Neural network-based full-process working condition turbine disc cavity temperature prediction and regulation method

Abstract

The application provides a neural network-based full-process working condition turbine disc cavity temperature prediction and regulation method, and is based on a turbine disc cavity temperature database obtained through numerical simulation of an aero-engine turbine disc cavity; a neural network method is used to establish a mapping relationship between turbine disc cavity rotating speed, disc cavity cooling air input parameters and turbine disc cavity temperature, a full-process working condition turbine disc cavity temperature rapid prediction model is constructed, and high-precision turbine disc cavity temperature can be obtained within milliseconds. At the same time, according to the temperature control requirement of the turbine disc cavity under stable working condition, a rapid regulation model of the turbine disc cavity temperature is constructed, and by means of a bee colony swarm intelligence optimization algorithm, the disc cavity air cooling source input parameters (cooling air pressure, temperature, etc.) are optimized and regulated, so that the real-time regulation of the full-process working condition turbine disc cavity temperature is achieved.

Description

Technical Field

[0001] This invention belongs to the field of aero-engine thermal management technology, and in particular relates to a method for predicting and controlling turbine disk cavity temperature under all operating conditions based on neural networks. Background Technology

[0002] As a core hot-end component of aero-engines, the turbine disk operates in an extremely harsh and complex aero-thermal environment. With the accumulation of thermal fatigue and creep damage, the mechanical properties of the material (mechanical strength, toughness, etc.) gradually decline, affecting the turbine disk's service life and ultimately jeopardizing the safe and stable operation of the aero-engine. Particularly noteworthy is the deformation of the turbine disk's sealing structure under thermal and mechanical loads when the engine is in hot operating conditions. This deformation leads to changes in the sealing clearance, affecting the flow of cooling air between chambers and ultimately impacting the turbine disk temperature. To ensure the turbine disk's service life and reliability, compressed air must be drawn from the compressor for cooling. However, the amount of air available for cooling is limited, and excessive consumption of cooling air directly leads to a decrease in gas turbine efficiency. These factors transform turbine disk temperature prediction into a multivariate parameter prediction problem. Therefore, developing an efficient and high-precision method for predicting turbine disk temperature applicable to multiple variables is an urgent problem to be solved. Currently, researchers have studied methods for calculating turbine disk temperature from the perspectives of one-dimensional mathematical models and two-dimensional / three-dimensional numerical simulations. Hou Shengping et al. used a one-dimensional network method to couple the air system and the solid domain. In the calculation, the solid domain was regarded as a special fluid with infinite flow resistance and solved simultaneously with the fluid network. This method was successfully applied to the fluid-structure simulation of aero-engine turbine disks (Hou Shengping, Tao Zhi, Han Shujun, et al. Exploration and application of integrated fluid-structure network simulation method [J]. Journal of Aerospace Power, 2010, 25(03):509-14.). However, using a single one-dimensional calculation ignores the non-uniformity of temperature and flow, especially in the solution of complex models, and the prediction accuracy cannot be well guaranteed. Therefore, some scholars have proposed cross-dimensional calculation. Guo Xiaojie used the fluid network method for the flow field and the two-dimensional finite element analysis for the solid turbine disk. By establishing the thermal analysis boundary and the calculation method of air temperature rise, the high-pressure turbine casing and turbine disk and its front cavity were selected for coupled calculation and analysis (Guo Xiaojie. Research on the coupling method of air system and thermal analysis of aero-engine [D], Shanghai Jiaotong University, 2014). Sun et al. used CFD to simulate the complex three-dimensional disk cavity flow and heat transfer, and used finite element analysis (FEA) to perform fluid-structure interaction analysis on the thermal response of the turbine disk in a full transient cycle. The transient turbine disk temperature distribution under three operating conditions of shutdown, maximum takeoff and cruise was calculated (SUN Z, CHEW JW, HILLS NJ, et al. Coupled Aerothermomechanical Simulation for a Turbine Disk Through a Full Transient Cycle [J]. Journal of Turbomachinery, 2011, 134(1). However, it also incurred greater costs compared to the above two methods.

[0003] While one-dimensional mathematical models offer rapid computation, many characteristics are difficult to reflect within them, limiting their accuracy and application. Conversely, two-dimensional or three-dimensional numerical simulations achieve high accuracy at the cost of extensive computation. Currently, these methods only calculate the turbine disk temperature field under a specific operating condition, neglecting the prediction and control of disk cavity temperature under different conditions. This makes it difficult to meet future performance requirements for turbine disk cavity temperature in terms of multivariable parameter prediction, high accuracy, and high efficiency. Therefore, it is necessary to propose a rapid prediction model for turbine disk cavity temperature to resolve the trade-off between efficiency and accuracy. Summary of the Invention

[0004] Objective of the Invention: The technical problem to be solved by this invention is to address the shortcomings of existing technologies by providing a method for predicting and controlling turbine disk cavity temperature throughout its entire operating process based on neural networks, comprising the following steps:

[0005] Step 1: Establish the physical model of the turbine disk cavity and the air system flow path model of the aero-engine (these two models are determined according to different aircraft models, and different aircraft models have different structures). The physical model of the turbine disk cavity and the air system flow path model of the aero-engine differ from one aircraft model, but they all include stators and rotors in general. Based on the operating parameters of the aero-engine throughout its entire life cycle, obtain the variation range of characteristic variables.

[0006] Step 2: Establish a simulation model of turbine disk cavity temperature. The simulation model includes a one-dimensional model of the air system flow path and a two-dimensional model of the disk cavity structure (the two-dimensional model of the disk cavity structure is a specific structural model, which is obtained by applying existing technology based on the physical model of the aero-engine turbine disk cavity according to different aircraft models). The one-dimensional model of the air system flow path is obtained by one-dimensional engineering calculation method, and the two-dimensional model of the disk cavity structure is modeled by two-dimensional finite element calculation method. The heat transfer boundary is calculated from the one-dimensional model of the air system flow path and loaded onto the two-dimensional model of the disk cavity structure to calculate the temperature value at typical locations of the turbine disk cavity.

[0007] Step 3: Generate a feature dataset by uniform sampling within the range of feature variable variation. After sampling, input the corresponding parameters in the feature dataset into the turbine disk cavity temperature simulation model. Calculate the turbine disk cavity temperature value from the initial conditions and establish a label dataset corresponding to the feature dataset.

[0008] Step 4: Randomly divide the feature dataset and label dataset obtained in Step 3 into training set, test set, and validation set, which are used for parameter fitting, optimization, and validation in the deep neural network model, respectively.

[0009] Step 5: Based on the deep neural network model, the training set is used as the input to the deep neural network model, the test set is used as the output of the deep neural network model, and the validation set is used to test the deep neural network model to improve its generalization ability. By training and testing the data using the deep neural network model, the mapping relationship between the turbine disk rotation speed and the cold source input parameters (temperature, pressure) and the turbine disk temperature is obtained. The temperature values ​​of typical characteristic points in the turbine disk cavity are obtained within milliseconds, ultimately resulting in a rapid turbine disk cavity temperature prediction model.

[0010] Step 6: Within the possible range of changes in air cold source temperature and pressure, select estimated values ​​for the air cold source input parameters (pressure, temperature, etc.) based on the optimization screening strategy of reverse design technology. Using the estimated values ​​obtained through screening and based on the trained turbine disk cavity temperature rapid prediction model, calculate the temperature TP at typical locations of the turbine disk under full-process operating conditions. pred(i) Because there are many typical locations on the turbine disk, there are more than two temperature points. The maximum number of temperature points to be predicted is n, and the nth temperature point is denoted as TP. pred(n) The value of i ranges from 1 to n;

[0011] Step 7, based on the actual temperature control requirements of the turbine disk cavity throughout the entire operating process (TP) exp(i) The temperature TP predicted in step 6 pred(i) Construct the objective function F for the inverse design problem. obj ;

[0012] Step 8, determine the objective function F obj Is it less than the set threshold φ? The threshold can generally be set to 10. -3 If so, the value of the air cooling source input parameter in step 7 will be used as the result; otherwise, steps 6 to 7 will be repeated, and the values ​​of the turbine disk air cooling source input parameters (pressure, temperature) will be re-assumed using reverse design techniques.

[0013] In step 1, the characteristic variables include the pressure, temperature, and turbine rotor speed of each air system flow path.

[0014] In step 2, the one-dimensional model of the air system flow path is based on the air system flow path model and is modeled through the orifice, the gap between the grates, and the stator temperature field, as follows:

[0015] hole:

[0016]

[0017] Where m1 is the flow rate of the orifice, and C d Here, ρ represents the flow coefficient, A represents the area of ​​the orifice, and ρ represents the flow rate coefficient. t,1 P is the air density at the orifice inlet. t,1 P is the total pressure at the orifice inlet. s,2γ represents the static pressure at the orifice outlet, and γ represents the air adiabatic index.

[0018] Tooth gap:

[0019]

[0020] Where m2 is the flow rate through the gap between the grates, A2 represents the sealing area between the grates, and n t It refers to the number of teeth on the comb, T. t,1 The gas flow rate (K) represents the inlet temperature of the grate gap, R represents the gas constant, and K represents the gas flow rate. co Z represents the flow resistance factor of the toothed grates. t,1 Z is the total pressure at the inlet of the tooth gap. s,2 The static pressure at the outlet of the tooth gap;

[0021] The stator temperature field is solved using the finite element method, as shown below:

[0022] [K T (T)]{T}={R} (3)

[0023] [K S (T)]{L}={F} (4)

[0024] Among them, [K T [T] is the temperature-dependent thermal conductivity matrix, {T} and {R} are the temperature and heat load vectors of discrete nodes, [K] S [(T)] is the temperature-dependent global stiffness matrix, and {L} and {F} are the displacement and load vectors of discrete nodes.

[0025] In step 5, the turbine disk cavity temperature field is fitted using a deep neural network model to obtain the disk cavity temperature, as shown in the following formula:

[0026]

[0027] Among them, f TP (x) is the temperature expression for the predicted point in the disk cavity, σ is the activation function, and w k Let x be the weight of the k-th layer of the neural network. k Let b be the variable input to the k-th layer of the neural network, b be the bias corresponding to each layer, k be the k-th neuron, and L be the number of neurons in each layer.

[0028] In step 6, the reverse design technique is a bee colony optimization algorithm, specifically including: the bee colony optimization algorithm divides the artificial bee colony into three categories by simulating the actual honey-collecting mechanism of bees: foraging bees, observation bees, and scout bees. The goal of the entire bee colony is to find the nectar source with the largest nectar yield. The solution is represented by the location of the nectar source, and the fitness value of the solution is represented by the amount of pollen in the nectar source. All bees are divided into three groups: hired bees, follower bees, and explorer bees. Hired bees are responsible for initially finding nectar sources, collecting nectar, and sharing information. Follower bees are responsible for staying in the hive and collecting nectar based on the information provided by the hired bees. Explorer bees are responsible for randomly finding new nectar sources to replace the original nectar sources after the original nectar sources are abandoned. Like other swarm intelligence algorithms, the bee colony algorithm is iterative.

[0029] After initializing the bee colony and nectar source, the following three stages are repeatedly executed: the hired bee stage, the follower bee stage, and the explorer bee stage, in order to find the optimal solution to the problem.

[0030] The stages of the hired bee phase, the follower bee phase, and the explorer bee phase are described as follows:

[0031] Initialization: During the initialization phase, honey sources are generated using the following formula:

[0032] x ij =x minj +rand[0,1](x maxj -x minj (6)

[0033] Where, x ij x represents the j-th dimension value of the i-th honey source. minj and x maxj These represent the minimum and maximum values ​​of the j-th dimension, respectively; rand[0,1] represents a random number generated between 0 and 1;

[0034] The hired bee stage: During the hired bee stage, the hired bees use the following formula to find new nectar sources:

[0035]

[0036] Among them, v ij Represents a new source of nectar. Represents a random number between -1 and 1; x kj This represents the value of a neighboring honey source in the j-th dimension.

[0037] Follower bee stage: The hired bees share nectar source information, the follower bees analyze the nectar source information, and use the roulette wheel strategy to select nectar sources to follow and mine, so as to ensure that the probability of mining nectar sources with higher fitness values ​​is greater. The mining process of the follower bees is the same as that of the hired bees. They use formula (7) to find new nectar sources and leave behind those with better adaptability.

[0038] Exploratory bee phase: If a nectar source is not updated after being mined more than twice, the nectar source is abandoned and the exploratory bee phase is started. The exploratory bees use formula (6) to randomly find new nectar sources to replace the abandoned nectar sources.

[0039] In step 7, the objective function of the inverse design problem is as follows:

[0040]

[0041] The present invention also provides a storage medium storing a computer program or instructions, which, when the computer program or instructions are run, implements the method for predicting and controlling turbine disk cavity temperature under full-process operating conditions based on neural networks.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] This invention constructs a rapid turbine disk temperature prediction model based on the relationship between turbine disk cavity temperature and thermodynamic boundary conditions throughout the entire operating cycle. This model enables the acquisition of typical turbine disk cavity temperature points within milliseconds. Based on this rapid prediction model, the impact of different cooling strategies on disk cavity temperature can be evaluated to meet practical requirements. Simultaneously, by employing a swarm intelligence optimization algorithm, the input parameters of the disk cavity air cooling source (cooling air pressure, temperature, etc.) are optimized and controlled to achieve real-time dynamic control of the turbine disk cavity temperature throughout the entire operating cycle, thereby obtaining the optimal temperature control scheme for the turbine disk. Attached Figure Description

[0044] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, and the advantages of the present invention in the above and / or other aspects will become clearer.

[0045] Figure 1 This is a schematic diagram of a typical turbine disk cavity structure.

[0046] Figure 2 Schematic diagram of the topology of the turbine disk cavity air system.

[0047] Figure 3 A proxy model for turbine disk cavity temperature.

[0048] Figure 4a This is a schematic diagram for verifying the accuracy of the turbine disk cavity prediction model.

[0049] Figure 4b This is a schematic diagram for verifying the accuracy of the turbine disk cavity prediction model.

[0050] Figure 5 The results show the temperature changes in the turbine disk cavity at different speeds. Detailed Implementation

[0051] This invention provides a method for predicting and controlling turbine disk cavity temperature throughout its entire operating cycle based on neural networks, comprising the following steps:

[0052] Step 1: Establish the physical model of the turbine disk cavity and the air system flow path model of the aero-engine (these two models are determined according to different aircraft models, and different aircraft models have different structures). The physical model of the turbine disk cavity and the air system flow path model of the aero-engine differ from one aircraft model, but they all include stators and rotors in general. Based on the operating parameters of the aero-engine throughout its entire life cycle, obtain the variation range of characteristic variables.

[0053] Step 2: Establish a simulation model of turbine disk cavity temperature. The simulation model includes a one-dimensional model of the air system flow path and a two-dimensional model of the disk cavity structure (the two-dimensional model of the disk cavity structure is a specific structural model, which is obtained by applying existing technology based on the physical model of the aero-engine turbine disk cavity according to different aircraft models). The one-dimensional model of the air system flow path is obtained by one-dimensional engineering calculation method, and the two-dimensional model of the disk cavity structure is modeled by two-dimensional finite element calculation method. The heat transfer boundary is calculated from the one-dimensional model of the air system flow path and loaded onto the two-dimensional model of the disk cavity structure to calculate the temperature value at typical locations of the turbine disk cavity.

[0054] Step 3: Generate a feature dataset by uniform sampling within the range of feature variable variation. After sampling, input the corresponding parameters in the feature dataset into the turbine disk cavity temperature simulation model. Calculate the turbine disk cavity temperature value from the initial conditions and establish a label dataset corresponding to the feature dataset.

[0055] Step 4: Randomly divide the feature dataset and label dataset obtained in Step 3 into training set, test set, and validation set, which are used for parameter fitting, optimization, and validation in the deep neural network model, respectively.

[0056] Step 5: Based on the deep neural network model, the training set is used as the input to the deep neural network model, the test set is used as the output of the deep neural network model, and the validation set is used to test the deep neural network model to improve its generalization ability. By training and testing the data using the deep neural network model, the mapping relationship between the turbine disk rotation speed and the cold source input parameters (temperature, pressure) and the turbine disk temperature is obtained. The temperature values ​​of typical characteristic points in the turbine disk cavity are obtained within milliseconds, ultimately resulting in a rapid turbine disk cavity temperature prediction model.

[0057] Step 6: Within the possible range of changes in air cold source temperature and pressure, select estimated values ​​for the air cold source input parameters (pressure, temperature, etc.) based on the optimization screening strategy of reverse design technology. Using the estimated values ​​obtained through screening and based on the trained turbine disk cavity temperature rapid prediction model, calculate the temperature TP at typical locations of the turbine disk under full-process operating conditions. pred(i)Because there are many typical locations on the turbine disk, there are more than two temperature points. The maximum number of temperature points to be predicted is n, and the nth temperature point is denoted as TP. pred(n) The value of i ranges from 1 to n;

[0058] Step 7, based on the actual temperature control requirements of the turbine disk cavity throughout the entire operating process (TP) exp(i) The temperature TP predicted in step 6 pred(i) Construct the objective function F for the inverse design problem. obj ;

[0059] Step 8, determine the objective function F obj Is it less than the set threshold φ? The threshold can generally be set to 10. -3 If so, the value of the air cooling source input parameter in step 7 will be used as the result; otherwise, steps 6 to 7 will be repeated, and the values ​​of the turbine disk air cooling source input parameters (pressure, temperature) will be re-assumed using reverse design techniques.

[0060] In step 1, the characteristic variables include the pressure, temperature, and turbine rotor speed of each air system flow path.

[0061] In step 2, the one-dimensional model of the air system flow path is based on the air system flow path model and is modeled through the orifice, the gap between the grates, and the stator temperature field, as follows:

[0062] hole:

[0063]

[0064] Where m1 is the flow rate of the orifice, and C d Here, ρ represents the flow coefficient, A represents the area of ​​the orifice, and ρ represents the flow rate coefficient. t,1 P is the air density at the orifice inlet. t,1 P is the total pressure at the orifice inlet. s,2 γ represents the static pressure at the orifice outlet, and γ represents the air adiabatic index.

[0065] Tooth gap:

[0066]

[0067] Where m2 is the flow rate through the gap between the grates, A2 represents the sealing area between the grates, and n t It refers to the number of teeth on the comb, T. t,1 The gas flow rate (K) represents the inlet temperature of the grate gap, R represents the gas constant, and K represents the gas flow rate. co Z represents the flow resistance factor of the toothed grates. t,1 Z is the total pressure at the inlet of the tooth gap. s,2 The static pressure at the outlet of the tooth gap;

[0068] The stator temperature field is solved using the finite element method, as shown below:

[0069] [K T (T)]{T}={R} (3)

[0070] [K S (T)]{L}={F} (4)

[0071] Among them, [K T [T] is the temperature-dependent thermal conductivity matrix, {T} and {R} are the temperature and heat load vectors of discrete nodes, [K] S [(T)] is the temperature-dependent global stiffness matrix, and {L} and {F} are the displacement and load vectors of discrete nodes.

[0072] In step 5, the turbine disk cavity temperature field is fitted using a deep neural network model to obtain the disk cavity temperature, as shown in the following formula:

[0073]

[0074] Among them, f TP (x) is the temperature expression for the predicted point in the disk cavity, σ is the activation function, and w k Let x be the weight of the k-th layer of the neural network. k Let b be the variable input to the k-th layer of the neural network, b be the bias corresponding to each layer, k be the k-th neuron, and L be the number of neurons in each layer.

[0075] In step 6, the reverse design technique is a bee colony optimization algorithm, specifically including: the bee colony optimization algorithm divides the artificial bee colony into three categories by simulating the actual honey-collecting mechanism of bees: foraging bees, observation bees, and scout bees. The goal of the entire bee colony is to find the nectar source with the largest nectar yield. The solution is represented by the location of the nectar source, and the fitness value of the solution is represented by the amount of pollen in the nectar source. All bees are divided into three groups: hired bees, follower bees, and explorer bees. Hired bees are responsible for initially finding nectar sources, collecting nectar, and sharing information. Follower bees are responsible for staying in the hive and collecting nectar based on the information provided by the hired bees. Explorer bees are responsible for randomly finding new nectar sources to replace the original nectar sources after the original nectar sources are abandoned. Like other swarm intelligence algorithms, the bee colony algorithm is iterative.

[0076] After initializing the bee colony and nectar source, the following three stages are repeatedly executed: the hired bee stage, the follower bee stage, and the explorer bee stage, in order to find the optimal solution to the problem.

[0077] The stages of the hired bee phase, the follower bee phase, and the explorer bee phase are described as follows:

[0078] Initialization: During the initialization phase, honey sources are generated using the following formula:

[0079] x ij =x minj +rand[0,1](xmaxj -x minj (6)

[0080] Where, x ij x represents the j-th dimension value of the i-th honey source. minj and x maxj These represent the minimum and maximum values ​​of the j-th dimension, respectively; rand[0,1] represents a random number generated between 0 and 1;

[0081] The hired bee stage: During the hired bee stage, the hired bees use the following formula to find new nectar sources:

[0082]

[0083] Among them, v ij Represents a new source of nectar. Represents a random number between -1 and 1; x kj This represents the value of a neighboring honey source in the j-th dimension.

[0084] Follower bee stage: The hired bees share nectar source information, the follower bees analyze the nectar source information, and use the roulette wheel strategy to select nectar sources to follow and mine, so as to ensure that the probability of mining nectar sources with higher fitness values ​​is greater. The mining process of the follower bees is the same as that of the hired bees. They use formula (7) to find new nectar sources and leave behind those with better adaptability.

[0085] Exploratory bee phase: If a nectar source is not updated after being mined more than twice, the nectar source is abandoned and the exploratory bee phase is started. The exploratory bees use formula (6) to randomly find new nectar sources to replace the abandoned nectar sources.

[0086] In step 7, the objective function of the inverse design problem is as follows:

[0087]

[0088] The present invention also provides a storage medium storing a computer program or instructions, which, when the computer program or instructions are run, implements the method for predicting and controlling turbine disk cavity temperature under full-process operating conditions based on neural networks.

[0089] Example

[0090] This invention uses a turbine disk cavity structure as an example to illustrate a method for predicting and controlling the temperature field of a turbine disk cavity based on a neural network. Figure 1 As shown, a turbine disk cavity structure can be simplified into a two-dimensional axisymmetric model, including a rotor disk, stator 1, and stator 2. Temperature measurements (TP) are performed on several typical structures on the turbine disk. This turbine disk cavity contains two cooling flow paths, with cold source inlets 1 and 2. The specific air system topology is shown in [reference needed]. Figure 2Based on the two models above, the pressure (p1) and temperature (T1) of the cold source inlet ①, the pressure (p2) and temperature (T2) of the cold source ② inlet, and the variation range of the turbine disk rotor (n) can be given.

[0091] Based on the above model, a numerical calculation model for the turbine disk cavity temperature can be established. The air system flow path is obtained using a one-dimensional engineering calculation method, while the turbine disk cavity is modeled using a two-dimensional finite element method. The heat transfer boundary is calculated from the air system flow path and applied to the turbine disk cavity finite element model to calculate the turbine disk cavity temperature. The turbine disk cavity rotor-stator temperature field is solved using the finite element method, as shown below:

[0092] [K T (T)]{T}={R} (1)

[0093] [K S (T)]{L}={F} (2)

[0094] Among them, [K T [T] is the temperature-dependent thermal conductivity matrix, {T} and {R} are the temperature and heat load vectors of discrete nodes, [K] S [(T)] is the temperature-dependent global stiffness matrix, and {L} and {F} are the displacement and load vectors of discrete nodes.

[0095] Based on the analysis of the range of input variables and the establishment of a numerical model for turbine disk cavity temperature, 315 feature datasets were generated within the research range of variables using uniform sampling. Temperature calculations were performed on typical feature points of the turbine disk cavity, resulting in labeled datasets for the temperatures of four typical components. These datasets were divided into training, validation, and test sets for training and testing the rapid prediction model of the disk cavity temperature field, with the training, test, and validation sets accounting for 70%, 15%, and 15% respectively.

[0096] Based on the above dataset, establish as follows Figure 3 The turbine disk cavity temperature model is shown. All disk cavity proxy models employ neural networks to handle regression problems, as shown in the following equation, achieving a nonlinear mapping. The prediction results from the neural network are compared with the labels, and the error calculated using the loss function is applied to update the learning parameters of the neural network until the error meets the accuracy requirements. The activation function, loss function, and optimizer respectively employ the tansig-purelin function, the mean squared loss function, and the Levenberg-Marquardt backpropagation optimization algorithm.

[0097]

[0098] f TP(x) is the temperature expression for the predicted point in the disk cavity, σ is the activation function, and w k Let x be the weight of the k-th layer of the neural network. k Let b be the variable input to the k-th layer of the neural network, b be the bias corresponding to each layer, k be the k-th neuron, and L be the number of neurons in each layer.

[0099] After obtaining the rapid prediction model, its accuracy and generalization ability still need to be tested. The results are as follows: Figure 4a , Figure 4b As shown, this prediction model can achieve high-precision turbine disk temperature prediction. The R-values ​​for the training, test, and validation sets are all above 0.98. Testing was conducted on the sample with the largest error. It can be seen that in the test set, the maximum relative error does not exceed 0.0175, and the prediction error for the vast majority of samples is within 0.0075. Furthermore, the prediction model can replace disk temperature calculation models and other numerical simulation methods, completing the calculation of multiple turbine disk temperatures within 0.1 seconds, meeting the needs of engineering calculations and providing support for subsequent performance analysis and control.

[0100] Based on the above prediction model, the turbine disk cavity temperature can be predicted under different operating conditions. The results of temperature variation with rotational speed at four typical points are as follows: Figure 5 As shown.

[0101] For turbine disk cavity temperature control, within the possible range of cooling input conditions (cold source input pressure and temperature) at a specific turbine disk speed, the estimated values ​​of the input boundary are selected using a bee colony optimization algorithm. The actual temperature control requirements (TP) of the turbine disk cavity are then determined based on the entire operating cycle. exp(i) , with the estimated temperature TP pred(i) (i = 1, 2, ..., n), construct the objective function for the inverse design problem:

[0102]

[0103] Obtaining the objective function F in reverse design techniques obj n is the number of temperature points to be measured;

[0104] Determine the objective function F obj If the input parameters of the air source in the turbine disk are less than the set threshold φ, then the values ​​of the input parameters of the air source in the turbine disk (temperature of cold source 1, pressure of cold source 1, temperature of cold source 2, pressure of cold source 2) are used as the result. Otherwise, the bee colony optimization algorithm is used to re-assume the values ​​of the input parameters of the air source in the turbine disk and recalculate until the optimal input conditions are obtained.

[0105] In its specific implementation, this application provides a computer storage medium and a corresponding data processing unit. The computer storage medium is capable of storing a computer program, which, when executed by the data processing unit, can run the invention's content regarding the full-process turbine disk cavity temperature prediction and control method based on neural networks, as well as some or all of the steps in various embodiments. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0106] Those skilled in the art will clearly understand that the technical solutions in the embodiments of the present invention can be implemented using computer programs and their corresponding general-purpose hardware platforms. Based on this understanding, the technical solutions in the embodiments of the present invention, or the parts that contribute to the prior art, can be embodied in the form of computer programs, i.e., software products. These computer program software products can be stored in a storage medium and include several instructions to cause a device containing a data processing unit (which may be a personal computer, server, microcontroller, MUU, or network device, etc.) to execute the methods described in various embodiments or certain parts of the embodiments of the present invention.

[0107] This invention provides a method for predicting and controlling turbine disk cavity temperature throughout its entire operating cycle based on neural networks. Many methods and approaches exist for implementing this technical solution; the above description is merely a preferred embodiment of the invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention. All components not explicitly stated in this embodiment can be implemented using existing technologies.

Claims

1. A method for predicting and controlling turbine disk cavity temperature under full-process operating conditions based on neural networks, characterized in that, Includes the following steps: Step 1: Establish the physical model of the turbine disk cavity and the air system flow path model of the aero-engine; obtain the variation range of characteristic variables based on the operating parameters of the aero-engine throughout its entire life cycle. Step 2: Establish a simulation model of turbine disk cavity temperature value. The simulation model of turbine disk cavity temperature value includes a one-dimensional model of air system flow path and a two-dimensional model of disk cavity structure. The one-dimensional model of air system flow path is obtained by one-dimensional engineering calculation method, and the two-dimensional model of disk cavity structure is modeled by two-dimensional finite element calculation method. The heat transfer boundary is calculated from the one-dimensional model of air system flow path and loaded onto the two-dimensional model of disk cavity structure to calculate the temperature value at typical locations of turbine disk cavity. Step 3: Generate a feature dataset by uniform sampling within the range of feature variable variation. After sampling, input the corresponding parameters in the feature dataset into the turbine disk cavity temperature simulation model. Calculate the turbine disk cavity temperature value from the initial conditions and establish a label dataset corresponding to the feature dataset. Step 4: Randomly divide the feature dataset and label dataset obtained in Step 3 into training set, test set, and validation set, which are used for parameter fitting, optimization, and validation in the deep neural network model, respectively. Step 5: Based on the deep neural network model, the training set is used as the input of the deep neural network model, the test set is used as the output of the deep neural network model, and the validation set is used to test the deep neural network model to improve the model's generalization ability; by training and testing the data through the deep neural network model, the mapping relationship between the turbine disk speed and the cold source input parameters and the turbine disk temperature is obtained, and the temperature values ​​of typical feature points of the turbine disk cavity are obtained in milliseconds, finally obtaining a rapid prediction model for the turbine disk cavity temperature; Step 6: Within the possible range of changes in air cold source temperature and pressure, select estimated values ​​for the air cold source input parameters based on the optimization screening strategy of reverse design technology. Using the estimated values ​​obtained through screening and a trained turbine disk cavity temperature rapid prediction model, calculate the temperature TP at typical locations of the turbine disk under full-process operating conditions. pred（i） The maximum number of temperature points to be predicted is n, and the nth temperature point is denoted as TP. pred（n） The value of i can be 1 to n; Step 7, based on the actual temperature control requirements of the turbine disk cavity throughout the entire operating process (TP) exp（i） The temperature TP predicted in step 6 pred（i） Construct the objective function F for the inverse design problem. obj ; Step 8, determine the objective function F obj Is it less than the set threshold? If so, take the value of the air cooling source input parameter in step 7 as the result; otherwise, repeat steps 6 to 7 and use reverse design technology to re-assume the value of the turbine disk air cooling source input parameter. In step 2, the one-dimensional model of the air system flow path is based on the air system flow path model and is modeled through the orifice, the gap between the grates, and the stator temperature field, as follows: hole: （1）， in, For the flow rate of the orifice, Here, A represents the flow coefficient and the area of ​​the orifice. The air density at the orifice inlet. Total pressure at the orifice inlet. γ represents the static pressure at the orifice outlet, and γ represents the air adiabatic index. Tooth gap: （2）， in, The flow rate is the distance between the tooth grates. This represents the sealing area between the teeth. It refers to the number of teeth on the comb. R represents the inlet temperature of the gap between the comb teeth, and R represents the gas constant. It represents the flow resistance factor of the toothed grates. The total pressure at the inlet of the tooth gap. The static pressure at the outlet of the tooth gap; The stator temperature field is solved using the finite element method, as shown below: （3）， （4）， Among them, [K T [T] is the temperature-dependent thermal conductivity matrix, where {T} and {R} are the temperature and heat load vectors of the discrete nodes, respectively. S [(T)] is the temperature-dependent global stiffness matrix, and {L} and {F} are the displacement and load vectors of the discrete nodes, respectively.

2. The method according to claim 1, characterized in that, In step 1, the characteristic variables include the pressure, temperature, and turbine rotor speed of each air system flow path.

3. The method according to claim 2, characterized in that, In step 5, the turbine disk cavity temperature field is fitted using a deep neural network model to obtain the disk cavity temperature, as shown in the following formula: （5）， in, (x) is the temperature expression for the predicted point in the disk cavity, σ is the activation function, and w k Let x be the weight of the k-th layer of the neural network. k Let b be the variable input to the k-th layer of the neural network, b be the bias corresponding to each layer, k be the k-th neuron, and L be the number of neurons in each layer.

4. The method according to claim 3, characterized in that, In step 6, the reverse design technique is a bee colony optimization algorithm, which specifically includes: the bee colony optimization algorithm divides the artificial bee colony into three categories by simulating the actual honey-collecting mechanism of bees: foraging bees, observation bees, and scout bees. The goal of the entire bee colony is to find the nectar source with the largest amount of nectar. The location of the nectar source is used to represent the solution, and the amount of pollen in the nectar source is used to represent the fitness value of the solution. All bees are divided into three groups: hired bees, follower bees, and explorer bees. Hired bees are responsible for initially finding nectar sources, collecting nectar, and sharing information. Follower bees are responsible for staying in the hive and collecting nectar based on the information provided by the hired bees. Explorer bees are responsible for randomly finding new nectar sources to replace the original nectar sources after the original nectar sources are abandoned. After initializing the bee colony and nectar source, the following three stages are repeatedly executed: the hired bee stage, the follower bee stage, and the explorer bee stage, in order to find the optimal solution to the problem.

5. The method according to claim 4, characterized in that, The stages of the hired bee phase, the follower bee phase, and the explorer bee phase are described as follows: Initialization: During the initialization phase, honey sources are generated using the following formula: （6）， in, Represents the j-th dimension value of the i-th honey source. and These represent the minimum and maximum values ​​of the j-th dimension, respectively. Represents a random number generated between 0 and 1; The hired bee stage: During the hired bee stage, the hired bees use the following formula to find new nectar sources: （7）， in, Represents a new source of nectar. Represents a random number between -1 and 1; This represents the value of a neighboring honey source in the j-th dimension. Follower bee stage: hired bees share nectar source information, follower bees analyze nectar source information, use roulette wheel strategy to select nectar sources for tracking and mining, use formula (7) to find new nectar sources, and leave behind better adaptors; Exploratory bee phase: If a nectar source is not updated after being mined more than twice, the nectar source is abandoned and the exploratory bee phase is started. The exploratory bees use formula (6) to randomly find new nectar sources to replace the abandoned nectar sources.

6. The method according to claim 5, characterized in that, In step 7, the objective function of the inverse design problem is as follows: （8）。 7. A storage medium, characterized in that, It stores a computer program or instructions that, when executed, implement the method as described in any one of claims 1 to 6.

Images