Training method and detection method of aero-engine performance detection model

Abstract

The invention provides a training method and a detection method of an aero-engine performance detection model, and the training method comprises the steps: generating a covariance matrix which represents the collaborative change degree between different sensor features according to a plurality of sensor features, collected through sensors, of each initial training sample in an initial training set; based on the covariance matrix, carrying out feature selection on a plurality of sensor features collected by a sensor of the plurality of initial training samples to obtain a target training set; for each target training sample, inputting the target training sample into an initial prediction model, and outputting performance prediction information of the aero-engine, the initial prediction model being constructed according to a transform network and an attention network; and adjusting model parameters of the initial prediction model based on the performance prediction information and the performance label of the target training sample to obtain a trained aero-engine performance detection model.

Description

Technical Field

[0001] This application relates to the field of aero-engine technology, and more specifically, to a training method and a testing method for an aero-engine performance testing model. Background Technology

[0002] Aero engines are the core propulsion system of aircraft, and their operational status directly affects flight safety and operational efficiency. Therefore, accurate condition monitoring and health management of engines are of significant engineering importance. Currently, engine health monitoring systems are widely used in airlines and maintenance, repair, and overhaul services. If key engine parameters such as thrust and remaining service life can be assessed and predicted in real time, such systems can facilitate proactive maintenance strategies. For example, based on condition prediction results, the replacement of critical components (such as turbine blades and sensors) or system calibration can be scheduled, thereby effectively reducing unplanned downtime and improving operational safety.

[0003] The relevant technologies suffer from low prediction accuracy and high computational resource consumption when predicting performance parameters such as thrust and remaining service life of aero engines. Summary of the Invention

[0004] In view of this, this application provides a training method and a testing method for an aero-engine performance testing model.

[0005] One aspect of this application provides a training method for an aero-engine performance testing model, comprising:

[0006] Based on multiple sensor features acquired by sensors for each initial training sample in the initial training set, a covariance matrix is ​​generated to characterize the degree of coordinated variation among different sensor features. These different sensor features represent different performance detection parameters corresponding to different components in the aforementioned aero-engine. Based on the covariance matrix, feature selection is performed on multiple sensor features acquired by sensors from the initial training samples to obtain a target training set. The target training samples in the target training set include multiple sensor features determined through feature selection, where the determined sensor features characterize key features of engine thermodynamics, mechanical dynamics, and aerodynamics. For each target training sample, the target training sample is input into an initial prediction model, which outputs performance prediction information for the aero-engine. This initial prediction model is constructed based on a transform network and an attention network. Based on the performance prediction information and the performance labels of the target training samples, the model parameters of the initial prediction model are adjusted to obtain a trained aero-engine performance detection model.

[0007] Another aspect of this application provides a method for detecting the performance of an aero-engine, comprising: acquiring multiple performance parameters to be detected collected by a target sensor in the aero-engine; inputting the multiple performance parameters to be detected into an aero-engine performance detection model, and outputting predicted performance information.

[0008] Another aspect of this application provides a training apparatus for an aero-engine performance detection model, comprising: a generation module, configured to generate a covariance matrix characterizing the degree of cooperative variation among different sensor features based on multiple sensor features acquired by sensors for each initial training sample in an initial training set, wherein the different sensor features characterize different performance detection parameters corresponding to different components in the aero-engine; a selection module, configured to perform feature selection on multiple sensor features acquired by sensors for multiple initial training samples based on the covariance matrix to obtain a target training set, wherein the target training samples in the target training set include multiple sensor features determined by feature selection, wherein the determined sensor features characterize key features of engine thermodynamics, mechanical dynamics, and aerodynamics; a prediction module, configured to input the target training sample into an initial prediction model for each target training sample and output performance prediction information of the aero-engine, wherein the initial prediction model is constructed based on a transform network and an attention network; and a parameter tuning module, configured to adjust the model parameters of the initial prediction model based on the performance prediction information and the performance labels of the target training samples to obtain a trained aero-engine performance detection model.

[0009] Another aspect of this application provides an aero-engine performance testing device, comprising: an acquisition module for acquiring multiple performance parameters to be tested collected by target sensors in the aero-engine; and a testing module for inputting the multiple performance parameters to be tested into an aero-engine performance testing model and outputting predicted performance information.

[0010] Another aspect of this application provides an electronic device comprising: one or more processors; and a memory for storing one or more programs, wherein, when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to perform the method as described above.

[0011] Another aspect of this application provides a computer-readable storage medium storing computer-executable instructions that, when executed, are used to implement the method described above.

[0012] Another aspect of this application provides a computer program product comprising computer-executable instructions which, when executed, are used to implement the method described above.

[0013] According to embodiments of this application, a covariance matrix characterizing the degree of coordinated change among different sensor features is generated from multiple sensor features collected by sensors in the initial training samples. Feature selection is then performed on the sensor features in the initial training samples based on the covariance matrix to obtain a target training set. The target training samples in the target training set are then used to train the model, thereby obtaining an aero-engine performance detection model. Since feature selection is performed on the sensor features in the initial training samples based on the covariance matrix of the degree of coordinated change among different sensor features, features with strong performance correlation can be filtered out. Therefore, model training based on the filtered sensor features can improve the model's prediction accuracy. Furthermore, using fewer sensor features for model training can increase the training speed and effectively reduce the computational resources required for model training. Attached Figure Description

[0014] The above and other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0015] Figure 1 An exemplary system architecture is shown that can be applied to training and testing methods for an aero-engine performance testing model according to embodiments of this application.

[0016] Figure 2 A flowchart is shown of a training method for an aero-engine performance testing model according to an embodiment of this application.

[0017] Figure 3 A heatmap of Pearson correlation coefficients between different sensor features in the initial training set according to an embodiment of this application is shown.

[0018] Figure 4 The remaining useful life prediction results based on the C-MAPSS dataset according to an embodiment of this application are shown.

[0019] Figure 5 A heatmap showing the Pearson correlation coefficients between sensor features of a 1000 kg-class aircraft turbofan engine according to an embodiment of this application is presented.

[0020] Figure 6 A data-enhanced schematic diagram of a 1000 kg-class aircraft turbofan engine according to an embodiment of this application is shown.

[0021] Figure 7 A schematic diagram showing the thrust comparison of a 1000 kg-class aircraft turbofan engine according to an embodiment of this application is provided.

[0022] Figure 8 A flowchart of a method for detecting the performance of an aircraft engine according to an embodiment of this application is shown.

[0023] Figure 9 A block diagram of a training apparatus for an aircraft engine performance testing model according to an embodiment of this application is shown.

[0024] Figure 10 A block diagram of an aircraft engine performance testing apparatus according to an embodiment of this application is shown.

[0025] Figure 11 A block diagram of an electronic device suitable for implementing the methods described above, according to an embodiment of this application, is shown. Detailed Implementation

[0026] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0027] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0028] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0029] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0030] In the embodiments of this application, the collection, updating, analysis, processing, use, transmission, provision, disclosure, and storage of data (e.g., including but not limited to user personal information) comply with relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. In particular, necessary measures have been taken to prevent unauthorized access to user personal information data and to safeguard user personal information security and network security.

[0031] In the embodiments of this application, the user's authorization or consent was obtained before obtaining or collecting the user's personal information.

[0032] Several key challenges remain when applying these technologies to real-world aero-engine engineering applications. First, while advanced sequence models, such as the Transformer, perform exceptionally well in various prediction tasks, their performance is highly dependent on large amounts of high-quality training data and extremely sensitive to hyperparameter configuration. However, in actual engine monitoring, full-cycle data from operation to failure is extremely scarce and costly to acquire, resulting in insufficient generalization ability of such data-driven models under limited sample conditions. Second, existing methods typically treat data dimensionality reduction and data augmentation as independent modules, lacking an integrated modeling framework for aero-engine prediction tasks. This fragmented approach makes it difficult to achieve stable and reliable performance in high-risk, small-sample engineering contexts. Finally, most existing research builds models based on idealized simulation data, while actual engine test data is affected by various engineering constraints and environmental fluctuations, exhibiting characteristics such as high noise, non-stationarity, and limited sample size. This leads to models performing well in simulation environments but experiencing a significant drop in prediction accuracy in real-world scenarios.

[0033] Existing aero-engine condition prediction algorithms mainly use simulation data as input. When used in real engine engineering applications, they have the following three problems:

[0034] 1. Due to the high cost and stringent testing conditions involved in full-cycle real engine testing, existing research relies heavily on simulation data to build models, resulting in insufficient adaptability and reliability of algorithms in actual engineering environments.

[0035] 2. Even if some studies incorporate real test data, the scale and completeness of the data are still far inferior to simulation data. Real data is affected by various uncertainties in engineering practice, and the acquisition process is complex and contains a lot of hidden noise, making it difficult for the model to extract robust features from it, thus causing the model's performance to degrade in practical applications.

[0036] 3. As models based on self-attention mechanisms, such as Transformer, become increasingly popular in time series prediction tasks, it is necessary to systematically analyze their performance differences and applicable boundaries with traditional long short-term memory (LSTM) networks and other models under aero-engine data conditions. This will provide a theoretical basis for model selection and achieve a balance between prediction accuracy and efficiency under limited computing resources.

[0037] In view of this, embodiments of this application provide a training method and a detection method for an aero-engine performance detection model. The training method includes generating a covariance matrix characterizing the degree of cooperative variation among different sensor features based on multiple sensor features collected by sensors for each initial training sample in an initial training set. These different sensor features characterize different performance detection parameters corresponding to different components in the aero-engine, including compressors, turbines, pipes, and fans. Based on the covariance matrix, feature selection is performed on multiple sensor features collected by sensors for multiple initial training samples to obtain a target training set. The target training samples in the target training set include multiple sensor features determined through feature selection, where the determined sensor features characterize key features of engine thermodynamics, mechanical dynamics, and aerodynamics. For each target training sample, the target training sample is input into an initial prediction model, which outputs aero-engine performance prediction information. The initial prediction model is constructed based on a transform network and an attention network. Based on the performance prediction information and the performance labels of the target training samples, the model parameters of the initial prediction model are adjusted to obtain a trained aero-engine performance detection model.

[0038] Figure 1 An exemplary system architecture for training and testing methods of an aero-engine performance testing model, according to embodiments of this application, is shown. It should be noted that... Figure 1 The examples shown are merely examples of system architectures that can be applied to the embodiments of this application, in order to help those skilled in the art understand the technical content of this application, but do not mean that the embodiments of this application cannot be used in other devices, systems, environments or scenarios.

[0039] like Figure 1 As shown, the system architecture 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing communication links between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired and / or wireless communication links, etc.

[0040] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, and / or social media platform software, etc. (for example only).

[0041] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0042] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0043] It should be noted that the training and testing methods for the aero-engine performance testing model provided in this application embodiment can generally be executed by server 105. Correspondingly, the training and testing devices for the aero-engine performance testing model provided in this application embodiment can generally be located in server 105. The training and testing methods for the aero-engine performance testing model provided in this application embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the training and testing devices for the aero-engine performance testing model provided in this application embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Alternatively, the training and testing methods for the aero-engine performance testing model provided in this application embodiment can also be executed by the first terminal device 101, the second terminal device 102, or the third terminal device 103, or by other terminal devices different from the first terminal device 101, the second terminal device 102, or the third terminal device 103. Correspondingly, the training and testing devices for the aero-engine performance testing model provided in this application embodiment can also be located in the first terminal device 101, the second terminal device 102, or the third terminal device 103, or in other terminal devices different from the first terminal device 101, the second terminal device 102, or the third terminal device 103.

[0044] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0045] Figure 2 A flowchart is shown of a training method for an aero-engine performance testing model according to an embodiment of this application.

[0046] like Figure 2 As shown, the training method for this aero-engine performance testing model includes operations S201~S204.

[0047] In operation S201, based on the sensor features collected by multiple sensors for each initial training sample in the initial training set, a covariance matrix is ​​generated to characterize the degree of coordinated change among different sensor features, where different sensor features characterize different performance detection parameters of different components in the aero-engine.

[0048] In operation S202, based on the covariance matrix, feature selection is performed on multiple sensor features collected by multiple sensors from multiple initial training samples to obtain the target training set. The target training samples in the target training set include multiple sensor features determined by feature selection. The determined sensor features characterize the key features of engine thermodynamics, mechanical dynamics and aerodynamics.

[0049] In operation S203, for each target training sample, the target training sample is input into the initial prediction model, and the performance prediction information of the aero-engine is output. The initial prediction model is constructed based on the transform network and the attention network.

[0050] In operation S204, based on performance prediction information and the performance labels of target training samples, the model parameters of the initial prediction model are adjusted to obtain a trained aero-engine performance detection model.

[0051] According to embodiments of this application, performance prediction information can reflect the performance of an aero-engine, such as working performance like engine thrust or health indicators like engine lifespan.

[0052] According to embodiments of this application, aero-engines contain a variety of components, including but not limited to high-pressure compressors, low-pressure compressors, low-pressure turbines, bypass pipes, and fans. For high-pressure compressors, sensor characteristics include, but are not limited to, the total outlet temperature and total pressure of the high-pressure compressor. For fans, sensor characteristics include, but are not limited to, the fan's physical speed and calibrated rotational speed. In addition to the above characteristics, other sensor characteristics such as the core engine's physical speed, core engine calibrated rotational speed, bypass ratio, exhaust enthalpy, cycle number, and operating parameters can also be collected. The sensor characteristics in the initial training samples involved in this embodiment refer to all characteristics related to aero-engines, which will not be described one by one here.

[0053] According to an embodiment of this application, an actual test dataset is obtained through an aircraft engine monitoring system and used as the initial training set. The initial training set is defined as follows: for:

[0054]

[0055] Where m represents the total number of initial training samples, and n represents the total number of sensor features. It represents the set of real numbers.

[0056] Single initial training sample Represented as:

[0057]

[0058] This represents the physical quantity monitored by all sensors at the i-th sampling time, i.e., the feature of the i-th sensor. Each of the n sensor features represents a different feature.

[0059] According to an embodiment of this application, based on the initial training set Multiple initial training samples A covariance matrix is ​​constructed to determine the degree of co-variance among different sensor features. Based on this covariance matrix, feature selection is performed on multiple sensor features acquired by sensors from multiple initial training samples to obtain the target training set. During this process, sensor features with low relevance to performance prediction can be removed, so that each target training sample in the target training set only includes the sensor features with high relevance to performance prediction determined by feature selection. These retained sensor features can characterize any key feature in engine thermodynamics, mechanical dynamics, and aerodynamics.

[0060] According to an embodiment of this application, after feature selection, each target training sample in the target training set can be input into the initial prediction model to output the performance prediction information of the aero-engine. Based on the performance prediction information and the performance labels of the target training samples, the model parameters of the initial prediction model are adjusted to obtain the trained aero-engine performance detection model.

[0061] According to embodiments of this application, a covariance matrix characterizing the degree of coordinated change among different sensor features is generated from multiple sensor features collected by sensors in the initial training samples. Feature selection is then performed on the sensor features in the initial training samples based on the covariance matrix to obtain a target training set. The target training samples in the target training set are then used to train the model, thereby obtaining an aero-engine performance detection model. Since feature selection is performed on the sensor features in the initial training samples based on the covariance matrix of the degree of coordinated change among different sensor features, features with strong performance correlation can be filtered out. Therefore, model training based on the filtered sensor features can improve the model's prediction accuracy. Furthermore, using fewer sensor features for model training can increase the training speed and effectively reduce the computational resources required for model training.

[0062] According to an embodiment of this application, a covariance matrix characterizing the degree of coordinated change among different sensor features is generated based on multiple sensor features acquired by sensors for each initial training sample in the initial training set. This includes: generating an average training sample based on multiple initial training samples, wherein the average training sample includes multiple average sensor features, and the average sensor features are determined based on the sensor features corresponding to the average sensor features in the multiple initial training samples; and generating a covariance matrix based on the number of average training samples and the number of initial training samples.

[0063] According to embodiments of this application, when constructing the covariance matrix, for each type of sensor feature in the initial training samples, the feature values ​​of that type of sensor feature can be extracted from multiple initial training samples in the initial training set. The average of these multiple feature values ​​is then calculated to obtain the average sensor feature of that sensor feature. Combining the average sensor features of different types yields the average training sample. .

[0064] According to embodiments of this application, a covariance matrix is ​​generated based on the average number of training samples and the number of initial training samples. As shown in the following formula:

[0065]

[0066] in, The initial number of training samples is T, where T represents the transpose. Represents the average training samples. This indicates the degree of coordinated change among the features of each sensor.

[0067] According to embodiments of this application, before generating the covariance matrix, the method further includes:

[0068] Based on the Pearson correlation coefficient, sensor features in multiple initial training samples are filtered to obtain multiple processed initial training samples.

[0069] According to embodiments of this application, in order to further improve the quality of target training samples in the target training set, feature screening can be performed based on the Pearson correlation coefficient. Preliminary screening is used to evaluate the linear correlation strength between sensor features and prediction performance, thereby obtaining multiple processed initial training samples.

[0070] According to an embodiment of this application, based on the Pearson correlation coefficient, feature filtering is performed on sensor features in multiple initial training samples to obtain multiple processed initial training samples. This includes: for any target feature among the multiple sensor features of the initial training samples, calculating the correlation coefficient between the target feature and the performance of the aero-engine based on multiple target features in the initial training set and the performance label corresponding to each initial training sample; if the correlation coefficient is less than a correlation threshold, deleting the target feature of each initial training sample in the initial training set to obtain processed initial training samples.

[0071] According to embodiments of this application, the relevant threshold can be set according to actual needs, for example, it can be set to 0.8.

[0072] According to embodiments of this application, for any target feature among multiple sensor features of the initial training sample... Based on multiple target features in the initial training set and the performance labels corresponding to each initial training sample, the correlation coefficient between the target features and the performance of the aero-engine is calculated. As shown in the following formula:

[0073]

[0074] Where, x i For the target features collected by the sensor, y i That is the corresponding performance tag. The average feature corresponding to the target feature can be calculated by averaging the sensor features corresponding to the target feature in different initial training samples. It is the average performance label, which can be calculated by averaging the performance labels of sensor features corresponding to the target features in different initial training samples.

[0075] According to an embodiment of this application, with a correlation threshold of 0.8, if the correlation coefficient of the target feature is... If the correlation value is less than the correlation threshold of 0.8, the target feature is deleted from each initial training sample, thus retaining features that are highly correlated with the prediction performance and removing irrelevant or noisy features, thereby obtaining the processed initial training samples.

[0076] According to an embodiment of this application, based on the covariance matrix, feature selection is performed on multiple sensor features collected by sensors from multiple initial training samples to obtain a target training set, including: calculating the projection variance based on the covariance matrix for any sensor feature among the multiple sensor features; calculating the unit feature vector based on the covariance matrix and the projection variance; and generating the target training set based on the multiple unit feature vectors and the initial training samples.

[0077] According to embodiments of this application, for any one of a plurality of sensor features, the projection variance is calculated based on the covariance matrix, as shown in the following formula:

[0078]

[0079] in, The eigenvalues ​​of the sensor features represent the variance of the data along the corresponding eigenvector direction, i.e., the projection variance, and I is the identity matrix.

[0080] According to an embodiment of this application, the unit eigenvector is calculated based on the covariance matrix and the projection variance, as shown in the following formula:

[0081]

[0082] in, Representing unit eigenvectors, they can form orthogonal bases and eliminate redundant coupling between preserved features.

[0083] According to embodiments of this application, a target training set is generated based on multiple unit feature vectors and initial training samples. .

[0084] According to an embodiment of this application, generating a target training set based on multiple unit feature vectors and initial training samples includes: generating an average training sample based on the initial training sample; selecting multiple target feature vectors from the multiple unit feature vectors according to the size of the unit feature vectors, and constructing a transformation matrix based on the multiple target feature vectors; and generating the target training set based on the transformation matrix and the average training sample.

[0085] According to an embodiment of this application, an average training sample is generated based on the initial training samples. Based on the magnitude of the unit eigenvectors, the first k largest unit eigenvectors are used as target eigenvectors, thus forming a transformation matrix from the target eigenvectors. Based on the transformation matrix and the average training samples, the target training set X is generated. PCA As shown in the following formula:

[0086]

[0087] in, This refers to the target training set after removing redundancies.

[0088] According to an embodiment of this application, before inputting the target training samples into the initial prediction model, the method further includes: processing sensor features and random noise using a generative adversarial network for any sensor feature in the target training set to obtain discrimination parameters; generating feature distribution information based on the target discriminator and sensor features obtained by taking partial derivatives of the discrimination parameters; and generating a data-enhanced target training set based on the feature distribution information corresponding to different sensor features.

[0089] According to embodiments of this application, for each sensor feature after dimensionality reduction, a generative adversarial network is used to expand the sample size, thereby obtaining discriminant parameters. As shown in the following formula:

[0090]

[0091] In this model, G is the generator, responsible for mapping random noise z into a synthesized sequence. D is the discriminator, which outputs probability scores to distinguish between real and fake data. Indicates sensor characteristics, This represents the noise distribution.

[0092] According to embodiments of this application, the theoretically optimal target discriminator is obtained by taking the partial derivative with respect to V(D,G). According to the target discriminator Sensor characteristics Generate feature distribution information As shown in the following formula:

[0093]

[0094] in, This generates the distribution of the data, i.e., the feature distribution information. Adversarial training is used to... This generates high-quality augmented data, which is the target training set for data augmentation.

[0095] According to an embodiment of this application, before generating a data-enhanced target training set based on feature distribution information, the method further includes: using a genetic algorithm to genetically optimize the feature distribution information to obtain optimized feature distribution information.

[0096] According to embodiments of this application, in order to further optimize the data distribution and make the data distribution closer to the real value range, genetic operations such as heredity, crossover, and mutation in genetic algorithms can be used to optimize the feature distribution information, thereby obtaining optimized feature distribution information. The value range of this feature distribution information is more accurate, and expanding the samples based on this feature distribution information can obtain target training samples that are closer to the real environment.

[0097] According to an embodiment of this application, the target training samples are input into an initial prediction model to output performance prediction information for an aero-engine, including: generating a dimensional space matrix based on the embedding weight matrix of the transform network and the target training samples; calculating the query vector, key vector, and value vector for any attention head in the attention network based on the dimensional space matrix, and calculating the attention head weights based on the query vector, key vector, and value vector; concatenating the weights of multiple attention heads to obtain fused features; and inputting the fused features into a feedforward network to output performance prediction information.

[0098] According to embodiments of this application, based on the embedding weight matrix of the transform network ( ) and target training samples, generate a dimensional space matrix, for a target training set including multiple target training samples. The dimensional space matrix of the target training set can be generated using the following formula. :

[0099]

[0100] in, To embed the weight matrix, The bias term maps the k-dimensional input to a high-dimensional space matrix. .

[0101] According to embodiments of this application, for each attention head, three independent weight matrices are used. , , Calculate the query (Q) vector, key (K) vector, and value (V) vector.

[0102]

[0103] Calculate the attention head weight based on the query vector, key vector, and value vector. As shown in the following formula:

[0104]

[0105] in, For the dimensions of query and key, The softmax function is used as a scaling factor to prevent gradient vanishing; the softmax function yields a normalized attention weight matrix.

[0106] The multi-head splicing is then performed using the following formula to obtain the fused features. :

[0107]

[0108] Where h is the number of read / write heads, W o To output the weight matrix, the representation fusion of different feature subspaces is achieved.

[0109] The fused features are input into a feedforward network (FFN) to enhance the nonlinear representation, as shown in the following formula:

[0110]

[0111] Where x represents the fused feature, W1 and W2 are the weights, and b1 and b2 are the bias terms. This indicates performance prediction information.

[0112] According to an embodiment of this application, the training method for an aero-engine performance testing model further includes: generating individual chromosomes based on multiple model parameters of the aero-engine performance testing model, and generating an initial population based on the value range of each model parameter in the individual chromosome, wherein the initial population includes multiple initial individuals; iteratively performing the following operations: Operation 1: For each initial individual, inputting multiple test samples into the aero-engine performance testing model including the initial individual, and outputting performance test information corresponding to each test sample; Operation 2: Calculating the individual fitness with respect to the initial individual based on the multiple performance test information and the test label of each test sample; Operation 3: Selecting a target individual from the multiple initial individuals based on the individual fitness for genetic operation, and using the individual obtained from the genetic operation as a new initial individual, and returning to execute Operation 1; If the individual fitness is less than a preset fitness threshold, updating the parameters of the aero-engine performance testing model based on the initial individual corresponding to the individual fitness, to obtain an updated aero-engine performance testing model.

[0113] According to embodiments of this application, model parameters may include, but are not limited to, the hidden layer dimension d. model Learning rate The model parameters, such as the time window length W, are mapped to gene sequences to form individual chromosomes. As shown in the following formula:

[0114]

[0115] Each parameter takes a value within its preset search space, forming the initial population. N is the population size, and the initial population. any of This represents an initial individual.

[0116] According to an embodiment of this application, for each initial individual, multiple test samples are input into an aero-engine performance testing model including the initial individual, and performance test information corresponding to each test sample is output. Based on the multiple performance test information and the test label of each test sample, the individual fitness with the initial individual is calculated.

[0117] According to embodiments of this application, individual fitness is based on a fitness function defined according to the root mean square error (RMSE). It is computationally quantifiable. Furthermore, to balance the model's accuracy on the training set with its generalization ability on the test set, a regularization term is introduced. As shown in the following formula:

[0118]

[0119] Where y is the test label. This is performance test information, M tr and M te These represent the number of training samples and the number of test samples, respectively. The smaller the value of this function (error), the higher the individual's fitness.

[0120] According to an embodiment of this application, a target individual is selected from multiple initial individuals based on individual fitness for genetic operation, and the individual obtained from the genetic operation is used as a new initial individual for genetic operation, such as inheritance, crossover, mutation, etc., so that the above operation can be repeated using the new initial individual obtained from the genetic operation.

[0121] According to an embodiment of this application, when the individual fitness is less than a preset fitness threshold, the parameters of the aero-engine performance testing model are updated based on the initial individual corresponding to the individual fitness, resulting in an updated aero-engine performance testing model. The preset fitness threshold can be set according to actual needs, for example, it can be 0.6.

[0122] Figure 3 A heatmap of Pearson correlation coefficients between different sensor features in the initial training set according to an embodiment of this application is shown. Figure 4 The remaining useful life prediction results based on the C-MAPSS dataset according to an embodiment of this application are shown.

[0123] In one specific embodiment, the publicly available C-MAPSS dataset is used as the initial training set. For multiple initial training samples in the initial training set, principal component analysis is used to calculate the correlation between sensor features, and a heatmap of the Pearson correlation coefficient is plotted as follows. Figure 3As shown in the heatmap, the matrix composed of sensor features includes the number of cycles, operating parameters (Op.1-Op.3), and sensor parameters (S1-S21). The color gradient intuitively reflects the strength of the correlation: dark blue indicates a strong positive correlation (coefficient close to 1), black indicates a strong negative correlation (coefficient close to -1), and light blue or white corresponds to a weak correlation (coefficient close to 0).

[0124] from Figure 3 As can be seen, Op.1 and Op.2 exhibit significant characteristics. They form a dense, deep blue area together with core sensors such as S2 (total outlet temperature of the low-pressure compressor), S3 (total outlet temperature of the high-pressure compressor), and S4 (total outlet temperature of the low-pressure turbine). The correlation coefficients are generally greater than 0.8. This indicates a strong synergistic effect between these parameters and the temperature state of the engine's core flow path. This makes them key variables for capturing the relationship between the control system and thermal response. Conversely, they show weak, light blue correlations (coefficients below 0.3) with sensor features such as S7 (total outlet pressure of the high-pressure compressor) and S10 (engine pressure ratio). This highlights their unique influence on different feature types (temperature and pressure / pressure ratio). Retaining these two features can fully characterize the drive logic at the system's operating end.

[0125] Regarding sensor characteristics, S2 (total outlet temperature of the low-pressure compressor), S3 (total outlet temperature of the high-pressure compressor), and S4 (total outlet temperature of the low-pressure turbine) form a strongly coupled cluster, with cross-correlation coefficients all exceeding 0.95. These three parameters all belong to the temperature monitoring parameters of the engine's core flow path, reflecting the heat transfer process from the compressor to the turbine, representing the coordinated monitoring of common physical quantities. Retaining this group allows for comprehensive coverage of flow path temperature characteristics while compressing the size. S6 (total pressure in the bypass duct), S8 (physical velocity of the blower), and S9 (physical velocity of the core engine) form another strongly correlated block (correlation coefficients ranging from 0.85 to 0.92). Among them, S8 and S9 are directly related to the dynamic state of the engine rotor, while S6 reflects the aerodynamic characteristics of the bypass duct. They collectively correspond to the linked operating state of the blower's core engine and serve as key indicators of mechanical and aerodynamic performance. Their inclusion ensures the ability to characterize core operating conditions with low-dimensional data. Although S11 (high-pressure compressor outlet static pressure), S13 (fan corrected speed), S14 (core engine corrected speed), and S15 (bypass ratio) do not form excessively large correlation clusters, they exhibit moderate correlation in local regions (summing up to 0.7 to 0.8). S11 complements the static pressure state of the high-pressure compressor, S13 and S14 are normalized representations of speed, and S15 reflects the flow distribution between the inner and outer pipes. These four parameters collectively cover key derived parameters of engine aerodynamic performance, and their preservation prevents the loss of inter-subsystem synergy due to excessive dimensionality reduction. Although S17 (exhaust enthalpy) has a weak correlation with most parameters (coefficients ranging from 0.2 to 0.4), its monitored exhaust energy state can supplement the operating information of engine auxiliary systems (such as the exhaust circuit), thereby improving the characterization of overall engine performance.

[0126] In summary, selecting 13 sets of sensor features—Op.1, Op.2, S2-S4, S6, S8, S9, S11, S13, S14, S15, and S17—as the sensor features in the target training samples has two advantages. First, it reduces the number of parameters by merging highly redundant parameters, for example, reducing S2-S4 from 24 to 13. Second, it fully preserves the important role of operating parameters, core flow path temperature and velocity characteristics, key aerodynamic performance parameters, and auxiliary system information. This ensures that despite the reduced scale, the target training samples can still accurately describe the key relevant logic of engine thermal, mechanical, and aerodynamic operations. This lays the foundation for efficient and comprehensive subsequent model training.

[0127] By predicting the remaining useful life using the aforementioned target training samples, we can obtain... Figure 4The prediction results shown in the figure are as follows: the black line represents the actual observation, while the colored line represents the prediction results under different configurations. By observation and comparison, it can be found that the model optimized by the genetic algorithm has a significantly improved fit between the colored line and the black line, which proves the role of automatic hyperparameter optimization in improving model stability. Figure 4 In this context, High-Dim represents high-dimensional data, i.e., the initial training set, and Low-Dim represents low-dimensional data, i.e., the target training set.

[0128] Root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and score were selected as evaluation metrics for RUL prediction. This experimental framework aims to systematically study model performance under different data complexities and optimization strategies. Specifically, two types of input data (a high-dimensional initial training set (24 parameters) and a low-dimensional target training set (13 parameters)) were fed into two baseline models: LSTM and Transformer. To further enhance model performance and explore the impact of optimization, eliminate the influence of randomness, and ensure the reliability of the results, each combination of model and data type (high-dimensional / low-dimensional) was tested through 10 iterations of repeated experiments.

[0129] Table 1 Assessment of Remaining Useful Life Prediction Results

[0130]

[0131] As can be seen from Table 1, the performance prediction information (i.e., remaining lifetime) predicted by the low-dimensional target training set in this embodiment has better accuracy than that of the high-dimensional initial training set.

[0132] Figure 5 A heatmap showing the Pearson correlation coefficients between sensor features of a 1000 kg-class aircraft turbofan engine according to an embodiment of this application is presented. Figure 6 A data-enhanced schematic diagram of a 1000 kg-class aircraft turbofan engine according to an embodiment of this application is shown. Figure 7 A schematic diagram showing the thrust comparison of a 1000 kg-class aircraft turbofan engine according to an embodiment of this application is provided.

[0133] In another specific embodiment, the method of this embodiment is illustrated using long-term test data of a 1000-kilogram-class aircraft turbofan engine as the initial training set.

[0134] The specific monitoring parameters of the initial training set are shown in Table 2. This dataset for actual engine testing contains a wide range of monitoring parameters (i.e., sensor characteristics). It includes atmospheric pressure indices (e.g., P0, P0.1, P0.2) and inlet-related conditions. These inlet conditions include total inlet temperature (Tt1), total inlet pressure (Pt1), inlet static pressure (Ps1), inlet air mass flow rate (Wai (kg / s)) and its corrected counterpart (Wai.cor), as well as parameters describing the inlet diameter and associated pressure and temperature distribution. Fuel-related parameters are also included, such as tank temperature (Tf), main fuel mass flow rate (Wf), and fuel supply pressure (Pf). Electrical and starting-related characteristics are reflected by the starting voltage (Ustart), starting current (Istart), excitation current, and the voltage and current of the 270V generator (Ugen, 270V and Igen, 270V). In addition, there are temperature parameters within the oil system (Toil, return oil temperature and supply oil temperature), motor and nozzle temperatures (Tmotor, 1, Tmotor, 2 for engine temperature and nozzle temperature, nozzle temperature 2, nozzle temperature 2), combustion chamber temperatures (combustion chamber temperature and combustion chamber after-temperature 2, combustion chamber temperature 2), fuel pressure at the nozzle (Pf, nozzle, 1, Pf, nozzle, 2), and exhaust temperature (Texhaust (daq)). These parameters comprehensively reflect the engine's operating status in multiple dimensions, enabling a comprehensive evaluation of its performance during testing.

[0135] Table 2

[0136]

[0137] According to embodiments of this application, a target dataset is formed using the method of this application, combined with... Figure 5The heatmap shown (where blue indicates a strong positive correlation and black indicates a strong negative correlation, both representing strong correlations) illustrates the logical relationship between thrust and other parameters as follows: In the intake system: S1 (P0_Atmosphere), S3 (Pt1_Intake), and S4 (Ps1_Intake) show a strong negative correlation with thrust, reflecting the complex inverse relationship between intake pressure parameters and thrust generation under actual operating conditions. In contrast, S5 (Wa1) has a strong positive correlation with thrust, reflecting the logical flow-driven combustion. Sufficient intake flow ensures uniform fuel-air mixing in the combustion chamber, stable energy release during combustion, and directly supports thrust enhancement. In the fuel supply system: S9 (Wf_Large) and S10 (Pf_Fuel Supply) show a strong positive correlation with thrust. They collaboratively control fuel supply, laying the fuel foundation for thrust generation. S39 (Pf_Nozzle1), S40 (Pf_Nozzle2), S41 (Pf-B), and S42 (Pf-A) are highly correlated with S9 (Wf_Large) and S10 (Pf_FuelSupply), regulating fuel injection uniformity and pressure stability, indirectly improving thrust output consistency. In environmental correction and pressure distribution parameters: S12 (Wa1cor), S13 (P0_1), and S14 (P0_2) are strongly positively correlated with thrust. S12 eliminates misjudgments caused by environmental interference, while S13 and S14 compensate for spatial differences in atmospheric pressure, improving the accuracy of the intake parameter-thrust correlation analysis. S18 (Ps01_1), S19 (Ps01_2), S20 (Ps01_3), and S21 (Ps01_4) are highly correlated with S3 (Pt1_Intake) and S4 (Ps1_Intake), reflecting the pressure uniformity of the intake manifold cross-section and ensuring stable combustion. Among the post-combustion and fuel pressure parameters: S37 (T afterburner 1), S36 (T afterburner 2), and S35 (T afterburner 3) are strongly positively correlated with thrust (blue in the heatmap), directly reflecting combustion intensity and energy conversion efficiency. S38 (Pf_J) is highly correlated with core fuel supply parameters, supporting the stability of post-combustion temperature. The exhaust temperature parameter S43 (exhaust temperature daq) is highly correlated with thrust. As the final feedback on combustion efficiency, it forms a closed-loop connection between intake, combustion, thrust, and exhaust.

[0138] Based on the above analysis, these parameters cover the entire process from intake pressure regulation → fuel supply coordination → environmental correction and adaptation → post-combustion effect verification. Through a strongly positive and negative correlation interconnection network, they accurately capture the physical mechanism of the engine from intake to thrust output. This application selected 22 parameters (S1, S3, S4, S5, S9, S10, S12, S13, S14, S18, S19, S20, S21, S35, S36, S37, S38, S39, S40, S41, S42, and S43) as model inputs (i.e., sensor features in the target training samples) to learn the "coordination-constraint" relationship between engine subsystems, accurately analyze the multi-parameter coupling law of thrust generation, and provide solid data support for engine thrust prediction.

[0139] The data augmentation analysis method is as follows: First, 22 sets of dimensionality-reduced sensor parameter data are extracted, each containing 10,000 data points. This data is used to construct a training dataset of size 22×10,000 (i.e., the target training set), which is then fed into LSTM and Transformer models for training, and optimized using GA. Subsequently, this training dataset is augmented to generate expanded datasets of sizes 22×20,000 and 22×30,000, see [link to documentation]. Figure 6 Meanwhile, a dataset of the same size was extracted from the original data as a control dataset. Figure 6 (a) and Figure 6 (c) The raw data exhibits significant noise fluctuations in the latter half, while Figure 6 (b) and Figure 6 (d) The enhanced data shows a more convergent and smooth sequence feature within the light green or orange boxes, which indicates that the generative adversarial network effectively suppresses random noise while preserving the original fluctuation pattern.

[0140] All test sets are 10% the size of their corresponding training sets, and the data split starts from a random point after the training set. This setup improves the reliability of the experiments. After constructing the dataset, training and test sets of different sizes are fed into the model for training. The generated predictions are then compared and analyzed. Due to the large number of experiments, using multiple evaluation metrics would significantly increase the workload of data processing. To focus on the training effect of the model, only RMSE is selected as the evaluation metric for model performance.

[0141] Following the design of the above scheme, an augmented dataset was generated. Compared to the original dataset (the first 10,000 sets of original data), the augmented data refined the noise details while retaining the basic fluctuation range of the original dataset. This refinement gives the data sequence greater regularity while preserving the inherent fluctuation characteristics of the original data. When compared with the unaugmented data (the latter part of the original data within the same set), as shown in Table 3, the augmented data clearly shows enhanced stability, characterized by a more convergent fluctuation range. Specifically, its fluctuation amplitude is reduced, and noise interference is mitigated. Simultaneously, the regularity of the sequence is strengthened: irregular fluctuations caused by random noise in the unaugmented data are effectively smoothed. This makes the augmented data sequence more coherent and more suitable for subsequent analysis tasks.

[0142] Table 3

[0143]

[0144] According to embodiments of this application, drawing is performed based on target training samples and corresponding samples. Figure 7 The diagram showing the thrust comparison is as follows. Figure 7 The figure demonstrates the thrust prediction performance based on augmented data. The blue dots represent the measured thrust on the test bench (True Thrust), and the yellow dashed line represents the predicted thrust (Predicted Thrust). The nearly overlapping curve characteristics of the two and the extremely low root mean square error (RMSE) fully verify that the method has high accuracy and reliability when processing complex real-world engineering data.

[0145] According to embodiments of this application, a physical logic-guided feature selection mechanism is used to perform eigenvalue decomposition and projection space reconstruction on high-dimensional sensor parameters with a correlation coefficient threshold greater than 0.8 using principal component analysis. This not only effectively eliminates redundant information and reduces computational overhead, but also preserves the key physical characteristics of engine thermal, mechanical, and aerodynamic coupling through linear combination of feature vectors, laying a high-quality data foundation for subsequent model training. Secondly, addressing the pain point of scarce and costly real-world test data in actual aero-engine engineering scenarios, a generative adversarial network (GAN) is introduced to enhance single-dimensional parameters. The minimax game between the generator and discriminator effectively refines data noise and expands the sample size, enabling the model to obtain more regular and stable training signals with limited economic investment, significantly enhancing the model's generalization ability and robustness in real-world engineering environments.

[0146] In terms of prediction architecture, this method fully leverages the potential of the Transformer architecture in capturing long-term temporal dependencies and combines it with a genetic algorithm for global intelligent evolutionary search of model hyperparameters, solving the problems of low efficiency and susceptibility to local optima in manual parameter tuning. Experimental results show that this invention performs excellently in predicting the remaining service life of the NASA C-MAPSS dataset and the thrust prediction of a 1000kg-class aero-engine test vehicle. Especially with an augmented dataset of 30,000 samples, the GA-Transformer framework exhibits superior prediction accuracy, with an extremely low root mean square error. Compared to traditional long short-term memory networks, it demonstrates greater adaptability and performance ceiling when handling complex, high-dimensional real-world engineering data. In summary, this method, through the integrated synergy of feature reduction, data augmentation, and intelligent optimization, overcomes the technical bottlenecks of data scarcity and high complexity in real-world scenarios, providing reliable data support for the production, maintenance, and repair of aero-engines. This has profound significance for improving aviation operational safety and optimizing proactive maintenance decisions.

[0147] Figure 8 A flowchart of a method for detecting the performance of an aircraft engine according to an embodiment of this application is shown.

[0148] like Figure 8 As shown, the methods for testing the performance of aero-engines include operations S801 to S802.

[0149] The S801 is used to acquire multiple performance parameters to be detected from the target sensors in the aero-engine.

[0150] When operating S802, multiple performance parameters to be tested are input into the aero-engine performance testing model, and predicted performance information is output.

[0151] According to embodiments of this application, the predicted performance information may include remaining service life after thrust is achieved. In actual testing, if the thrust of an aero-engine needs to be tested, 22 performance parameters to be tested can be collected, including S1, S3, S4, S5, S9, S10, S12, S13, S14, S18, S19, S20, S21, S35, S36, S37, S38, S39, S40, S41, S42, and S43 (see Table 2). If the remaining service life of the aero-engine needs to be tested, 13 performance parameters to be tested can be collected, including Op.1, Op.2, S2-S4, S6, S8, S9, S11, S13, S14, S15, and S17.

[0152] According to an embodiment of this application, the above-mentioned performance parameters to be tested are input into the aero-engine performance testing model, and the predicted performance information such as the thrust or remaining service life of the aero-engine is output.

[0153] According to embodiments of this application, a covariance matrix characterizing the degree of coordinated change among different sensor features is generated from multiple sensor features collected by sensors in the initial training samples. Feature selection is then performed on the sensor features in the initial training samples based on the covariance matrix to obtain a target training set. The target training samples in the target training set are then used to train the model, thereby obtaining an aero-engine performance detection model. Since feature selection is performed on the sensor features in the initial training samples based on the covariance matrix of the degree of coordinated change among different sensor features, features with strong performance correlation can be filtered out. Therefore, model training based on the filtered sensor features can improve the model's prediction accuracy. Furthermore, using fewer sensor features for model training can increase the training speed and effectively reduce the computational resources required for model training.

[0154] Figure 9 A block diagram of a training apparatus for an aircraft engine performance testing model according to an embodiment of this application is shown.

[0155] like Figure 9 As shown, the training device 900 for the aero-engine performance testing model includes a generation module 910, a selection module 920, a prediction module 930, and a parameter tuning module 940.

[0156] The generation module 910 is used to generate a covariance matrix that characterizes the degree of coordinated change among different sensor features based on multiple sensor features collected by sensors for each initial training sample in the initial training set. The different sensor features characterize different performance detection parameters corresponding to different components in the aero-engine.

[0157] The selection module 920 is used to perform feature selection on multiple sensor features acquired by multiple sensors from multiple initial training samples based on the covariance matrix to obtain a target training set. The target training samples in the target training set include multiple sensor features determined by feature selection. The determined sensor features characterize key features of engine thermodynamics, mechanical dynamics and aerodynamics.

[0158] The prediction module 930 is used to input the target training sample into the initial prediction model for each target training sample and output the performance prediction information of the aero-engine. The initial prediction model is constructed based on the transform network and the attention network.

[0159] The parameter tuning module 940 is used to adjust the model parameters of the initial prediction model based on performance prediction information and the performance labels of the target training samples, so as to obtain a trained aero-engine performance detection model.

[0160] According to embodiments of this application, a covariance matrix characterizing the degree of coordinated change among different sensor features is generated from multiple sensor features collected by sensors in the initial training samples. Feature selection is then performed on the sensor features in the initial training samples based on the covariance matrix to obtain a target training set. The target training samples in the target training set are then used to train the model, thereby obtaining an aero-engine performance detection model. Since feature selection is performed on the sensor features in the initial training samples based on the covariance matrix of the degree of coordinated change among different sensor features, features with strong performance correlation can be filtered out. Therefore, model training based on the filtered sensor features can improve the model's prediction accuracy. Furthermore, using fewer sensor features for model training can increase the training speed and effectively reduce the computational resources required for model training.

[0161] Figure 10 A block diagram of an aircraft engine performance testing apparatus according to an embodiment of this application is shown.

[0162] like Figure 10 As shown, the aircraft engine performance testing device 1000 includes an acquisition module 1010 and a testing module 1020.

[0163] The acquisition module 1010 is used to acquire multiple performance parameters to be detected collected by the target sensor in the aero-engine.

[0164] The detection module 1020 is used to input multiple performance parameters to be detected into the aero-engine performance detection model and output predicted performance information.

[0165] According to embodiments of this application, a covariance matrix characterizing the degree of coordinated change among different sensor features is generated from multiple sensor features collected by sensors in the initial training samples. Feature selection is then performed on the sensor features in the initial training samples based on the covariance matrix to obtain a target training set. The target training samples in the target training set are then used to train the model, thereby obtaining an aero-engine performance detection model. Since feature selection is performed on the sensor features in the initial training samples based on the covariance matrix of the degree of coordinated change among different sensor features, features with strong performance correlation can be filtered out. Therefore, model training based on the filtered sensor features can improve the model's prediction accuracy. Furthermore, using fewer sensor features for model training can increase the training speed and effectively reduce the computational resources required for model training.

[0166] Any one or more of the modules, submodules, units, and subunits according to the embodiments of this application, or at least part of the functions of any one or more of them, can be implemented in one module. Any one or more of the modules, submodules, units, and subunits according to the embodiments of this application can be implemented by dividing them into multiple modules. Any one or more of the modules, submodules, units, and subunits according to the embodiments of this application can be at least partially implemented as hardware circuits, such as field-programmable gate arrays (FPGAs), programmable logic arrays (PLAs), systems-on-a-chip, systems-on-a-substrate, systems-on-package, application-specific integrated circuits (ASICs), or implemented by hardware or firmware in any other reasonable manner by integrating or packaging circuits, or implemented in any one of software, hardware, and firmware, or in a suitable combination of any of these. Alternatively, one or more of the modules, submodules, units, and subunits according to the embodiments of this application can be at least partially implemented as computer program modules, which, when run, can perform corresponding functions.

[0167] For example, any and multiple modules among the generation module 910, selection module 920, prediction module 930, parameter tuning module 940, or acquisition module 1010 and detection module 1020 can be combined into one module / unit / subunit, or any one of these modules / units / subunits can be split into multiple modules / units / subunits. Alternatively, at least some of the functions of one or more of these modules / units / subunits can be combined with at least some of the functions of other modules / units / subunits and implemented in one module / unit / subunit. According to embodiments of this application, at least one of the generation module 910, selection module 920, prediction module 930, parameter tuning module 940, or acquisition module 1010, detection module 1020 can be at least partially implemented as hardware circuits, such as field-programmable gate arrays (FPGAs), programmable logic arrays (PLAs), systems-on-a-chip, systems-on-a-substrate, systems-on-package, application-specific integrated circuits (ASICs), or any other reasonable means of integrating or packaging circuits, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the generation module 910, selection module 920, prediction module 930, parameter tuning module 940, or acquisition module 1010, detection module 1020 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0168] It should be noted that the training device and testing device of the aero-engine performance testing model in the embodiments of this application correspond to the training method and testing method of the aero-engine performance testing model in the embodiments of this application. For a detailed description of the training device and testing device of the aero-engine performance testing model, please refer to the training method and testing method of the aero-engine performance testing model, which will not be repeated here.

[0169] Figure 11 A block diagram of an electronic device suitable for implementing the methods described above, according to an embodiment of this application, is shown. Figure 11 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0170] like Figure 11 As shown, an electronic device 1100 according to an embodiment of this application includes a processor 1101, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1102 or a program loaded from a storage portion 1108 into a random access memory (RAM) 1103. The processor 1101 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1101 may also include onboard memory for caching purposes. The processor 1101 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0171] RAM 1103 stores various programs and data required for the operation of electronic device 1100. Processor 1101, ROM 1102, and RAM 1103 are interconnected via bus 1104. Processor 1101 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 1102 and / or RAM 1103. It should be noted that the programs may also be stored in one or more memories other than ROM 1102 and RAM 1103. Processor 1101 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0172] According to embodiments of this application, the electronic device 1100 may further include an input / output (I / O) interface 1105, which is also connected to a bus 1104. The electronic device 1100 may also include one or more of the following components connected to the input / output (I / O) interface 1105: an input section 1106 including a keyboard, mouse, etc.; an output section 1107 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1108 including a hard disk, etc.; and a communication section 1109 including a network interface card such as a LAN card, modem, etc. The communication section 1109 performs communication processing via a network such as the Internet. A drive 1110 is also connected to the input / output (I / O) interface 1105 as needed. A removable medium 1111, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 1110 as needed so that computer programs read from it can be installed into the storage section 1108 as needed.

[0173] According to embodiments of this application, the method flow according to embodiments of this application can be implemented as a computer software program. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable storage medium, the computer program containing program code for performing the method shown in the flowchart. In such embodiments, the computer program can be downloaded and installed from a network via communication section 1109, and / or installed from removable medium 1111. When the computer program is executed by processor 1101, it performs the functions defined in the system of embodiments of this application. According to embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0174] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0175] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium. Examples include, but are not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0176] For example, according to embodiments of this application, a computer-readable storage medium may include the ROM 1102 and / or RAM 1103 described above and / or one or more memories other than ROM 1102 and RAM 1103.

[0177] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods provided in the embodiments of this application. When the computer program product is run on an electronic device, the program code is used to enable the electronic device to implement the methods provided in the embodiments of this application.

[0178] When the computer program is executed by the processor 1101, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0179] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 1109, and / or installed from the removable medium 1111. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0180] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0181] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions. Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations are not explicitly described in this application. In particular, without departing from the spirit and teachings of this application, the features described in the various embodiments of this application can be combined and / or combined in various ways. All such combinations and / or combinations fall within the scope of this application.

[0182] The embodiments of this application have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of this application. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. This application does not depart from its scope, and those skilled in the art can make various substitutions and modifications, all of which should fall within the scope of this application.

Claims

1. A training method for an aero-engine performance testing model, characterized in that, include: Based on the sensor features collected by multiple sensors for each initial training sample in the initial training set, a covariance matrix is ​​generated to characterize the degree of coordinated change among different sensor features, wherein different sensor features characterize different performance detection parameters of different components in the aero-engine. Based on the covariance matrix, feature selection is performed on multiple sensor features collected by sensors from multiple initial training samples to obtain a target training set. The target training samples in the target training set include multiple sensor features determined by feature selection. The determined sensor features characterize key features of engine thermodynamics, mechanical dynamics and aerodynamics. For each target training sample, the target training sample is input into an initial prediction model, and the performance prediction information of the aero-engine is output. The initial prediction model is constructed based on a transform network and an attention network. Based on the performance prediction information and the performance labels of the target training samples, the model parameters of the initial prediction model are adjusted to obtain a trained aero-engine performance detection model.

2. The method according to claim 1, characterized in that, Based on the sensor features acquired by multiple sensors for each initial training sample in the initial training set, a covariance matrix characterizing the degree of coordinated variation among different sensor features is generated, including: An average training sample is generated based on a plurality of initial training samples, wherein the average training sample includes a plurality of average sensor features, and the average sensor features are determined based on sensor features corresponding to the average sensor features in the plurality of initial training samples; The covariance matrix is ​​generated based on the average number of training samples and the number of initial training samples.

3. The method according to claim 1, characterized in that, Before generating the covariance matrix, the following steps are also included: Based on the Pearson correlation coefficient, sensor features in multiple initial training samples are filtered to obtain multiple processed initial training samples.

4. The method according to claim 3, characterized in that, Based on the Pearson correlation coefficient, feature filtering is performed on the sensor features in multiple initial training samples to obtain multiple processed initial training samples, including: For any target feature among the multiple sensor features of the initial training sample, the correlation coefficient between the target feature and the performance of the aero-engine is calculated based on the multiple target features in the initial training set and the performance label corresponding to each initial training sample. If the correlation coefficient is less than the correlation threshold, the target features of each initial training sample in the initial training set are deleted to obtain the processed initial training sample.

5. The method according to claim 1 or 2, characterized in that, Based on the covariance matrix, feature selection is performed on multiple sensor features acquired by sensors from multiple initial training samples to obtain a target training set, including: For any sensor feature among multiple sensor features, the projection variance is calculated based on the covariance matrix; Calculate the unit eigenvector based on the covariance matrix and the projection variance; The target training set is generated based on multiple unit feature vectors and the initial training samples.

6. The method according to claim 5, characterized in that, The target training set is generated based on multiple unit feature vectors and the initial training samples, including: Based on the initial training samples, generate average training samples; According to the magnitude of the unit feature vector, select multiple target feature vectors from multiple unit feature vectors, and construct a transformation matrix based on the multiple target feature vectors; The target training set is generated based on the transformation matrix and the average training samples.

7. The method according to claim 1, characterized in that, Before inputting the target training samples into the initial prediction model, the method further includes: For any sensor feature in the target training set, a generative adversarial network is used to process the sensor feature and random noise to obtain discrimination parameters; Based on the target discriminator obtained by taking the partial derivative of the discrimination parameters and the sensor features, feature distribution information is generated; Generate a data-enhanced target training set based on the feature distribution information corresponding to different sensor features; This process, which begins before generating the data-enhanced target training set based on feature distribution information, also includes: The feature distribution information is genetically optimized using a genetic algorithm to obtain the optimized feature distribution information.

8. The method according to claim 1, characterized in that, The target training samples are input into the initial prediction model, and the performance prediction information of the aero-engine is output, including: A dimensional space matrix is ​​generated based on the embedding weight matrix of the transform network and the target training samples; For any attention head in the attention network, the query vector, key vector, and value vector are calculated based on the dimensional space matrix, and the attention head weights are calculated based on the query vector, key vector, and value vector. The weights of multiple attention heads are concatenated to obtain a fused feature; The fused features are input into a feedforward network, which outputs the performance prediction information.

9. The method according to claim 1, characterized in that, Also includes: Individual chromosomes are generated based on multiple model parameters of the aero-engine performance testing model, and an initial population is generated based on the value range of each model parameter in the individual chromosome, wherein the initial population includes multiple initial individuals; Perform the following operations iteratively: Operation 1: For each initial individual, input multiple test samples into the aero-engine performance testing model including the initial individual, and output the performance test information corresponding to each test sample; Step 2: Calculate the individual fitness with the initial individual based on the multiple performance test information and the test label of each test sample; Operation 3: Select a target individual from multiple initial individuals based on individual fitness, perform genetic operations, and use the individual obtained from the genetic operations as a new initial individual to return to the execution of Operation 1; If the individual fitness is less than a preset fitness threshold, the parameters of the aero-engine performance testing model are updated based on the initial individual corresponding to the individual fitness, resulting in an updated aero-engine performance testing model.

10. A method for testing the performance of an aero-engine, characterized in that, include: Acquire multiple performance parameters to be detected by target sensors in an aero-engine; Multiple performance parameters to be detected are input into an aero-engine performance detection model, and predicted performance information is output. The aero-engine performance detection model is trained based on the method of any one of claims 1 to 9.

Images