Gas path parameter prediction method and system based on aero-engine component-level performance reduction digital twin model

Abstract

The invention belongs to the technical field of aero-engines, and discloses a gas path parameter prediction method and system based on an aero-engine component-level performance reduction digital twin model. When the performance of a certain part is reduced, replacing the part mechanism model of the part with the reduced performance in the mechanism model with an embedded mechanism model, and enabling other parts to be still part-level mechanism models; the embedded mechanism model is a data driving model based on a neural network and is obtained by training fault component data; and training the data driving model according to historical data of the whole engine, and filtering and fusing gas path parameter prediction of the replaced whole mechanism model and the data driving model through Kalman filtering. The method effectively solves the problem that parameter estimation is difficult to carry out under the condition that the performance of a single component is reduced in a traditional method. And performing data simulation and data acquisition and storage on the performance reduction parameters, and performing model training, gas path parameter estimation and gas path parameter visualization on the performance reduction data for aero-engine parameter estimation.

Description

Technical Field

[0001] The present invention relates to the technical field of aero-engines, and in particular to a method and system for predicting gas path parameters based on a digital twin model of aero-engine component-level performance degradation. Background Art

[0002] In the process of predicting the parameters of the air path system of an aircraft engine, the traditional method is to use a mechanism model or a data-driven model to simulate the air path parameters. Therefore, when the performance of a single component degrades, the model cannot be modified in real time to address the performance degradation, resulting in a performance deviation in the estimated air path parameters of the system. Using a data-driven model to replace the faulty component to predict data parameters and combining it with an embedded digital twin framework can address this problem well. In view of the above situation, it is necessary to study an air path parameter prediction platform based on the digital twin model of performance degradation at the component level of an aircraft engine. Summary of the invention

[0003] In order to solve the problems existing in the background technology, the purpose of the present invention is to provide a gas path parameter prediction method and system based on the digital twin model of aircraft engine component-level performance degradation, which is used to predict the gas path parameters of the aircraft engine and perform a visual demonstration of the gas path parameters of the aircraft engine when the component-level performance is degraded. It can be used for estimating the parameters of the aircraft engine gas path system, training data-driven models, and displaying the real-time working status of the aircraft engine.

[0004] The technical solution adopted by the present invention is as follows: a method for predicting gas path parameters based on a digital twin model of performance degradation at the component level of an aero-engine, establishing an embedded digital twin framework for predicting gas path parameters; the embedded digital twin framework includes a mechanism model, a data-driven model, and a Kalman filter module; the mechanism model and the data-driven model predict gas path parameters respectively, and the Kalman filter module filters and fuses the gas path parameter predictions of the mechanism model and the data-driven model and outputs them;

[0005] When the performance of a certain component deteriorates, part of the mechanism model of the component with deteriorated performance in the mechanism model is replaced by an embedded mechanism model, and the remaining components remain component-level mechanism models; the embedded mechanism model is a data-driven model based on a neural network, which is trained with the data of the faulty component; the data-driven model is trained according to the historical data of the entire engine, and the gas path parameter predictions of the replaced overall mechanism model and the data-driven model are filtered and fused through Kalman filtering.

[0006] The neural network-based data-driven model includes a self-attention mechanism network layer and a long short-term memory network layer; the self-attention mechanism network layer calculates the mutual self-attention scores of the input quantities of the gas path parameters of the component-level mechanism model to obtain a self-attention weight sequence; the long short-term memory network layer receives the self-attention weight sequence processed by the self-attention mechanism network layer based on the long short-term memory unit, and outputs the predicted value of the gas path parameter of the next mechanism component.

[0007] The component-level mechanism model is a component-level model established according to the component-level operating mechanism of the aircraft engine. It is a nonlinear model. The Newton-Raphson solution method is used to solve the model parameters of each component at each time step. The working point of each component is determined according to the model input to predict the gas path parameters; the data-driven model is based on a neural network. The data-driven model is trained through historical data. It is mainly used to fit the actual engine's gas path parameter operating state. Given specific input parameters, the gas path parameters of each section of the aircraft engine are fitted.

[0008] A gas path parameter prediction system based on a digital twin model of aircraft engine component-level performance degradation, comprising:

[0009] An embedded digital twin framework module is used for predicting gas path parameters; the embedded digital twin framework module includes a mechanism model, a data-driven model, and a Kalman filter module; the mechanism model and the data-driven model predict gas path parameters respectively, and the Kalman filter module filters and fuses the gas path parameter predictions of the mechanism model and the data-driven model and outputs them;

[0010] The gas path parameter prediction and monitoring module is used to monitor components with degraded performance and replace the corresponding partial excitation model with the corresponding embedded mechanism model.

[0011] The gas path parameter prediction and monitoring module also includes simulating the performance degradation of components to generate performance degradation data, collecting the input and output data of the faulty components and the control input and parameter output data of the overall engine; the input and output data of the faulty components are used to train the embedded mechanism model; the control input and parameter output data of the overall engine are used to train the overall data-driven model.

[0012] The beneficial effect of the present invention is: by replacing the performance-degraded components in the mechanism model with a neural network model based on a self-attention mechanism and a long short-term memory network, the model is trained by fault data, and the neural network model is embedded into the mechanism model, and the overall parameter estimation neural network model is trained according to the historical data of the entire engine, and the gas path parameter predictions of the two models are filtered and fused through Kalman filtering, and integrated into a whole, which is called an embedded digital twin framework. The digital twin framework can perform air path parameter estimation and gas path parameter visualization of aircraft engines.

[0013] This method can be used to estimate gas path parameters when the performance parameters of aircraft engine components are degraded, and can be used for gas path parameter visualization, which has broad application potential. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a tree diagram of the main system structure of the gas path parameter estimation platform.

[0015] Figure 2 It is the overall operating architecture of the gas path parameter prediction system based on the digital twin model of aircraft engine component-level performance degradation.

[0016] Figure 3 It is a hierarchical structure of the gas path parameter prediction system based on the digital twin model of aircraft engine component-level performance degradation, which is composed of three parts: hardware layer, system layer and application layer.

[0017] Figure 4 It is the system block diagram of the aircraft engine embedded mechanism model.

[0018] Figure 5 It is the system block diagram of the data-driven model.

[0019] Figure 6 It is the system block diagram of the embedded digital twin framework.

[0020] Figure 7 It is a schematic diagram of the output trend of the faulty component (low-pressure compressor); (a)-(e) are the flow rate, total enthalpy, total temperature, total pressure, and torque output by the data-driven model that replaces the LPC component.

[0021] Figure 8 It is a schematic diagram of the prediction trend of the gas path parameters of the embedded mechanism model; (a)-(l) are respectively the total temperature of engine section 23, the total temperature of section 3, the total temperature of section 4, the total temperature of section 45, the total temperature of section 5, the total pressure of section 23, the total pressure of section 3, the total pressure of section 4, the total pressure of section 45, the total pressure of section 5, and the low-pressure shaft speed and the high-pressure shaft speed.

[0022] Fig. 9 It is a schematic diagram of the prediction trend of the gas path parameters of the embedded digital twin model; (a)-(l) are the total temperature of engine section 23, section 3, section 4, section 45, section 5, the total pressure of section 23, section 3, section 4, section 45, section 5, the low-pressure shaft speed and the high-pressure shaft speed, respectively. DETAILED DESCRIPTION

[0023] The present invention will be further described below with reference to the accompanying drawings and implementation examples with respect to the situation where the performance of LPC is degraded.

[0024] like Figure 4 , Figure 5 , Figure 6 As shown in the figure, the embedded digital twin framework model consists of an aircraft engine mechanism model, a data-driven model, and a Kalman filter module. The aircraft engine mechanism model, also known as the component-level mechanism model, is modeled by the input and output mechanism of the gas path system of the actual aircraft engine components. The Newton-Raphson method is used to iteratively solve the working state of each component, and then predict the parameters of the aircraft engine gas path system; the data-driven model, also known as the neural network model, is based on a recurrent neural network. By setting the model architecture to train the historical data of the engine, it can fit the output characteristics of the engine. This model is parallel to the engine mechanism model; the Kalman filter module uses Kalman filtering technology to filter and fuse the parameter estimates of the two parallel models to form the overall embedded digital twin framework model.

[0025] The embedded mechanism model consists of a self-attention mechanism network layer and a long short-term memory network layer. The self-attention mechanism network layer is used to calculate the correlation between input parameters, perform self-attention weighting on the input parameters, and generate attention weight input; the long short-term memory network layer uses its unique memory unit structure to effectively capture long-term dependencies, can process sequence data containing long-term delay information, and retain important historical information during the fitting process.

[0026] Component-level mechanistic models model non-degraded components in this way and can reliably predict performance parameters under non-degraded conditions.

[0027] Here’s how it works:

[0028] Aircraft engine component-level fault data acquisition and component-level data-driven model training (such as Figure 4 shown):

[0029] Step 0: Follow Figure 4 The mechanical model of the various components of the aircraft engine connected as shown, including the air inlet, fan, low-pressure compressor, high-pressure compressor, combustion chamber, high-pressure turbine, low-pressure turbine, tail nozzle, etc. Here, it is assumed that the low-pressure compressor fails, that is, the efficiency of the low-pressure compressor decreases to 90% of the original value at a constant rate after the system stabilizes.

[0030] Step 1: Run the mechanism model and save the input and output data of the low-pressure compressor for subsequent neural network training and fitting.

[0031] Step 2: Use the data saved in step 1 to train a long short-term memory neural network fitting model with multiple self-attention mechanisms.

[0032] Step 3: Test the trained model, give the model the air path data of the low-pressure compressor inlet interface input, and observe the difference between the model data and the actual output, such as Figure 7 shown.

[0033] Data-driven model embedding mechanism model:

[0034] Step 0: To obtain an embedded mechanism model, the existing mechanism model needs to be modified and the mechanism model of the faulty component is replaced by a neural network model.

[0035] Step 1: Connect the neural network model, set the normalization parameters of the model input and output, and cancel the Newton-Raphson solver outside the original mechanism model to solve the component.

[0036] Step 2: Based on the original input, the existing embedded mechanism model can make global parameter estimates for the performance-degraded components.

[0037] Step 3: Parameter estimation trends of the aircraft engine data-driven model are as follows Figure 8 shown.

[0038] Embedded digital twin parameter prediction framework (such as Figure 4 shown):

[0039] Step 0: The overall embedded digital twin parameter prediction framework is divided into two parallel parts. One is the embedded mechanism model for data-driven model embedding of faulty components, and the other is the data-driven model for overall gas path parameter estimation trained using recurrent neural networks and historical fault data.

[0040] Step 1: Import the embedded mechanism model and the data-driven model.

[0041] Step 2: Connect the variables of the prediction parameters of the two models, connect each group to the Kalman filter module, and set the input parameters of the two models.

[0042] Step 3: Set the simulation step size, simulation time, control amount, and Kalman filter coefficient, and start the simulation. At this time, the overall model data is the gas path parameter prediction output of the embedded digital twin framework after filtering and fusion by the Kalman filter algorithm.

Claims

1. A method for predicting gas path parameters based on a digital twin model of aircraft engine component-level performance degradation, characterized in that: Establishing an embedded digital twin framework for gas path parameter prediction; the embedded digital twin framework includes a mechanism model, a data-driven model, and a Kalman filter module; the mechanism model and the data-driven model respectively predict gas path parameters, and the Kalman filter module filters and fuses the gas path parameter predictions of the mechanism model and the data-driven model and outputs them; When the performance of a certain component deteriorates, part of the mechanism model of the component with deteriorated performance in the mechanism model is replaced by an embedded mechanism model, and the remaining components remain component-level mechanism models; the embedded mechanism model is a data-driven model based on a neural network, which is trained with the data of the faulty component; the data-driven model is trained according to the historical data of the entire engine, and the gas path parameter predictions of the replaced overall mechanism model and the data-driven model are filtered and fused through Kalman filtering.

2. The method for predicting gas path parameters based on the digital twin model of aircraft engine component-level performance degradation according to claim 1 is characterized in that: The neural network-based data-driven model includes a self-attention mechanism network layer and a long short-term memory network layer; the self-attention mechanism network layer calculates the mutual self-attention scores of the input quantities of the gas path parameters of the component-level mechanism model to obtain a self-attention weight sequence; the long short-term memory network layer receives the self-attention weight sequence processed by the self-attention mechanism network layer based on the long short-term memory unit, and outputs the predicted value of the gas path parameter of the next mechanism component.

3. The method for predicting gas path parameters based on the digital twin model of aircraft engine component-level performance degradation according to claim 1 is characterized in that: The component-level mechanism model is a component-level model established according to the component-level operating mechanism of the aircraft engine. It is a nonlinear model that uses the Newton-Raphson solution method to calculate the model parameters of each component at each time step, and determines the working point of each component according to the model input to predict the gas path parameters; The data-driven model is based on a neural network and is trained through historical data. It is mainly used to fit the actual engine's gas path parameter operating status. Given specific input parameters, the gas path parameters of each section of the aircraft engine are fitted.

4. A gas path parameter prediction system based on the aircraft engine component-level performance degradation digital twin model according to any one of claims 1 to 3, characterized in that: include: Embedded digital twin framework module for gas path parameter prediction; The embedded digital twin framework module includes a mechanism model, a data-driven model, and a Kalman filter module; The mechanism model and the data-driven model respectively predict the gas path parameters, and the Kalman filter module filters and fuses the gas path parameter predictions of the mechanism model and the data-driven model for output; The gas path parameter prediction and monitoring module is used to monitor components with degraded performance and replace the corresponding partial excitation model with the corresponding embedded mechanism model.

5. The gas path parameter prediction system based on the aircraft engine component-level performance degradation digital twin model according to claim 4 is characterized in that: The gas path parameter prediction and monitoring module also includes simulating the performance degradation of components to generate performance degradation data, collecting the input and output data of the faulty components and the control input and parameter output data of the overall engine; the input and output data of the faulty components are used to train the embedded mechanism model; the control input and parameter output data of the overall engine are used to train the overall data-driven model.

Images