Engine flight thrust estimation method fusing adversarial self-coding and sparse attention

Abstract

The invention provides an engine flight thrust estimation method fusing adversarial self-coding and sparse attention, and belongs to the crossing field of aerospace engineering and artificial intelligence. The method comprises the following steps that ground test run data and flight state data of the turbofan aero-engine are collected and preprocessed, and steady-state sensor data are obtained; inputting the steady-state sensor data and the corresponding working condition label into the trained adversarial auto-encoder for feature extraction to obtain a potential feature vector; s3, inputting the potential feature vector into a sparse attention module to obtain a fusion feature; inputting the fusion features into a multi-scale time sequence modeling module, and obtaining multi-scale fusion features through time sequence feature extraction of the multi-scale time sequence modeling module; and inputting the multi-scale fusion features into a trained thrust regression module to obtain an aero-engine flight thrust estimated value. The method can achieve the precise estimation of the thrust in the flight state, and has a remarkable practical engineering application value and a deployment prospect.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of aerospace engineering and artificial intelligence, and particularly relates to an engine flight thrust estimation method fusing adversarial auto-encoding and sparse attention. BACKGROUND

[0002] As the power core of aircraft, the health state evaluation of aero-engine has always been a research hotspot in academia and industry. Thrust, as a key indicator of engine performance, especially in the field of military engines, has important reference value in health state evaluation. However, the engine thrust cannot be directly measured during flight, and usually relies on ground or high-altitude test beds, which requires the cooperation of special equipment and professional personnel, resulting in high testing costs and long testing cycles. Therefore, conducting thrust estimation based on on-board monitoring data not only helps to achieve rapid and low-cost engine health evaluation, but also provides important support for on-condition maintenance.

[0003] In recent years, data-driven methods have been widely concerned in engine thrust estimation. For example, Song Hanqiang et al. built an extreme learning machine model based on ground test bed data, taking the monitorable parameters as input and the ground measured thrust as label, and achieved a relative error of ±3% in the steady-state multi-speed operating condition. Wang Zhao et al. proposed a solid rocket ramjet engine thrust estimation method based on PSO optimized BP neural network, and verified that the SNPSO optimization method can control the relative error of thrust within 5% while ensuring simplicity. Zhou Ting proposed a thrust estimation method based on similarity transformation and deep dynamic neural network, which effectively solved the problem of insufficient data under large flight envelope and variable state, and greatly reduced the thrust estimation error in the non-training area. To overcome the problem of repeated training in multi-condition modeling, Zou Yuhang et al. used simulation data, combined with model fine-tuning and improved domain adversarial neural network, to improve the adaptability of the model in different operating conditions.

[0004] Although the above researches continue to improve in the method level, there are still two core problems as follows: Firstly, the thrust during flight cannot be directly obtained, which leads to the fact that existing methods are heavily dependent on simulation data or ground test data as labels. However, the thrust label in simulation data is based on idealized assumptions and lacks credibility; there are significant differences between ground test conditions and flight conditions, which restricts the generalization ability of supervised learning models. Secondly, existing deep learning methods mostly use sequential or grid structure, which is difficult to effectively capture the complex physical coupling relationship between multiple components inside the engine, such as the aerodynamic-mechanical interaction between low-pressure turbine pressure and rotor speed, limiting the further improvement of estimation accuracy. Therefore, it is urgent to develop a thrust estimation method that is suitable for unlabeled scenarios, has structure perception ability and can fuse multi-scale time series information, in order to realize accurate estimation of engine thrust in real flight environment and further improve the health state evaluation level and engineering feasibility. SUMMARY

[0005] The technical problem solved by the present application is that: as the core component of the aircraft, the thrust of the aero-engine is directly related to the flight performance and safety. However, in the flight state, the thrust cannot be directly measured due to the limitations of the on-board sensor. The existing method relies on labeled ground test data or simulation data to train the model, which is difficult to directly migrate to the flight condition. Especially in the face of large differences in working conditions and missing thrust labels, traditional supervised learning and shallow models have poor generalization and poor robustness.

[0006] The technical solution adopted by the present application to solve the above technical problem is: fusion of generative adversarial feature extraction, sparse attention mechanism and multi-scale neural network, used for thrust estimation of aero-engine in flight state, aiming to improve the modeling ability of complex working conditions and the robustness of unmeasurable variables.

[0007] To solve the above technical problem, the specific technical solution of the present application is as follows:

[0008] An engine flight thrust estimation method fusing adversarial auto-encoding and sparse attention, the method comprising the following steps:

[0009] Step S1: collecting ground test data and flight state data of a turbofan aero-engine, and preprocessing the collected ground test data and flight state data to obtain steady-state sensor data;

[0010] Step S2: inputting the steady-state sensor data and corresponding working condition label into the trained adversarial auto-encoder for feature extraction to obtain a latent feature vector;

[0011] Step S3: inputting the latent feature vector into the sparse attention module, and after joint processing of sparse compression by the Linformer sparse attention module and Gumbel-Softmax gating, obtaining a fusion feature;

[0012] Step S4: inputting the fusion feature into the multi-scale time series modeling module, and after time series feature extraction by the multi-scale time series modeling module, obtaining a multi-scale fusion feature;

[0013] Step S5: inputting the multi-scale fusion feature into the trained thrust regression module to obtain an aero-engine flight thrust estimation value.

[0014] Further, step S1 comprises the following steps:

[0015] Step S11: collecting ground test data of a turbofan aero-engine, the ground test data including sensor data, working condition label and thrust label;

[0016] Step S12: Collecting flight state data of a turbofan aero-engine, the flight state data being composed of sensor data and working condition labels;

[0017] Step S13: Unifying the ground test data and the flight state data into a time sequence format;

[0018] Step S14: Normalizing all the collected ground test data and flight state data;

[0019] Step S15: Dividing the normalized sensor data of the ground test data and the flight state data into steady state and transition state to obtain steady state sensor data and transition state sensor data.

[0020] Further, the adversarial autoencoder in step S2 is constructed and trained in the following manner:

[0021] Step S21: Building an adversarial autoencoder, the adversarial autoencoder being built from three sub-modules of an encoder, a decoder and an adversarial discriminator;

[0022] Step S22: Building a loss function, the loss function including a reconstruction loss and an adversarial loss;

[0023] Step S23: Using the loss function to adversarially train the adversarial autoencoder to obtain a trained adversarial autoencoder.

[0024] Further, the sparse attention module in step S3 mainly includes five parts of a linear projection sub-module, a low-rank sparse compression sub-module, an attention calculation sub-module, a differentiable channel selection sub-module and a feature fusion sub-module. The sparse attention module performs the following operations:

[0025] Step S31: The linear projection sub-module uses three sets of linear transformation weights to project the input latent feature vector Z into a query matrix, a key matrix and a value matrix, respectively;

[0026] Step S32: The low-rank sparse compression sub-module uses a Linformer sparse compression mechanism to perform low-rank projection on the key matrix and the value matrix to obtain a low-rank key matrix and a low-rank value matrix;

[0027] Step S33: The attention calculation sub-module uses the low-rank key matrix and the low-rank value matrix to calculate an attention score matrix to obtain an attention representation;

[0028] Step S34: The differentiable channel selection sub-module performs channel-level screening on the input latent feature vector Z through a Gumbel-Softmax gating mechanism, and finally outputs the screened gating features;

[0029] ​Step S35: the attention representation is weighted and fused with the gating feature to obtain a fused feature.

[0030] Further, the multi-scale time sequence modeling module in step S4 includes a short-time transient module and a long-time transient module, and the following operations are performed:

[0031] Step S41: the short-time transient module adopts a one-dimensional convolutional neural network as a short-time modeler, the fused feature is input into the short-time transient module, and a short-time output is obtained through convolution operation processing of the short-time transient module;

[0032] Step S42: the long-time transient module is composed of a Transformer encoder composed of 4 layers of stacks, and the input is a fused feature vector; a global stable trend lasting for 60 seconds is extracted through the long-time transient module to obtain a long-time output;

[0033] Step S43: the short-time output and the long-time output are fused in a multi-scale manner to obtain a multi-scale fused feature.

[0034] Further, step S43 includes the following steps:

[0035] Step S431: the short-time output and the long-time output are spliced in the feature dimension to obtain a spliced output;

[0036] Step S432: the spliced output is input into a gated recurrent unit network for fusion to obtain a multi-scale fused feature.

[0037] Further, the thrust regression module in step S5 is constructed and trained in the following manner:

[0038] Step S51: a fully connected regression network is constructed, and the fully connected regression network is a 3-layer multilayer perceptron; the input is a multi-scale fused feature, and the output is a thrust estimate value;

[0039] Step S52: a weighted mean square error is used as a regression loss function, and the trained thrust regression module is obtained through training on thrust label data.

[0040] The present application has the following beneficial effects: 1) an end-to-end, interpretable, and transferable aero-engine thrust estimation system is constructed. 2) The method can realize precise estimation of thrust in flight state by using test data knowledge transfer without flight labels, and has significant practical engineering application value and deployment prospect. 3) It is suitable for engine state evaluation and health monitoring in various flight conditions. BRIEF DESCRIPTION OF DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings described below are some embodiments of the present application, and the other drawings can be obtained by those skilled in the art without creative effort on the basis of these drawings.

[0042] Figure 1 is the flow chart of the aero-engine thrust estimation of the present application.

[0043] Figure 2 is the division diagram of steady state and transition state of the present application.

[0044] Figure 3 is the air thrust estimation result diagram of the simulation data of the present application. DETAILED DESCRIPTION

[0045] The technical solutions in the embodiments of the present application will be described clearly and completely in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present application.

[0046] The present application proposes an engine flight thrust estimation method fusing adversarial auto-encoding and sparse attention, as shown in Figure 1 The method comprises the following steps:

[0047] Step S1: Collecting the ground test data and flight state data of a turbofan aero-engine, and pre-processing the collected ground test data and flight state data to obtain steady-state sensor data.

[0048] Step S11: Collecting the ground test data of a turbofan aero-engine, and the ground test data comprising sensor data, working condition label and thrust label.

[0049] The ground test bench can set different control parameters to a certain extent to approximate simulate the cruise process in flight. In the present application, by setting the control quantities such as engine inlet temperature, pressure, speed and throttle angle, three typical cruise working conditions are defined as the target domain data source: working condition 1: low-altitude subsonic cruise (high-altitude air density); working condition 2: high-altitude subsonic cruise (standard mission); working condition 3: high-altitude supersonic cruise (extreme state). The working condition label is represented by , ; the working condition label is 1 corresponding to working condition 1, the working condition label is 2 corresponding to working condition 2, and the working condition label Corresponds to the case 3 when t = 3.

[0050] During ground test, high-precision sensor network is deployed at each cross section and key components of the engine to collect sensor data in real time. The key components refer to the high-pressure rotor, low-pressure rotor, combustion chamber, compressor, outer duct, low-pressure turbine, tail nozzle, throttle lever, and thrust frame. The test personnel set the case parameters under the corresponding case label by adjusting the intake temperature, pressure, speed, and throttle angle. The thrust label can be accurately obtained through the thrust sensor on the adapter frame. The collected sensor data covers multiple typical working stages, and the specific collection parameters are shown in Table 1:

[0051] Table 1 Collection sensor parameter table

[0052]

[0053] The sensor data, case label, and thrust label jointly constitute the ground test data. Among them, the sensor data is collected in real time by the sensors arranged on the key components of the engine and recorded through the data acquisition module in the test control system; the case parameters under the corresponding case label are set by the test personnel by adjusting the intake temperature, pressure, speed, and throttle angle, and different combinations of settings correspond to different simulated flight conditions; the thrust label is accurately measured by the thrust sensor installed on the adapter frame of the test platform, and the axial net thrust output of the engine under the condition is obtained as the supervision signal of the regression model.

[0054] Step S12: Collecting flight state data of a turbofan aero-engine, the flight state data is jointly constituted by sensor data and case label.

[0055] The sensor data of the turbofan aero-engine under the flight task condition is collected by the airborne monitoring system (Engine Monitoring System, EMS). The system integrates multiple sensors, is distributed on the key components of the engine, and can monitor the speed, temperature, pressure, and fuel flow in real time. The types of collected parameters are consistent with the ground test data, and details are shown in Table 1. The case label is dynamically determined according to the flight state information (such as altitude, speed, and ambient temperature), and is recorded synchronously with the sensor data.

[0056] Step S13: Unifying the ground test data and the flight state data into time series format.

[0057] The ground test data and the flight state data are both in time series format, and the flight state data does not have a thrust label. Each piece of data constitutes a complete sample, representing the running state of the engine in a certain continuous time window. The format of each sample is unified into a triple form: , represents the sensor data, represents a working condition label, represents a thrust label. The thrust label only exists in the ground test data, and the thrust label of the flight state data is null. Through sample format standardization, the quality and scale of the input data can be ensured to be uniform, and the requirements of the downstream neural network module for the time series structure can be met.

[0058] The data sampling frequency is set to 5 Hz, that is, 5 groups of data points are collected per second. The duration of each piece of data is set to 60 seconds, corresponding to a sample sequence length of 300.

[0059] Step S14: Normalize all collected ground test data and flight state data.

[0060] All collected ground test data and flight state data are normalized to eliminate the influence between different dimensions. The normalization operation is applied to the sensor data of each sample and the thrust label (only exists in the ground test data) is also normalized in the training stage to improve the numerical stability of the regression model; in the inference stage, the actual thrust value is restored through the inverse normalization operation.

[0061] In the present application, represents a time series sensor sample data (300 time steps, d-dimensional sensor data), and the sensor data X represents a sample set. The sensor data normalization is calculated by the following formula:

[0062]

[0063] wherein, is the normalized sensor data; x is the sensor sample data; is the minimum value of the sensor sample data; is the maximum value of the sensor sample data.

[0064] represents a time series thrust sample label (300 time steps), and the thrust label represents a sample set. The thrust label normalization is calculated by the following formula:

[0065]

[0066] wherein, is the normalized thrust label; is the thrust sample label; is the minimum value of the thrust sample label; is the maximum value of the thrust sample label.

[0067] Step S15: dividing the sensor data after the ground test data and the normalized flight state data into steady state and transition state to obtain steady state sensor data and transition state sensor data.

[0068] The process of an aircraft engine changing from one operating state to another is called a transitional operating process, and the transitional operating process is called a transient state. Other stable operating states can be considered steady states. The main purpose of this invention is to obtain sensor data in the steady state during flight. Therefore, it is necessary to classify sensor data into two categories: steady state and transition state. First, the rate of change of all normalized sensor data is calculated:

[0069]

[0070] in, For the The rate of change of data points (eigenvectors); For the The sensor data after normalization of the data points, that is, the sensor data after normalization at the current sampling moment; For the The sensor data after normalization of the data point, that is, the sensor data after normalization of the previous sampling moment; is the sampling time interval.

[0071] Based on the rate of change, the k-means clustering algorithm is used to partition the data set into K different clusters, so that the data points within each cluster are as similar as possible, while the data differences between different clusters are as large as possible. The goal of this algorithm is to minimize the sum of the squares of the distances from the data points to the cluster centers, as shown in the following formula:

[0072]

[0073] Among them, J is the loss function, that is, the sum of squared errors within the cluster, which is used to measure the clustering effect; K is the number of clusters; represents the kth cluster; is the center of the kth cluster. Based on the above clustering results, the clusters with smaller mean and lower fluctuation amplitude are determined as steady-state clusters, and the corresponding samples are used as steady-state sensor data. Clusters with large mean values ​​and dramatic changes are identified as transitional clusters, and the corresponding samples are used as transitional sensor data. Finally, sensor samples and operating condition labels within the stable clusters are used as input for the subsequent adversarial autoencoder. Figure 2 The division results of transition state and steady state are shown.

[0074] Step S2: Steady-state sensor data And the corresponding working condition label input trained adversarial autoencoder for feature extraction, and obtain the potential feature vector .

[0075] The adversarial autoencoder is used as a feature extraction front-end module to realize label-free latent space modeling and migration in the flight state. The adversarial autoencoder is trained jointly by reconstruction loss and adversarial loss. Through adversarial optimization training, the ground test data and flight state data can be uniformly distributed in the latent space, thereby improving the label-free generalization ability of the model.

[0076] The adversarial autoencoder is constructed and trained in the following way:

[0077] Step S21: build an adversarial autoencoder, which is built by an encoder, a decoder and an adversarial discriminator.

[0078] The encoder is used to receive input steady-state sensor data and working condition labels , and output a latent feature vector . represents the number of sensors, represents the number of samples, represents the number of working conditions, represents the dimension size of the latent space.

[0079] The decoder receives the latent feature vector and the working condition label , and attempts to reconstruct the original input steady-state sensor data.

[0080] The discriminator is used to distinguish whether the input latent feature vector is generated by the encoder (fake sample) or from the standard normal distribution (real sample).

[0081] Step S22: construct a loss function, which includes a reconstruction loss and an adversarial loss.

[0082] The reconstruction loss is represented as follows:

[0083] ℒ r e c =  x * [ ‖ x * − D θ ( E ϕ ( x * , c ) , c ) ‖ 2 ]

[0084] wherein, is the reconstruction loss; represents the latent feature vector output by the encoder according to the normalized sensor data and the working condition label , that is ; represents the reconstruction result of the decoder on the latent feature vector and the working condition label , that is ;  x * [ · ] denotes the expected (average) value over all training samples.

[0085] The adversarial loss includes the discriminator loss and the encoder fooling loss:

[0086] ℒ a d v T = −  z ~  ( 0 , I ) [ l o g T ψ ( z ) ] −  x * [ l o g ( 1 − T ψ ( E ϕ ( x * , c ) ) ]

[0087] ℒ a d v E = −  x * [ l o g T ψ ( E ϕ ( x * , c ) ) ]

[0088] where, is the discriminator loss; is the encoder fooling loss; the random variable follows a standard normal distribution with mean 0 and variance , , is a real sample; denotes the latent feature vector output by the encoder from the normalized sensor data and the working condition label . denotes the discriminator's judgment on whether the input latent feature vector is generated by the encoder (fake sample) or from the standard normal distribution (real sample);  x * [ · ] denotes the expected (average) value over all training samples.  z ~  ( 0 , I ) [ · ] denotes the expected (average) value of the random variable following the standard normal distribution.

[0089] The total training objective is to jointly minimize the following joint loss function:

[0090]

[0091] where, is a weight hyperparameter used to balance the reconstruction accuracy and the latent distribution alignment performance.

[0092] Step S23: Adversarial training is performed on the adversarial autoencoder to obtain a trained adversarial autoencoder.

[0093] The training of the adversarial autoencoder adopts an alternating optimization strategy, including the following steps:

[0094] Step S231: The discriminator is updated with fixed encoder parameters to minimize the discriminator loss to enhance the discrimination ability.

[0095] Step S231: The encoder is updated with fixed discriminator parameters to minimize the encoder fooling loss ​Make the fake samples closer to the real distribution.

[0096] Step S232: Update the encoder + decoder (reconstruction) while minimizing the reconstruction loss , improve the restoration accuracy of the original input.

[0097] Step S233: Joint optimization jointly minimizes the joint loss , achieve the balance between latent space alignment and input information preservation.

[0098] Step S234: Freeze the trained adversarial autoencoder to obtain the trained adversarial autoencoder as a feature extractor.

[0099] Input all steady-state sensor data and corresponding working condition labels into the trained adversarial autoencoder to obtain the latent feature vector .

[0100] Step S3: Input the latent feature vector into the sparse attention module, and after joint processing of sparse compression and Gumbel-Softmax gating in the Linformer sparse attention module, the fusion feature is obtained.

[0101] Based on the sparse attention module, the sparse attention mechanism is introduced, and the latent feature vector is input into the sparse attention module to model the nonlinear dependency relationship between sensor channels. In order to further reduce the interference of redundant input, the sparse attention module also includes a differentiable channel selection module, which uses the Gumbel-Softmax strategy to dynamically filter the channel combination that significantly affects the thrust from the input channels, and shields irrelevant or noise channels.

[0102] The sparse attention module mainly includes a linear projection submodule, a low-rank sparse compression submodule, an attention calculation submodule, a differentiable channel selection submodule, and a feature fusion submodule. The sparse attention module performs the following operations:

[0103] Step S31: The linear projection submodule uses three sets of linear transformation weights to project the input latent feature vector into a query matrix (Query), a key matrix (Key) and a value matrix (Value) respectively, represented as follows:

[0104]

[0105] Where, is the input latent feature vector; , , are learnable parameters, , , Query matrix, key matrix, value matrix, respectively.

[0106] Step S32: The low-rank sparse compression submodule uses the Linformer sparse compression mechanism to perform low-rank projection on the key matrix and the value matrix to obtain a low-rank key matrix and a low-rank value matrix .

[0107] The Linformer sparse compression mechanism can reduce the attention calculation complexity from to , significantly improving the calculation efficiency.

[0108] Step S33: The attention calculation submodule uses the low-rank key matrix and the low-rank value matrix to calculate the attention score matrix, and obtains the attention representation, and the attention score matrix calculation formula is as follows:

[0109]

[0110] wherein, is the attention representation, is the feature dimension size of the low-rank key matrix , is a scaling factor to prevent gradient explosion; represents the attention representation of each channel to other channels, represents the final attention matrix dimension; represents transposition; represents an activation function, which functions to convert an arbitrary real number value vector input into a probability distribution vector with a value range of [0, 1] and a sum of 1.

[0111] Step S34: The differentiable channel selection submodule performs channel-level screening on the input latent feature vector Z through the Gumbel-Softmax gating mechanism, and finally outputs the screened gating feature .

[0112] The differentiable channel selection submodule is the core component of the sparse attention module, which performs channel-level screening on the input latent feature vector Z through the Gumbel-Softmax gating mechanism. Its working logic is: using the differentiable Gumbel-Softmax sampling strategy, dynamic weights are generated for each sensor channel, automatically identifying the channel combination that significantly affects the thrust estimation, suppressing the interference of redundant or noisy channels, and finally outputting the screened gating feature , realizing the strengthening of key information and the filtering of irrelevant information.

[0113] Step S35: The attention representation is combined with the gating feature The weighted fusion is performed to obtain a fused feature. The fusion manner is represented as follows:

[0114]

[0115] wherein, α ∈ [ 0 , 1 ] is a learnable fusion weight; and the final output fused feature , represents a sample number, represents a fused feature dimension.

[0116] Step S4: inputting the fused feature into a multi-scale time sequence modeling module, and performing time sequence feature extraction through the multi-scale time sequence modeling module to obtain a multi-scale fused feature.

[0117] To simultaneously model short-time transient response and long-time stable trend in the sensor sequence of the aero-engine, a multi-scale time sequence modeling module is used to complete time sequence feature extraction. Through the multi-scale time sequence modeling module, instantaneous nonlinear disturbance and long-time stable trend can be simultaneously perceived, so that more comprehensive engine time sequence modeling is realized, and the thrust estimation accuracy is improved.

[0118] The multi-scale time sequence modeling module includes a short-time transient module and a long-time transient module, and performs the following operations:

[0119] Step S41: the short-time transient module uses a one-dimensional convolutional neural network (1D CNN) as a short-time modeler, the fused feature is input into the short-time transient module, and a short-time output is obtained through convolution operation processing of the short-time transient module.

[0120] The network depth of the one-dimensional convolutional neural network of the short-time transient module is set to 5 layers, each layer uses a causal convolution with a sliding window; the time step of each input sample is set to 300, corresponding to a 60-second time sequence window (a sampling frequency of 5 Hz); the length of each convolution kernel is set to 10, and the step is 1, so as to ensure covering the local gradient features corresponding to typical transient phenomena; the output of the short-time transient module is a short-time output , which represents the transient response features of each sensor channel in the short term; which represents the short-time hidden dimension of each layer.

[0121] Step S42: the long-time transient module is composed of a Transformer encoder stacked by 4 layers, the input is a fused feature vector, and a long-time output is obtained through the long-time transient module to extract a global stable trend for 60 seconds.

[0122] The long-time transient module is used to extract a global stable trend for 60 seconds, and position encoding is added in the time dimension to preserve the order of working condition changes; the number of attention heads of the Transformer encoder is set to 4, and the output is a long-time output , denotes the long-term per-layer hidden dimension.

[0123] Step S43: Multi-scale feature fusion is performed on the short-term output and the long-term output to obtain multi-scale fusion features.

[0124] Step S431: The short-term output is concatenated with the long-term output in the feature dimension to obtain a concatenated output . Z fused = [ Z short , Z long ] .

[0125] Step S432: The concatenated output is input into a gated recurrent unit (GRU) network for fusion to obtain multi-scale fusion features .

[0126] The GRU models the time dynamics and unifies the feature dimension output; the final output is the multi-scale fusion features , i.e., the fused multi-scale time modeling representation, which serves as the input of the subsequent thrust estimation module.

[0127] Step S5: The multi-scale fusion features are input into the trained thrust regression module to obtain the aero-engine flight thrust estimate value.

[0128] The thrust regression module is used to map the multi-scale fusion features to the aero-engine flight thrust estimate value. Through the thrust regression module, high-precision thrust prediction can be achieved, which helps to ensure the reliability of critical flight missions.

[0129] The thrust regression module is constructed and trained in the following manner:

[0130] Step S51: A fully connected regression network is constructed, which is a 3-layer multi-layer perceptron (MLP); the input is the multi-scale fusion features , and the output is the thrust estimate value .

[0131] The network structure of the fully connected network is as follows:

[0132] First layer MLP: the number of neurons is set to 2 , and the activation function uses GeLU (Gaussian Error Linear Unit), whose formula is:

[0133]

[0134] where x represents the input vector of the hidden layer in the fully connected regression network, The activation function is a cumulative distribution function of a standard normal distribution, which can better capture nonlinear characteristics and alleviate the "dead neuron" problem of ReLU.

[0135] The second layer MLP: the number of neurons is reduced to The activation function is ReLU (Rectified Linear Unit), and the formula is:

[0136]

[0137] Wherein, x represents the input vector of the hidden layer in the full connection regression network.

[0138] The output layer is a linear layer (without activation function), and the number of neurons is 1, which directly outputs the normalized thrust estimation value , and the actual thrust value is restored through denormalization.

[0139] Step S52: using the weighted mean square error as the regression loss function, training on the labeled data (ground test), obtaining the trained thrust regression module.

[0140] The regression loss function is represented as follows:

[0141] ℒ r e g =  ( x * , y * ) [ w ( x * ) ⋅ ( y ^ − y * ) 2 ]

[0142] Wherein, is the thrust estimation value; is the normalized true thrust label (only exists in the ground test data); is a weight function adjusted according to the working condition parameters, used to balance the error contribution of high thrust and low thrust stages;  ( x * , y * ) [ · ] represents the expected (average) value of all training samples; The final regression loss function is used to train the thrust regression module, and the trained thrust regression module is obtained.

[0143] The adversarial autoencoder, the sparse attention module, the multi-scale time sequence modeling module and the thrust regression module constitute the aircraft engine flight thrust estimation module.

[0144] The simulation experiment of the application is based on F-119 engine test and flight data, and the in-flight thrust estimation result of the simulation data is as shown in Figure 3 , wherein the solid line is the true engine reasoning value, and the gray line is the thrust value predicted by the method. The results show that the engine flight thrust estimation method fusing adversarial autoencoder and sparse attention proposed in the application still has good estimation accuracy and generalization ability under unsupervised flight state, and 95% error falls within ±3kN interval, An index average of more than 0.75 shows a higher potential for practical application.

[0145] The above descriptions are only the preferred embodiment of the application, not intended to limit the protection scope of the application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the application shall be included in the protection scope of the application.

Claims

1. A method for estimating engine flight thrust by integrating adversarial autoencoding and sparse attention, characterized in that: The method comprises the following steps: Step S1: collecting ground test data and flight status data of a turbofan aircraft engine, and preprocessing the collected ground test data and flight status data to obtain steady-state sensor data; Step S2: Input the steady-state sensor data and the corresponding working condition labels into the trained adversarial autoencoder for feature extraction to obtain a potential feature vector; Step S3: The latent feature vector is input into the sparse attention module, and the fusion feature is obtained through the joint processing of sparse compression of the sparse attention module Linformer and Gumbel-Softmax gating; Step S4: inputting the fusion features into the multi-scale time series modeling module, and extracting the time series features of the multi-scale time series modeling module to obtain the multi-scale fusion features; Step S5: Input the multi-scale fusion features into the trained thrust regression module to obtain the estimated value of the aircraft engine flight thrust.

2. The engine flight thrust estimation method integrating adversarial autoencoding and sparse attention according to claim 1 is characterized in that: Step S1 includes the following steps: Step S11: collecting ground test data of a turbofan aircraft engine, where the ground test data includes sensor data, operating condition labels, and thrust labels; Step S12: collecting flight status data of a turbofan aircraft engine, where the flight status data is composed of sensor data and operating condition labels; Step S13: unifying the ground test data and the flight status data into a time series format; Step S14: normalizing all collected ground test data and flight status data; Step S15: dividing the sensor data after the ground test data and the normalized flight state data into steady state and transition state to obtain steady state sensor data and transition state sensor data.

3. The engine flight thrust estimation method integrating adversarial autoencoding and sparse attention according to claim 1 is characterized in that: The adversarial autoencoder in step S2 is constructed and trained as follows: Step S21: Building an adversarial autoencoder, which consists of three submodules: encoder, decoder, and adversarial discriminator; Step S22: constructing a loss function, which includes reconstruction loss and adversarial loss; Step S23: Using the loss function to perform adversarial training on the adversarial autoencoder to obtain a trained adversarial autoencoder.

4. The engine flight thrust estimation method integrating adversarial autoencoding and sparse attention according to claim 1 is characterized in that: The sparse attention module in step S3 mainly includes five parts: linear projection submodule, low-rank sparse compression submodule, attention calculation submodule, differentiable channel selection submodule, and feature fusion submodule; The sparse attention module performs the following operations: Step S31: The linear projection submodule uses three sets of linear transformation weights to project the input latent feature vector into a query matrix, a key matrix, and a value matrix respectively; Step S32: The low-rank sparse compression submodule uses the Linformer sparse compression mechanism to perform low-rank projection on the key matrix and the value matrix to obtain a low-rank key matrix and a low-rank value matrix; Step S33: The attention calculation submodule calculates the attention score matrix using the low-rank key matrix and the low-rank value matrix to obtain the attention representation; Step S34: The differentiable channel selection submodule performs channel-level screening on the input potential feature vector Z through the Gumbel-Softmax gating mechanism, and finally outputs the screened gated features; Step S35: Perform weighted fusion of the attention representation and the gated feature to obtain the fused feature.

5. The engine flight thrust estimation method integrating adversarial autoencoding and sparse attention according to claim 1 is characterized in that: The multi-scale time series modeling module in step S4 includes a short-term transient module and a long-term transient module, and performs the following operations: Step S41: The short-term transient module uses a one-dimensional convolutional neural network as a short-term modeler, fuses the features into the short-term transient module, and obtains a short-term output after the convolution operation of the short-term transient module; Step S42: The long-term transient module is composed of a 4-layer stacked Transformer encoder. The input is the fused feature vector. The long-term transient module extracts the global stable trend of up to 60 seconds and obtains the long-term output. Step S43: Perform multi-scale feature fusion on the short-term output and the long-term output to obtain multi-scale fusion features.

6. The engine flight thrust estimation method integrating adversarial autoencoding and sparse attention according to claim 5 is characterized in that: Step S43 includes the following steps: Step S431: concatenate the short-term output and the long-term output in the feature dimension into a concatenated output; Step S432: The spliced ​​output is input into the gated recurrent unit network for fusion to obtain multi-scale fusion features.

7. The engine flight thrust estimation method integrating adversarial autoencoding and sparse attention according to claim 1 is characterized in that: The thrust regression module in step S5 is constructed and trained as follows: Step S51: Construct a fully connected regression network, which is a 3-layer multi-layer perceptron; the input is multi-scale fusion features, and the output is thrust estimation; Step S52: Use the weighted mean square error as the regression loss function and perform training on the thrust label data to obtain a trained thrust regression module.

Images