CN114841032B - Design method for life stability of thermal component of gas turbine - Google Patents

Abstract

The invention discloses a design method for life robustness of a thermal component of a gas turbine, in particular to the technical field of gas turbines, which comprises the following steps of S1: a flow-heat-solid coupling analysis method is adopted to obtain a temperature field, and a transient finite element analysis method is adopted to obtain a cyclic stress-strain curve from the starting of the gas turbine to the checking point of the heat component in the stopping process; step S2: establishing a low cycle fatigue life agent model of the thermal component by adopting a response surface method; step S3: setting a robustness design variable of a thermal component, and establishing a quantile optimization model and constraint conditions of low cycle fatigue life; step S4: and solving the quantile optimization model by adopting a SPEA-II multi-objective optimization algorithm to obtain an objective function value. In the design stage of the thermal component of the gas turbine, the design variable which ensures the service life of the thermal component to have robustness can be searched in a larger design space, the design service life of the thermal component is prolonged, and the maintenance cost in the whole life cycle is reduced.

Description

A Design Method for Life Robustness of Gas Turbine Thermal Components

Technical field

The present invention relates to the technical field of gas turbines, and in particular, to a robust design method for the life of gas turbine thermal components.

Background technique

Thermal components of gas turbines, especially turbine blades, combustion chambers, retaining rings, etc., are high-temperature key components of ground gas turbines. Gas turbine thermal components serve in harsh environments such as high temperature, high pressure, and high speed, and their reliability is crucial to the stable operation of the gas turbine. Due to changes in operating conditions caused by frequent starts, hot components will bear large centrifugal loads with low frequency changes and large thermal loads. Therefore, low cycle fatigue is an important factor that limits the service life of hot components. Therefore, during the design phase, ensure that hot components have adequate low cycle fatigue life.

The existing life prediction of hot parts of ground gas turbines is mostly based on the calculation results of the temperature field and flow field of hot parts. Finite element analysis of the determined structure is carried out to obtain the stress/strain distribution, and then Manson-Coffin is used to predict the life of hot parts. ) formula to obtain the thermal component life. However, during the manufacturing, processing and service process of hot parts, there are varying degrees of uncertainty in material parameters, geometric structure parameters, and operating loads. The above factors make the fatigue life of hot parts have a large probability range. Existing design methods for thermal components have not considered the impact of fluctuations in the above factors on the life of the thermal components, resulting in the inability to accurately estimate the design life of the thermal components.

Therefore, it is necessary to reasonably control and optimize the factors that affect the life reliability of hot components. Carrying out robust design of the life of gas turbine hot parts can improve the service life of hot parts, reduce the sensitivity of the fatigue life of hot parts to random variables such as load and material parameters, improve the accuracy and reliability of life prediction, and ultimately provide accurate information for the formulation of gas turbine overhaul plans. in accordance with.

Chinese patent CN105608316B discloses a method for calculating the actual service life of the flame tube of the main combustion chamber of an engine. The method for calculating the actual service life of the engine main combustion chamber flame tube includes step 1: obtaining the design point state life, take-off point state life and high temperature take-off point state life of the comparison engine; step 2: obtaining the design point state of the comparison engine life, take-off point state life and high-temperature take-off point state life; Step 3: Obtain the design point state life, take-off point state life and high-temperature take-off point state life of the engine to be tested under the second load spectrum; Step 4: Obtain the second load spectrum The design point state life, take-off point state life and high-temperature take-off point state life of the engine to be tested under the theoretical first turning period under the second load spectrum; Step 5: Calculate the actual service life of the flame tube of the engine to be tested. This invention does not take into account uncertain factors, such as temperature load, material parameters, etc., which will lead to the failure to accurately predict the design life of the flame tube.

Chinese patent CN201910433274.2 discloses a turbine blade fatigue-creep damage coupling probabilistic life prediction calculation method. The method includes the following steps: S1. Collect turbine blade attributes; S2. Determine the assessment location; S3. Perform finite element simulation on the turbine blade to obtain stress and strain information of the turbine blade assessment point; S4. Calculate fatigue damage: through low-cycle fatigue life The model calculates fatigue life and fatigue damage information; S5. Calculate creep damage: Calculate creep life and creep damage information through the creep life model; S6. Calculate overall damage and perform life distribution fitting; S7. Based on cumulative damage theory , combining the life information of multiple operating conditions to obtain the final probability life distribution of the blade. Although the patent considers the impact of materials, loads, and geometric dimensions on life, it does not disclose how to solve the technical problem of improving the robustness of turbine blade life design.

Chinese patent CN107895088B involves a method for predicting the life of an aeroengine combustion chamber, including: CFD analysis of the aeroengine combustion chamber; elasto-plastic static analysis of the aeroengine combustion chamber; compilation of the load spectrum of the aeroengine combustion chamber; fatigue test pieces of the aeroengine combustion chamber matrix alloy Design: design of Hastelloy alloy creep fatigue test standard parts; aircraft engine combustion chamber matrix alloy fatigue test load design; aircraft engine combustion chamber matrix alloy test; using a method combining support vector machine (SVM) and genetic algorithm (GA), Establish a damage prediction model for the matrix alloy of the aero-engine combustion chamber; predict the life of the aero-engine combustion chamber. Although the patent discloses the algorithms of CFD and support vector machines, the invention solves the problem of how to predict the life of the combustion chamber, and does not disclose how to solve the technical problem of improving the robustness of the combustion chamber and the life of the flame tube.

Contents of the invention

In order to solve the problem of low lifetime robustness of gas turbine thermal components, the present invention provides a design method for the lifetime robustness of gas turbine thermal components.

In order to achieve the purpose of the present invention, the technical solutions adopted by the present invention are as follows:

A design method for the life robustness of gas turbine thermal components, including

Step S1: Use the flow-thermal-solid coupling analysis method to obtain the temperature field, and use the transient finite element analysis method to obtain the cyclic stress-strain curve of the thermal component assessment point during the gas turbine startup to shutdown process;

Step S2: Obtain the strain amplitude of the thermal component assessment point through the cyclic stress-strain curve. According to the Manson-Coffin formula of the thermal component material and the obtained strain amplitude, calculate the low cycle fatigue life of the thermal component; and use the response surface method to establish the thermal component Low cycle fatigue life surrogate model;

Step S3: Set the robustness design variables of the thermal component, including controllable variables and noise variables, set the maximum mean and minimum standard deviation of the low cycle fatigue life of the thermal component as the design goal, determine the constraints, and establish a design based on quantile Parametric design optimization model;

Furthermore, for different thermal components, the constraints are different;

Step S4: Use the SPEA-II multi-objective optimization algorithm to solve the quantile-based parameter design optimization model to obtain the objective function value and corresponding design variables.

Further, the stress-strain curve acquisition step in step S1 is:

Step 1: Calculate the thermal component temperature field calculation boundary condition parameters through thermodynamic formulas;

Step 2: Use CFD software to solve the flow field and solid temperature field of the thermal component to obtain the distribution of the temperature field of the thermal component from startup to shutdown;

Step 3: Directly map the temperature on the solid domain node obtained in the CFD calculation to the solid domain node of the thermal component calculated by ANSYS to calculate the stress and strain of the thermal component;

Step 4: Determine the assessment point and obtain the cyclic stress-strain curve of the assessment point in the solid domain of the thermal component.

Further, the assessment point is an area where structural strength failure occurs in the thermal component.

Further, the step of obtaining the low cycle fatigue life surrogate model in step S2:

Step 11: Select the design variable Calculated using the Mason-Coffin formula: The low cycle fatigue life N _f is obtained, and the calculated low cycle fatigue life N _f is called the original sample point;

Step 12: Transform the design variables: where x _i is the i-th sample of the design variable X, μ _i and σ _i are the mean and standard deviation of the design variable, x′ _i is the initial sample point after the spatial size is changed;

Step 13: Randomly select n1 initial sample points, conduct machine learning through response surface method, and obtain the learning model;

Step 14: Take the remaining n-n1 initial sample points as detection sample points and monitor the accuracy of the learning model. If the error between the low cycle fatigue life calculated by the learning model and the initial sample points is greater than a%, repeat step 13. and step 14, if the accuracy of the monitored learning model meets the error between the low cycle fatigue life calculated by the learning model and the initial sample point ≤ a%, proceed to step 15;

Further, the error a≤5;

Step 15: Use the Monte Carlo sampling method to sample the agent model to obtain the low cycle fatigue life distribution curve of the assessment point.

Further, σ′ _f in step 11 is the fatigue strength coefficient, σ _m is the average stress, ε′ _f is the fatigue ductility coefficient, Δε is the strain amplitude, b is the fatigue strength index, and c is the fatigue ductility index.

Further, setting robustness design variables in step S3 and establishing a quantile optimization model and constraints for low cycle fatigue include:

Thermal component life N=f(y,z), where y,z are robust design variables;

The design goal is The constraint condition is y _L ≤ y ≤ y _U ;

g _j (y,z)≤0;

Further, N ^0.5 is the lower quantile with probability 0.5, is the difference between the lower quantiles with probabilities P2 and P1, μ is the life mean optimization target, y is the controllable variable, y _U is the upper limit of the controllable variable, y _L is the lower limit of the controllable variable; 1≤ j≤m, g _j (y, z) is the j-th constraint among m constraints, and z is the noise variable.

Further, the steps of the SPEA-II multi-objective optimization algorithm include:

Step a: Set the robustness design variable to the evolved individual, and set the maximum evolutionary generation T, population size N, and archive set size M. Initialize to generate the first-generation evolutionary population P0 and empty archive set Q0. The evolution generation t=0, assuming Determine the maximum evolutionary generation number T;

Step b: Use the SPEA-II multi-objective optimization algorithm to calculate the fitness value F(i) of all individuals i in the current evolutionary population Pt and archive set Qt;

Step c: Use the next generation archive set Qt+1 to save the current evolutionary group Pt and all non-dominated individuals in the current archive set Qt, and compare the number of non-dominated individuals in the archive set Qt+1 with the set archive set size M The size follows the principle of withdrawing more and making up less, so that the number of individuals in the archive set Qt+1 is equal to M;

Step d: If the evolution algebra t ≥ T, or other termination conditions are met, the evolution stops, the non-dominated solution in Qt+1 is stored in the non-dominated solution set NDSet, and the NDSet and objective function values are output;

Step e: Use the binary tournament selection method to conduct tournament selection on the next generation archive set Qt+1, and select appropriate individuals to enter the matching library;

Step f: Perform crossover and mutation operations in the evolutionary algorithm in the paired library, and save the results to the next generation evolutionary population set Pt+1, let the evolutionary generation t=t+1, repeat steps b to e until termination is satisfied. Conditional step d.

Further, hot parts include parts in gas turbines that work in high temperature and high pressure environments, specifically including turbine blades, flame tubes, retaining rings, discs and other parts.

Compared with the prior art, the beneficial effects of the present invention are specifically reflected in:

(1) In the design stage of gas turbine thermal components, the present invention introduces noise factors as design variables in the early stage of design, thus further expanding the design space. That is, the present invention can find ways to ensure a robust life span of thermal components in a larger design space. Sexual design variables;

(2) One of the optimization goals of the present invention is to minimize the standard deviation, that is, to minimize the fluctuation range, and reduce the life dispersion interval of gas turbine hot parts, so that the low-cycle fatigue life of hot parts can be more accurately estimated and formulated for the gas turbine overhaul cycle. Provide accurate basis;

(3) The present invention prolongs the service life of thermal components, reduces the number of maintenance times, reduces maintenance costs in the entire life cycle, and enhances the market competitiveness of the gas turbine.

Description of drawings

Figure 1 is a flow chart of the method for calculating the life robustness of gas turbine thermal components according to the present invention;

Figure 2 is a schematic diagram of characteristic dimensions of a gas turbine blade in Embodiment 1;

Figure 3 shows the robustness analysis results of the gas turbine blades of Embodiment 1;

Figure 4 is a schematic diagram of the characteristic dimensions of the gas turbine flame tube in Embodiment 2;

Figure 5 is a comparison of the life robustness design results of the gas turbine flame tube in Example 2;

Detailed ways

In order to make the purpose and technical solution of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments.

Example 1

According to the robust design method of gas turbine thermal component life shown in Figures 1 to 3, the low cycle fatigue life of a ground gas turbine turbine blade is robustly designed. The specific steps include:

Step S1: Perform CFD calculations on the flow field and solid temperature field of the turbine blades to obtain the temperature field distribution of the turbine blades. At the same time, use the transient finite element analysis method to analyze the solid domain of the turbine blades, and determine the air film hole edge of the blade body through the calculation results. As the assessment point, the cyclic stress-strain curve of the air film hole edge of the blade body is obtained;

Specifically, this region has high stress levels due to the large temperature gradient.

Step S2: Obtain the strain amplitude of the turbine blade at the assessment point from startup to shutdown through the cyclic stress-strain curve. Based on the Manson-Coffin curve of the material IN738LC and the obtained strain amplitude, calculate the low cycle fatigue life of the blade; and use the response The surface method is used to establish the low cycle fatigue life surrogate model of the blade;

Step S3: Set the life robustness design variables of the blade, and set the maximum mean and minimum standard deviation of the low cycle fatigue life of the blade as the design goal, determine the constraints, and establish a quantile-based parameter design optimization model;

Specifically, the robust design variables are controllable random variables and noise variables. The purpose of establishing a quantile-based parameter design optimization model is to calculate the maximum mean and ensure that the fluctuation range of the maximum mean is minimized;

Step S4: Use the SPEA-II multi-objective optimization algorithm to optimize the quantile-based parameter design optimization model to obtain the output design variables and objective function values.

Specifically, the steps for obtaining the stress-strain curve in step S1 are:

Step 1: Obtain the boundary condition parameters for blade three-dimensional temperature field calculation through calculation of the overall characteristics of the gas turbine;

Step 2: Use CFD software to solve the flow field and solid temperature field of the blade, and obtain the distribution cloud diagram of the flow field and temperature field in the combustion chamber during the startup-rated operating conditions-stop process of the blade;

Step 3: Use the same solid domain grid as in the CFD calculation to directly map the temperature on the blade solid domain grid nodes obtained in the CFD calculation to the flame tube solid domain grid calculated by ANSYS; conduct transient stress on the blade Strain analysis, cyclic loading 3 to 4 times;

Step 4: According to the calculation results, determine the edge of the blade air film hole as the assessment point, and obtain the stress-strain curve of the assessment point.

Specifically, the assessment point in step 4 is the area where the structural strength of the thermal component fails.

Specifically, the step of obtaining the low cycle fatigue life proxy model in step S2 is:

Step 11: Select the design variables, set the number of original samples to 32, and use the Latin hypercube sampling method to obtain the original samples. Conduct finite element analysis on the turbine blades to obtain the average stress and strain amplitude of the assessment point, and pass the average stress Calculated using the modified Mason-Coffin formula: The low cycle fatigue life N _f is obtained, and the calculated low cycle fatigue life N _f is called the original sample point; where σ′ _f is the fatigue strength coefficient, σ _m is the average stress, ε′ _f is the fatigue ductility coefficient, Δε is the strain amplitude, b is the fatigue strength index, and c is the fatigue ductility index.

Step 12: Convert design variables: where x _i is _{the i} -th sample of the design _variable It is to reduce the rounding error of the computer and improve the stability and generalization of the training support vector regression machine. The basic design variables are used as input data due to different physical meanings and dimensions, resulting in large differences in their respective value ranges. In support vector regression Machine training is prone to instability, so the basic design variables need to be sized so that they have equal importance in training;

Step 13: Randomly select 70% of the initial sample points, conduct machine learning through the response surface method, and obtain the learning model;

Step 14: Take the remaining 30% of the initial sample points as detection sample points and monitor the accuracy of the learning model. If the error between the low cycle fatigue life calculated by the learning model and the initial sample points is greater than 5%, repeat steps 13 and Step 14, if the accuracy of the monitored learning model meets the error between the low cycle fatigue life calculated by the learning model and the initial sample point ≤ 5%, proceed to step 15;

Specifically, the error reflects the error between the actual response value and the predicted value of the original sample point;

Step 15: Use the Monte Carlo sampling method to sample the agent model to obtain the low cycle fatigue life distribution curve of the assessment point;

Specifically, setting robustness design variables in step S3 and establishing a quantile optimization model and constraint conditions for low cycle fatigue include:

Thermal component life N=f(y,z), where y,z are robust design variables;

The design goal is The specific N ^0.5 is the lower quantile with a probability of 0.5,/> is the difference between the lower quantile of probabilities P2 and P1, μ is the life mean optimization target, where P2 is 0.9999 and P1 is 0.0001,

The constraints are:

The maximum temperature does not exceed the design value: Ti<T0,

The average cross-section temperature does not exceed the allowable temperature, Tavg<T1;

y is a controllable variable, y includes the air film hole diameter d, the air mold hole inclination angle a, the air mold hole composite angle β, the air film hole circumferential spacing L1, the air film hole axial spacing L2, and the elastic modulus E of In738LC ;

Z is the noise variable, which includes blade inlet gas temperature T and blade inlet gas pressure P

The steps of the SPEA-II multi-objective optimization algorithm specifically include:

Step (1) Initialization: Set the size of the evolutionary population N, the size of the archive set M, and the maximum evolutionary generation T. Under the premise that the maximum temperature Ti<T0 and the cross-sectional average temperature Tavg<T1 are satisfied, the variable d, Randomly initialize an evolutionary group, namely point set P0 and empty archive set Q0, in the 8-dimensional variable space composed of a, β, L1, L2, E, T, and P. Let the evolutionary algebra t=0 and set the maximum evolutionary algebra T.

Step (2) Fitness allocation: Substitute the points in the union of point sets P0 and Q0 into the two objective functions , dominated solutions and non-dominated solutions can be distinguished based on the relative position of each point in the objective space composed of objective function values. According to the relationship between domination and dominance, the intensity value S(i) of individual i is calculated, and an original fitness value R(i) is preliminarily calculated based on the intensity value S(i), and then the point i is calculated in the target space and The Euclidean distance of other points can be used to obtain the density value D(i) of the point, and the fitness value F(i)=R(i)+D(i) of all individuals in the current evolutionary group Pt and the current archive set Qt is calculated.

Step (3) Environment selection: Use the next generation archive set Qt+1 to save all non-dominated individuals in Pt and Qt. If the number of individuals in Qt+1 exceeds M, use the relative position of the individuals in the target space to eliminate individuals in Qt+1 that are too close to other individuals one by one, that is, remove some individuals from dense areas in the target space; If the number of individuals in Qt+1 is smaller than M, then according to the individual fitness value, the dominant individuals in Pt and Qt are selected in order to fill Qt+1;

Step (4) End condition: If t≥T, or other end conditions are met, save the non-dominated individuals in Qt+1 to NDSet as the final result. At the end of the evolution process, NDSet is the final archive set, and NDSet and the corresponding objective function value;

Step (5) Pairing selection: Perform a binary tournament selection on Qt+1, that is, randomly select two individuals in Qt+1, and the individuals with good fitness enter the matching library, and the selection operation is repeated until the matching library is filled.

Step (6) Evolution operation: For individuals in Qt+1, pairs are randomly paired and binary coding is used. By performing a crossover operation, individuals can also randomly mutate in binary digits (0 becomes 1, 1 becomes 0), Perform the mutation operation and save the evolution result to Qt+1, let t=t+1, and go to step (2).

Example 2

According to the robust design method of gas turbine thermal component life shown in Figure 1, Figure 4 and Figure 5, the low cycle fatigue life of a ground gas turbine flame tube is robustly designed. The specific steps include:

Step S1: Use the flow-thermal-solid coupling analysis method to perform CFD calculations on the flow field and solid temperature field of the flame tube to obtain the temperature field distribution of the flame tube. At the same time, use the transient finite element analysis method to analyze the solid domain of the flame tube in the combustion chamber. , through the calculation results, the cyclic stress-strain curve of the flame tube gas film hole edge during the gas turbine startup to shutdown process is obtained;

Specifically, this region has high stress levels due to the large temperature gradient.

Step S2: Obtain the strain amplitude of the flame tube at the assessment point from startup to shutdown through the cyclic stress-strain curve. Calculate the low cycle fatigue of the flame tube based on the Manson-Coffin formula of the flame tube material Hastelloy-X and the obtained strain amplitude. life; and the response surface method is used to establish a proxy model for the low cycle fatigue life of the flame tube;

Step S3: Set the life robustness design variables of the flame tube, including controllable variables and noise variables. Set the maximum mean and minimum standard deviation of the low cycle fatigue life of the flame tube as the design goals, determine the constraints, and establish a quantile-based Several parameter design optimization models;

Specifically, the robust design variables are controllable random variables and noise variables. The purpose of establishing a quantile-based parameter design optimization model is to calculate the maximum mean and ensure that the fluctuation range of the maximum mean is minimized;

Step S4: Use the SPEA-II multi-objective optimization algorithm to solve the quantile-based parameter design optimization model to obtain the output design variables and objective function values.

Specifically, the steps for obtaining the stress-strain curve in step S1 are:

Step 1: Calculate the boundary condition parameters for the three-dimensional temperature field calculation of the flame tube through thermodynamic formula calculation;

Step 2: Use CFD software to solve the flow field and solid temperature field of the combustion chamber flame tube, and obtain the distribution cloud diagram of the flow field and temperature field in the combustion chamber during the startup-rated operating conditions-stop process of the flame tube;

Step 3: Use the same solid domain grid as in the CFD calculation to directly map the temperatures on the flame tube solid domain grid nodes obtained in the CFD calculation to the flame tube solid domain grid calculated by ANSYS; perform instantaneous calculation of the flame tube. State stress and strain analysis, cyclic loading 3 to 4 times;

Step 4: According to the calculation results, determine the edge of the air film hole as the assessment point, and obtain the stress-strain curve of the assessment point.

Specifically, the assessment point in step 4 is the area where the structural strength of the thermal component fails.

Further, the step of obtaining the low cycle fatigue life surrogate model in step S2:

Step 11: Select the design variables, set the number of original samples to 32, and use the Latin hypercube sampling method to obtain the original samples. Conduct finite element analysis on the flame tube to obtain the average stress and strain amplitude of the assessment point, and pass the average stress Calculated using the modified Mason-Coffin formula: The low cycle fatigue life N _f is obtained, and the calculated low cycle fatigue life N _f is called the original sample point; where σ′ _f is the fatigue strength coefficient, σ _m is the average stress, ε′ _f is the fatigue ductility coefficient, Δε is the strain amplitude, b is the fatigue strength index, and c is the fatigue ductility index.

Specifically, the design variables include flame tube structural parameters, material parameters and load boundary conditions, including the diameter of the air film hole d, the circumferential spacing of the air film holes L1, the axial spacing of the air film holes L2, the air film hole arrangement angle β, The flame tube wall thickness t, the elastic modulus E of H-X, the flame tube inlet gas temperature T, the flame tube inlet gas pressure P, and the fuel flow rate m.

Step 12: Convert design variables: where x _i is _{the i} -th sample of the design _variable It is to reduce the rounding error of the computer and improve the stability and generalization of the training support vector regression machine. The basic design variables are used as input data due to different physical meanings and dimensions, resulting in large differences in their respective value ranges. In support vector regression Machine training is prone to instability, so the basic design variables need to be sized so that they have equal importance in training;

Step 13: Randomly select 70% of the initial sample points, conduct machine learning through the response surface method, and obtain the learning model;

Step 14: Take the remaining 30% of the initial sample points as detection sample points and monitor the accuracy of the learning model. If the error between the low cycle fatigue life calculated by the learning model and the initial sample points is greater than 5%, repeat steps 13 and Step 14, if the accuracy of the monitored learning model meets the error between the low cycle fatigue life calculated by the learning model and the initial sample point ≤ 5%, proceed to step 15;

Specifically, the error reflects the error between the actual response value and the predicted value of the original sample point;

Step 15: Use the Monte Carlo sampling method to sample the agent model to obtain the low cycle fatigue life distribution curve of the assessment point;

Specifically, setting robustness design variables in step S3 and establishing a quantile optimization model and constraint conditions for low cycle fatigue include:

Thermal component life N=f(y,z), where y,z are robust design variables;

The design goal is The specific N ^0.5 is the lower quantile with a probability of 0.5,/> is the difference between the lower quantile of probabilities P2 and P1, μ is the life mean optimization target, where P2 is 0.9999 and P1 is 0.0001,

The constraints are:

The maximum allowable wall temperature does not exceed the design value: Ti<T0,

The wall thickness does not exceed the set value t<t0;

y is a controllable variable, y includes the air film hole diameter d1, the air film hole circumferential spacing L1, the air film hole axial spacing L2, the air film hole arrangement angle β, the flame tube wall thickness d2, and the elastic modulus E of H-X ;

The noise design variables z are: flame tube inlet gas temperature T, flame tube inlet gas pressure P, and fuel flow rate m;

The steps of the SPEA-II multi-objective optimization algorithm specifically include:

Step (1) Initialization: Set the size of the evolutionary population N, the size of the archive set M, and the maximum evolutionary generation T. Under the premise of satisfying the constraint conditions Ti<T0 and wall thickness t<t0, set the variables d, L1, L2 ,In the 9-dimensional variable space composed of β,t,E,T,P,m, an evolutionary group, namely the point set P0 and the empty archive set Q0, is randomly initialized, the evolutionary generation t=0, and the maximum evolution generation T is set.

Step (2) Fitness allocation: Substitute the points in the union of point sets P0 and Q0 into the two objective functions , dominated solutions and non-dominated solutions can be distinguished based on the relative position of each point in the objective space composed of objective function values. According to the relationship between domination and dominance, the intensity value S(i) of individual i is calculated, and an original fitness value R(i) is preliminarily calculated based on the intensity value S(i), and then the point i is calculated in the target space and The Euclidean distance of other points can be used to obtain the density value D(i) of the point, and the fitness value F(i)=R(i)+D(i) of all individuals in the current evolutionary group Pt and the current archive set Qt is calculated.

Step (3) Environment selection: Use the next generation archive set Qt+1 to save all non-dominated individuals in Pt and Qt. If the number of individuals in Qt+1 exceeds M, use the relative position of the individuals in the target space to eliminate individuals in Qt+1 that are too close to other individuals one by one, that is, remove some individuals from dense areas in the target space; If the number of individuals in Qt+1 is smaller than M, then according to the individual fitness value, the dominant individuals in Pt and Qt are selected in order to fill Qt+1;

Step (4) End condition: If t≥T, or other end conditions are met, save the non-dominated individuals in Qt+1 to NDSet as the final result. At the end of the evolution process, NDSet is the final archive set, and NDSet and the corresponding objective function value;

Step (5) Pairing selection: Perform a binary tournament selection on Qt+1, that is, randomly select two individuals in Qt+1, and the individuals with good fitness enter the matching library, and the selection operation is repeated until the matching library is filled.

Step (6) Evolution operation: For individuals in Qt+1, pairs are randomly paired and binary coding is used. By performing a crossover operation, individuals can also randomly mutate in binary digits (0 becomes 1, 1 becomes 0), Perform the mutation operation and save the evolution result to Qt+1, let t=t+1, and go to step (2).

The above are only embodiments of the present invention. The descriptions are relatively specific and detailed, but should not be construed as limiting the patent scope of the present invention. It should be noted that, for those of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention.

Claims (3)

1. A design method for the life robustness of gas turbine thermal components, which is characterized by including: Step S1: Use the flow-thermal-solid coupling analysis method to obtain the temperature field, and use the transient finite element analysis method to obtain the cyclic stress-strain curve of the thermal component assessment point during the gas turbine startup to shutdown process; Step S2: Obtain the strain amplitude of the thermal component assessment point through the cyclic stress-strain curve. According to the Manson-Coffin formula of the thermal component material and the obtained strain amplitude, calculate the low cycle fatigue life of the thermal component; and use the response surface method to establish the thermal component Low cycle fatigue life surrogate model; Step S3: Set the robustness design variables of the thermal component, including controllable variables and noise variables, set the maximum mean and minimum standard deviation of the low cycle fatigue life of the thermal component as the design goal, determine the constraints, and establish a design based on quantile Parametric design optimization model; Step S4: Use the SPEA-II multi-objective optimization algorithm to solve the quantile-based parameter design optimization model to obtain the objective function value and corresponding design variables; Among them, the steps to obtain the stress-strain curve in step S1 are: Step 1: Calculate the thermal component temperature field calculation boundary condition parameters through thermodynamic formulas; Step 2: Use CFD software to solve the flow field and solid temperature field of the thermal component to obtain the distribution of the temperature field of the thermal component from startup to shutdown; Step 3: Directly map the temperature on the solid domain node obtained in the CFD calculation to the solid domain node of the thermal component calculated by ANSYS to calculate the stress and strain of the thermal component; Step 4: Determine the assessment points and obtain the cyclic stress-strain curve of the assessment points in the solid domain of the thermal component; Steps to obtain the low cycle fatigue life surrogate model in step S2: Step 11: Select the design variable Calculated using the Mason-Coffin formula: , the low cycle fatigue life N _f is obtained, and the calculated low cycle fatigue life N _f is called the original sample point; in the formula, /> is the fatigue strength coefficient,/> is the average stress, is the fatigue ductility coefficient,/> is the strain amplitude, b is the fatigue strength index, c is the fatigue ductility index; Step 12: Transform the design variables: , where x _i is the i-th sample of design variable X,/> and/> are the mean and standard deviation of the design variables,/> is the initial sample point after the spatial size is changed; Step 13: Randomly select n1 initial sample points, conduct machine learning through response surface method, and obtain the learning model; Step 14: Take the remaining n-n1 initial sample points as detection sample points and monitor the accuracy of the learning model. If the error between the low cycle fatigue life calculated by the learning model and the initial sample points is greater than a%, repeat step 13. and step 14, if the accuracy of the monitored learning model meets the error between the low cycle fatigue life calculated by the learning model and the initial sample point ≤ a%, proceed to step 15; Step 15: Use the Monte Carlo sampling method to sample the agent model to obtain the low cycle fatigue life distribution curve of the assessment point; In step S3, setting robust design variables and establishing a quantile optimization model and constraints for low cycle fatigue include: Thermal component life N=f(y,z), where y,z are robust design variables; The design goal is ; The constraints are ; g _j (y,z)≤0; is the lower quantile with probability 0.5,/> is the difference between the lower quantiles with probabilities P2 and P1,/> is the life mean optimization objective, y is the controllable variable, y _U is the upper limit of the controllable variable, y _L is the lower limit of the controllable variable; 1≤j≤m, gj(y,z) is the m constraint The jth constraint, z is the noise variable; The steps of the SPEA-II multi-objective optimization algorithm in step S4 include: Step (1) Initialization: Set the size of the evolutionary population N, the size of the archive set M, and the maximum evolutionary generation T. Under the premise that the maximum temperature Ti<T0 and the cross-sectional average temperature Tavg<T1 are satisfied, the variable d, Randomly initialize an evolutionary group, namely point set P0 and empty archive set Q0, in the 8-dimensional variable space composed of a, β, L1, L2, E, T, and P. Let the evolution generation t=0 and set the maximum evolution generation T; Step (2) Fitness allocation: Substitute the points in the union of point sets P0 and Q0 into the two objective functions In , the dominant solution and the non-dominated solution are distinguished according to the relative position of each point in the target space composed of the objective function value; according to the relationship between the dominant and the dominated, the intensity value S(i) of the individual i is calculated, and according to the intensity value S(i) initially calculates an original fitness value R(i), and then calculates the Euclidean distance between point i and other points in the target space to obtain the density value D(i) of the point, and calculates the current evolutionary group Pt and The fitness value F(i)=R(i)+D(i) of all individuals in the current archive set Qt; Step (3) Environment selection: Use the next generation archive set Qt+1 to save all non-dominated individuals in Pt and Qt; if the number of individuals in Qt+1 exceeds M, use the relative position of the individuals in the target space to eliminate Qt one by one. Individuals in +1 that are too close to other individuals in Euclidean distance, that is, some individuals are removed from dense areas in the target space; if the number of individuals in Qt+1 is smaller than M, then according to the individual fitness value, some individuals are removed from Pt and Qt The dominant individual is selected in order and filled with Qt+1; Step (4) End condition: If t ≥ T, or other end conditions are met, save the non-dominated individuals in Qt+1 to NDSet as the final result. At the end of the evolution process, NDSet is the final archive set, and output NDSet and the corresponding objective function value; Step (5) Pairing selection: Perform a binary tournament selection on Qt+1, that is, randomly select two individuals in Qt+1, and the individuals with good fitness enter the matching library, and the selection operation is repeated until the matching library is filled; Step (6) Evolution operation: Randomly pair individuals in Qt+1 using binary coding. By performing crossover operations, individuals can also perform mutation operations through random mutation on binary digits, and save the evolution results to Qt. In +1, let t=t+1 and go to step (2). 2. The design method for the life robustness of gas turbine thermal components according to claim 1, characterized in that the assessment point is an area where structural strength failure of the thermal component occurs. 3. The design method for the lifetime robustness of gas turbine thermal components according to claim 1, characterized in that the thermal components at least include turbine blades, flame tubes, retaining rings, and disks.