Prediction method of transient temperature field of turbine guide vanes based on multi-layer feedforward neural network

Through the method based on multi-layer feedforward neural network, the problem of complex calculation and low accuracy of the transition state temperature field prediction method of the traditional turbine guide vane is solved, and the rapid and accurate prediction of the temperature field of the turbine guide vane is achieved, which satisfies modern turbines Efficient forecast demand for engines.

CN118761334BActive Publication Date: 2024-11-05NANJING UNIV OF AERONAUTICS & ASTRONAUTICS
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411251842.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-09
Publication Date
2024-11-05
Estimated Expiration
2044-09-09

AI Technical Summary

Technical Problem

The traditional turbine guide vane transition state temperature field prediction method has complex calculations, high resource consumption, and poor prediction accuracy, making it difficult to meet the requirements of modern turbine engines for multivariate parameter prediction, high accuracy and high efficiency.

Method used

Using a multi-layer feedforward neural network method, the transition state temperature field data set is established by determining the working state and key temperature measurement points of the guide vane, and a multi-layer feedforward neural network model is constructed to quickly extract the input features and outputs. The nonlinear mapping relationship of the turbine guide vane can achieve a rapid prediction of the transition state temperature field.

Benefits of technology

The characteristic representation of the transition state temperature field of the turbine guide vane is effectively extracted, which realizes rapid prediction of the temperature field of the turbine guide vane under different operating conditions, improves the prediction accuracy and efficiency, and meets the practical application needs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118761334B_ABST
    Figure CN118761334B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for estimating the transient temperature field of a turbine guide vane based on a multi-layer feedforward neural network. First, the transient experimental data of the temperature of key points on the turbine guide vane when different operating conditions are switched are discretized into multiple state points, and a database of the temperature field of the turbine guide vane at each moment in the transient state is established, which is used as a training sample. Then, a multi-layer feedforward neural network is introduced to train the training samples, and the nonlinear mapping relationship between the input features and the corresponding outputs is quickly extracted through the combination of multi-layer fully connected layers, batch normalization layers, and activation layers, and a rapid prediction model for the transient temperature field of the turbine guide vane when different operating conditions change is obtained. The present invention can quickly and accurately predict the dynamic changes of the temperature of key points on the guide vane when the operating conditions change, and provides an efficient and reliable prediction tool for the transient temperature field of the turbine guide vane.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of aero-engine thermal management, and in particular relates to a method for estimating a turbine guide vane transient temperature field based on a multi-layer feedforward neural network. Background Art

[0002] Turbine guide vanes are key components of turbine engines. During a single flight mission, the engine undergoes multiple unstable operating states, including starting, acceleration, cruising, deceleration, and shutdown. During these unstable operating states, the flame temperature within the combustion chamber fluctuates rapidly, causing the turbine guide vanes to be exposed to a non-uniform temperature environment. With repeated thermal cycling, cracks gradually form at key locations on the guide vane surface, ultimately leading to guide vane failure. The transient temperature field distribution of the turbine guide vanes significantly impacts their performance and lifespan. Therefore, accurately estimating the transient temperature field distribution of the turbine guide vanes is crucial for the design, optimization, and operation of turbine engines.

[0003] Traditional methods for predicting the transient temperature field of turbine guide vanes rely primarily on numerical simulation and experimental testing. Numerical simulation predicts the temperature field by establishing complex mathematical models and solving the corresponding equations. However, this method consumes a large amount of computing time and resources and is extremely sensitive to changes in geometric parameters and boundary conditions. Experimental testing, on the other hand, requires a large number of test samples and is costly. These traditional methods suffer from high computational complexity, high resource consumption, and suboptimal prediction accuracy. They are unable to meet the requirements of modern turbine guide vane transient temperature field prediction for multivariable parameter prediction, high accuracy, and high efficiency.

[0004] In recent years, neural network models have achieved considerable application in temperature field prediction. Deep neural networks, in particular, have demonstrated significant advantages in handling complex nonlinear problems. However, when it comes to predicting the transient temperature field of turbine guide vanes, traditional neural network models still suffer from limitations, such as poor timing, information loss, and low prediction accuracy. Therefore, a method for predicting the transient temperature field of turbine guide vanes based on a multi-layer feedforward neural network is urgently needed to improve prediction accuracy and efficiency to meet practical application requirements. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the purpose of the present invention is to provide a method for predicting the transient temperature field of the turbine guide vane based on a multi-layer feedforward neural network, which can effectively extract the characteristic representation of the transient temperature field of the turbine guide vane, realize the solution of complex nonlinear problems from input to output, and ultimately enable the rapid prediction of the transient temperature field of the turbine guide vane.

[0006] The purpose of the present invention is achieved through the following technical solutions:

[0007] A method for estimating the transient temperature field of a turbine guide vane based on a multi-layer feedforward neural network comprises the following steps:

[0008] Step 1: Determine the working state of the guide vanes in the turbine engine as input variables, including the duration of the transition state of the turbine guide vanes when the working conditions are switched, the pressure, temperature and flow of the mainstream at the initial and final states, and the temperature and flow of the cooling gas at the initial and final states, and other key data.

[0009] Step 2: Identify key temperature measurement points on the turbine guide vanes. By collecting data from these measurement points, we can analyze the temperature distribution of the turbine guide vanes under high-temperature and high-pressure environments. The locations of key temperature measurement points vary for different turbine guide vane configurations, and the location and number of these measurement points directly impact the accuracy and comprehensiveness of the temperature field measurement.

[0010] Step 3: Establish a transitional temperature field dataset for the turbine guide vanes during operating mode switching. By discretizing the temperature field information at each moment in the transitional state, we obtain the state points at each moment in the transitional state. The time of each state point and the turbine guide vane operating condition at that moment are used as the input dataset, and the temperature of the key measuring points of the guide vanes at that moment is used as the output dataset.

[0011] Step 4: randomly divide the input data set and output data set obtained in step 3 into a training set, a test set, and a validation set, which are used for fitting, optimizing, and validating the parameters in the multi-layer feedforward neural network model, respectively;

[0012] Step 5: Construct a multi-layer feedforward neural network model. This neural network model is a deep neural network. It is composed of a combination of multiple fully connected layers, batch normalization layers, and activation layers to quickly extract the nonlinear mapping relationship between input features and corresponding outputs. The training set obtained in step 4 is used to train the multi-layer feedforward neural network model, and the validation set and test set are used to verify and test the multi-layer feedforward neural network model to improve the generalization ability of the model. The nonlinear mapping relationship between input features and corresponding outputs is obtained by training and testing the data with the multi-layer feedforward neural network model, and a rapid prediction model of the transition temperature field of the turbine guide vane under different operating conditions is obtained.

[0013] In step 1, the input variables include the duration of the transition state when the turbine guide vane operating condition switches T , the initial pressure of the main flow through the turbine guide vanes P ms1 , mainstream final state pressure P ms2 , mainstream initial state temperature T ms1 , mainstream final state temperature T ms2 , the initial flow rate of the cooling gas from the inlet 1 cooling guide vaneM 1_1 , final state flow M 2_1 , the initial flow rate of the cooling gas from the inlet 2 cooling guide vane M 1_2 , final state flow M 2_2 , the initial temperature of the cooling gas T c1 , final state temperature T c1 .

[0014] In step 2, key temperature measurement points on the turbine guide vanes are acquired. To conserve computing resources, during actual measurement and data analysis, several measurement points with similar temperatures are typically combined into a single measurement point. Furthermore, to provide a more comprehensive understanding of the guide vane's temperature distribution, the average temperature of the guide vane's mid-axis surface is also measured. This comprehensive approach allows for more representative turbine guide vane temperature field information to be obtained while ensuring data accuracy.

[0015] In step 3, the transition state temperature field data set includes data collected every 0.1 second under different time histories.

[0016] In step 5, the multi-layer feedforward neural network includes an input layer, multiple fully connected layers, a batch normalization layer, an activation layer, and a regression output layer. The specific number of layers depends on the dimensions of the input and output parameters and the required model accuracy.

[0017] In step 5, the output of the fully connected layer is as follows.

[0018] (1)

[0019] in, For the k The output expression of a fully connected layer is: σ is the activation function, w k For the neural network k The weights of the layers, x k Input to the neural network k The variables of the layer, b is the bias corresponding to each layer of the network, and L is the number of input features.

[0020] The present invention can quickly extract input features (transition state duration when operating condition switching) through a multi-layer feedforward neural network based on the relationship between the transition state temperature field of the turbine guide vane and the thermodynamic boundary conditions when operating condition switching. T , mainstream initial state pressure P ms1 , mainstream final state pressureP ms2 , mainstream initial state temperature T ms1 , mainstream final state temperature T ms2 , the initial flow rate of cooling gas from the cooling guide vane at the inlet 1 M 1_1 , final state flow M 2_1 , the initial flow rate of cooling gas from the 2 inlet cooling guide vanes M 1_2 , final state flow M 2_2 , the initial temperature of the cooling gas T c1 , final state temperature T c1 ) and the corresponding output (temperature of key measuring points of turbine guide vanes), a rapid prediction model for the transient temperature field of turbine guide vanes under different operating conditions was obtained.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] The present invention is based on a multi-layer feedforward neural network, which can effectively extract the characteristic representation of the turbine guide vane temperature field at each moment when the operating condition switches, realize the solution of complex nonlinear problems from input to output, and obtain a rapid prediction model of the turbine guide vane transition temperature field under different operating condition switching, and ultimately realize the rapid estimation of the turbine guide vane transition temperature field. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 Schematic diagram of the three-dimensional model simulation of the turbine guide vanes in the present invention.

[0024] Figure 2 Schematic diagram of the locations of key temperature measurement points of the turbine guide vanes in the present invention.

[0025] Figure 3 Schematic diagram of the turbine guide vane transition temperature field prediction method based on a multi-layer feedforward neural network.

[0026] Figure 4 Transition state duration T Comparison chart of the predicted and actual values ​​of the guide vane temperature at 10 seconds. DETAILED DESCRIPTION

[0027] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0028] Example

[0029] Taking a turbine guide vane as an example, the method for estimating the transient temperature field of the turbine guide vane based on a multi-layer feedforward neural network is described. The schematic diagram of the three-dimensional model simulation of the turbine guide vane is shown in the figure. Figure 1 The model shows the airflow environment around a guide vane in actual operation. The main flow enters from the front of the guide vane, while the cooling air enters from below. Through the airflow path inside the guide vane, the cooling air removes some heat, helping the guide vane maintain a lower temperature in a high-temperature environment.

[0030] Step 1: Determine the operating state of the guide vanes in the turbine engine as input variables, including key data such as the duration of the turbine guide vane transition state when switching between operating modes, the initial and final pressure, temperature, and flow rate of the main flow, and the initial and final temperature and flow rate of the cooling gas. The input parameters of the model are described as follows:

[0031] (1) Duration of the transition state from state 1 to state 2 T , which is used to describe the time scale for the system to transition from state 1 to state 2.

[0032] (2) Mainstream state 1 pressure P ms1 (kPa) and mainstream state 2 pressure P ms2 (kPa), these two parameters represent the pressure values ​​of the mainstream at the initial state and the final state of the system under the duration T.

[0033] (3) Mainstream state 1 temperature T ms1 (K) and mainstream state 2 temperature T ms2 (K), these two parameters represent the temperature values ​​of the mainstream at the initial and final states of the system under the duration T.

[0034] (4) State 1 ingress 1 traffic M 1_1 (g / s) and state 2 inlet 1 flow M 2_1 (g / s), these two parameters represent the initial and final state flow rates of inlet 1 of the system under duration T.

[0035] (5) State 1 ingress 2 traffic M 1_2 (g / s) and State 2 Inlet 2 Flow M 2_2 (g / s), these two parameters represent the initial and final state flow rates of inlet 2 of the system under duration T.

[0036] (6) State 1 Air Conditioning Temperature Tc_1 (K) and state 2 air conditioning temperature T c_2 (K), these two parameters represent the initial and final cooling air temperatures of the system under duration T.

[0037] Step 2: Identify key temperature measurement points on the turbine guide vanes. By collecting data from these measurement points, we can analyze the temperature distribution of the turbine guide vanes under high-temperature and high-pressure environments. The locations of key temperature measurement points vary for different turbine guide vane configurations, and the location and number of these measurement points directly impact the accuracy and comprehensiveness of the temperature field measurement.

[0038] For the key temperature measurement points of a turbine guide vane, Figure 2 As shown in the figure, the guide vane model diagram is marked with multiple key measuring points, which are represented by letters a to g. The marked measuring points are distributed in different positions of the guide vane. The positions of these measuring points are very important because they are used to monitor the temperature distribution of the guide vane in actual use. Since the temperature difference between temperature measuring points with similar positions is small, in order to save computing resources, in the actual measurement and data analysis process, two measuring points with similar temperatures are usually merged into one measuring point. Therefore, the measuring points b, d, e, f, g and the average temperature of the guide vane mid-axis surface are used as output parameters. These six temperature points can accurately reflect the overall temperature distribution of the guide vane and provide basic data for further analysis and optimization.

[0039] Step 3: Establish a transitional temperature field dataset for the turbine guide vanes during operating mode switching. By discretizing the temperature field information at each moment in the transitional state, we obtain the state points at each moment in the transitional state. The time of each state point and the turbine guide vane operating condition at that moment are used as the input dataset, and the temperature of the key measuring points of the guide vanes at that moment is used as the output dataset.

[0040] In this example, the goal is to discretize existing transition state data from different time histories. A database containing 11 input variables and 6 output variables is constructed at each state point. Data from turbine guide vane operating conditions and transition state temperature measurements at three different time histories, spanning 20, 60, and 100 seconds, are used. For each time histories, a set of data is collected every 0.1 seconds to ensure data continuity and integrity. After integrating this data, a dataset of 10,800 state points is obtained.

[0041] In step 4, the obtained input data set and output data set are randomly divided into a training set, a test set, and a validation set, which are used for fitting, optimizing, and validating the parameters in the multi-layer feedforward neural network model, respectively.

[0042] These data sets are divided into training set, validation set and test set for training and testing of the dynamic rapid estimation model of guide vane temperature field, where the proportions of training set, test set and validation set are 70%, 15% and 15% respectively.

[0043] Step 5: Construct a multi-layer feedforward neural network model. This neural network model is a deep neural network. It is composed of a combination of multiple fully connected layers, batch normalization layers, and activation layers to quickly extract the nonlinear mapping relationship between input features and corresponding outputs. The training set obtained in step 4 is used to train the multi-layer feedforward neural network model, and the validation set and test set are used to verify and test the multi-layer feedforward neural network model to improve the generalization ability of the model. The nonlinear mapping relationship between input features and corresponding outputs is obtained by training and testing the data with the multi-layer feedforward neural network model, and a rapid prediction model of the transition temperature field of the turbine guide vane under different operating conditions is obtained.

[0044] Based on the above dataset, we establish Figure 3 The model for predicting the transient temperature field of the turbine guide vane is shown in the figure. The temperature field prediction model is constructed through a multi-layer feedforward neural network, which consists of an input layer, multiple fully connected layers, a batch normalization layer, an activation layer, and a regression output layer to achieve a nonlinear mapping from input parameters to output temperature fields. There are 11 input parameters, including T 、 P ms1 、 P ms2 、 T ms1 、 T ms2 、 M 1_1 、 M 2_1 、 M 1_2 、 M 2_2 、 T c1 、 T c1 , the 11 input parameters pass through 5 layers of fully connected layers with the number of nodes being 120, 240, 480, 240, and 120 respectively, 5 layers of batch normalization layers, 5 layers of activation layers with the activation function being ReLU, and a regression output layer containing 6 nodes to obtain the output temperature field dynamic prediction model.

[0045] The fully connected layer satisfies the following formula:

[0046] (1)

[0047] in, For the k The output expression of a fully connected layer is: σ is the activation function, w k is the weight of the k-th layer of the neural network, x k is the variable input to the kth layer of the neural network, b is the bias corresponding to each layer of the network, and L is the number of input features.

[0048] Several training hyperparameters were set for this multi-layer feedforward neural network, including a maximum number of iterations of 4000, a mini-batch size of 256, a validation frequency of 500, an initial learning rate of 0.01, a learning rate decay period of 100, and a learning rate decay factor of 0.5. The training process used the Adam optimization algorithm, which includes forward and backward propagation of data, calculation of the loss function, and updating of model parameters. After each training cycle, model performance was evaluated using validation data, and the best-performing model on the validation set was selected as the final model.

[0049] After obtaining the rapid prediction model of the turbine guide vane transition temperature field, the accuracy of the prediction model needs to be tested. The results are as follows: Figure 4 shown. Figure 4 The comparison between the predicted and true values ​​of the six temperature values ​​on the guide vane within 10 seconds is shown. Within the 10-second time history, the trends of the predicted and true values ​​of the measuring points are generally good, especially at multiple key measuring points, where the change trends of the predicted and true values ​​are almost identical. The maximum average relative error of each measuring point is 0.15%, indicating that the rapid estimation model can accurately predict the guide vane temperature changes within the transition state.

[0050] The present invention provides a method for estimating the transient temperature field of turbine guide vanes based on a multi-layer feedforward neural network. There are numerous methods and approaches for implementing this technical solution. The above is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art may make various improvements and modifications without departing from the principles of the present invention, and such improvements and modifications are also within the scope of protection of the present invention. Any components not specified in this embodiment may be implemented using existing technologies.

Claims

1. A method for estimating the transient temperature field of a turbine guide vane based on a multi-layer feedforward neural network, characterized in that: The following steps are involved: Step 1, determining the working state of the guide vane in the turbine engine as an input variable, including the duration of the transition state of the turbine guide vane when the working state is switched, the pressure, temperature and flow rate of the mainstream at the initial state and the final state, and the temperature and flow rate data of the cooling gas at the initial state and the final state; Step 2, determine the key temperature measuring points on the turbine guide vanes, and analyze the temperature distribution of the turbine guide vanes under high temperature and high pressure environment by collecting data from the measuring points; Step 3, establishing a transitional temperature field data set of the turbine guide vane when the operating condition is switched, obtaining the state point at each moment in the transition state by discretizing the temperature field information at each moment in the transition state, taking the moment of each state point and the operating condition of the turbine guide vane at that moment as the input data set, and taking the temperature of the key measuring point of the turbine guide vane at that moment as the output data set; Step 4, randomly dividing the input data set and the output data set obtained in step 3 into a training set, a test set, and a validation set, which are used for fitting, optimizing, and validating the parameters in the multi-layer feedforward neural network model, respectively; Step 5, construct a multi-layer feedforward neural network model, which is a deep neural network. It is composed of a combination of multi-layer fully connected layers, batch normalization layers and activation layers to quickly extract the nonlinear mapping relationship between input features and corresponding outputs. The training set obtained in step 4 is used to train the multi-layer feedforward neural network model, and the validation set and test set are used to verify and test the multi-layer feedforward neural network model to improve the generalization ability of the model. The nonlinear mapping relationship between input features and corresponding outputs is obtained by training and testing the data with the multi-layer feedforward neural network model, and a prediction model for the transition temperature field of the turbine guide vane when different operating conditions change is obtained.

2. The method for estimating the transient temperature field of a turbine guide vane based on a multi-layer feedforward neural network according to claim 1, characterized in that: In step 1, the input variables include the duration of the transition state when the turbine guide vane operating condition is switched T , the initial pressure of the main flow through the turbine guide vanes P ms1 , Mainstream final state pressure P ms2 , mainstream initial state temperature T ms1 , mainstream final state temperature T ms2 , the initial flow rate of cooling gas from inlet 1 cooling guide vane M 1_1 , final state flow M 2_1 , the initial flow rate of cooling gas from inlet 2 cooling guide vane M 1_2 , final state flow M 2_2 , the initial temperature of the cooling gas T c1 , final state temperature T c1 .

3. The method for estimating the transient temperature field of a turbine guide vane based on a multi-layer feedforward neural network according to claim 1, characterized in that: In step 2, key temperature measurement points on the turbine guide vane are obtained. During the actual measurement and data analysis process, several measurement points with similar temperatures are combined into one measurement point representation, and the average temperature measurement of the axial surface of the turbine guide vane is added to obtain more representative turbine guide vane temperature field information.

4. The method for estimating the transient temperature field of a turbine guide vane based on a multi-layer feedforward neural network according to claim 1, characterized in that: In step 3, the transition state temperature field data set includes data collected every 0.1 second under different time histories.

5. The method for estimating the transient temperature field of a turbine guide vane based on a multi-layer feedforward neural network according to claim 1, characterized in that: In step 5, the multi-layer feedforward neural network includes an input layer, multiple fully connected layers, a batch normalization layer, an activation layer and a regression output layer; the specific number of layers is determined according to the dimensions of the input parameters and the output parameters and the required accuracy of the model.

6. The method for estimating the transient temperature field of a turbine guide vane based on a multi-layer feedforward neural network according to claim 5, characterized in that: The fully connected layer satisfies the following formula: The output of the fully connected layer is as follows (1) in, For the k The output expression of the fully connected layer is: σ is the activation function, w k For the neural network k The weight of the layer, x k As input to the neural network k The variables of the layer, b is the bias corresponding to each layer of the network, and L is the number of input features.

Citation Information

Patent Citations

  • Method for predicting outlet flow non-uniformity coefficient of turbine gas collection cavity

    CN111079920A

  • Turbine transition state blade tip clearance estimation method based on long short-term memory neural network

    CN116822296A

  • Machine control and method for configuring the machine control

    EP4300356A1