Process parameter recommendation method for improving fretting fatigue performance of blade tenon of gas turbine by AI

Abstract

The invention discloses a process parameter recommendation method for improving the fretting fatigue performance of a blade tenon of a gas turbine through AI. According to the method, a tenon characteristic parameter-process parameter mapping database is established through ultrasonic vibration rolling processing experiment data, wherein a characteristic parameter combination is dynamically acquired in real time by using an established adaptive sensor network detection system; obtaining a physical constraint based on the established physical control equation of each characteristic parameter to construct a PINN model; a physical field of complete space coordinates is inverted by using a PINN model trained by a tenon feature parameter-process parameter mapping database and is screened, Pareto solution sets of different positions after the complete space coordinates are screened are generated based on an NSGA-II algorithm, and five optimal process parameter combinations at the positions of the space coordinates are obtained through screening based on a TOPSIS decision-making method. According to the method, intelligent recommendation of the ultrasonic vibration rolling strengthening process parameters of the gas turbine blade tenon can be realized, and the recommendation result is relatively excellent.

Description

Technical Field

[0001] The present invention belongs to the technical field of the fretting fatigue performance of gas turbine blade tenons, and particularly relates to a process parameter recommendation method for improving the fretting fatigue performance of gas turbine blade tenons empowered by AI. Background Art

[0002] As a key connecting component of the rotor system, the gas turbine blade tenon is subjected to the combined action of centrifugal load, aerodynamic excitation and high-temperature environment for a long time. The proportion of its fretting fatigue failure accounts for more than 60% of the blade failures. Therefore, it is necessary to strengthen the processing of the gas turbine blade tenon to improve the fretting fatigue performance of the gas turbine blade tenon. However, in the existing traditional strengthening processing, the optimization of process parameters mainly relies on the empirical trial-and-error method, and there are two major technical bottlenecks: First, the efficiency and accuracy of multi-physical field data acquisition are insufficient. The existing methods use a single fixed sensor layout (such as a single strain gauge or thermocouple), which is difficult to capture the coupling effect of transient stress (>500Hz) and local high-temperature gradient (>100°C / mm) on the tenon contact surface, resulting in an identification error of the key area exceeding 30%. Second, the mapping relationship modeling between process parameters and fatigue performance is limited. Traditional finite element analysis requires thousands of core hours of computing resources, and it is difficult to integrate the synergistic effects of material nonlinear surface integrity. The parameter recommendation results generally deviate from the Pareto optimal solution. Although some studies have tried to introduce machine learning in recent years, due to the lack of physical constraints, the generated process parameters often violate the law of conservation of energy or cause machining chatter, and the qualified rate of actual application is less than 70%. Therefore, it is urgent to develop an intelligent recommendation system that integrates high-precision sensing, multi-field coupling modeling and physical enhanced optimization to break through the bottleneck of empirical dependence and achieve the improvement of the fretting fatigue life of gas turbine blade tenons. Summary of the Invention

[0003] The purpose of the present invention is to overcome the deficiencies of the prior art and propose a process parameter recommendation method for improving the fretting fatigue performance of gas turbine blade tenons empowered by AI.

[0004] To achieve the above purpose, the present invention adopts the following technical solutions:

[0005] The process parameter recommendation method for improving the fretting fatigue performance of gas turbine blade tenons by AI of the present invention is as follows:

[0006] S1. Establish a three-dimensional model of the gas turbine blade.

[0007] S2. Determine the key monitoring area of the tenon tooth surface of the gas turbine blade tenon through finite element analysis.

[0008] S3. Build an adaptive sensor network detection system: Arrange a mobile detection unit composed of a laser displacement sensor and an infrared thermal imager carried by a multi-degree-of-freedom robotic arm on the test fixture base, and arrange a fixed detection unit composed of an acoustic emission sensor array on the test fixture base.

[0009] S4. Conduct ultrasonic vibration rolling experiments, and use the adaptive sensor network detection system to dynamically collect various characteristic parameters of the key monitoring areas of the tenon head and tenon teeth surfaces in real time, obtaining a data set composed of spatial coordinates (x, y, z), ultrasonic vibration rolling time t, process parameter combinations, and characteristic parameter combinations. Use the data set to establish a mapping database of tenon head characteristic parameters - process parameters; among them, the process parameter combinations are composed of ultrasonic frequency f, static load p, and feed speed v, and the characteristic parameter combinations are composed of temperature field T, stress field σ, and displacement field u. The temperature field T is measured by an infrared thermal imager, and the stress field σ and displacement field u are measured by an acoustic emission sensor array.

[0010] S5. Establish a physical control equation system for each characteristic parameter in a three-dimensional rectangular coordinate system: establish a heat conduction equation based on the temperature distribution, establish a stress balance equation based on the stress distribution, and establish a vibration wave equation based on the displacement distribution.

[0011] S6. Construct a PINN network architecture with spatial coordinates (x, y, z), process parameter combinations, and ultrasonic vibration rolling time t as inputs and characteristic parameter combinations as outputs.

[0012] S7. Physical-guided data generation and generalization: For the key monitoring areas of the tenon head and tenon teeth surfaces, combined with the mapping database of tenon head characteristic parameters - process parameters, create spherical neighborhoods with the coordinates of each grid point in the key monitoring areas of the tenon head and tenon teeth surfaces as the centers of the spheres, and generate new grid points through Latin hypercube sampling; input the coordinates of each new grid point, the process parameter combinations at the corresponding center grid points of the spheres, and the preset ultrasonic vibration rolling time into the trained PINN network model, and retain those that meet |R T | < 1 °C / s, |R σ | < 5 MPa and |R u | < 10 km / s 2 of the predicted values, the corresponding new grid point coordinates, and the process parameter combinations at the corresponding center grid points of the spheres.

[0013] S8. Online recommendation and adaptive adjustment: According to the characteristic parameter combinations retained in step S7, call the NSGA-II algorithm to generate a Pareto solution set. The Pareto solution set is composed of multiple candidate process parameter combinations, and based on the TOPSIS decision-making method, screen out 5 optimal process parameter combinations at different spatial coordinate positions in the key monitoring areas of the tenon head and tenon teeth surfaces except for the spatial coordinate positions in the data set.

[0014] Preferably, in step S1, a three-dimensional model of the gas turbine blade is established through three-dimensional modeling software, or the surface point cloud data of the gas turbine blade is obtained by a three-dimensional scanner. The surface point cloud data of the gas turbine blade is sequentially denoised and surface reconstructed through Geomagic Design X software to establish a three-dimensional model of the gas turbine blade.

[0015] Preferably, the specific process of step S2 is as follows: Import the three-dimensional model of the gas turbine blade into finite element software, perform mesh division on the three-dimensional model of the gas turbine blade, calculate the curvature information of each grid point on the tenon tooth surface of the gas turbine blade tenon using a curvature analysis tool, apply aerodynamic load, centrifugal load, and temperature boundary conditions to the three-dimensional model of the gas turbine blade, simulate the working conditions of the gas turbine blade, solve the stress field and temperature field of each grid point on the tenon tooth surface through thermo-mechanical coupling finite element analysis, divide the area where the curvature change rate of each grid point on the tenon tooth surface is greater than 15% / mm into an aerodynamic sensitive area, the area where the strain gradient is greater than 200 με / mm into a structural danger area, and the area where the temperature gradient is greater than 80 °C / cm into a thermal load area, so as to obtain the key monitoring area on the tenon tooth surface composed of each aerodynamic sensitive area, each structural danger area, and each thermal load area.

[0016] Preferably, the specific process of step S4 is as follows: Set the ranges of each process parameter, select multiple process parameter nodes within each process parameter range for orthogonal design to obtain multiple process parameter combinations. Then set the amplitude A, and conduct ultrasonic vibration rolling strengthening processing experiments on the key monitoring area of the tenon tooth surface under different process parameter combinations. And after each ultrasonic vibration rolling strengthening processing experiment on the tenon of the gas turbine blade is completed using a process parameter combination, replace the new tenon of the gas turbine blade, and use another process parameter combination to conduct ultrasonic vibration rolling strengthening processing experiments on the new tenon of the gas turbine blade. When conducting ultrasonic vibration rolling strengthening processing experiments under each process parameter combination, use an adaptive sensor network detection system to dynamically collect each characteristic parameter of the key monitoring area of the tenon tooth surface in real time, so as to obtain a data set composed of spatial coordinates (x, y, z), ultrasonic vibration rolling time t, process parameter combinations, and characteristic parameter combinations, and establish a mapping database of tenon characteristic parameters - process parameters using the data set.

[0017] More preferably, when conducting ultrasonic vibration rolling strengthening processing experiments under a process parameter combination and using an adaptive sensor network detection system to dynamically collect each characteristic parameter of the key monitoring area of the tenon tooth surface in real time, the process is as follows:

[0018] ① The multi-degree-of-freedom robotic arm drives the laser displacement sensor to scan the tenon tooth surface of the gas turbine blade tenon on the workpiece fixture clamped on the test fixture base at a preset speed, and based on the ResNet-18 deep learning model, identify the positions of each aerodynamic sensitive area, each structural danger area, and each thermal load area on the tenon tooth surface;

[0019] ②Use the corresponding process parameter combination to perform ultrasonic vibration rolling strengthening on the key monitoring area of the gas turbine blade tenon, and at the same time start the enhanced acquisition mode: the multi-degree-of-freedom robotic arm drives the infrared thermal imager to move along the preset path, so that the infrared thermal imager can detect the temperature distribution of the key monitoring area in real time, and adjust the relative pose of the infrared thermal imager probe and the tenon surface, so that the infrared thermal imager probe always maintains a perpendicular incident measurement attitude to the tenon tooth surface of the tenon, and at the same time the acoustic emission sensor array detects the stress distribution and displacement distribution of the tenon tooth surface in real time.

[0020] Preferably, the heat conduction equation established based on the temperature distribution is

[0021]

[0022] In the formula, ρ is the material density of the gas turbine blade tenon, c p is the specific heat capacity of the material, k is the thermal conductivity of the material, is the vector differential operator, is the divergence of the heat flux density, Q ultrasonic is the ultrasonic energy input term, and Q ultrasonic = 2πfηA 2 , η is the transducer efficiency, Q friction is the frictional heat generation term, and Q friction = μpv, μ is the friction coefficient;

[0023] The stress balance equation established based on the stress distribution is

[0024]

[0025] In the formula, is the divergence of the stress field, representing the net stress per unit volume, F ultrasonic is the dynamic load caused by ultrasonic vibration, and F ultrasonic = 2π 2 f 2 Aρ.

[0026] The vibration wave equation established based on the displacement distribution is

[0027]

[0028] In the formula, c is the sound speed in the material of the gas turbine blade tenon, is the Laplace operator of the displacement field, S(t) is the ultrasonic transducer excitation source term, β is the attenuation coefficient, is the initial phase angle.

[0029] More preferably, the PINN network architecture in step S6 is as follows:

[0030] ① Input layer: spatial coordinates (x, y, z), process parameter combination, and ultrasonic vibration rolling time t;

[0031] ② Main network: an 8-layer fully connected network, and the output prediction value is the predicted value of the characteristic parameter combination;

[0032] ③ Calculation of differential terms: calculate the predicted temperature field gradient predicted stress field divergence and predicted displacement field Laplacian term

[0033] ④ Physical residual constraint: calculate the heat residual R T 、stress residual R σ and vibration residual R u ;

[0034] ⑤ Hybrid loss function:

[0035] L = w data (L T + L σ + L u ) + w phys (‖R T ‖2 + ‖R σ ‖2 + ‖R u ‖2)

[0036] In the formula, N is the number of samples in the predicted temperature field or the actual temperature field T, is the predicted temperature value of the i-th sample in the predicted temperature field , and is the actual temperature value of the i-th sample in the actual temperature field T; M is the number of samples in the predicted stress field or the actual stress field σ, is the predicted stress value of the i-th sample in the predicted stress field , and is the actual stress value of the i-th sample in the actual stress field σ; is the number of samples in the predicted displacement field or the actual displacement field u, is the predicted displacement value of the i-th sample in the predicted displacement field , and is the actual displacement value of the i-th sample in the actual displacement field u; w data and w phys are both weight coefficients; among them, the actual temperature field T, stress field σ, and displacement field u all come from the tenon feature parameter - process parameter mapping database.

[0037] More preferably, in the step S6, the thermal residual R T , stress residual R σ and vibration residual R u are calculated by the formula

[0038]

[0039] Preferably, in the step S7, in addition to the known domain strengthening for the key monitoring area of the tenon and mortise surface, unknown domain extrapolation is also performed for other areas of the tenon and mortise surface except the key monitoring area. The unknown domain extrapolation process is as follows: construct a geometric similarity index Γ:

[0040]

[0041] In the formula, α, β and γ are all weight coefficients, and R, K t and Q are the curvature radius, tangential curvature and area of the grid points selected for other areas of the tenon and mortise surface except the key monitoring area respectively, and R0, K t0 and Q0 are the curvature radius, tangential curvature and area of the grid points selected within the key monitoring area of the tenon and mortise surface;

[0042] For other areas of the tenon and mortise surface except the key monitoring area, the process parameter combination of each grid point with Γ value > 85% is taken as the control group corresponding to the grid point of the key monitoring area of the tenon and mortise surface. The process parameter combination (f0, p0, v0) of the control group is scaled proportionally to form the process parameter combination (f new , p new , v new ) of the grid points for other areas of the tenon and mortise surface except the key monitoring area. The grid point coordinates of other areas of the tenon and mortise surface except the key monitoring area, the corresponding process parameter combination and the preset ultrasonic vibration rolling time are input into the PINN network model, and the predicted values and the corresponding grid point coordinates and process parameter combinations that meet |R T | < 3 °C / s, |R σ | < 10 MPa and |R u | < 30 km / s 2 are retained. Among them, the process parameters of each grid point for other areas of the tenon and mortise surface except the key monitoring area are

[0043]

[0044] More preferably, according to the characteristic parameter combination retained by the unknown domain extrapolation process, the NSGA-II algorithm is called to generate a Pareto solution set, which consists of multiple candidate process parameter combinations. Based on the TOPSIS decision-making method, 5 optimal process parameter combinations at different spatial coordinate positions in other areas of the tenon-mortise surface except the key monitoring area are screened, and then 5 optimal process parameter combinations at all spatial coordinate positions of the tenon-mortise surface except the spatial coordinate positions in the dataset are obtained.

[0045] The present invention has the following beneficial effects:

[0046] The present invention can realize the intelligent recommendation of the ultrasonic vibration rolling strengthening process parameters of the gas turbine blade tenon. Compared with the traditional empirical process parameter combination, it can significantly improve the fretting fatigue performance of the gas turbine blade tenon. Specifically, the present invention establishes a tenon characteristic parameter-process parameter mapping database through the characteristic parameter combination obtained from the ultrasonic vibration rolling processing experiment, establishes the physical control equation of each characteristic parameter, and constructs a PINN network model based on the physical control equation of each characteristic parameter to obtain physical constraints, realizing the integration of physical laws and data-driven. The PINN network model trained with the tenon characteristic parameter-process parameter mapping database can invert the physical field (characteristic parameter combination) of the complete spatial coordinates from the sparse spatial coordinates of the key monitoring area, so that the process parameter combination data can be screened according to the physical field of each spatial coordinate to meet the thermo-mechanical-vibration coupling law, solving the problems of data fragmentation, high cost and low efficiency caused by the traditional single-sensor point-by-point measurement; further, the NSGA-II algorithm is used to generate the Pareto solution set at different positions after the complete spatial coordinate data screening, and 5 optimal process parameter combinations at different spatial coordinate positions of the tenon-mortise surface are screened based on the TOPSIS decision-making method. Compared with the traditional empirical recommended process parameter combination, it can significantly improve the fretting fatigue performance of the gas turbine blade tenon. Among them, the characteristic parameter combination is detected by the built adaptive sensor network detection system. The adaptive sensor network detection system adopts a mobile detection unit composed of a multi-degree-of-freedom robotic arm carrying a laser displacement sensor and an infrared thermal imager, and a fixed detection unit composed of an acoustic emission sensor array. The fixed detection unit and the mobile detection unit form a spatial topological association, which can realize the collaborative capture of multi-physical parameters in the full spatial domain, and use the millisecond-level clock synchronization technology, thereby ensuring the spatio-temporal consistency of each characteristic parameter data, improving the detection accuracy, and making the output results of the PINN network model and the NSGA-II algorithm more accurate. Description of the Drawings

[0047] Figure 1 It is a flowchart of the method of the present invention. Detailed Embodiments

[0048] The present invention will be further described below in conjunction with embodiments.

[0049] The method for recommending process parameters of the present invention for improving the fretting fatigue performance of the tenon of a gas turbine blade is as Figure 1 shown, and the specific steps are as follows:

[0050] S1. Establish a three-dimensional model of the gas turbine blade: Establish a three-dimensional model of the gas turbine blade through three-dimensional modeling software, or use a three-dimensional scanner (such as GOM ATOS Q three-dimensional scanner) to obtain the surface point cloud data of the gas turbine blade, and perform point cloud denoising and surface reconstruction on the surface point cloud data of the gas turbine blade in sequence through Geomagic DesignX software to establish a three-dimensional model of the gas turbine blade.

[0051] S2. Determine the key monitoring areas of the tenon tooth surface of the gas turbine blade tenon: Import the three-dimensional model of the gas turbine blade into finite element software, perform mesh division on the three-dimensional model of the gas turbine blade, calculate the curvature information of each grid point on the tenon tooth surface of the gas turbine blade tenon using the curvature analysis tool, and apply aerodynamic load, centrifugal load and temperature boundary conditions to the three-dimensional model of the gas turbine blade to simulate the working conditions of the gas turbine blade. Solve the stress field and temperature field of each grid point on the tenon tooth surface through thermo-mechanical coupling finite element analysis. Divide the areas where the curvature change rate of each grid point on the tenon tooth surface is greater than 15% / mm into aerodynamic sensitive areas, the areas where the strain gradient is greater than 200 με / mm into structural dangerous areas, and the areas where the temperature gradient is greater than 80 °C / cm into thermal load areas, so as to obtain the key monitoring areas on the tenon tooth surface composed of each aerodynamic sensitive area, each structural dangerous area and each thermal load area. In this embodiment, hexahedral meshes are used during mesh division, the rotational speed is set to 15000 rpm when applying centrifugal load, and the aerodynamic pressure is set to 2 MPa when applying aerodynamic load.

[0052] S3. Build an adaptive sensor network detection system: Arrange a mobile detection unit composed of a laser displacement sensor and an infrared thermal imager carried by a multi-degree-of-freedom robotic arm on the test fixture base, and arrange a fixed detection unit composed of an 8-channel acoustic emission sensor array on the test fixture base. The fixed detection unit and the mobile detection unit form a spatial topological association, which can realize the collaborative capture of multi-physical parameters in the full spatial domain.

[0053] S4. Establish a mapping database of tenon feature parameters - process parameters: Set the ranges of various process parameters, select multiple process parameter nodes within each process parameter range for orthogonal design to obtain multiple process parameter combinations. Then set the amplitude A and conduct ultrasonic vibration rolling strengthening processing experiments on the key monitoring areas of the tenon tooth surface of the tenon under different process parameter combinations. After each ultrasonic vibration rolling strengthening processing experiment on the tenon of the gas turbine blade is completed using a process parameter combination, replace it with a new tenon of the gas turbine blade (with the same structure and dimensions), and conduct ultrasonic vibration rolling strengthening processing experiments on the new tenon of the gas turbine blade using another process parameter combination. When conducting ultrasonic vibration rolling strengthening processing experiments under each process parameter combination, achieve microsecond-level clock synchronization through the industrial Ethernet, and use the adaptive sensor network detection system to dynamically collect various feature parameters of the key monitoring areas of the tenon tooth surface in real time, so as to obtain a data set composed of spatial coordinates (x, y, z), ultrasonic vibration rolling time t, process parameter combinations, and feature parameter combinations. Use the data set to establish a mapping database of tenon feature parameters - process parameters. Among them, the value range of the amplitude A is 5μm to 20μm, and the process parameter combination consists of ultrasonic frequency f, static load p, and feed speed v. In this embodiment, the value range of the ultrasonic frequency f is 20kHz to 40kHz, the value range of the static load p is 0.5MPa to 3MPa, and the value range of the feed speed is 50mm / min to 300mm / min; the feature parameter combination consists of three feature parameters: temperature field T, stress field σ, and displacement field u; the process of conducting ultrasonic vibration rolling strengthening processing experiments under a process parameter combination and using the adaptive sensor network detection system to dynamically collect various feature parameters of the key monitoring areas of the tenon tooth surface in real time is as follows:

[0054] ① The multi-degree-of-freedom robotic arm drives the laser displacement sensor to scan the tenon tooth surface of the gas turbine blade on the workpiece fixture clamped on the test fixture base at a speed of 20mm / s, and based on the ResNet-18 deep learning model, identify the positions of each aerodynamic sensitive area, each structural dangerous area, and each thermal load area on the tenon tooth surface; among them, a laser displacement sensor with an accuracy of ±0.01mm is used to perform high-density laser scanning of 0.1mm×0.1mm on the tenon tooth surface. After the ResNet-18 deep learning model is trained, when inputting displacement data, it outputs the positions of the aerodynamic sensitive areas, each structural dangerous area, and each thermal load area.

[0055] ②Use the corresponding process parameter combination to perform ultrasonic vibration rolling strengthening on the key monitoring area of the gas turbine blade tenon, and at the same time start the enhanced acquisition mode: the multi-degree-of-freedom robotic arm drives the infrared thermal imager to move along the preset path, so that the infrared thermal imager can detect the temperature distribution of the key monitoring area in real time, and adjust the relative position and pose of the infrared thermal imager probe and the tenon surface, so that the infrared thermal imager probe always maintains a perpendicular incident measurement attitude to the tenon tooth surface of the tenon, reducing the measurement error caused by the angle deviation, ensuring the accuracy of the temperature data, and at the same time the 8-channel acoustic emission sensor array detects the stress distribution and displacement distribution of the tenon tooth surface in real time; among them, an infrared thermal imager with a temperature measurement range of 400-1200 °C is used to perform 100 Hz high-frequency thermal imaging on the tenon surface.

[0056] S5. Establish a physical control equation system for each characteristic parameter in the three-dimensional rectangular coordinate system:

[0057] ①Based on the temperature distribution, establish the heat conduction equation:

[0058]

[0059] In the formula, is the temperature field, ρ is the density of the gas turbine blade tenon material, c p is the specific heat capacity of the material, k is the thermal conductivity of the material, is the vector differential operator, is the divergence of the heat flux density (scalar), used to describe the rate of heat energy diffusion, Q ultrasonic is the ultrasonic energy input term, and Q ultrasonic = 2πfηA 2 , η is the transducer efficiency (determined by calibration experiment), Q friction is the frictional heat generation term, and Q friction = μpv, μ is the friction coefficient (calibrated by friction and wear experiment).

[0060] ②Based on the stress distribution, establish the stress balance equation:

[0061]

[0062] In the formula, σ is the stress field (obeying the Johnson-Cook constitutive model, and the parameters are obtained through the split Hopkinson bar experiment), is the divergence of the stress field (three-dimensional vector), indicating the net stress per unit volume, F ultrasonic is the dynamic load caused by ultrasonic vibration, and F ultrasonic = 2π 2 f 2 Aρ.

[0063] ③Based on the displacement distribution, establish the vibration wave equation:

[0064]

[0065] Wherein, c is the sound speed in the tenon material of the gas turbine blade (measured by an ultrasonic flaw detector), is the Laplacian operator (scalar) of the displacement field u, which is used to describe the spatial diffusion characteristics of vibration, S(t) is the excitation source term of the ultrasonic transducer, and β is the attenuation coefficient (determined by the material damping characteristics), is the initial phase angle (which can be calibrated by a laser vibrometer) and is usually set to 0.

[0066] S6. Construct a PINN network model:

[0067] ① Input layer: the machining point coordinates (x, y, z), the process parameter combination, and the preset ultrasonic vibration rolling time t;

[0068] ② Main network: an 8-layer fully connected network (with 256 nodes in each layer and the activation function being swish), and the output prediction value is the predicted value of the characteristic parameter combination (the characteristic parameter combination includes the predicted temperature field predicted stress field predicted displacement field );

[0069] ③ Calculation of differential terms: Calculate through automatic differentiation (gradient of the predicted temperature field), (divergence of the predicted stress field), and (Laplacian term of the predicted displacement field);

[0070] ④ Physical residual constraint: thermal residual R T , stress residual R σ , and vibration residual R u are respectively

[0071]

[0072] ⑤ Hybrid loss function:

[0073] L = w data (L T + L σ + L u ) + w phys (‖R T ‖2 + ‖R σ ‖2 + ‖R u ‖2)

[0074] Wherein, N is the number of samples in the predicted temperature field or the actual temperature field T (the number of samples in the predicted temperature field is equal to the number of samples in the actual temperature field T), is the predicted temperature field and is the predicted temperature value of the i-th sample in the while is the actual temperature value of the i-th sample in the actual temperature field T; M is the predicted stress field or the number of samples in the actual stress field σ (the number of samples in the predicted stress field is equal to the number of samples in the actual stress field σ), is the predicted stress field and is the predicted stress value of the i-th sample in it, while is the actual stress value of the i-th sample in the actual stress field σ; K is the predicted displacement field or the number of samples in the actual displacement field u (the number of samples in the predicted displacement field is equal to the number of samples in the actual displacement field u), is the predicted displacement field and is the predicted displacement value of the i-th sample in it, while is the actual displacement value of the i-th sample in the actual displacement field u; w data and w phys are both weight coefficients, and w data / w phys = 0.7 / 0.3; where the actual temperature field T, stress field σ, and displacement field u all come from the tenon feature parameter - process parameter mapping database.

[0075] S7. Physical-guided data generation and generalization:

[0076] Known domain enhancement: For the key monitoring area of the tenon tooth surface, combined with the tenon feature parameter - process parameter mapping database, create a 0.1 mm spherical neighborhood with the coordinates of each grid point (i.e., grid microelement) in the key monitoring area of the tenon tooth surface as the center of the sphere, and generate new grid points with a data volume more than 100 times through Latin hypercube sampling; input the coordinates of each new grid point, the combination of process parameters at the corresponding center grid point, and the preset ultrasonic vibration rolling time into the trained PINN network model, and retain those that meet |R T | < 1 °C / s, |R σ | < 5 MPa, and |R u | < 10 km / s 2 of the predicted values, the corresponding new grid point coordinates, and the combination of process parameters at the corresponding center grid point.

[0077] Unknown domain extrapolation: For other areas of the tenon tooth surface except the key monitoring area, construct a geometric similarity index Γ:

[0078]

[0079] In the formula, α, β, and γ are all weight coefficients. In this embodiment, α = 0.4, β = 0.4, γ = 0.2, R, K t and Q are the radius of curvature, tangential curvature, and area of the grid points selected from other areas of the tenon tooth surface except the key monitoring area, respectively. R0, K t0 and Q0 are the radius of curvature, tangential curvature, and area of the grid points selected within the key monitoring area of the tenon tooth surface, which is used to quantify the geometric similarity degree of the radius of curvature of the two selected grid points; which is used to measure the geometric similarity in terms of curvature of the two selected grid points, which is used to reflect the geometric similarity of the areas of the two selected grid points.

[0080] For the process parameter combinations of each grid point in the area of the tenon tooth surface except the key monitoring area, the process parameter combination of the grid point in the key monitoring area of the tenon tooth surface corresponding to Γ value > 85% is taken as the control group. The process parameter combination (f0, p0, v0) of the control group is scaled proportionally to form the process parameter combination (f new , p new , v new ) of the grid points in the area of the tenon tooth surface except the key monitoring area. The grid point coordinates in the area of the tenon tooth surface except the key monitoring area, the corresponding process parameter combination, and the preset ultrasonic vibration rolling time are input into the PINN network model. The predicted values that meet |R T | < 3 °C / s, |R σ | < 10 MPa, and |R u | < 30 km / s 2 and the corresponding grid point coordinates and process parameter combinations are retained. Among them, the process parameters of each grid point in the area of the tenon tooth surface except the key monitoring area are

[0081]

[0082]

[0083] S8, Online Recommendation and Adaptive Adjustment: Based on the output of the PINN network model (output feature parameter combination, i.e., physical field) at different positions of the tenon tooth surface, the NSGA-II algorithm is called to generate a Pareto solution set (200 groups of candidate process parameter combinations are set in the Pareto solution set of this embodiment), and 5 optimal process parameter combinations at all spatial coordinate positions of the tenon tooth surface except the spatial coordinate positions in the dataset are screened based on the TOPSIS decision-making method. During the actual machining process, the selection of the process parameter combination at each spatial coordinate position of the tenon tooth surface can be manually selected or the first optimal process parameter combination can be selected by default by the computer.

[0084] Furthermore, to verify the improvement of the combined effect of the process parameters obtained by using the present invention on the fretting fatigue performance of the gas turbine blade tenons, an effect evaluation is carried out. The evaluation methods are as follows:

[0085] A. Virtual verification: Construct a machining-fatigue coupling simulation environment in finite element software, simulate the ultrasonic vibration rolling strengthening machining of the three-dimensional model of the gas turbine blade, and compare the life distributions under the combined process parameters obtained by using the present invention and the traditional empirical process parameters;

[0086] B. Physical verification: Use a high-frequency fatigue testing machine (such as Instron 8802) to sequentially load the tenons of two gas turbine blades machined with the combined process parameters obtained by using the present invention and the traditional empirical process parameters, monitor and compare the crack initiation cycles of the tenons of these two gas turbine blades, and conduct SEM microscopic morphology analysis and comparison.

[0087] In this embodiment, physical verification is adopted. Through high-frequency fatigue tests and microscopic morphology analysis, it is verified that, compared with the crack initiation cycle of the gas turbine blade tenon under the traditional empirical process parameter combination, the crack initiation cycle of the gas turbine blade tenon under the process parameter combination of the present invention is significantly delayed, indicating that the gas turbine blade tenon under the process parameter combination of the present invention has enhanced resistance to damage under alternating loads, the fatigue failure process is delayed, and the fretting fatigue life is improved.

Claims

1. A process parameter recommendation method for improving the fretting fatigue performance of gas turbine blade tenons by AI, characterized in that: The details are as follows: S1. Establish a three-dimensional model of the gas turbine blade; S2. Determine the key monitoring areas of the tenon and mortise surfaces of the gas turbine blade through finite element analysis; S3. Build an adaptive sensor network detection system: Arrange a mobile detection unit composed of a laser displacement sensor and an infrared thermal imager carried by a multi-degree-of-freedom robotic arm on the test fixture base, and arrange a fixed detection unit composed of an acoustic emission sensor array on the test fixture base; S4. Conduct ultrasonic vibration rolling machining experiments, and use the adaptive sensor network detection system to dynamically collect various characteristic parameters of the key monitoring areas of the tenon and mortise surfaces in real time, obtain a data set composed of spatial coordinates (x, y, z), ultrasonic vibration rolling time t, process parameter combinations, and characteristic parameter combinations, and establish a mapping database of tenon characteristic parameters - process parameters using the data set; among them, the process parameter combination consists of ultrasonic frequency f, static load p, and feed speed v, and the characteristic parameter combination consists of temperature field T, stress field σ, and displacement field u. The temperature field T is measured by the infrared thermal imager, and the stress field σ and displacement field u are measured by the acoustic emission sensor array; S5. Establish a physical control equation system for each characteristic parameter in a three-dimensional rectangular coordinate system: Establish a heat conduction equation based on the temperature distribution, establish a stress balance equation based on the stress distribution, and establish a vibration wave equation based on the displacement distribution; S6. Construct a PINN network architecture with spatial coordinates (x, y, z), process parameter combinations, and ultrasonic vibration rolling time t as inputs and characteristic parameter combinations as outputs; S7. Physical guidance data generation and generalization: For the key monitoring area of the tenon-mortise surface, combined with the tenon feature parameter - process parameter mapping database, create a spherical neighborhood with the coordinates of each grid point in the key monitoring area of the tenon-mortise surface as the center of the sphere, and generate new grid points through Latin hypercube sampling; input the coordinates of each new grid point, the combination of process parameters at the corresponding center grid point, and the preset ultrasonic vibration rolling time into the trained PINN network model, and retain those that meet |R T | < 1 °C / s, |R σ | < 5 MPa and |R u | < 10 km / s 2 of the predicted values, the corresponding new grid point coordinates, and the combination of process parameters at the corresponding center grid point; S8. Online recommendation and adaptive adjustment: According to the characteristic parameter combinations retained in step S7, call the NSGA-II algorithm to generate a Pareto solution set, which consists of multiple candidate process parameter combinations, and screen 5 optimal process parameter combinations at different spatial coordinate positions on the key monitoring areas of the tenon and mortise surfaces except for the spatial coordinate positions in the data set based on the TOPSIS decision-making method.

2. The process parameter recommendation method for improving the fretting fatigue performance of the gas turbine blade tenon according to claim 1, wherein: In step S1, the three-dimensional model of the gas turbine blade is established through three-dimensional modeling software, or the surface point cloud data of the gas turbine blade is obtained by a three-dimensional scanner, and the surface point cloud data of the gas turbine blade is sequentially denoised and surface reconstructed by Geomagic Design X software to establish the three-dimensional model of the gas turbine blade.

3. The process parameter recommendation method for improving the fretting fatigue performance of gas turbine blade tenons by AI according to claim 1, characterized in that: The specific process of step S2 is as follows: Import the three-dimensional model of the gas turbine blade into the finite element software, perform mesh division on the three-dimensional model of the gas turbine blade, calculate the curvature information of each grid point on the tenon and mortise surfaces of the gas turbine blade using the curvature analysis tool, apply aerodynamic load, centrifugal load, and temperature boundary conditions to the three-dimensional model of the gas turbine blade to simulate the working conditions of the gas turbine blade, solve the stress field and temperature field of each grid point on the tenon and mortise surfaces through thermo-mechanical coupled finite element analysis, divide the areas with a curvature change rate greater than 15% / mm among the grid points on the tenon and mortise surfaces into aerodynamic sensitive areas, divide the areas with a strain gradient greater than 200 με / mm into structural dangerous areas, and divide the areas with a temperature gradient greater than 80 °C / cm into thermal load areas, so as to obtain the key monitoring areas on the tenon and mortise surfaces composed of each aerodynamic sensitive area, each structural dangerous area, and each thermal load area.

4. The process parameter recommendation method for improving the fretting fatigue performance of gas turbine blade tenons by AI according to claim 1, wherein: The specific process of step S4 is as follows: Set the ranges of various process parameters, select multiple process parameter nodes within each range for orthogonal design to obtain multiple process parameter combinations. Then, set the amplitude A and conduct ultrasonic vibration rolling strengthening experiments on the key monitoring areas of the tenon and mortise surfaces under different process parameter combinations. After each ultrasonic vibration rolling strengthening experiment on the gas turbine blade tenon using a process parameter combination is completed, replace the new gas turbine blade tenon and conduct an ultrasonic vibration rolling strengthening experiment on the new gas turbine blade tenon using another process parameter combination. When conducting an ultrasonic vibration rolling strengthening experiment under each process parameter combination, use the adaptive sensor network detection system to dynamically collect various characteristic parameters of the key monitoring area of the tenon and mortise surface in real time, and then obtain a data set composed of spatial coordinates (x, y, z), ultrasonic vibration rolling time t, process parameter combinations, and characteristic parameter combinations. Use the data set to establish a mapping database of tenon characteristic parameters - process parameters.

5. The process parameter recommendation method for improving the fretting fatigue performance of a gas turbine blade tenon according to claim 4, characterized in that: The process of conducting an ultrasonic vibration rolling strengthening experiment under a process parameter combination and using the adaptive sensor network detection system to dynamically collect various characteristic parameters of the key monitoring area of the tenon and mortise surface in real time is as follows: ① The multi-degree-of-freedom robotic arm drives the laser displacement sensor to scan the tenon and mortise surface of the gas turbine blade tenon clamped on the workpiece fixture of the test fixture base at a preset speed, and based on the ResNet-18 deep learning model, identify the positions of each aerodynamic sensitive area, each structural dangerous area, and each thermal load area on the tenon and mortise surface. ② Use the corresponding process parameter combination to conduct ultrasonic vibration rolling strengthening on the key monitoring area of the gas turbine blade tenon, and at the same time start the enhanced acquisition mode: the multi-degree-of-freedom robotic arm drives the infrared thermal imager to move along a preset path, so that the infrared thermal imager can detect the temperature distribution of the key monitoring area in real time, and adjust the relative pose of the infrared thermal imager probe and the tenon surface to keep the infrared thermal imager probe in a perpendicular incident measurement posture for the tenon and mortise surface. At the same time, the acoustic emission sensor array detects the stress distribution and displacement distribution of the tenon and mortise surface in real time.

6. The process parameter recommendation method for improving the fretting fatigue performance of a gas turbine blade tenon by AI according to claim 1, wherein: The heat conduction equation established based on the temperature distribution is where ρ is the density of the material of the turbine blade tenon, c p is the specific heat capacity of the material, k is the thermal conductivity of the material, is the vector differential operator, is the divergence of the heat flux density, Q ultrasonic is the ultrasonic energy input term, and Q ultrasonic = 2πfηA 2 , η is the transducer efficiency, Q friction is the frictional heat generation term, and Q friction = μpv, μ is the friction coefficient; The stress balance equation established based on the stress distribution is In the formula, is the divergence of the stress field, representing the net stress per unit volume, and F ultrasonic is the dynamic load caused by ultrasonic vibration, and F ultrasonic = 2π 2 f 2 Aρ; The vibration wave equation established based on the displacement distribution is where c is the sound speed in the material of the gas turbine blade tenon, is the Laplacian operator of the displacement field, S(t) is the excitation source term of the ultrasonic transducer, and β is the attenuation coefficient, is the initial phase angle.

7. The process parameter recommendation method for improving the fretting fatigue performance of gas turbine blade tenons by AI according to claim 6, characterized in that: The PINN network architecture in step S6 is as follows: ① Input layer: spatial coordinates (x, y, z), process parameter combinations, and ultrasonic vibration rolling time t; ② Main network: an 8-layer fully connected network, and the output prediction value is the predicted value of the characteristic parameter combination; ③Differential term calculation: Calculate the gradient of the predicted temperature field through automatic differentiation Predict the divergence of the stress field and the Laplacian term of the predicted displacement field ④ Physical residual constraint: Calculate the thermal residual R T , stress residual R σ and vibration residual R u ; ⑤ Hybrid loss function: L = w data (L T + L σ + L u ) + w phys (‖R T ‖2 + ‖R σ ‖2 + ‖R u ‖2) Wherein, N is the number of samples in the predicted temperature field or the actual temperature field T, is the predicted temperature value of the i-th sample in the predicted temperature field ; and is the actual temperature value of the i-th sample in the actual temperature field T; M is the number of samples in the predicted stress field or the actual stress field σ, is the predicted stress value of the i-th sample in the predicted stress field ; and is the actual stress value of the i-th sample in the actual stress field σ; K is the number of samples in the predicted displacement field or the actual displacement field u, is the predicted displacement value of the i-th sample in the predicted displacement field ; and is the actual displacement value of the i-th sample in the actual displacement field u; w data and w phys are both weight coefficients; wherein, the actual temperature field T, stress field σ and displacement field u are all from the tenon feature parameter - process parameter mapping database.

8. The process parameter recommendation method for improving the fretting fatigue performance of gas turbine blade tenons by AI according to claim 7, characterized in that: The thermal residual R in the step S6 T , the stress residual R σ and the vibration residual R u are calculated by the formula 9. The process parameter recommendation method for improving the fretting fatigue performance of the blade tenon of a gas turbine by AI according to claim 1, wherein: In step S7, in addition to the known domain strengthening for the key monitoring area of the tenon and mortise surface, unknown domain extrapolation is also carried out for other areas of the tenon and mortise surface except the key monitoring area. The unknown domain extrapolation process is: construct a geometric similarity index Γ: Wherein, α, β, and γ are all weight coefficients, and R, K t and Q are the radius of curvature, tangential curvature, and area of the grid points selected from other areas of the tenon tooth surface of the tenon except for the key monitoring area, respectively, and R0, K t0 and Q0 are the radius of curvature, tangential curvature, and area of the grid points selected within the key monitoring area of the tenon tooth surface of the tenon; For the process parameter combinations of each grid point in the areas other than the key monitoring area of the tenon head and tenon tooth surface, the process parameter combinations of the grid points in the key monitoring area of the tenon head and tenon tooth surface corresponding to Γ value > 85% are taken as the control group. The process parameter combinations (f0, p0, v0) of the control group are scaled proportionally to form the process parameter combinations (f new , p new , v new ) of the grid points in the areas other than the key monitoring area of the tenon head and tenon tooth surface. The grid point coordinates of the areas other than the key monitoring area of the tenon head and tenon tooth surface, the corresponding process parameter combinations, and the preset ultrasonic vibration rolling time are input into the PINN network model, and the predicted values and the corresponding grid point coordinates and process parameter combinations that meet |R T | < 3 °C / s, |R σ | < 10 MPa and |R u | < 30 km / s 2 are retained. Among them, the process parameters of each grid point in the areas other than the key monitoring area of the tenon head and tenon tooth surface are 10. The process parameter recommendation method for improving the fretting fatigue performance of the gas turbine blade tenon according to claim 9, wherein: According to the combination of characteristic parameters retained in the unknown domain extrapolation process, the NSGA-II algorithm is called to generate a Pareto solution set. The Pareto solution set consists of multiple candidate process parameter combinations, and based on the TOPSIS decision-making method, 5 optimal process parameter combinations at different spatial coordinate positions in other regions of the tenon and mortise surface except the key monitoring area are screened, and then 5 optimal process parameter combinations at all spatial coordinate positions of the tenon and mortise surface except the spatial coordinate positions in the dataset are obtained.

Images