A method for rapid estimation and control of turbine tip clearance after performance degradation based on machine learning

Abstract

This invention discloses a rapid prediction and control method for turbine tip clearance after performance degradation based on machine learning. Turbine tip clearance is closely related to the performance, safety, and lifespan of aero-engines, requiring prediction and control under multiple operating conditions. Current methods cannot balance computational accuracy and efficiency, failing to meet the engineering computational requirements for tip clearance. This invention, based on a two-dimensional tip clearance database obtained from numerical simulation of aero-engine turbine components, employs machine learning to establish a mapping relationship between engine operating time, turbine thermodynamic boundary conditions, and tip clearance, constructing a rapid tip clearance prediction model capable of obtaining high-precision tip clearance within milliseconds. Based on this rapid prediction model, a rapid tip clearance control method is developed to address the impact of performance degradation on tip clearance, enabling the assessment of the influence of different cooling strategies on tip clearance and rapid recovery of tip clearance.

Description

Technical Field

[0001] This invention belongs to the field of engineering thermophysics technology, and specifically relates to a method for rapid prediction and control of turbine blade tip clearance after performance degradation based on machine learning. Background Art

[0002] Turbine tip clearance is defined as the clearance between rotating and stator components. During aero-engine operation, the tip clearance varies with operating conditions due to different thermal loads on turbine components. A reasonable tip clearance is closely related to aero-engine performance, while an unreasonable tip clearance will lead to higher leakage flow or blade wear, reducing engine performance or causing safety accidents. Predicting tip clearance is of great significance for the development of high-performance aero-engines. However, due to the complexity of the turbine structure and turbine component air system, turbine material properties, air system thermal boundaries, and mechanical loads all affect turbine component deformation, thus altering the tip clearance. This makes tip clearance prediction a multivariate parameter prediction problem. Therefore, developing an efficient and high-precision tip clearance prediction method applicable to multiple variables is an urgent problem to be solved.

[0003] Currently, researchers have conducted studies on tip clearance prediction methods from the perspectives of one-dimensional mathematical models, two-dimensional / three-dimensional numerical simulations, and experimental research. Peng et al. established a one-dimensional mathematical model of tip clearance, simplifying the casing, impeller, and blades into one-dimensional components. Based on this model, they studied the sensitivity of design parameters to tip clearance (Peng K, Fan D, Yang F, et al. Active generalized predictive control of turbine tip clearance for aero-engines. Chinese Journal of Aeronautics, 2013, 26(5): 1147-1155.). Bordo et al. established a two-dimensional analysis model of tip clearance using the finite element method, and considered the three-dimensional deformation effect of the blade component surface in this method, further improving the simulation accuracy (Bordo L, Bruzzone S, Perrone A, et al. Prediction of clearance in industrial gas turbine validated by field operation data. TurboExpo: Power for Land, Sea, and Air. American Society of Mechanical Engineers, 2012, 44731: 675-683.). Kumar et al. conducted three-dimensional simulations of turbine components, obtaining relatively rich information on blade tip clearance and turbine component deformation, but this method was more costly than the two methods mentioned above (Kumar R, Kumar VS, Butt MM, et al. Thermo-mechanical analysis and estimation of turbine blade tipclearance of a small gas turbine engine under transient operating conditions. Applied Thermal Engineering, 2020, 179: 115700.). In terms of experiments, Xu Yijun et al. conducted experimental research on the deformation of the casing components. They simulated the heating effect of the mainstream gas using a radiant heat source and measured the casing temperature and deformation to evaluate the feasibility of the cooling strategy (Xu Yijun, Mao Junkui, Wang Pengfei, et al. Casing model test of high-pressure turbine active clearance control system. Journal of Aerospace Power, 2016, 31(07): 1591-1601.).

[0004] Currently, these methods are insufficient to meet the performance requirements of future tip clearance prediction in terms of multivariate parameter prediction, high accuracy, and high efficiency. For experimental methods, obtaining sufficient tip clearance information in high-temperature turbine components is extremely difficult and expensive. For numerical methods, one-dimensional mathematical models offer very fast computation speeds, but many characteristics are difficult to reflect in a one-dimensional model, limiting its accuracy and application. Conversely, two-dimensional or three-dimensional numerical simulations achieve high-precision results at the cost of extensive computation. Therefore, it is necessary to propose a new tip clearance prediction model to resolve the trade-off between efficiency and accuracy.

[0005] With the development of machine learning, data-driven surrogate models will provide a solution for tip clearance prediction models with low computational cost and high accuracy. Fei et al. conducted sensitivity analysis of turbine disk parameters based on the extreme response surface method and studied the influence of material nonlinearity and dynamic changes in thermal load on tip clearance (Fei CW, Tang WZ, Bai GC. Nonlinear dynamic probabilistic design of turbine disk-radial deformation using extremum response surface method-based support vector machine of regression. Proceedings of the Institution of Mechanical Engineers, Part G: Journal of Aerospace Engineering, 2015, 229(2):290-300.). Schmidt et al. explored the design of two-dimensional casings using the Gaussian regression method (Schmidt T, Gümmer V, Konle M. Potential of surrogate modelling in compressor casing design focussing on rapid tipclearance assessments. The Aeronautical Journal, 2021, 125(1291):1587-1610.). Huang et al. used deep neural networks to diagnose blade damage based on sensor measurement data (Huang X, Zhang X, Xiong Y, et al. A novel intelligent fault diagnosis approach for earlycracks of turbine blades via improved deep belief network using three-dimensional blade tip clearance. IEEE Access, 2021, 9: 13039-13051.).

[0006] These surrogate models have achieved initial success in predicting turbine component deformation. However, none of them considered the impact of aero-engine performance degradation on tip clearance during their modeling process, which limits their ability to predict tip clearance after considering performance degradation. He Hui et al. pointed out that turbine gas temperature degradation and blade creep have a significant impact on turbine tip clearance (He Hui, Mao Junkui, Liu Fangyuan, et al. Research on Turbine Tip Clearance Prediction Method Considering Engine Performance Degradation. Propulsion Technology, 2020, 41(10):2283-2291.). Therefore, it is necessary to develop a rapid tip clearance prediction model suitable for considering performance degradation. Based on this model, the impact of performance degradation and cooling strategies on tip clearance can be explored, so as to quickly restore the tip clearance to the state before performance degradation. Summary of the Invention

[0007] This invention addresses the need for rapid prediction of turbine tip clearance. Considering that current methods often fail to balance accuracy and efficiency, and do not account for the impact of performance degradation on tip clearance, their prediction and control are limited. This invention utilizes a two-dimensional tip clearance database obtained from numerical simulations of aero-engine turbine components. Employing machine learning, it establishes a mapping relationship between engine operating time, turbine thermodynamic boundary conditions, and tip clearance, constructing a rapid tip clearance prediction model that considers performance degradation. This model can obtain high-precision tip clearance predictions within milliseconds. Based on this rapid prediction model, a rapid tip clearance control method is developed to address the impact of performance degradation on tip clearance. This method can assess the impact of different cooling strategies on tip clearance and quickly restore it.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0009] A method for rapid prediction and control of turbine blade tip clearance after performance degradation based on machine learning, characterized by the following steps:

[0010] Step 1: Establish a physical model of the turbine components of the aero-engine and the air system flow path. Based on the operating history parameters of the aero-engine, obtain the flow rate variation range, temperature variation range, rotor speed variation range, and aero-engine usage time variation range for each air system flow path.

[0011] Step 2: Establish a blade tip clearance calculation model and introduce gas temperature degradation and blade creep to correct the blade tip clearance after performance degradation. The casing and impeller are calculated using a two-dimensional axisymmetric finite element method, while the blades are calculated using a one-dimensional engineering method. The blade tip clearance is obtained from the initial clearance, casing deformation, impeller deformation, and blade deformation.

[0012] Step 3: Based on Latin hypercube sampling (LHS), sampling is performed within the range of feature variables in Step 1 to generate a feature dataset. The corresponding parameters in the dataset are then substituted into the tip clearance calculation model established in Step 2 to calculate the initial clearance, casing deformation, wheel deformation, and blade deformation, thereby obtaining the tip clearance and establishing a label dataset corresponding to the feature dataset.

[0013] 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 fitting, optimizing, and validating the parameters in the leaf tip gap fast prediction model, respectively.

[0014] Step 5: Based on the deep neural network model (Multilayer Perceptron, MLP), the feature dataset and label dataset of the training set and test set in Step 4 are used as the input and output of the deep neural network model, respectively. Through multiple training and testing, a fast tip clearance prediction model can be obtained, and the mapping relationship between the thermodynamic parameters of turbine components and tip clearance can be established. The tip clearance value can be obtained in milliseconds.

[0015] Step 6: Input the data from the validation set in Step 4 into the fast tip clearance prediction model in Step 5 to verify the accuracy and generalization ability of the fast tip clearance prediction model. This will result in a tip clearance prediction model that is both efficient and accurate.

[0016] Step 7: Based on the rapid tip clearance prediction model obtained in Steps 5 and 6, the tip clearance after performance degradation can be obtained by inputting different aero-engine usage times under various operating conditions such as idle, takeoff, cruise, and acceleration.

[0017] Step 8: For the tip clearance after performance degradation in Step 7, combined with the tip clearance rapid prediction model obtained in Step 5 and Step 6, the tip clearance under different cooling strategies can be quickly obtained by inputting different cooling strategy parameters, so as to select an appropriate tip clearance rapid adjustment strategy.

[0018] In step two, the tip clearance was corrected by introducing combustion gas temperature degradation and blade creep, as shown in the following formula:

[0019] δ=δ0+L c -L d -L b

[0020] δ0=δ0 0-no degradation -δ ε

[0021]

[0022]

[0023] η = 0.0017 × K / 1000 + 0.0018

[0024] Where δ is the tip clearance, δ0 is the initial clearance, and L c For the deformation of the casing, L d For the deformation of the wheel, L b for

[0025] Blade deformation, δ 0-no degradation δ represents the initial gap without considering performance degradation. ε This is due to blade creep deformation. Let T' be the blade creep rate, K be the engine operating time, η be the temperature degradation factor used to correct for the mainstream exhaust temperature, and T' be the temperature degradation rate. h T represents the mainstream gas temperature after performance degradation. h The mainstream gas temperature is not considered when performance degradation is not taken into account.

[0026] In step five, four deep neural networks are used to fit the initial gap, casing deformation, wheel deformation, and blade deformation respectively, thereby obtaining the blade tip clearance, as shown in the following formula:

[0027]

[0028]

[0029]

[0030]

[0031] in This is the initial gap fitting formula. The fitting formula for casing deformation is as follows: The formula for fitting the deformation of the wheel is as follows: Let σ be the blade deformation fitting formula, x be the input variable for each model, w be the weights of each network layer, b be the biases of each network layer, k be the kth neuron, and L be the number of neurons in each layer.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] This invention can construct a rapid prediction model for turbine blade tip clearance considering performance degradation based on the relationship between the blade tip clearance value after performance degradation and the thermodynamic boundary conditions. This model can achieve high-precision blade tip clearance within milliseconds. Based on this rapid prediction model, the impact of different cooling strategies and performance degradation on blade tip clearance can be evaluated, and the blade tip clearance can be quickly restored to the preset value. Attached Figure Description

[0034] Figure 1 This is a typical turbine component structure;

[0035] Figure 2 This is the scope of research on turbine gas thermodynamic parameters;

[0036] Figure 3 A process for rapid prediction and control of blade tip clearance;

[0037] Figure 4 The flow chart for calculating tip clearance;

[0038] Figure 5 This is a schematic diagram for calculating blade deformation. (a) is a schematic diagram of blade division, and (b) is a schematic diagram of blade heat conduction.

[0039] Figure 6 This is a machine learning-based model for predicting blade tip clearance.

[0040] Figure 7 These are the parameters for a machine learning-based tip clearance prediction model.

[0041] Figure 8 Verification of the accuracy of the blade tip clearance prediction model;

[0042] Figure 9 To account for the tip clearance prediction results after performance degradation;

[0043] Figure 10 The results show the tip clearance control under takeoff conditions. Among them: (a) is the effect of casing cooling flow rate, (b) is the effect of inner casing cooling temperature, and (c) is the effect of outer ring cooling temperature.

[0044] Figure 11 The results show the tip clearance control under climbing conditions. Among them: (a) is the effect of casing cooling flow rate, (b) is the effect of inner casing cooling temperature, and (c) is the effect of outer ring cooling temperature. Detailed Implementation

[0045] The present invention will be further described below with reference to embodiments.

[0046] Example 1

[0047] This invention uses a typical turbine structure as an example to illustrate a method for rapid prediction and control of turbine tip clearance after performance degradation based on machine learning. Figure 1As shown, a typical turbine structure can be simplified to a two-dimensional axisymmetric model, including a turbine disk, blades, and a double-layer casing (i.e., an inner casing and an outer ring). The tip clearance (δ) is the clearance between the rotor and the stator. This turbine component includes four cooling flow paths and one gas flow path. Flow path 1 is responsible for cooling the casing; due to different cooling temperatures, flow path 1 includes two branches, namely flow path branches ① and ②. Flow path 2 is responsible for cooling the left side of the turbine disk; similarly, due to different cooling temperatures, flow path 2 also includes two branches, namely flow path branches ③ and ④. Flow path 3 is responsible for cooling the turbine disk shaft center and the lower right edge; its branch is numbered ⑤. Flow path 4, i.e., flow path branch numbered ⑥, is responsible for cooling the upper right edge of the turbine disk. The gas flow path heats the blade surfaces, the gas side of the casing outer ring, and the gas side of the turbine disk. The constraints on the turbine component are also shown in the diagram. Figure 1 There are three locations: A, B, and C.

[0048] Based on the above model, the following can be summarized: Figure 2 The parameters related to blade tip clearance and their variation range during aero-engine operation are shown. arrive T represents the dimensionless cooling flow rate for cooling paths 1 to 4. 1-1 and T 1-2 Defined as the cooling temperature of branches ① and ② in flow path 1. In flow path 2, T 2-1 and T 2-2 This indicates the cooling temperatures of flow path branches ③ and ④. T3 and T4 are the cooling temperatures of flow paths 3 and 4, respectively. For the gas flow path, the heat transfer coefficients for the gas side of the turbine disc, the blade surface, and the outer ring of the casing are directly given, respectively as h. d h b h c The qualitative temperature used for each test is the turbine inlet temperature T. h N is the rotor speed. K is the engine operating time cycle, one cycle is 2400s.

[0049] By sorting out the variables, we can establish something like... Figure 3 This paper demonstrates a machine learning-based method for rapid prediction and control of blade tip clearance. Based on a blade tip clearance model and boundary conditions, the blade tip clearance considering performance degradation is calculated, and a relevant database is established. Using this database, a deep neural network is employed to fit the variables to the blade tip clearance, achieving rapid prediction and control.

[0050] The process of establishing the blade tip clearance database mainly involves the blade tip clearance calculation process, such as... Figure 4 As shown. Without considering the performance degradation of the aero-engine, the temperature field of the turbine component is first calculated based on the input thermal boundary. Then, the component deformation (L) is solved using the component temperature field and displacement constraints. c 、L d and L b(These are the deformations of the casing, rotor, and blades, respectively). Then, by calculating the relationship between the initial clearance (δ0) and the component deformation, δ is obtained. Considering the performance degradation of the aero-engine after long-term operation, gas temperature degradation and blade creep are introduced to correct the tip clearance. The detailed calculation method is shown below:

[0051] Method for calculating tip clearance:

[0052] δ=δ0+L c -L d -L b (1)

[0053] Where δ is the tip gap, L c 、L d and L b These represent the deformations of the casing, disc, and blades, respectively, with δ0 being the initial clearance.

[0054] The deformation of the wheel and casing was solved using the finite element method, as shown below:

[0055] [K T (T)]{T}={R} (2)

[0056] [K S (T)]{L}={F} (3)

[0057] 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.

[0058] The blades are calculated using a one-dimensional engineering method, dividing them into many segments, such as... Figure 5 As shown, the deformation of the entire blade is obtained by solving the deformation of each segment separately, as shown below:

[0059]

[0060]

[0061] ε l =σ l / E (6)

[0062] L bc-i =l×ε l (7)

[0063] Q = Q m +Q i-1 -Q i (8)

[0064] Q×Δt=cMΔT (9)

[0065] L bt-i =α×ΔT×l (10)

[0066] Among them, L bc For centrifugal deformation, L bt For thermal deformation, m is the number of blade segments, l is the length of the blade segments, and ε is... l Let σ be the strain rate. l Let E be the centrifugal tensile stress, E be Young's modulus, ΔP be the centrifugal force on each cross section, A be the cross-sectional area, and Q be the increase in energy of the leaf segment. m Q is the heat introduced through the blade surface. i-1 Q represents the heat introduced through segment i-1. i Let denoted as i, represent the heat output through segment i, Δt as the time step, c as the specific heat capacity of the blade, M as the mass of the blade segment, Δt as the temperature rise of the blade segment, and α as the coefficient of thermal expansion.

[0067] After considering performance degradation, the main modifications to the combustion gas temperature and blade creep are made based on the above, as shown below:

[0068] δ0=δ 0-no degradation -δ ε (11)

[0069]

[0070]

[0071] η=0.0017×K / 1000+0.0018 (14)

[0072] Where δ0 is the initial gap, δ 0-no degradation δ represents the initial gap without considering performance degradation. ε This is due to blade creep deformation. Let T' be the blade creep rate, K be the engine operating time, η be the temperature degradation factor used to correct for the mainstream exhaust temperature, and T' be the temperature degradation rate. h T represents the mainstream gas temperature after performance degradation. h The mainstream gas temperature is not considered when performance degradation is not taken into account.

[0073] Based on the tip gap calculation model considering performance degradation, a Latin hypercube sampling method was used to generate 2000 feature datasets within the variable study range. Tip gap calculations were then performed to obtain corresponding label datasets. These datasets were divided into training, validation, and test sets for training and testing the fast tip gap prediction model, with 1400, 200, and 400 sets respectively.

[0074] Based on the above dataset, establish as follows Figure 6 The tip clearance prediction model shown is a fast prediction model. The tip clearance prediction model consists of four surrogate models: the δ0 prediction model, the L... c Predictive models, L d Predictive models and L b Prediction Models. All four surrogate models employ deep neural networks to handle regression problems, as shown in equations (16) to (18), achieving 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 ReLU function, mean squared loss function, and Adam optimization algorithm.

[0075]

[0076]

[0077]

[0078]

[0079] in This is the initial gap fitting formula. The fitting formula for casing deformation is as follows: The formula for fitting the deformation of the wheel is as follows: Let σ be the blade deformation fitting formula, x be the input variable for each model, w be the weights of each network layer, b be the biases of each network layer, k be the kth neuron, and L be the number of neurons in each layer.

[0080] After obtaining the rapid prediction model, its accuracy and generalization ability still need to be tested. The results are as follows: Figure 7 As shown, this prediction model can obtain high-precision tip clearance, with prediction errors within 5% for most samples. For the training, validation, and test datasets, samples with prediction accuracy within 5% accounted for 93.12%, 92%, and 91.75% of the total samples, respectively. Furthermore, the prediction model can replace tip clearance calculation models and other numerical simulation methods, completing the calculation of 400 sets of degraded tip clearances within 0.1 seconds, meeting the needs of engineering calculations and providing support for subsequent performance analysis and control.

[0081] Based on the above prediction model, the tip clearance of aero-engines after performance degradation under different operating conditions can be predicted, and the results are as follows: Figure 8 As shown in the figure, the prediction results indicate that the blade tip clearance will decrease after considering the performance degradation of the aero-engine, and the magnitude of the performance change is closely related to the operating parameters of the aero-engine.

[0082] For tip clearance control after performance degradation, a rapid prediction model can be used to obtain changes in tip clearance under various cooling strategies, thereby identifying the optimal control scheme. For example... Figure 9 As shown, the distribution of blade tip clearance under three different cooling strategies for the casing and different engine operating times during takeoff is illustrated. The black dashed line represents the ideal blade tip clearance value, the black solid line represents the acceptable blade tip clearance control margin, and the white dotted line represents the initial cooling strategy. From the graph, we can search for the changes in blade tip clearance while maintaining the initial cooling strategy, as well as the control margin of the cooling strategy within the ideal blade tip clearance range. Furthermore, we can compare the control margins of different cooling strategies to obtain the optimal control variable. Figure 9 It can be seen that, within the entire operating range, the tip clearance can be controlled by adjusting the cooling flow and temperature of the casing.

[0083] Reference example 1

[0084] To verify the reliability of the machine learning-based method for rapid prediction and control of blade tip clearance, it was also validated under the climb condition of an aero-engine. This model can also obtain the blade tip clearance distribution under different cooling strategies and different aero-engine operating times within milliseconds, as shown in the results. Figure 10 As shown. Same Figure 9 Similarly, the black dashed line represents the ideal tip clearance value, the black solid line represents the acceptable tip clearance control margin, and the white dotted line represents the initial cooling strategy. From the graph, we can also find the changes in tip clearance while maintaining the initial cooling strategy, as well as the control margin of the cooling strategy within the ideal tip clearance range. Furthermore, we can compare the control margins of different cooling strategies to obtain the optimal control variable. (Comparison) Figure 9 It can be observed that, under climbing conditions, adjusting the temperature cannot achieve the ideal tip clearance across the entire operating range, and the robustness of adjusting the cooling temperature is lower than that of adjusting the cooling flow rate.

[0085] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for rapid prediction and control of turbine tip clearance after performance degradation based on machine learning, characterized in that, Includes the following steps: Step 1: Establish a physical model of the turbine components of the aero-engine and the air system flow path. Based on the operating history parameters of the aero-engine, obtain the range of characteristic variables for each air system flow path. Step 2: Establish a tip clearance calculation model and incorporate combustion gas temperature degradation and blade creep to correct for performance degradation of the tip clearance, as shown in the following formula: δ=δ0+L c -L d -L b δ0=δ 0-nodegradation -d ε η = 0.0017 × K / 1000 + 0.0018 Where δ is the tip clearance, δ0 is the initial clearance, and L c For the deformation of the casing, L d For the deformation of the wheel, L b For blade deformation, δ 0-nodegradation δ represents the initial gap without considering performance degradation. ε This is due to blade creep deformation. Where η is the blade creep rate, K is the engine operating time, η is the temperature degradation factor used to correct for the mainstream exhaust temperature, and T is the engine operating time. h ′ represents the mainstream gas temperature after performance degradation, T h This does not take into account the mainstream gas temperature when performance degradation occurs; Step 3: Sample the feature variables from Step 1 to generate a feature dataset. Substitute the corresponding parameters from the dataset into the blade tip clearance calculation model established in Step 2 to calculate the initial clearance, casing deformation, wheel deformation, and blade deformation, thereby obtaining the blade tip clearance and establishing 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 fitting, optimizing, and validating the parameters in the leaf tip gap fast prediction model, respectively. Step 5: Based on the deep neural network model, the feature dataset and label dataset of the training set and test set in Step 4 are used as the input and output of the deep neural network model, respectively. Through multiple training and testing, a fast prediction model for leaf tip gap is obtained. Step 6: Input the data from the validation set in Step 4 into the fast tip clearance prediction model in Step 5 to verify the accuracy and generalization ability of the fast tip clearance prediction model; Step 7: Based on the trained blade tip clearance rapid prediction model, for different aero-engine usage times under idle, takeoff, cruise and acceleration conditions, obtain the blade tip clearance after performance degradation. Step 8: For the tip clearance after performance degradation in Step 7, combine the trained tip clearance fast prediction model, and obtain the tip clearance under different cooling strategies by inputting different cooling strategy parameters, so as to select an appropriate tip clearance fast adjustment strategy.

2. The method for rapid prediction and control of turbine tip clearance after performance degradation based on machine learning according to claim 1, characterized in that: The range of characteristic variables mentioned in step one includes: the range of flow rate variation, the range of temperature variation, the range of rotor speed variation, and the range of aero-engine usage time variation.

3. The method for rapid prediction and control of turbine tip clearance after performance degradation based on machine learning according to claim 1, characterized in that: In step two, the casing and the wheel are calculated using a two-dimensional axisymmetric finite element method, while the blades are calculated using a one-dimensional engineering method. The blade tip clearance is obtained from the initial clearance, casing deformation, wheel deformation, and blade deformation.

4. The method for rapid prediction and control of turbine tip clearance after performance degradation based on machine learning according to claim 1, characterized in that: In step three, sampling is performed within the range of the characteristic variables from step one, based on Latin hypercube sampling.

5. The method for rapid prediction and control of turbine tip clearance after performance degradation based on machine learning according to claim 1, characterized in that: In step five, four deep neural networks are used to fit the initial gap, casing deformation, wheel deformation, and blade deformation respectively, thereby obtaining the blade tip clearance, as shown in the following formula: in This is the initial gap fitting formula. The fitting formula for casing deformation is as follows: The formula for fitting the deformation of the wheel is as follows: Let σ be the blade deformation fitting formula, x be the input variable for each model, w be the weights of each network layer, b be the biases of each network layer, k be the kth neuron, and L be the number of neurons in each layer.

Images