An accuracy improvement method for aero-engine performance digital twinning

Abstract

The application particularly relates to a precision improvement method for aero-engine performance digital twinning, which comprises the following steps: establishing a compensation model; setting two output channels to respectively output two output values; comparing and calculating the two output values with corresponding performance prediction target values respectively to obtain two loss parameters; iteratively inputting the loss parameters into the basic digital twinning model and the compensation model respectively until the error converges; and predicting the aero-engine performance parameters, and correspondingly inputting the monitoring data of the engine into the converged basic digital twinning model to obtain target performance parameter prediction values. The purpose is to gradually eliminate errors through the coupling training of the compensation model and the basic model, directly couple the compensation effect of the compensation model into the basic digital twinning model without increasing the complexity of the basic model, improve the prediction precision without increasing the prediction time, and ensure the real-time performance.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of aero-engine fault diagnosis, and particularly relates to a precision improvement method for aero-engine performance digital twinning. BACKGROUND

[0002] There are few measurable performance parameters of an aero-engine on board, and there are many other monitoring parameters, and it is of great significance to establish an accurate engine performance real-time prediction model for flight state evaluation, flight safety guarantee and engine fault diagnosis. Before each engine is installed, a test run will be carried out to accumulate a large amount of data, and a variety of parameters will be monitored in the test run, including environmental parameters: environmental temperature, environmental pressure, environmental humidity, etc.; state parameters: compressor inlet guide vane angle, rotor speed, throttle lever position, etc.; performance parameters: thrust, fuel flow, low-pressure turbine afterburner gas total temperature, etc.; based on these test data, an engine performance parameter prediction model can be established. At present, a large number of researches have been made on this work, mainly based on data-driven artificial intelligence methods, a single prediction model is established, a mapping relationship between various monitoring data and target performance parameters is found, and the target performance parameters are predicted. However, for a single prediction model, based on Taylor expansion, it can be found that the prediction error always exists, and it is necessary to eliminate the error generated in the prediction to make the final prediction result more accurate. SUMMARY

[0003] The application provides a precision improvement method for aero-engine performance digital twinning to solve the error problem caused by a single prediction model, which provides a compensation model, and gradually iteratively eliminates errors through coupling training of the compensation model and the basic model, directly couples the compensation effect of the compensation model into the basic prediction model without increasing the complexity of the basic model, improves the prediction precision without increasing the prediction time, and guarantees the real-time performance.

[0004] To achieve the above application purposes, the technical solution adopted by the application is as follows: a precision improvement method for aero-engine performance digital twinning, comprising the following steps:

[0005] S1: establishing an engine performance prediction basic digital twinning model: screening engine operation data, and adopting a time-series artificial intelligence algorithm to construct an engine performance prediction basic digital twinning model;

[0006] S2: establishing a compensation model: the structure and algorithm of the compensation model are consistent with those of the engine performance prediction basic digital twinning model, or the compensation model is a non-time-series algorithm;

[0007] S3: Establish model output channels: Set up two output channels, one of which is the output channel of the basic digital twin model for engine performance prediction, and the other is the sum output channel of the basic digital twin model for engine performance prediction and the compensation model.

[0008] S4: Backpropagation of training error: Compare the two output values ​​obtained in step S3 with the corresponding performance prediction target values ​​to obtain two loss parameters. Iterate the loss parameters to the basic digital twin model and compensation model for engine performance prediction for training, and repeat steps S3 and S4 until the error converges.

[0009] S5: Predict the performance parameters of the aero-engine: Input the data monitored under any operating state of the engine into the basic digital twin model for engine performance prediction that has been converged in step S4 to obtain the predicted values ​​of the target performance parameters.

[0010] Preferably, in step S1, the selected engine operating data includes: environmental parameters, working status parameters of each component, component performance parameters, and overall engine performance parameters. After processing the selected data, a training dataset is established.

[0011] Preferably, in step S1, the temporal artificial intelligence algorithm includes a long short-term memory network, a wave network, and an attention mechanism network. Based on the data selected in step S1, the number of input layer nodes, output layer nodes, network layers, and network layer nodes are determined to build a basic digital twin model for performance prediction.

[0012] Preferably, in step S4, the loss parameter is calculated by comparing the two output values ​​obtained in step S3 with the corresponding performance prediction target values ​​to obtain two loss values, loss1 and loss2. Loss1 is the loss value calculated through the output channel of the engine performance prediction basic digital twin model, and loss2 is the loss value calculated through the output channel of the sum of the engine performance prediction basic digital twin model and the compensation model. The two values ​​after the sum of loss2, loss1 and loss2 constitute the loss parameter, which is used as the loss parameter of the compensation model and the engine performance prediction basic digital twin model, respectively.

[0013] Preferably, the loss value is calculated using the cross-entropy loss function:

[0014]

[0015] Where H(Y,y) is the true value of the target parameter Y(x) i ) and model predicted value y(x) i The cross-entropy loss function value.

[0016] Preferably, the loss value calculation method adopts a mean square error loss function:

[0017]

[0018] Wherein MSE(Y,y) is the mean square error loss function value of the target parameter true value Y(x i ) and the model predicted value y(x i ).

[0019] Preferably, the training algorithm in step S4 includes RMSPorp and Adam.

[0020] Preferably, the engine arbitrary working state in step S5 includes starting, accelerating, cruising, boosting and decelerating states; the sensor detection data is extracted as the input of the engine performance prediction basic digital twin model converged in step S4.

[0021] An electronic device comprises:

[0022] One or more processors;

[0023] Storage device for storing one or more programs;

[0024] When the one or more programs are executed by the one or more processors, the one or more processors implement the precision improvement method for the aero-engine performance digital twin as described above.

[0025] A computer readable medium, the readable medium stores a computer program, the computer program is executed by a processor to implement the precision improvement method for the aero-engine performance digital twin as described above.

[0026] The present application has the following beneficial effects:

[0027] 1、The method of the present application introduces a compensation model to eliminate the first-order error that cannot be eliminated in a single prediction model. The model prediction error caused by sensor measurement error and engine performance degradation is compensated by the compensation model. The compensation model is coupled with the basic model, so that in the subsequent operation process, the prediction accuracy is improved while the operation speed is guaranteed, and the real-time performance of the prediction is ensured.

[0028] 2、In the present application, after coupling the compensation model with the basic model, the data can be directly input into the coupled basic model to complete the performance prediction, without the need to input the parameters into the compensation model. The model can be used on board to predict the engine performance parameters in real time. BRIEF DESCRIPTION OF DRAWINGS

[0029] Figure 1 The flow chart of the method of the present application;

[0030] Figure 2 A schematic diagram of the loss function propagation path for the coupled training method;

[0031] Figure 3 It is a model structure. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Unless otherwise specified, the technical means used in the embodiments are conventional means well known to those skilled in the art.

[0033] In the description of this invention, it should be understood that the terms "longitudinal", "lateral", "up", "down", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this invention, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention.

[0034] When building a single prediction model, Taylor expansion reveals that a single prediction model always has a prediction error, as shown in the following equation:

[0035]

[0036] In the formula, F represents the target performance parameter and each characteristic parameter x1, x2, ..., x3. n The true mapping relationship between them, f(v1,v2,...,v n ) represents the measured values ​​(v1, v2, ..., v) of the corresponding feature parameters of a single prediction model. n The predicted values ​​generated on the engine may be non-zero due to measurement errors in monitoring parameters and engine performance degradation. This can lead to non-zero first-order and second-order terms in subsequent expansions, resulting in prediction errors.

[0037] Since a single prediction model will always have error terms such as first-order error and second-order error, and it can be found that as the order increases, the error value of the corresponding order decreases exponentially. Therefore, the compensation model established in this invention mainly compensates for the first-order error term of the basic prediction model, and does not consider subsequent error terms, thereby improving the prediction accuracy.

[0038] like Figures 1-2 As shown, the method for improving the accuracy of digital twins for aero-engine performance according to the present invention specifically includes the following steps:

[0039] S1: Establish an engine performance prediction basic digital twin model (basic model): screen engine operating data, which can include: environmental parameters, such as: ambient temperature, ambient pressure, ambient humidity, etc.; component working state parameters, such as: fan speed, high pressure rotor speed, throttle lever position, etc.; component performance parameters, such as: compressor pressure ratio, combustor oil-gas ratio, turbine pressure ratio, etc.; whole machine performance parameters, such as: thrust, specific fuel consumption, etc.; after screening and processing suitable parameters, a time series artificial intelligence algorithm is used to build an engine performance prediction basic digital twin model; data processing mainly removes obviously incorrect data in the data, such as non-numeric, out-of-limit values, infinite values, etc., and establishes a training data set.

[0040] The time series artificial intelligence algorithm includes long short-term memory network, wave network and attention mechanism network. When building the engine performance prediction basic digital twin model, the physical structure of the aero-engine is imitated, each component network layer is built, each component network layer is connected according to the physical connection structure of each component of the engine, and the number of input layer nodes, the number of output layer nodes, the number of network layers and the number of network layer nodes are determined based on the selected feature parameter dimensions of each component.

[0041] S2: Establish a compensation model: the structure and algorithm of the compensation model are consistent with those of the engine performance prediction basic digital twin model; the compensation model is used to fit the first order error term of the engine performance prediction basic digital twin model. The model expression ability requirement of the compensation model can be lower than that of the engine performance prediction basic digital twin model, so the algorithm of the compensation model can be consistent with that of the engine performance prediction basic digital twin model, or a simpler non-time series algorithm can be used.

[0042] The establishment of the compensation model can refer to the construction of the engine performance prediction basic digital twin model, and the corresponding number of network layers and nodes of each layer are set. The structure of the compensation model matches that of the engine performance prediction basic digital twin model, and the model structures of the two cannot be too complex or too simple at the same time, so as to avoid overfitting or underfitting problems.

[0043] S3: Establish a model output channel: set two output channels, one of which is the output channel of the engine performance prediction basic digital twin model, and the other is the sum output channel of the engine performance prediction basic digital twin model and the compensation model; output two different values through the two output channels, one is the separate output y1 of the engine performance prediction basic digital twin model, and the other is the sum output y2 of the engine performance prediction basic digital twin model and the compensation model. In each iteration step of the training process, the two output values are output at the same time.

[0044] S4: training error back propagation: the two output values obtained in step S3 are compared with the corresponding performance prediction target values respectively to obtain two loss parameters, the loss parameters are iterated into the engine performance prediction basic digital twin model and the compensation model respectively for training, and steps S3 and S4 are repeated until the error converges.

[0045] The two output values obtained in step S3 are compared with the corresponding performance prediction target values respectively to obtain two loss values loss1 and loss2, wherein loss1 is a loss value calculated through an output channel of the engine performance prediction basic digital twin model, and loss2 is a loss value calculated through an output channel of the engine performance prediction basic digital twin model and the compensation model; the two values of loss2, loss1 and loss2 after addition constitute loss parameters, which are respectively taken as loss parameters of the compensation model and the engine performance prediction basic digital twin model.

[0046] In the present application, two loss value calculation methods are provided, which are cross-entropy loss function or mean square error loss function respectively.

[0047] The cross-entropy loss function is adopted:

[0048]

[0049] Wherein, H(Y, y) is the cross-entropy loss function value of the target parameter true value Y(x i ) and the model prediction value y(x i ).

[0050] Alternatively, the mean square error loss function is adopted:

[0051]

[0052] Wherein, MSE(Y, y) is the mean square error loss function value of the target parameter true value Y(x i ) and the model prediction value y(x i ).

[0053] Based on the training data set in step S1, and the loss parameter, the engine performance prediction basic digital twin model and the compensation model are trained in combination with a training algorithm. The training algorithm can adopt RMSPorp (Root Mean Square prop) and Adam (Adaptive moment estimation). All parameters of the engine performance prediction basic digital twin model and the compensation model are trained and updated, and the above steps are repeated for continuous iterative training until the error converges. In the present application, whether the error converges is determined by setting an error threshold, or the number of iterations is set. When the number of iterations reaches the set number, the training is completed, and the final engine performance prediction basic digital twin model and the compensation model are obtained.

[0054] S5: Predicting the performance parameters of the aero-engine: The data monitored in any working state of the engine is input into the engine performance prediction basic digital twin model converged in step S4 to obtain the predicted value of the target performance parameter.

[0055] The engine working state includes starting, accelerating, cruising, boosting, and decelerating states. The sensor detection data is extracted and input into the final engine performance prediction basic digital twin model converged in step S4 to obtain the predicted value of the target performance parameter. In this step, the parameter does not need to be input into the compensation model because the compensation effect has been coupled into the basic model through the coupling training method in step six. The model can also be used on board to predict the engine performance parameters in real time. The engine performance parameter accurate prediction value can be obtained in real time by inputting the engine monitoring parameters into the basic model in real time.

[0056] Embodiment

[0057] S1: Establishing an engine performance prediction basic digital twin model: Selecting environmental parameters, component working state parameters, component performance parameters, and whole machine performance parameters, the data is filtered as model input, the parameters are coupled and output, and a time series artificial intelligence algorithm is selected. In the present application, a long short-term memory network (LSTM) is preferably adopted. The algorithm can integrate the performance degradation of the aero-engine over time in the training data set into the model. The model is built by imitating the physical structure of the aero-engine, and the network layers of each component are built, as shown in Figure 3 The model structure is provided. Other existing model structures can also be used for building. This is not described in detail because it is prior art.

[0058] S2: Establishing a compensation model: The structure of the compensation model matches the engine performance prediction basic digital twin model, or the compensation model is a non-time series algorithm. The model structures of the two cannot be too complex or too simple at the same time, so as to avoid overfitting or underfitting.

[0059] S3: Establishing a model output channel: on the basis of step S1 and step S2, coupling is performed, and two output channels are set, one is the separate output y1 of the engine performance prediction basic digital twin model, and the other is the sum output y2 of the engine performance prediction basic digital twin model and the compensation model.

[0060] S4: Training error back propagation: y1 and y2 are calculated with the performance prediction target value respectively, the performance prediction target value is an aero-engine performance parameter, for example, the measured value of the thrust, the measured value of the fuel consumption rate, the measured value of the low-pressure turbine exhaust gas temperature, etc., which can be selected according to actual needs. y1 is the value output by the output channel of the engine performance prediction basic digital twin model, and y2 is the value output by the output channel of the compensation model; y1 and the predicted target true value are calculated to obtain a loss value loss1, y2 and the predicted target true value are calculated to obtain a loss value loss2, loss2 is set as the loss value of the compensation model; after adding loss1 and loss2, the loss value of the engine performance prediction basic digital twin model is set, since the output of the compensation model only appears in the coupled output channel y2, the separate output channel y1 does not contain the compensation model output, therefore, after error calculation, only the error loss2 calculated based on y2 can act on the training of the compensation model, so loss2 is taken as the loss value of the compensation model. Through the loss value, the parameters of the engine performance prediction basic digital twin model and the compensation model in steps S1 and S2 are updated at the same time, and iteration is continuously performed until the error is a set threshold value, or after completing the set iteration steps, the training is completed, and the final engine performance prediction basic digital twin model and compensation model are obtained.

[0061] S5: Predicting an aero-engine performance parameter: selecting an arbitrary working state (including starting, accelerating, cruising, boosting, decelerating, etc.), extracting sensor monitoring data, and corresponding real-time inputting to the engine performance prediction basic digital twin model trained in step S4 to obtain a target performance parameter prediction value.

[0062] The application also provides an electronic device, comprising:

[0063] one or more processors;

[0064] a storage device configured to store one or more programs;

[0065] When the one or more programs are executed by the one or more processors, the one or more processors implement the precision improvement method for the aero-engine performance digital twin as described above.

[0066] The application further provides a computer readable medium, wherein the computer readable medium stores a computer program, and the computer program is executed by a processor to implement the precision improvement method for the digital twin of an aero-engine performance.

[0067] The above-described embodiments are only used to describe the preferred modes of the application, and are not used to limit the scope of the application. Without departing from the design spirit of the application, various modifications, variations, modifications and replacements of the technical solutions of the application made by those skilled in the art shall fall within the protection scope of the claims of the application.

Claims

1. A method for precision improvement of an aero-engine performance digital twin, characterized in that, The method comprises the following steps: S1: establishing an engine performance prediction basic digital twin model: screening engine operation data, and constructing an engine performance prediction basic digital twin model using a time-series artificial intelligence algorithm; S2: establishing a compensation model: the structure and algorithm of the compensation model are consistent with those of the engine performance prediction basic digital twin model, or the compensation model is a non-time-series algorithm; S3: establishing a model output channel: two output channels are set, one of which is an output channel of the engine performance prediction basic digital twin model, and the other is an additive output channel of the engine performance prediction basic digital twin model and the compensation model; S4: training error back propagation: the two output values obtained in step S3 are compared and calculated with the corresponding performance prediction target values respectively to obtain two loss parameters, which are iterated into the engine performance prediction basic digital twin model and the compensation model for training, and steps S3 and S4 are repeated until the error converges; S5: predicting the performance parameters of an aero-engine: the data monitored in any working state of the engine are input into the engine performance prediction basic digital twin model converged in step S4 to obtain target performance parameter prediction values.

2. The precision enhancement method for aero-engine performance digital twinning according to claim 1, characterized in that: In step S1, the screened engine operation data include environmental parameters, component working state parameters, component performance parameters, and whole machine performance parameters. After the screened data are processed, a training data set is established.

3. The precision enhancement method for aero-engine performance digital twinning according to claim 1, characterized in that: In step S1, the time-series artificial intelligence algorithm includes a long short-term memory network, a wave network, and an attention mechanism network. According to the data screened in step S1, the number of input layer nodes, the number of output layer nodes, the number of network layers, and the number of network layer nodes are determined to build the performance prediction basic digital twin model.

4. The precision enhancement method for aero-engine performance digital twinning according to claim 1, characterized in that: In step S4, the loss parameters are obtained by comparing and calculating the two output values obtained in step S3 with the corresponding performance prediction target values to obtain two loss values loss1 and loss2. Loss1 is the loss value calculated through the engine performance prediction basic digital twin model output channel, and loss2 is the loss value calculated through the additive output channel of the engine performance prediction basic digital twin model and the compensation model. The two values of loss2, loss1, and the sum of loss2 constitute the loss parameters, which are used as the loss parameters of the compensation model and the engine performance prediction basic digital twin model respectively.

5. The precision enhancement method for aero-engine performance digital twinning according to claim 4, characterized in that: The loss value is calculated by using a cross-entropy loss function: Where H(Y,y) is the true value of the target parameter Y(x) i ) and model predicted value y(x) i The cross-entropy loss function value.

6. The precision enhancement method for aero-engine performance digital twinning according to claim 4, characterized in that: The loss value is calculated by using a mean square error loss function: where MSE(Y, y) is the mean squared error loss function value of the target parameter true value Y(x i ) and the model predicted value y(x i ).

7. The precision enhancement method for aero-engine performance digital twinning according to claim 1, characterized in that: The algorithm for training in step S4 includes RMSPorp and Adam.

8. The precision enhancement method for aero-engine performance digital twinning according to claim 1, characterized in that: In step S5, the engine working state includes starting, accelerating, cruising, boosting, and decelerating states. Sensor detection data are extracted as the input of the engine performance prediction basic digital twin model converged in step S4.

9. An electronic device, comprising: It comprises: one or more processors; a storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement a precision improvement method for aero-engine performance digital twinning as claimed in any one of claims 1-8.

10. A computer readable medium storing a computer program, characterized in that: The computer program, when executed by a processor, implements a precision improvement method for aero-engine performance digital twinning as claimed in any one of claims 1-8.

Images