method is implemented.

[0046] The present invention also provides an electronic device, including a memory and a processor, the memory stores a computer program, and when the processor executes the program, the method is implemented.

[0047] Beneficial effects: Through the design of parallel operation of the physical model reconstruction and deep learning reconstruction dual channels, when one party has an increase in error or fails, it can be automatically switched to the other party, and the smoothness of the signal during the switching process is ensured through weight matrix calculation, greatly improving the reliability of sensor signal reconstruction; innovatively integrating the high-precision characteristics of the physical model and the fast response advantages of the deep learning model, while ensuring accuracy, the real-time performance of signal reconstruction is improved; through the sliding window error analysis and dynamic weight adjustment mechanism, the system can automatically optimize the fusion strategy according to the actual working conditions. Description of the Drawings

[0048] Figure 1 It is a schematic diagram of the airborne fusion reconstruction method for the fault temperature and pressure sensor signals of an aeroengine;

[0049] Figure 2 It is a schematic diagram of the calculation of the reconstructed total temperature of the compressor outlet gas path;

[0050] Figure 3 It is a schematic diagram of the calculation of the reconstructed total pressure of the low-pressure turbine outlet gas path;

[0051] Figure 4 It is the network structure diagram of the airborne reconstruction sub-module of the Transformer architecture;

[0052] Figure 5 It is the comparison chart of the accuracy and time consumption of two reconstruction modules;

[0053] Figure 6 It is the dynamic simulation diagram of the reconstructed sensor signal;

[0054] Figure 7 It is the schematic diagram of the time consumption of the reconstructed sensor signal. Specific implementation manners

[0055] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0056] Embodiment 1

[0057] The specific implementation manner of the present invention takes a dual-rotor turbofan engine as the research object, Figure 1 It is the flowchart of the airborne fusion reconstruction method for the fault temperature and pressure sensor signals of an aeroengine. The design of this signal reconstruction scheme includes the following steps:

[0058] Step 1), introduce the measured sensor parameters to simplify the complete component-level model, establish the corresponding model-based temperature and pressure airborne reconstruction sub-module for different reconstructed sensor signal types, and realize the model-based sensor signal reconstruction.

[0059] Step 2), construct a multi-dimensional training set according to the flight conditions and the measured sensor data, train and establish the temperature and pressure airborne reconstruction sub-module under the Transformer architecture, design an output layer that matches the dual-physical quantity measurement characteristics, and realize the synchronous deep learning reconstruction of the temperature and pressure signals;

[0060] Step 3), analyze and compare the dual-parameter accuracy and time consumption of the two signal reconstruction methods, design a switching logic based on the joint evaluation of the change rate of the dual-channel error and the maximum absolute error within the sliding window, and design a dual-channel independent weighted coefficient calculation strategy for the temperature and pressure signal characteristics, and realize the smooth fusion of the multi-parameter reconstructed signals through the weight matrix operation.

[0061] The detailed steps of step 1) are as follows:

[0062] Step 1.1), introduce the total temperature and total pressure measurement values T 22,real , P 22,real at the fan outlet, the total pressure measurement value P 3,real at the compressor outlet, the total pressure measurement value P 6,real at the low-pressure turbine outlet, the high and low pressure rotor speeds NH real , and reconstruct the compressor outlet pressure sensor T3.

[0063] Taking the dual-rotor mixed-exhaust turbofan engine as the research object, the definitions of each cross-section are as follows in the table.

[0064] Table 1 Definitions of Each Cross-Section of the Turbofan Engine

[0065]

[0066]

[0067] When reconstructing the compressor outlet pressure sensor T3, the low-pressure turbine outlet pressure P 6,real is introduced. According to the static pressure conservation at the outlet of the bypass duct and the outlet of the low-pressure turbine, the gas flow rate through the bypass duct is calculated. In summary, the fan outlet temperature sensor T 22,real and the fan outlet pressure sensor P 22,real are selected and introduced, the compressor outlet pressure sensor P 3,real , the low-pressure turbine outlet pressure P 6,real , the low-pressure rotor speed NL real and the high-pressure rotor speed NH real are used for signal reconstruction.

[0068] The fan inlet temperature T2 and the fan inlet pressure P2 are calculated according to the flight altitude and Mach number. After introducing the fan outlet pressure sensor P 22,real and the compressor outlet pressure sensor P 3,real , the pressure ratios of the fan and compressor components can be calculated. Then, according to the low-pressure rotor speed, the fan component pressure ratio, and the inlet and outlet temperature and pressure, the gas flow rate of the fan component is calculated through the fan component model. According to the low-pressure turbine outlet pressure P 6,real and the bypass duct component, the partial flow rate through the bypass duct is calculated, and then the gas flow rate of the compressor component is calculated. The calculated pressure ratio, inlet section temperature, pressure, and gas flow rate of the compressor component are input into the aerodynamic and thermodynamic model of the compressor component to calculate the measured value T3 of the compressor outlet pressure sensor to be reconstructed, as shown in the following formula:

[0069] T3 = f(T 22 , P 22 , W 22 , π comp , NH)

[0070] When reconstructing the signal of the compressor outlet temperature sensor using the mainstream parameter model, iterative solution is not required, which greatly reduces the time consumed for signal reconstruction and improves the real-time performance. The reconstruction gas path calculation process is as Figure 2 shown.

[0071] Step 1.2), introduce the measured values of the total temperature and total pressure sensors at the inlet and outlet of the compressor T 22,real , P 22,real , T3,real , P 3,real , the measured value T of the total temperature at the low-pressure turbine outlet 6,real , the measured value NL of the low-pressure rotor speed sensor real and the measured value NH of the high-pressure rotor speed sensor real , reconstruct the low-pressure turbine outlet pressure sensor P6.

[0072] Calculate the total temperature T2 and total pressure P2 at the fan inlet based on the flight altitude and Mach number, introduce the total pressure sensor P at the fan outlet 22,real and the total pressure sensor P at the compressor outlet 3,real to calculate the pressure ratios of the fan and compressor components, the same as Equation (3.3). Based on the low-pressure rotor speed, the fan component pressure ratio, and the total temperature and total pressure at the inlet and outlet, calculate the gas flow rate of the fan component through the fan component model. Based on the high-pressure rotor speed, the compressor component pressure ratio, and the total temperature and total pressure at the inlet and outlet of the compressor, calculate the gas flow rate of the compressor component through the compressor component model.

[0073]

[0074] Calculate the gas flow rate in the bypass duct according to the mass continuity equation, as shown in the following formula:

[0075] W 13 = W 22 - W 25

[0076] Given the inlet flow rate, total temperature, and total pressure in the bypass duct, calculate the static pressure at the bypass duct outlet through the bypass duct aerodynamics model, which is balanced with the static pressure at the low-pressure turbine outlet, as shown in the following formula:

[0077] P S6 = P S16 = f(W 13 , P 13 , T 13 )

[0078] For the low-pressure turbine component, the flow continuity in the core duct indicates the outlet flow rate of the low-pressure turbine. The cross-sectional area at the low-pressure turbine outlet is a non-variable design parameter, from which the total pressure at the low-pressure turbine outlet cross-section can be deduced, as shown in the following formula:

[0079]

[0080] Among them, R is the gas constant, and A6 is the cross-sectional area at the low-pressure turbine outlet.

[0081] When reconstructing the signal of the total pressure sensor at the low-pressure turbine outlet using the mainstream parameter model, there is no need for iterative solution, which greatly reduces the time consumed for reconstructing the signal and improves the real-time performance. The reconstruction gas path calculation process is as Figure 3 shown.

[0082] The detailed steps of step 2) are as follows:

[0083] Step 2.1), traverse the fuel flow within the control plan under different envelope operating points, synchronously collect measurable sensor signal data, perform multi-parameter joint data preprocessing and data set partitioning, and construct an analytical redundancy deep learning network model data set for temperature and pressure dual-physical quantity reconstruction.

[0084] Traverse the fuel flow from low to high within the control plan under different envelope operating points, determine model input parameters such as the adjustable area according to the control plan, and synchronously collect all measurable sensor data. During the traversal process, the fuel flow steps up randomly by an amplitude of {2%, 3%, 4%}, and the fuel remains unchanged after each step until the model reaches a steady state operation. Continuously collect the temperature and pressure dual-parameter data during the dynamic process and at steady state until the fuel flow reaches the maximum value of the control plan.

[0085] After the data collection is completed, for the magnitude differences between the values of different types of sensors, use the min-max normalization method to normalize the collected data to avoid the pressure parameter with a larger value from dominating the network training. The calculation formula is as follows:

[0086]

[0087] Among them, X norm represents the normalized data, X represents the data before normalization, and X min represents the minimum value of the data before normalization, and X max represents the maximum value of the data before normalization.

[0088] When constructing the data set, use a sliding window to partition the data. The feature data includes the measurable sensor time series data, flight altitude, Mach number, and fuel flow at the current and the previous 9 moments; the label data is set as the measured values of the temperature and pressure dual channels at the current moment. Use the sliding window method to extract time series features to ensure the time series correlation of the temperature and pressure signal reconstruction.

[0089] Step 2.2), design an analytical redundancy deep learning network model based on the Transformer network architecture, construct a temperature and pressure dual-parameter parallel output layer, and realize the dual-channel signal synchronous data-driven reconstruction adapted to the multi-parameter characteristics of the temperature and pressure sensors.

[0090] The Transformer encoder architecture is adopted to process the input and output data. Its self-attention mechanism can effectively capture the long-range dependencies of the sensor sequence through parallel computing, which is applicable to the processing of long sequences of flight parameters while improving the computing efficiency. The network input includes fuel flow, afterburner fuel flow, flight altitude, Mach number, and measurable parameters other than the target sensor, and the output is the true value of the temperature and pressure sensors at the current moment.

[0091] The model architecture is as Figure 4 shown, and the core modules include the multi-head self-attention layer and the parameter regression layer.

[0092] The multi-head self-attention layer is composed of h self-attention units in parallel. Each unit converts the input feature vector X' (including positional encoding) into Q (query), K (key-value), and V (value) matrices through a learnable weight matrix, as shown in the following formula:

[0093] Q = W Q X

[0094] K = W K X

[0095] V = W V X

[0096] Among them, W Q , W K , W V are all weight matrices, which belong to learnable parameters and are updated with the backpropagation algorithm during the training of the network. X is the feature vector matrix obtained by encoding the original input with positional encoding.

[0097] Next, calculate the attention weights according to the following formula, as shown in the following formula:

[0098]

[0099] Among them, d k is the number of columns of the Q and K matrices, which is calculated by the following formula

[0100]

[0101] Among them, h represents the number of self-attention layers.

[0102] The parameter regression layer first inputs the equal-dimensional feature matrix output by the encoder into the fully connected layer for non-linear transformation, as shown in the following formula:

[0103] f3(x) = W3x + b3

[0104] Among them, W and b are both learnable parameters and are updated with the backpropagation algorithm during the training of the network.

[0105] Then, feature dimensionality reduction is achieved through average pooling operations with step size s and window k to ensure that the output dimension matches the dual-parameter prediction requirements, as shown in the following formula:

[0106] The calculation formula for the pooling layer is as follows:

[0107]

[0108] where i is the index after pooling, x i·s+j is an element in the input vector, k is the size of the pooling window, and s is the step size of the pooling.

[0109] The network training uses the flight conditions and sensor measurement data within a time-sliding window to complete the synchronous reconstruction of the temperature and pressure dual-channel signals through a data-driven approach. This architecture omits the decoder module and directly realizes the mapping from the input features to the dual-parameter output through the encoder.

[0110] The detailed steps of step 3) are as follows:

[0111] Step 3.1), conduct a comparative evaluation of the dual-parameter accuracy and calculation time consumption between the dual-channel output signals of the airborne reconstruction sub-module of the Transformer architecture and the model-based airborne reconstruction sub-module. According to the evaluation results, select the airborne reconstruction sub-module of the Transformer architecture as the main operating unit. When the error of this channel exceeds the preset threshold or the error change trend deteriorates, trigger the dual-channel synchronous switch to the model-based reconstruction module.

[0112] Conduct a comparative evaluation of the dual-parameter accuracy and time consumption between the temperature and pressure dual-channel output signals of the airborne reconstruction sub-module of the Transformer architecture and the model-based module (as Figure 5 shown). The results show that although the accuracy of the dual-channel output of the Transformer module is slightly lower than that of the model-based module, it meets the dual-parameter tolerance requirements and has better real-time performance. Therefore, this module is preferentially operated when the errors of both channels do not exceed the limit; when the error change trend of either the temperature or pressure channel deteriorates or exceeds the error limit threshold, switch to the model-based module. To maintain the smoothness of the dual-channel signal switch, design an independent weight calculation strategy based on the dual-parameter error index.

[0113] Step 3.2), count the independent signal errors of the temperature and pressure dual-channels of the airborne reconstruction sub-module of the Transformer architecture within the time window, synchronously calculate the percentage change rate and maximum absolute error value of the errors of each channel, and generate a switching decision signal according to the combined evaluation index of the error change rate trend and the error threshold.

[0114] Count the independent signal errors of the temperature and pressure dual-channels of the airborne reconstruction sub-module of the Transformer architecture within the time window. The calculation formula is as follows:

[0115]

[0116] Among them, among them, respectively represent the true signals of temperature and pressure, respectively represent the output signals of the temperature and pressure Transformer modules, e t T ,e t P respectively represent the temperature and pressure errors, and the subscript t represents the moment.

[0117] Secondly, calculate the percentage change rate of the error between adjacent moments within the time window, as shown in the following formula:

[0118]

[0119] Among them, ò is a very small positive number set to prevent division by zero, and r is the percentage change rate of the error.

[0120] Next, statistically calculate the maximum percentage change rate of the error and the maximum absolute error value within the window, as shown in the following formula:

[0121]

[0122] Among them, r max is the maximum percentage change rate of the error within the window, e max is the maximum error within the window, and the superscripts T and P represent temperature and pressure respectively.

[0123] Finally, make a joint decision to switch or maintain the module according to the dual-channel joint evaluation index, as shown in the following formula:

[0124]

[0125] Among them, θ is the set threshold, the subscripts r and e represent the maximum percentage change rate of the error and the maximum error threshold respectively, and the superscripts T and P represent temperature and pressure respectively.

[0126] Step 3.3), design a dual-channel independent weight calculation strategy, construct dynamic weighting coefficients for the temperature and pressure signals respectively, and achieve a smooth transition of the dual-channel reconstructed signal from the Transformer architecture module to the model base module through weight matrix operations, and finally output the fused temperature and pressure signals.

[0127] To achieve a smooth transition of the temperature and pressure dual-channel signals, design a dual-channel independent weighted fusion strategy, as shown in the following formula:

[0128]

[0129] Among them, W tis a two-dimensional diagonal weight matrix. The superscripts T and P represent temperature and pressure respectively. The superscripts Trans and Model represent the output signals of the Transformer airborne reconstruction sub-module and the output signals of the model-based airborne reconstruction module respectively.

[0130] The weight calculation during the switching process is shown in the following formula:

[0131]

[0132] where Δ is the set step size of the weight change process.

[0133] The final fusion signal strategy is shown in the following formula:

[0134]

[0135] The performance of the three signal reconstruction methods is verified. When the main fuel flow rate at the ground point increases linearly with time, the accuracy and time consumption of the three sensor signal reconstruction methods are verified under the fuel dynamic change process. The simulation results are as Figure 6 、 Figure 7 shown.

[0136] In summary, the designed airborne fusion reconstruction method for the fault temperature and pressure sensor signals of aero-engines can provide reconstructed signals well, reflect the true working state of the engine, and has a positive significance for the provision of sensor signals.

[0137] Example 2

[0138] This example provides an airborne reconstruction system for the fault temperature and pressure sensors of an aero-engine, including:

[0139] A data acquisition unit for obtaining engine operation parameters and sensor data in real time;

[0140] A first reconstruction unit, connected to the data acquisition unit, for performing the functions of the first reconstruction module;

[0141] A second reconstruction unit, connected to the data acquisition unit, for performing the functions of the second reconstruction module;

[0142] A fusion control unit, connected to the first reconstruction unit and the second reconstruction unit respectively, for performing the functions of error evaluation and signal fusion;

[0143] where the output of the data acquisition unit is input to the first reconstruction unit and the second reconstruction unit simultaneously. The outputs of the first reconstruction unit and the second reconstruction unit are jointly input to the fusion control unit, and the output of the fusion control unit is used as the final output of the system.

[0144] The fusion control unit includes:

[0145] An error calculation module, which is used to calculate the error metrics output by two reconstruction units;

[0146] A weight decision module, connected to the error calculation module, which is used to determine the fusion weights according to the error metrics;

[0147] A signal fusion module, connected to the weight decision module, which is used to perform weighted fusion calculations;

[0148] Among them, the error calculation module receives the outputs of two reconstruction units, the weight decision module generates weight parameters according to the error calculation results, and the signal fusion module completes signal fusion according to the weight parameters.

[0149] Embodiment 3

[0150] This embodiment provides an aeroengine control system, including:

[0151] A sensor group, which is used to monitor the temperature and pressure of each part of the engine;

[0152] A control unit, which is used to adjust the engine operation parameters according to the monitoring data;

[0153] The airborne reconstruction system as described above, connected to the sensor group and the control unit;

[0154] Among them, when a fault occurs in the sensor group, the reconstruction signal output by the airborne reconstruction system automatically replaces the output of the faulty sensor, ensuring that the control unit continuously obtains reliable monitoring data.

[0155] Embodiment 4

[0156] This embodiment provides a computer-readable storage medium, storing a computer program, and when the program is executed by a processor, the following processing flow is implemented:

[0157] First, call the physical model reconstruction algorithm to obtain the first set of reconstruction values;

[0158] At the same time, call the deep learning model to obtain the second set of reconstruction values;

[0159] Then perform real-time error evaluation on the two sets of reconstruction values;

[0160] Finally, dynamically adjust the fusion weights according to the evaluation results, and output the final reconstructed signal after weighted fusion;

[0161] Among them, the execution of the physical model reconstruction algorithm and the deep learning model are independent of each other, and the error evaluation and the calculation of the fusion weights depend on the output results of both.

[0162] Embodiment 5

[0163] This embodiment provides an electronic device, including a memory and a processor. The memory stores a computer program, and when the processor executes the program, the following system functions are implemented:

[0164] Obtain engine sensor data through a data acquisition interface;

[0165] Run the physical model reconstruction algorithm and the deep learning reconstruction algorithm in parallel;

[0166] Compare the output results of the two algorithms in real time;

[0167] Dynamically adjust the algorithm weights and output the fused reconstruction signal;

[0168] Among them, the physical model reconstruction algorithm and the deep learning reconstruction algorithm are executed in parallel. The weight adjustment module monitors the output performance of both in real time, and the signal output module obtains the final reconstruction signal according to the weight parameters.

[0169] It should be noted that the above is only the specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes and substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. An airborne fusion reconstruction method for the signals of an aeroengine fault temperature and pressure sensor, characterized in that It includes the following steps: (1) Establish a first reconstruction module based on a physical model, and obtain a first temperature reconstruction value and a first pressure reconstruction value through the calculation of engine aerodynamic parameters; (2) Construct a second reconstruction module based on deep learning, process the sensor time-series data through a trained neural network model, and output a second temperature reconstruction value and a second pressure reconstruction value; (3) Conduct real-time error evaluation on the outputs of the first reconstruction module and the second reconstruction module, and dynamically adjust the fusion weights of the two modules according to the evaluation results; (4) Based on the adjusted fusion weights, perform weighted fusion on the reconstruction values of the two modules, and output the final temperature reconstruction signal and pressure reconstruction signal; Among them, the first reconstruction value obtained in step (1) serves as the benchmark input for the error evaluation in step (3), the second reconstruction value obtained in step (2) serves as the comparison input for the error evaluation in step (3), and the output result of step (3) directly determines the weight allocation in step (4).

2. The method according to claim 1, characterized in that Step (1) specifically includes: (1.1) Collect parameters such as fan outlet temperature, fan outlet pressure, compressor outlet pressure, turbine outlet pressure, and rotor speed; (1.2) Input the collected parameters into a pre-established aerodynamic thermodynamics model, and sequentially calculate the fan flow rate and the core flow rate; (1.3) Based on the calculated flow parameters, obtain the first temperature reconstruction value and the first pressure reconstruction value through the compressor characteristic curve.

3. The method according to claim 1, wherein Step (2) specifically includes: (2.1) Construct a training data set including engine operation parameters and sensor time-series data; (2.2) Design a neural network architecture with two output branches, where the first branch is for temperature reconstruction and the second branch is for pressure reconstruction; (2.3) Use the training data set to train the neural network to obtain a deep learning model that can synchronously output the second temperature reconstruction value and the second pressure reconstruction value.

4. The method according to claim 1, characterized in that Step (3) specifically includes: (3.1) Set a sliding time window, and calculate the temperature error and pressure error output by the first reconstruction module and the second reconstruction module in real time; (3.2) Calculate the error change rate of adjacent time steps and analyze the error change trend; (3.3) When the error of any parameter exceeds the preset threshold or the error change rate continues to increase, generate a reconstruction module switching instruction; (3.4) According to the error evaluation results, calculate the real-time fusion weights of the first reconstruction module and the second reconstruction module.

5. The method according to claim 1, characterized in that, Step (4) specifically includes: (4.1) Receive the fusion weights calculated in step (3); (4.2) Perform weighted averaging on the first temperature reconstruction value and the second temperature reconstruction value according to the weights to obtain the final temperature reconstruction signal; (4.3) Perform weighted averaging on the first pressure reconstruction value and the second pressure reconstruction value according to the weights to obtain the final pressure reconstruction signal; (4.4) Output the fused temperature reconstruction signal and pressure reconstruction signal to the engine control system.

6. An airborne reconstruction system for an aero-engine fault temperature and pressure sensor, characterized in that, It includes: A data acquisition unit for obtaining engine operation parameters and sensor data in real time; A first reconstruction unit connected to the data acquisition unit for performing the functions of the first reconstruction module described in claims 1-2; A second reconstruction unit connected to the data acquisition unit for performing the functions of the second reconstruction module described in claims 1 and 3; The fusion control unit, which is respectively connected to the first reconstruction unit and the second reconstruction unit, is configured to perform the error evaluation and signal fusion functions described in claims 1, 4 - 5; Among them, the output of the data acquisition unit is input to both the first reconstruction unit and the second reconstruction unit simultaneously. The outputs of the first reconstruction unit and the second reconstruction unit are jointly input to the fusion control unit, and the output of the fusion control unit serves as the final output of the system.

7. The system according to claim 6, wherein The fusion control unit includes: An error calculation module, configured to calculate the error metrics of the outputs of the two reconstruction units; A weight decision module, connected to the error calculation module, configured to determine the fusion weights according to the error metrics; A signal fusion module, connected to the weight decision module, configured to perform weighted fusion calculation; Among them, the error calculation module receives the outputs of the two reconstruction units, the weight decision module generates weight parameters according to the error calculation results, and the signal fusion module completes signal fusion according to the weight parameters.

8. An aeroengine control system, characterized in that, It includes: A sensor group, configured to monitor the temperature and pressure of various parts of the engine; A control unit, configured to adjust the engine operation parameters according to the monitored data; The airborne reconstruction system described in claims 6 - 7, connected to the sensor group and the control unit; Among them, when a fault occurs in the sensor group, the reconstruction signal output by the airborne reconstruction system automatically replaces the output of the faulty sensor to ensure that the control unit continuously obtains reliable monitored data.

9. A computer-readable storage medium storing a computer program, characterized in that, When the program is executed by the processor, it implements the method described in any one of claims 1 - 5.

10. An electronic device, comprising a memory and a processor, characterized in that, The memory stores a computer program, and when the processor executes the program, it implements the method described in any one of claims 1 - 5.

Images