Deep Learning-Based Turbine Temperature Distribution Prediction Method

Abstract

A method for predicting turbine temperature distribution based on deep learning, which is used for predicting the evolution of the temperature distribution in the solid domain of a gas turbine during a transient experiment over time, includes the following steps: constructing an experimental geometric model of a turbine cascade and generating a computational grid for the turbine cascade, and manufacturing a turbine cascade test piece according to the experimental geometric model of the turbine cascade; constructing a gas turbine transient experimental device, determining the transient experimental conditions and boundary conditions of the turbine cascade, and constructing a recurrent neural network encoded by a transient heat conduction equation; conducting a gas turbine transient experiment, collecting the time series data of the total temperature at the mainstream inlet, the time series data of the surface temperature of the turbine cascade test piece, and the material thermal physical properties of the turbine cascade test piece to form experimental data, training the recurrent neural network to complete the inversion of the experimental conditions; using the trained recurrent neural network to predict the temperature distribution in the solid domain of the turbine cascade during the entire stable operation stage. The present invention can shorten the transient experiment time and significantly reduce the transient experiment cost.

Description

Technical Field

[0001] The present invention belongs to the technical field of aero-engine turbine thermal protection, and particularly relates to a turbine temperature distribution prediction method based on deep learning, which is used for predicting the evolution of the temperature distribution in the solid domain over time during a gas turbine transient experiment. Background Art

[0002] In the design of gas turbines, accurate prediction of the operating life is a major challenge. With the increase in thermal load, thermal stress fatigue has attracted more and more attention. Considering thermal stress in the high-pressure turbine heat transfer design, the most important requirement is to accurately predict the metal temperature distribution of the turbine cascade. However, current high-temperature measurement techniques can only measure the surface temperature of the turbine cascade or the temperature at limited discrete points, and cannot predict the temperature field of the entire solid domain. This undoubtedly causes difficulties in accurately evaluating the turbine life. To meet this requirement and obtain the time-series evolution data of the temperature field of the entire solid domain, researchers have proposed a series of methods, but these methods have the following advantages and disadvantages:

[0003] (1) The heat transfer problem between the turbine cascade and the mainstream is a typical problem of the third kind of boundary condition, and the heat transfer coefficient, recovery temperature, and adiabatic temperature during the operation of the turbine are unknown quantities, so it is impossible to directly use the third kind of thermal boundary condition to achieve the time-series prediction of the temperature field of the entire solid domain. For this reason, researchers usually use CFD for three-dimensional simulation, that is, constructing a computational model of the fluid domain and the solid domain, making it have the same inlet and outlet boundary conditions and geometric conditions as the experimental conditions, and then simulating and solving the computational model through the conjugate heat transfer model. However, due to the complex flow structure inside the turbine cascade, if a low-precision turbulence model, such as the RANS / URANS model, etc., is used, the flow structure cannot be accurately predicted, which will affect the accuracy of the prediction results of the temperature field of the solid domain; if a high-precision turbulence model, such as the SAS / LES model, etc., is used, although the prediction accuracy can be improved, its high computational cost limits the use of researchers.

[0004] (2) To avoid the problem of selecting a turbulence model in CFD, some researchers calculate the surface heat flux density or the surface convective heat transfer coefficient through the time-series data of the surface temperature, and then input the surface heat flux density or the convective heat transfer coefficient as the boundary condition into the CFD simulation calculation. Among them, the LRM method proposed by Virginia Tech and the IRM method proposed by the University of Oxford are typical representatives of calculating the surface heat flux density or the surface convective heat transfer coefficient and are widely used. However, the disadvantage of the above methods is that they have high requirements for the measurement accuracy of the time-series temperature of the turbine surface. A measurement uncertainty of 2K may even cause a deviation of 50% in the surface heat flux density or the surface convective heat transfer coefficient, which will greatly affect the prediction of the temperature field of the solid domain.

[0005] (3) In the LRM method proposed by Virginia Tech and the IRM method proposed by the University of Oxford, there are relatively high restrictions on solid heat conduction and the data processing window period. For this reason, Northwestern Polytechnical University proposed a method for obtaining the convective heat transfer coefficient based on the numerical solution of transient solid heat conduction. This method can significantly reduce the restrictions on thermal properties such as solid thermal conductivity. However, in its calculation process, the surface temperature of the cascade is directly used as the first kind of boundary condition for calculating the temperature field, which still poses relatively high requirements for the measurement technology of the cascade surface temperature, that is, the time resolution and measurement accuracy of the surface cascade. And due to the adoption of the first kind of thermal boundary condition, this method cannot be extrapolated, that is, if the full-cycle temperature field evolution needs to be obtained, full-cycle experiments and full-cycle surface temperatures need to be provided, which undoubtedly limits the use of this method. Summary of the Invention

[0006] In order to overcome the above-mentioned disadvantages of the prior art, the purpose of the present invention is to provide a method for predicting the turbine temperature distribution based on deep learning. As a means for processing gas turbine experimental data, this method effectively realizes the prediction of the temporal variation of the temperature distribution inside the solid domain of the turbine, greatly reduces the cost of experimental equipment and the difficulty of experimental operation, and obtains more accurate experimental data, which better meets the needs of gas turbine experimental personnel.

[0007] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0008] A method for predicting the turbine temperature distribution based on deep learning, comprising the following steps:

[0009] Step 1, construct a geometric model of the turbine cascade experiment, generate a computational grid for the turbine cascade, and process the turbine cascade test piece according to the geometric model of the turbine cascade;

[0010] Step 2, construct a gas turbine transient experimental device, wherein the turbine cascade test piece is installed in the wind tunnel test section of the gas turbine transient experimental device, and the gas turbine transient experimental device is used to simulate the cooling characteristics and heat transfer characteristics experiments of the turbine cascade when the upstream incoming flow instantaneously reaches a certain operating condition;

[0011] Step 3, determine the transient experimental conditions and boundary conditions of the turbine cascade, and construct a recurrent neural network encoded by the transient heat conduction equation in combination with the computational grid of the turbine cascade;

[0012] Step 4, conduct a gas turbine transient experiment, collect the time-series data of the total temperature at the mainstream inlet, the time-series data of the surface temperature of the turbine cascade test piece, and the material thermal property parameters of the turbine cascade test piece to form experimental data;

[0013] Step 5, use the experimental data to train the recurrent neural network to complete the inversion of the experimental conditions;

[0014] Step 6: Use the trained recurrent neural network to predict the temperature distribution at all stages of the stable operation of the turbine cascade solid domain.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0016] (1) The third type of thermal boundary condition is directly inverted through the surface temperature of the turbine cascade and the mainstream inlet boundary condition, which is applicable to any turbine cascade and avoids the influence of the accuracy of complex flow prediction in the turbine cascade experiment.

[0017] (2) The present invention couples the idea of full three-dimensional solid heat conduction, enabling it to be applicable to the cascade test pieces processed from any material and eliminating the limitation on the thermal physical properties of the processed material.

[0018] (3) The present invention integrates the idea of linear regression, greatly reducing the requirements for the time resolution and measurement accuracy of the temperature measurement technology, reducing the experimental cost and improving the experimental accuracy.

[0019] (4) The present invention can effectively realize the inversion of the third type of thermal boundary condition, not only reducing the limitation of the window period during experimental operation, but also enabling the extrapolation prediction of the entire stable operation period, reducing the operation duration of the experiment.

[0020] (5) The present invention can achieve the inversion of experimental conditions within a few minutes and quickly predict the evolution within the set prediction period. Description of the Drawings

[0021] Figure 1 is a flowchart of a method for predicting the turbine temperature distribution based on deep learning according to the present invention.

[0022] Figure 2 is a gas turbine transient experimental device of a preferred embodiment of the present invention.

[0023] Figure 3 is a method for operating an experimental device of a preferred embodiment of the present invention.

[0024] Figure 4 is a computational model of a preferred embodiment of the present invention.

[0025] Figure 5 is a computational grid of a preferred embodiment of the present invention.

[0026] Figure 6 is a schematic diagram of the hard coding method adopted by the present invention.

[0027] Figure 7 is a schematic diagram of a recurrent neural network encoded by the transient heat conduction equation of the present invention.

[0028] Figure 8is the prediction deviation of the heat transfer coefficient of a preferred embodiment of the present invention.

[0029] Figure 9 is the prediction effect diagram of the solid domain temperature distribution of a preferred embodiment of the present invention. Detailed implementation manners

[0030] The implementation manners of the present invention will be described in detail below in conjunction with the accompanying drawings and embodiments.

[0031] A method for predicting the temperature distribution of a turbine based on deep learning according to the present invention has the function of inverting the operating condition parameters of the turbine and can be used as an effective data processing tool for turbine cascade experiments; and has the ability to predict the temperature distribution of the solid domain of the turbine cascade, which can shorten the transient experiment time and significantly reduce the transient experiment cost. Refer to Figure 1 As shown, it mainly includes the following steps:

[0032] Step 1: Construct a geometric model of the turbine cascade experiment, generate a computational grid of the turbine cascade, and machine a turbine cascade test piece according to the geometric model of the turbine cascade.

[0033] In this step, a three-dimensional modeling software is used to construct a geometric model of the turbine cascade experiment. The specific operations are as follows:

[0034] Generate a geometric model of the turbine cascade according to the profile data of the turbine cascade through three-dimensional modeling instructions, generate a geometric model of the connection structure according to the assembly structure required for the wind tunnel test section of the gas turbine transient experimental device, and fuse the geometric model of the turbine cascade and the geometric model of the connection structure through the combination instructions of the three-dimensional modeling software to obtain the geometric model of the turbine cascade experiment.

[0035] In this step, the turbine cascade test piece is machined from the turbine cascade manufacturing material (such as metal material) by additive manufacturing or casting.

[0036] Among them, the turbine cascade manufacturing material needs to have complete thermal property parameters, that is, its specific heat capacity Cp, density ρ, and thermal conductivity λ are known. If it is a room temperature experiment (that is, the mainstream temperature is 0°C - 100°C), the influence of temperature on the thermal properties can be ignored, and the above thermal property parameters can be set as fixed values. If it is a medium or high temperature experiment, the variation relationship of the above physical properties with temperature is still required.

[0037] On this basis, a three-dimensional computational grid generation software is used to mesh the geometric model of the turbine cascade experiment to obtain a computational grid of the turbine cascade. This computational grid simulates the role of the turbine cascade test piece in the gas turbine transient experiment, that is, the solid domain part.

[0038] Step 2, construct a gas turbine transient experimental device, where the turbine cascade test piece is installed in the wind tunnel test section of the gas turbine transient experimental device. The gas turbine transient experimental device is used to simulate the cooling characteristics and heat transfer characteristics experiments of the turbine cascade when the upstream incoming flow instantaneously reaches a certain working condition, and can measure the surface temperature of the turbine cascade test piece and the total temperature and total pressure of the mainstream during the process.

[0039] The gas turbine transient experimental device and the operating method of the experimental device of the present invention can refer to patents such as CN112414739A, CN112683943A, and CN118443350A, etc. This device can complete gas turbine transient experiments and measure, collect, and record the time-series data of the mainstream total temperature, total pressure, and turbine surface temperature.

[0040] Figure 2 A preferred structure of the gas turbine transient experimental device is shown, which mainly includes a wind tunnel, a cold air supply system, a total temperature probe, a total pressure probe, a transient infrared thermal imager, an infrared window, and a collection system, etc.; the wind tunnel includes an intake section, an experimental section, and an exhaust section, where the experimental section is the core part of the wind tunnel, and the turbine cascade test piece is installed in the wind tunnel experimental section; the cold air supply system is only installed during the cooling characteristics experiment of the turbine cascade (i.e., the cooling experiment), and does not need to be installed during the heat transfer characteristics experiment of the turbine cascade (i.e., the heat transfer experiment). The cold air supply system is connected to the experimental section at the bottom interface of the turbine cascade test piece; the total temperature probe and the total pressure probe are installed in the experimental section and upstream of the turbine cascade test piece for measuring the total temperature and total pressure of the mainstream; the infrared window is installed at the position of the wind tunnel experimental section facing the turbine cascade test piece; the transient infrared thermal imager measures the surface temperature of the turbine cascade test piece through the infrared window; the collection system collects and records the data measured by the total temperature probe, the total pressure probe, and the transient infrared thermal imager.

[0041] Step 3, determine the transient experimental conditions and boundary conditions of the turbine cascade, and construct a recurrent neural network encoding the transient heat conduction equation in combination with the turbine cascade calculation grid.

[0042] The transient experimental conditions of the turbine cascade in this step mainly include: the Mach numbers at the inlet and outlet of the turbine cascade, the total temperature, the total pressure, the cold air inlet mass flow rate, and the cold air inlet total temperature, etc. These experimental conditions are used to set the mainstream flow parameters, cold air flow parameters of the gas turbine transient experimental device, and the construction of mainstream, cold air, and boundary information in the recurrent neural network. It should be noted that the cold air flow parameters (inlet mass flow rate and inlet total temperature) are only required when the gas turbine experimental device is used for the cooling experiment, and do not need to be set for the heat transfer experiment.

[0043] This step uses the turbine cascade computational grid to construct the geometric information in the recurrent neural network. That is, the computational domain of the recurrent neural network is set as the solid domain part through the turbine cascade computational grid, so that the computational domain of the recurrent neural network accurately simulates the turbine cascade test piece, thereby achieving the purpose of predicting the temperature distribution inside the turbine cascade test piece, that is, in the solid domain. The mainstream and boundary information of the recurrent neural network is set using the hard-coding method to be consistent with the experimental conditions and boundary conditions, thereby achieving the purpose of accurately inverting the experimental conditions.

[0044] The recurrent neural network constructed in this step includes multiple recurrent units. The recurrent units at different time steps are connected through the recurrent operator to achieve the recurrent advancement of the temporal prediction of the temperature distribution inside the turbine cascade test piece, that is, in the solid domain. Each recurrent unit includes an exponential operator module and a CNe operator module, or includes an exponential operator module, a CNe operator module, and an LNe operator module.

[0045] Among them, the exponential operator module is used to calculate the change in the temperature at any spatial position x in the solid domain within the time step h. The CNe operator module and the LNe operator module are used to calculate the influence of the temperature at the spatial position x' on the temperature change at the spatial position x. The difference is that when the LNe operator module is further used on the basis of the CNe operator module, higher accuracy can be obtained.

[0046] In this step, the construction of the recurrent neural network only considers the turbine cascade test piece (i.e., the solid domain) as the computational domain. That is, the boundary conditions of the turbine cascade test piece (i.e., the solid domain) under the experimental conditions are inverted by collecting the mainstream inlet total temperature, the cold air inlet total temperature (only during the cooling experiment), and the surface temperature of the turbine cascade test piece. Therefore, the transient heat conduction equation is encoded into the construction of the recurrent neural network, and the transient heat conduction equation can be selected according to the three-dimensional effect of the turbine cascade test piece in the gas turbine transient experiment, that is, one-dimensional transient heat conduction, two-dimensional transient heat conduction, and three-dimensional transient heat conduction.

[0047] The transient heat conduction equation is expressed as follows:

[0048]

[0049] In the formula, T represents the temperature of the turbine cascade solid domain, t represents time, and α represents the diffusion coefficient;

[0050] The CNe operator module is used to discretize the formula (1) in time and space, and r i and A i are introduced to simplify the equation representation. The resulting equation is as follows:

[0051]

[0052] In the formula, T in+1 represents the temperature at the i-th grid point at the (n + 1)-th moment, T i n represents the temperature at the i-th grid point at the n-th moment represents the temperature at the adjacent grid points (excluding the i-th grid point) of the i-th grid point at the n-th moment, h is the time step, r i is a definition form of a commonly used grid ratio, A i is a compact form of adjacent grid point information

[0053] According to the calculation accuracy requirement, the result obtained by the CNe operator module is used as the calculation result or the predicted value of the calculation result

[0054] When a higher requirement for calculation accuracy is needed, the LNe operator module is used to discretize formula (1) in time and space to obtain

[0055]

[0056] In the formula, T i pred is the predicted value of the calculation result obtained by the CNe operator module or the LNe operator module last time is a compact form of adjacent grid point prediction information, that is, prediction-correction is realized through the LNe operator module to achieve higher calculation accuracy

[0057] Both the CNe and LNe algorithms are explicit time formats, which are beneficial to parallel computing. For the convenience of programming, the present invention decomposes the CNe and LNe algorithms into an exponential operator module (exp Conv), a CNe operator module (CNe Block), and an LNe operator module (LNe Block) according to their action methods. There can be multiple CNe operator modules and LNe operator modules in each loop unit. More CNe operator modules and LNe operator modules mean higher calculation accuracy and greater calculation cost (slower calculation speed and the memory / video memory space required for calculation).

[0058] The hard-coding method of the present invention refers to processing the boundary conditions by introducing virtual nodes (i.e., ghost nodes). This process utilizes the padding function of the convolution kernel of the convolutional neural network to fill the corresponding number of grid padding outside the boundary according to the requirements of the spatial discretization format in the calculation domain boundary. The filled value is the result calculated by using the boundary node algebraic equation method or the virtual boundary method according to the actual physical boundary conditions and the spatial discretization format of the virtual nodes

[0059] Among them, for the fluid-structure interface where the mainstream contacts the turbine cascade test piece, the third type of boundary condition (Robin boundary condition) is adopted, and the first type (isothermal) or second type (adiabatic) boundary condition is adopted for the surfaces of other turbine cascade test pieces; when the mainstream is in a steady flow, the heat transfer coefficient at the fluid-structure interface is a specific constant. The idea of linear regression is introduced into the third type of boundary condition and rewritten as:

[0060]

[0061] In the formula, q represents the convective heat flux density at the fluid-structure interface, T t represents the mainstream temperature, T w represents the wall temperature, T r represents the recovery temperature, h represents the convective heat transfer coefficient at the fluid-structure interface, which is an unknown variable and is obtained through neural network training; due to the introduced idea of linear regression, formula (4) can be regarded as a linear equation with T t -T w as the independent variable x and q as the dependent variable y. The slope a of the linear equation calculated through linear regression is the convective heat transfer coefficient h, and the intercept b is h(T r -T t ).

[0062] Step 4: Conduct a gas turbine transient experiment, collect the time-series data of the total temperature at the mainstream inlet, the time-series data of the surface temperature of the turbine cascade test piece, and the material thermal physical properties of the turbine cascade test piece to form experimental data.

[0063] For the specific operation of this step, reference can be made to Figure 3 , adjust the total pressure at the mainstream inlet through the flow control device of the wind tunnel, adjust the total temperature at the mainstream inlet through the temperature control device of the wind tunnel, and make them reach the predetermined inlet conditions respectively. If it is a cooling experiment, the mass flow rate of the cold air inlet and the cold air inlet temperature also need to be adjusted through the flowmeter and temperature control device of the cold air supply system respectively to make them reach the predetermined cold air flow conditions; during the experiment, ensure the total pressure at the mainstream inlet, that is, make the mainstream reach a steady flow state. If it is a cooling experiment, the stability of the above cold air flow parameters also needs to be ensured; the acquisition system collects and records the data measured by the total temperature probe, the total pressure probe, and the transient infrared thermal imager. Among them, the inlet total pressure is used to timely adjust the mainstream or cold air inlet flow through negative feedback to ensure the requirement of flow stability; the time-series data of the total temperature at the mainstream inlet (the total temperature of the cold air inlet if it is a cooling experiment) and the surface temperature of the turbine cascade measured by the transient infrared thermal imager are used for the training of the recurrent neural network encoded by the constructed transient heat conduction equation.

[0064] Step 5: Use the obtained experimental data to train the recurrent neural network to complete the inversion of the experimental conditions.

[0065] For the fluid-structure interface where the mainstream in the present invention contacts the turbine cascade test piece, the third type of boundary condition (Robin boundary condition) is adopted, and the first type (isothermal) or second type (adiabatic) boundary condition can be adopted for the surfaces of other turbine cascade test pieces. The first type (isothermal) and second type (adiabatic) boundary conditions are both known boundary conditions, while the convective heat transfer coefficient h at the fluid-structure interface in the third type of boundary condition is an unknown number, that is, it is set as the parameter to be trained in the recurrent neural network encoded by the transient heat conduction equation. Taking the surface temperature time series data of the turbine cascade test piece obtained in step 4 as the true value, and taking the surface temperature result predicted based on the guessed value h' as the predicted value, the mean square error (MSE) is used to measure the deviation between the predicted value and the true value.

[0066]

[0067] In the formula, N is the number of measurement points of the surface temperature at the fluid-structure interface, T s,i is the surface temperature of the i-th measurement point, and f(h i ’) is the surface temperature result predicted by the recurrent neural network encoded by the transient heat conduction equation based on the guessed value h i ’.

[0068] The Adam optimizer is used to update the guessed value h' of the convective heat transfer coefficient to make it approach the true value h of the convective heat transfer coefficient. The learning rate during the training process is set to 2, and the StepLR equal-spacing learning rate adjustment strategy is adopted. In StepLR, the adjustment interval is set to 100, the adjustment magnification is set to 0.9, and the training cycle is 1000 epochs.

[0069] Through the above operations, the training of the recurrent neural network encoded by the transient heat conduction equation can be completed. At this time, the obtained recurrent neural network can accurately represent the boundary conditions of the cascade test piece (i.e., the solid domain) in the gas turbine transient experiment carried out, that is, the inversion of the experimental conditions of the gas turbine transient experiment carried out is completed.

[0070] Step 6, use the trained recurrent neural network to predict the temperature distribution of the turbine cascade solid domain during the entire stable operation stage.

[0071] Through the accurate expression of the boundary conditions of the cascade test piece in step 5, the recurrent unit of the recurrent neural network encoded by the transient heat conduction equation can be used to obtain the prediction of the temperature distribution of the turbine cascade solid domain at any time and the evolution of the temperature distribution of the turbine cascade solid domain during the stable operation period.

[0072] In a preferred embodiment of the present invention, to reflect the accuracy of the method, the actual experimental operation process is analogized through a CFD example, and the high-precision prediction result of the CFD is used as the true value to illustrate the present invention.

[0073] Select a CFD turbulence model in the CFD software to simulate the turbine heat transfer / cooling experiment, and set the CFD data recording format, the range of recorded data, and the duration of recorded data. The above operations are respectively analogous to the construction of the gas turbine transient experimental device and the settings for data measurement, acquisition, and recording.

[0074] The calculation model and calculation grid adopted in this embodiment are as Figure 4 , Figure 5 shown. The calculation model is a fluid-structure interaction model of a flat plate in cross flow. The length of the fluid-structure interaction region is 50 mm, and the thickness is 5 mm. The material of the solid domain of this model is selected as the haynes 230 superalloy. Use the calculation grid generation software to divide the grid, and adopt a uniform orthogonal structured grid scheme. The grid cell size: Δx = Δy = Δz = 0.25 mm. The initial state of this embodiment is that the inlet Mach number Ma = 0.2, and the inlet total temperature is the room temperature of 300 K. Then, the flow conditions change transiently, that is, the inlet total temperature steps up to 800 K.

[0075] Use the uniform orthogonal structured grid divided in the above operations to construct the computational domain of the recurrent neural network encoded by the transient heat conduction equation. The divided structured grid characterizes the internal region and geometry of the computational domain.

[0076] Generate virtual nodes (i.e., ghost nodes) for the structured grid through the padding function of the convolutional neural network. For the fluid-structure interface of the mainstream, the third kind of boundary condition is adopted, and the mainstream total temperature information is encoded into the boundary condition. The schematic diagram is as Figure 6 shown in (b). It should be noted that when the mainstream is in a steady flow, the heat transfer coefficient at the fluid-structure interface is a specific constant. Therefore, the linear regression idea is introduced into the third kind of boundary condition and rewritten as:

[0077]

[0078] In the formula, q represents the convective heat flux density at the fluid-structure interface, h represents the convective heat transfer coefficient at the fluid-structure interface, T t represents the mainstream temperature, T w represents the wall temperature, T r represents the recovery temperature. h is an unknown variable and needs to be obtained through neural network training. In this embodiment, the mainstream inlet total temperature is 800 K and is substituted into the above formula.

[0079] For other solid domain surfaces, the second kind of boundary condition is adopted, and the adiabatic information is encoded into the boundary condition, that is, the filled value is the temperature at the center of the first layer of grid on the boundary. The schematic diagram is as Figure 6 shown in (a).

[0080] The recurrent neural network encoded by the transient heat conduction equation realizes the cyclic advancement of the temporal prediction of the temperature distribution in the solid domain of the turbine cascade by connecting the recurrent units through the recurrent operator. Based on the divided uniform orthogonal structured grid, the convolution values of the exponential module, CNe module, and LNe module are calculated respectively to construct the recurrent unit. After testing, in this embodiment, using 1 exponential operator, 1 CNe operator module, and 4 LNe operator modules can achieve the best calculation accuracy and calculation efficiency, and its schematic diagram is as shown in Figure 7 shown.

[0081] The haysen 230 superalloy is used as the solid domain material of the model in this embodiment, and the initial operating condition is an inlet Mach number of 0.2 and an inlet total temperature of 300K. After meeting the conditions of flow stability and solid domain thermal equilibrium, the inlet total temperature is step-adjusted to 800K, and other flow conditions are ensured to remain unchanged. Taking h = 0.01s as the time step, the surface temperature of the solid domain at the fluid-solid interface is measured, collected, and recorded.

[0082] As described above, h is an unknown variable and needs to be obtained through neural network training. In this embodiment, according to the high-precision calculation results of CFD, the mean value of the heat transfer coefficient Therefore, the initial value of h is assigned as U(400, 600) to obtain the guessed value h', and taking h = 0.01s as the time step, the experimental operation interval is predicted to obtain the temporal data T of the predicted surface temperature of the solid domain w,pred . Taking the temporal data T of the surface temperature of the solid domain collected as the true value, the mean square error (MSE) is used to measure the deviation between the predicted value T w and the true value T w,pred . w The deviation.

[0083]

[0084] In the formula, N is the number of measurement points of the surface temperature at the fluid-solid interface, T w,i is the surface temperature of the i-th measurement point, and T w,pred,i is the surface temperature result predicted by the recurrent neural network encoded by the transient heat conduction equation based on the guessed value h i '.

[0085] The Adam optimizer is used to update the guessed value h' of the convective heat transfer coefficient to make it approach the true value h of the convective heat transfer coefficient. The learning rate during the training process is set to 2, and the StepLR equal-spacing learning rate adjustment strategy is adopted. In StepLR, the adjustment interval is set to 100, the adjustment magnification is set to 0.9, and the training period is 1000 epochs.

[0086] Through the above operations, the training of the recurrent neural network encoded by the transient heat conduction equation can be completed. At this time, the trained recurrent neural network encoded by the transient heat conduction equation can accurately represent the boundary conditions of the cascade test piece (i.e., the solid domain) in the gas turbine transient experiment carried out, that is, the inversion of the experimental conditions of the gas turbine transient experiment carried out is completed. Figure 8 It shows the deviation distribution between the predicted heat transfer coefficient h’ and the true value h after the training of the recurrent neural network encoded by the transient heat conduction equation is completed through the above operations. From Figure 8 It can be seen that the deviation of the heat transfer coefficient inverted by the trained recurrent neural network encoded by the transient heat conduction equation is in the interval (-5, 5), and it can accurately invert the actual operating conditions.

[0087] Using the trained recurrent neural network encoded by the transient heat conduction equation, the temperature distribution of the solid domain at t = 10s is predicted with a time step of h = 0.01s. Figure 9 It shows the comparison between the high-precision CFD calculation results and the trained recurrent neural network encoded by the transient heat conduction equation. From Figure 9 It can be seen that the trained recurrent neural network encoded by the transient heat conduction equation can accurately predict the temperature distribution inside the solid domain, and the maximum deviation is only 0.07K, which proves the accuracy of the present invention in predicting the temperature distribution inside the solid domain.

Claims

1. A turbine temperature distribution prediction method based on deep learning, characterized in that: The following steps are involved: Step 1, constructing a turbine blade cascade experimental geometric model, generating a turbine blade cascade computational grid, and processing a turbine blade cascade test piece according to the turbine blade cascade geometric model; Step 2, constructing a gas turbine transient test device, wherein a turbine blade cascade test piece is installed in a wind tunnel test section of the gas turbine transient test device, and the gas turbine transient test device is used to simulate the cooling characteristics and heat transfer characteristics of the turbine blade cascade when the upstream incoming flow instantaneously reaches a certain operating condition; Step 3, determining the transient experimental working conditions and boundary conditions of the turbine blade grid, and constructing a recurrent neural network encoding the transient heat conduction equation in combination with the turbine blade grid calculation grid; Step 4, conduct a gas turbine transient experiment, collect the mainstream inlet total temperature time series data, the turbine blade test piece surface temperature time series data and the material thermophysical property parameters of the turbine blade test piece, and form the experimental data; Step 5, using the experimental data to train the recurrent neural network to complete the inversion of the experimental conditions; Step 6, using the trained recurrent neural network, predicting the temperature distribution of the solid domain of the turbine blade cascade during the whole stage of stable operation; In the step 3, the calculation domain of the recurrent neural network is set as the solid domain part through the turbine blade grid calculation grid, so that the calculation domain of the recurrent neural network accurately simulates the turbine blade grid test piece, thereby achieving the purpose of predicting the temperature distribution inside the turbine blade grid test piece, that is, the solid domain; the mainstream and boundary information of the recurrent neural network are set by a hard coding method to make them consistent with the experimental working conditions and boundary conditions, thereby achieving the purpose of accurately inverting the experimental working conditions; The hard coding method processes the boundary conditions by introducing virtual nodes. The process utilizes the filling function of the convolution kernel of the convolutional neural network to fill a corresponding number of grids outside the boundary of the computational domain according to the requirements of the spatial discrete format. The filled values ​​are the results calculated by the boundary node algebraic equation method or the virtual boundary method according to the actual physical boundary conditions and the virtual node spatial discrete format. Among them, the third type of boundary conditions are used for the fluid-solid interface where the mainstream contacts the turbine blade test piece, and the first or second type of boundary conditions are used on the surfaces of other turbine blade test pieces. The first type of boundary conditions are isothermal boundary conditions, the second type of boundary conditions are adiabatic boundary conditions, and the third type of boundary conditions are Robin boundary conditions. When the mainstream is a stable flow, the heat transfer coefficient of its fluid-solid interface is a specific constant. The linear regression idea is introduced into the third type of boundary conditions and rewritten as: Where q represents the heat flux density of fluid-solid interface convection heat transfer, T t represents the mainstream temperature, T w represents the wall temperature, T r represents the recovery temperature, h represents the convective heat transfer coefficient of the fluid-solid interface, which is an unknown variable and is obtained through neural network training; Formula (4) is regarded as T t -T w The linear equation with x as the independent variable and y as the dependent variable is obtained by linear regression. The slope a of the linear equation is the convective heat transfer coefficient h, and the intercept b is h(T r -T t ).

2. The turbine temperature distribution prediction method based on deep learning according to claim 1 is characterized in that: The step 1 is to construct a turbine blade cascade experimental geometric model using a three-dimensional modeling software, and to obtain a turbine blade cascade test piece by processing the turbine blade cascade manufacturing material through additive manufacturing or casting; The turbine blade cascade experimental geometric model is meshed using three-dimensional computational mesh generation software to obtain the turbine blade cascade computational mesh, which simulates the solid domain part of the turbine blade cascade test piece in the gas turbine transient experiment.

3. The turbine temperature distribution prediction method based on deep learning according to claim 1 is characterized in that: The gas turbine transient test device includes a wind tunnel, a cold air supply system, a total temperature probe, a total pressure probe, a transient infrared thermal imager, an infrared window and a collection system; the wind tunnel includes an air intake section, a test section and an exhaust section, and the turbine blade test piece is installed in the wind tunnel test section; the cold air supply system is only installed when simulating the cooling characteristics experiment of the turbine blade, and does not need to be installed when simulating the heat transfer characteristics experiment of the turbine blade, and the cold air supply system is connected to the test section at the bottom interface of the turbine blade test piece; the total temperature probe and the total pressure probe are installed in the test section and located upstream of the turbine blade test piece, and are used to measure the total temperature and total pressure of the mainstream; the infrared window is installed in the wind tunnel test section at a position directly opposite to the turbine blade test piece; the transient infrared thermal imager measures the surface temperature of the turbine blade test piece through the infrared window; the collection system collects and records the data measured by the total temperature probe, the total pressure probe and the transient infrared thermal imager.

4. The turbine temperature distribution prediction method based on deep learning according to claim 1 is characterized in that: The transient experimental working conditions of the turbine blade grid include: the turbine blade grid inlet and outlet Mach number, total temperature, total pressure, cold air inlet mass flow rate and cold air inlet total temperature; The experimental conditions are used to set mainstream flow parameters, cold air flow parameters of a gas turbine transient experimental device and to construct mainstream, cold air and boundary information in a recurrent neural network.

5. The turbine temperature distribution prediction method based on deep learning according to claim 1 is characterized in that: In the recurrent neural network, recurrent units of different time steps are connected by recurrent operators to realize the recurrent advancement of the time series prediction of the temperature distribution inside the turbine blade test piece, i.e., the solid domain; the recurrent unit includes an exponential operator module and a CNe operator module, or includes an exponential operator module, a CNe operator module and an LNe operator module; the exponential operator module is used to calculate the change of the temperature at any spatial position x in the solid domain within the time step h; the CNe operator module and the LNe operator module are used to calculate the influence of the temperature at the spatial position x' on the temperature change at the spatial position x.

6. The turbine temperature distribution prediction method based on deep learning according to claim 5 is characterized in that: The transient heat conduction equation is as follows: Where T represents the solid domain temperature of the turbine blade, t represents time, and α represents the diffusion coefficient; The CNe operator module is used to discretize formula (1) in time and space, and r is introduced i and A i Expressed as a simplified equation, the resulting equation is as follows: Where, T i n+1 represents the temperature at the i-th grid point at the n+1th time, T i n represents the temperature at the ith grid point at the nth time, represents the temperature of the adjacent grid points of the ith grid point at the nth time, h is the time step, r i is the grid ratio, A i It is a compact form of information about adjacent grid points; According to the calculation accuracy requirement, the result calculated by the CNe operator module is used as the calculation result or the predicted value of the calculation result; When there is a higher requirement for calculation accuracy, the LNe operator module is used to discretize formula (1) in time and space to obtain: Where, T i pred It is the predicted value of the calculation result obtained by the CNe operator module or the LNe operator module last time. A compact form of prediction information for neighboring grid points.

7. The turbine temperature distribution prediction method based on deep learning according to claim 1 is characterized in that: In step 4, the inlet total pressure of the mainstream is adjusted by the flow control device of the wind tunnel, and the inlet total temperature of the mainstream is adjusted by the temperature control device of the wind tunnel, so that they respectively reach the predetermined inlet conditions. If it is a cooling experiment, the flow meter and temperature control device of the cold air supply system are also required to adjust the cold air inlet mass flow rate and the cold air inlet temperature respectively, so that they respectively reach the predetermined cold air flow conditions; during the experiment, the inlet total pressure of the mainstream is guaranteed, that is, the mainstream reaches a stable flow state. If it is a cooling experiment, the stability of the above-mentioned cold air flow parameters must also be guaranteed; the acquisition system collects and records the data measured by the total temperature probe, the total pressure probe and the transient infrared thermal imager, wherein the inlet total pressure is used to timely adjust the mainstream or cold air inlet flow through negative feedback to ensure the flow stability requirements; The mainstream inlet total temperature and the turbine blade surface temperature time series data measured by the transient infrared thermal imager are used to train the recurrent neural network encoded by the constructed transient heat conduction equation.

8. The turbine temperature distribution prediction method based on deep learning according to claim 1 is characterized in that: In the step 5, the surface temperature time series data of the turbine blade test piece obtained in step 4 is taken as the true value, the surface temperature result predicted based on the guessed value h' is taken as the predicted value, and the Adam optimizer is used to update the guessed value h' of the convective heat transfer coefficient to make it close to the true value h of the convective heat transfer coefficient, thereby completing the training of the recurrent neural network.

Images