A design method for the contact profile of bolt-connected combined piston skirt

Abstract

The present invention discloses a method for designing a special-shaped contact surface of a bolt-connected combined piston skirt. The method first establishes a finite element simulation model of the piston and obtains the radial and circumferential surface deformations of the skirt contact; then the surface deformations are fitted to obtain initial radial and circumferential profiles, respectively; then the optimized target parameters are selected and sampled to obtain profile samples; then the finite element simulation model corresponding to the profile samples is simulated to train a machine learning proxy model; after verification, a trained machine learning proxy model with accuracy that meets the requirements is obtained; then an iterative solution is performed on this model to obtain the optimal target parameters; finally, the optimal profile is obtained based on the optimal target parameters, and a three-dimensional curved surface formed by linear superposition is used as the optimal profile. The present invention innovatively applies artificial intelligence algorithms to the design of special-shaped contact surfaces of bolt-connected combined piston skirts, with the help of a data-driven method of machine learning.

Description

Technical Field

[0001] The invention relates to the technical field of pistons, in particular to a method for designing a bolt-connected combined piston skirt top contact special-shaped surface. Background Art

[0002] As engine enhancement indicators continue to improve, engine power density and burst pressure are also increasing, and pistons are facing higher thermal and mechanical loads. Currently, the contact methods for the skirt top of piston structures on the market include welding, bolting, and one-piece casting of the skirt top. Due to process limitations and considerations of welding and casting defects, many styles of pistons still use the skirt top bolt connection method. However, with the high-load performance requirements of the engine, after the bolts are tightened, large deformation and uneven contact are common near the tightening part. On this basis, under the action of burst pressure during actual operation, the inconsistent radial deformation inside and outside will further lead to more uneven force on the skirt top contact surface.

[0003] At present, the main design scheme for the skirt top contact surface is a linear slope contact surface structure based on experience. Although the contact characteristics of the skirt top can be improved to a certain extent, since different contact pressure distributions of pistons with different structures are not taken into account, the empirical design method will inevitably lead to failure of the skirt top contact structure. For example, the document with application number 201911321834.1 discloses a method and device for determining the pin hole profile. According to the radial deformation of each preset sampling point of the pin hole profile, the pin hole profile is optimized to obtain the target pin hole profile. According to the target pin hole profile, the pin hole structure of the piston structure model is adjusted to obtain the target piston structure model, and the target piston structure model is subjected to thermo-mechanical coupling calculation to obtain the distribution results of physical quantities such as contact pressure and stress of the pin hole structure. When the physical quantity distribution results meet the preset distribution conditions, the final target pin hole profile is determined. The document with application number 202311383310.1 discloses a piston pin hole profile design optimization method, device, electronic device and storage medium, wherein the design optimization method includes: obtaining the cross-sectional radial deformation of the piston pin under preset working conditions; based on the cross-sectional radial deformation, fitting the piston pin surface deformation curve, and using the piston pin surface deformation curve as the piston pin hole axial profile; constructing a piston pin hole simulation analysis model, and the piston pin hole profile in the piston pin hole simulation analysis model is obtained based on the piston pin hole axial profile and the preset ellipticity mapping of the piston pin hole section; performing simulation calculations to obtain the physical quantity distribution results of the piston pin hole; and judging whether it meets the preset distribution requirements.

[0004] Considering the design cost and design effect, the above method inevitably has the following technical defects:

[0005] (1) Long design cycle: The design methods in the above literature use finite element modeling and simulation or extract the corresponding deformation through experiments in the early stage. Considering factors such as modeling and processing, a single round of modeling optimization takes several days. In addition, the correlation between the preliminary model lines and the actual engine load is not strong. Therefore, the optimization cannot be completed in the first round of analysis, and multiple rounds of optimization are required. Considering the trial and error time cost of empirical design, the time will be longer. The current fast-paced design of automotive engines makes it often difficult to implement this method. In addition, the preliminary line range of this scheme is large and there is no precise correlation with the engine load. The preliminary design line is quite different from the actual one, resulting in more optimization times and further lengthening the design cycle.

[0006] (2) High test cost: The design method in the above literature collects the deformation of the contact part through testing, which involves component-level testing of the piston, and extracts the deformation value based on the component-level test, and pre-processes the data value. In this process, it involves not only the design of test tooling, the construction of the hydraulic test system, the formulation of the test plan and the purchase of corresponding supporting facilities, etc., but even in the test of the main engine factory with hydraulic test conditions, this process for one of the pistons will cost hundreds of thousands. This is very costly for the fast-paced design requirements of the engine, so there is an urgent need for a low-cost and accurate design method.

[0007] (3) It is difficult to guarantee the design effect: In the design methods of the above literature, the contact deformation between the piston pin hole and the piston pin is extracted by experimental means, which often leads to errors in the radial deformation data of the pin hole at the sampling point. In addition, the thermo-mechanical coupling model does not propose effective methods and measures for the accuracy of the contact modeling between the pin hole and the piston pin. The pin hole profile first proposed by this method lacks smoothness. Due to the problem of the accuracy of the contact modeling between the pin hole and the piston pin, the physical quantity results of the pin hole structure calculated subsequently are also prone to abnormalities, which brings difficulties to the optimization evaluation. In addition, the radial deformation of the pin hole calculated by thermo-mechanical coupling is limited, and the design cannot be completed in one go, resulting in more subsequent optimization rounds and a longer cycle. It is difficult to adapt to the current fast-paced design and is difficult to apply in practice. Summary of the Invention

[0008] In view of the deficiencies in the prior art, the technical problem to be solved by the present invention is to provide a method for designing a bolt-connected combined piston skirt top contact profile.

[0009] The technical solution of the present invention to solve the above technical problem is to provide a method for designing a bolt-connected combined piston skirt top contact profile surface, characterized in that the method comprises the following steps:

[0010] Step 1: Establish a finite element simulation model of the piston;

[0011] Step 2: Under preset simulation conditions, perform a thermal-mechanical coupling simulation on the finite element simulation model of step 1 to obtain the radial and circumferential surface deformations of the piston skirt top contact;

[0012] Step 3: fitting the radial and circumferential surface deformations of the piston skirt top contact obtained in step 2 to obtain initial radial and circumferential profiles of the skirt top contact, respectively; wherein the initial circumferential profile of the skirt top contact is characterized by a fitting function that is most similar to the surface deformation state of the skirt top contact;

[0013] Step 4: In the radial direction of the skirt top contact, the positions and indentations of the key deformation points in the initial radial profile of the skirt top contact obtained in Step 3 are selected as the target parameters for optimization; in the circumferential direction of the skirt top contact, the eigenvalues ​​of the fitting function of the initial circumferential profile obtained in Step 3 are selected as the target parameters for optimization;

[0014] Step 5: Sampling the radial and circumferential target parameters of the skirt top contact according to the sampling sample interval to obtain n1 groups of profile line samples;

[0015] Step 6: Based on the n1 groups of profile samples from step 5, adjust the finite element simulation model from step 1 to obtain finite element simulation models of n1 groups of different profiles of the same scale; then perform finite element simulation on each finite element simulation model to obtain simulation results as output values ​​for training the machine learning proxy model, and use the corresponding profile samples as input values ​​for training the machine learning proxy model to obtain a trained machine learning proxy model; the simulation results include contact pressure and stress;

[0016] Step 7: Repeat steps 5 and 6 to extract n2 groups of profile samples, and then perform finite element simulation according to step 6. The obtained simulation results and their corresponding profile samples are used as the validation set of the trained machine learning proxy model obtained in step 6. The accuracy of the trained machine learning proxy model is verified by selecting the corresponding accuracy verification index. If the accuracy does not meet the requirements, adjust the sampling method of step 5, retrain the machine learning proxy model until the accuracy meets the requirements, and save the trained machine learning proxy model that meets the accuracy requirements to replace the finite element simulation process.

[0017] Step 8: Based on the machine learning agent model obtained in step 7, the contact pressure and stress of the skirt top contact are used as optimization indicators, the position and indentation of the key deformation points selected in the radial profile of the skirt top contact, and the eigenvalues ​​of the fitting function of the circumferential direction of the skirt top contact are used as optimization target parameters, and the optimization indicators are continuously iteratively solved through a multi-objective optimization algorithm to obtain the optimal target parameters;

[0018] Step 9: Refit the optimal target parameters obtained in step 8 to obtain the optimal radial and circumferential profiles of the skirt top contact, respectively. Then, the three-dimensional surface formed by linearly superimposing the optimal radial and circumferential profiles is used as the optimal profile at the contact surface of the piston top or piston skirt.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] (1) The present invention utilizes a data-driven machine learning approach to innovatively apply artificial intelligence algorithms to the design of the contact profile of bolted combined piston skirts. To address the blindness and inefficiency of previous designs for piston skirt contact profiles, the present invention uses a small sample simulation to train a high-precision proxy model in the early stages of design, and optimizes the skirt profile scheme using a multi-objective optimization algorithm. This design method can achieve the design of the skirt contact profile of any bolted combined piston, provided the contact material is known, and more efficiently obtain the optimal profile scheme.

[0021] (2) In the radial direction of the skirt top contact surface, the present invention preferably proposes a three-segment skirt top contact radial profile scheme, that is, the piston skirt top surface is composed of a curved segment-a straight segment-a curved segment. This will significantly improve the common stress concentration phenomenon at both ends of the contact surface of the piston skirt top.

[0022] (3) In the circumferential direction of the skirt top contact surface, the present invention preferably proposes to obtain the profile scheme of the skirt top circumferential scheme by using a function fitting method. The present invention preferably recommends using a trigonometric function to fit the circumferential profile line. This design concept can reduce the excessive mutual extrusion of the skirt top contact surface caused by the bolt preload, and can improve the service life and reliability of the piston.

[0023] (4) The profile design method proposed in the present invention does not involve experiments, thus avoiding the time cost and expense of experimental design, procurement, experimental testing, and data processing related processes. Instead, it uses machine learning algorithms to greatly improve design efficiency and design effects while ensuring design accuracy. It can improve design efficiency, shorten the design cycle while ensuring design effects, reduce design costs, and quickly obtain skirt contact profile design solutions based on different piston structures.

[0024] (5) The initial profile of the present invention is formed by fitting the deformation of the piston skirt top contact surface on the basis of the finite element simulation of the piston. The radial and circumferential surface deformation of the skirt top contact is obtained by loading the preset simulation working conditions on the piston assembly. The radial and circumferential surface deformation obtained by this method is closer to the indentation of the actual skirt top contact profile, thus getting rid of the problem that the current piston design for the skirt top contact profile is overly dependent on design experience and database, and the profile deformation is corrected by function fitting, which not only achieves the smoothness of the profile but also improves the design accuracy and is more convenient for profile processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is the overall flow chart of the present invention;

[0026] Figure 2 This is a schematic structural diagram of a bolt-connected assembled piston according to the present invention;

[0027] Figure 3 Schematic diagram of the radial three-segment profile of the piston skirt top contact profile of the present invention;

[0028] Figure 4 It is a schematic diagram of the circumferential fitting function profile of the piston skirt top contact surface of the present invention.

[0029] In the figure, there is a piston top 1, a piston skirt 2, a positioning pin hole 3, and a bolt hole 4. DETAILED DESCRIPTION

[0030] The specific embodiments of the present invention are given below. The specific embodiments are only used to further illustrate the present invention and do not limit the scope of protection of the present invention.

[0031] The present invention provides a method for designing a bolt-connected assembled piston skirt top contact profile (hereinafter referred to as the method), characterized in that the method comprises the following steps:

[0032] Step 1: Establish a finite element simulation model of the combined piston;

[0033] Preferably, in step 1, the finite element simulation model includes a piston body, a piston pin, a cylinder liner and a connecting rod, which more accurately simulates the actual working condition of the piston.

[0034] Preferably, in step 1, during the modeling process, the skirt top contact (i.e., the contact between the piston top 1 and the piston skirt 2) is established in a point-to-point manner. The subsequent contact surface is obtained by mapping, and the contact surface nodes are copied one-to-one. This can reduce the contact pressure and stress concentration and stress singular points caused by node mismatch during the simulation process, thereby ensuring the accuracy of the finite element simulation.

[0035] Step 2: Under preset simulation conditions, perform a thermo-mechanical coupling simulation (i.e., thermal stress and mechanical stress coupling simulation) on the finite element simulation model of step 1 to obtain radial and circumferential surface deformations of the piston skirt top contact;

[0036] Preferably, in step 2, on the load boundary, a preset simulation working condition is set to take into account the thermal load and mechanical load borne by the piston during actual operation; the thermal load is the thermal load generated under the steady-state temperature field of the piston, and the mechanical load is the multi-source mechanical load borne by the piston under the maximum explosion pressure, which is the most dangerous working condition of the piston;

[0037] On the constraint boundary, the preset simulation working condition fully constrains the piston and cylinder sleeve part, and constrains the connecting rod body direction and the vertical connecting rod swing direction of the piston connecting rod big end part.

[0038] Among them, for the steady-state temperature field of the piston, if there is no actual measured temperature of the piston test, the steady-state temperature field of the piston is set according to the maximum temperature of the piston temperature design limit; if there is an actual measured temperature of the piston test, the final steady-state temperature field is compared with the actual measured temperature of the test, so that the error value between the steady-state temperature field obtained by simulation and the actual measured temperature of the test is controlled within 10%.

[0039] Among them, the multi-source mechanical loads include bolt preload, gas pressure generated under maximum explosion pressure, inertia force and lateral force given by the cylinder liner.

[0040] Step 3: Fitting the radial and circumferential surface deformations of the piston skirt top contact obtained in step 2 to achieve the purpose of smoothing the profile, thereby improving the design accuracy and obtaining the radial and circumferential initial profiles of the skirt top contact, respectively. This solves the problem that the current design of the initial profile of the piston skirt top contact is overly dependent on design experience and databases. The initial profile of the circumferential contact of the skirt top is characterized by a fitting function that is most similar to the surface deformation state of the circumferential contact of the skirt top.

[0041] Preferably, in step 3, the initial radial profile of the skirt top contact is a three-segment profile. Figure 3 As shown, a structural form of a straight line segment in the middle and curved segments on both sides (i.e., a profile formed by connecting the first curve, the straight line segment, and the second curve end to end) is adopted, which does not affect the load-bearing effect while reducing the stress concentration phenomenon at the radial contact parts at both ends.

[0042] Preferably, in step 3, in this embodiment, as Figure 4 As shown, point G is the bolt hole 4, point H is the locating pin hole 3, point I is the bolt hole 4, and point J is the locating pin hole 3. The bolt hole 4 and the locating pin hole 3 are spaced apart and 90° apart in the circumferential direction. The circumferential surface deformation of the piston skirt top contact is fitted to obtain a fitting function, wherein the contact parts around the locating pin hole 3 and the bolt hole 4 adopt trigonometric functions or high-order power functions.

[0043] Step 4: In the radial direction of the skirt top contact, the positions and indentations of the key deformation points in the initial radial profile of the skirt top contact obtained in Step 3 are selected as the target parameters for optimization; in the circumferential direction of the skirt top contact, the eigenvalues ​​of the fitting function of the initial circumferential profile obtained in Step 3 are selected as the target parameters for optimization;

[0044] Preferably, in step 4, the key deformation points of the radial initial profile that the skirt top contacts are the endpoints on both sides of the profile and the middle inflection point.

[0045] Preferably, in step 4, in this embodiment, in the radial direction, taking the three-segment profile as an example, Figure 3 As shown, select points B, C, D and E as key deformation points, and indent point B by H. B 、E point indentation H E , radial position R of point C C and the radial position R of point D D as the target parameter for optimization.

[0046] Preferably, in step 4, in this embodiment, in the circumferential direction, as Figure 4 As shown, considering the magnitude of the deformation around the positioning pin hole 3 and the bolt hole 4, the peaks and wavelengths of the trigonometric functions at the four positions G, H, I, and J are respectively used as target parameters for optimization.

[0047] Step 5: Sampling the radial and circumferential target parameters of the skirt top contact according to the sampling sample interval to obtain n1 groups of profile line samples;

[0048] Preferably, in step 5, the sampling interval is from displacement × 50% to displacement × 150%.

[0049] Preferably, in step 5, Latin hypercube sampling is used for sampling. The Latin hypercube sampling method enables every small interval within the interval to be sampled, and is the preferred sampling method for training machine learning agent models. Due to the uniformity of sampling, the problem of insufficient fitting of the trained model can be reduced.

[0050] Preferably, in step 5, in order to balance computation time and model accuracy, n1=30.

[0051] Step 6: Based on the n1 groups of profile samples from step 5, adjust the finite element simulation model from step 1 to obtain finite element simulation models of n1 groups of different profiles of the same scale; then perform finite element simulation on each finite element simulation model to obtain simulation results as output values ​​for training the machine learning proxy model, and use the corresponding profile samples as input values ​​for training the machine learning proxy model to obtain a trained machine learning proxy model; the simulation results include contact pressure and stress;

[0052] Preferably, in step 6, the adjustment method includes a grid mapping method, a method for adjusting the contact gap, and a method for adjusting the coordinates of the grid nodes on the surface of the contact part in the finite element simulation model.

[0053] Among them, the mesh mapping method is to establish a corresponding reference fitting surface during the mesh pre-processing process, and project the nodes on the contact surface into the established plane to achieve the adjustment of the finite element simulation model.

[0054] The method of adjusting the contact gap is to set different initial contact gaps for different nodes, thereby adjusting the finite element simulation model.

[0055] Among them, the method of adjusting the surface grid node coordinates of the contact part in the finite element simulation model is to recalculate the position coordinates of the nodes on the surface where the contact surface is located, and redefine the node positions in the finite element simulation model, thereby realizing the adjustment of the finite element simulation model.

[0056] Preferably, in step 6, the contact pressure is the maximum contact pressure of the piston skirt top contact surface, and the stress includes the maximum Mises stress of the piston skirt top contact surface, the maximum Mises stress at the locating pin hole 3, and the maximum Mises stress at the bolt hole 4.

[0057] Preferably, in step 6, the machine learning agent model is a response surface model, a Kriging model or a neural network model. In view of the possible nonlinearity between the profile parameters and the corresponding simulation results, the Kriging model and the neural network model are preferred, which have better fitting effects on nonlinear sample data and can obtain more accurate calculation results.

[0058] Step 7: Repeat steps 5 and 6 to extract n2 groups of profile samples, and then perform finite element simulation according to step 6. The obtained simulation results and their corresponding profile samples are used as the validation set of the trained machine learning proxy model obtained in step 6. The corresponding accuracy verification index is selected to verify the accuracy of the trained machine learning proxy model. If the accuracy does not meet the requirements (that is, the error exceeds the preset tolerance range), the sampling method of step 5 is adjusted, and the machine learning proxy model is retrained until the accuracy meets the requirements. The trained machine learning proxy model with the required accuracy is saved to replace the finite element simulation process.

[0059] Preferably, in step 7, n2 = 10 to ensure the representativeness of the validation set while minimizing computational cost. The primary function of the validation set is to test the performance of the machine learning agent model on unseen data to assess its generalization ability and accuracy. The validation set should be independent of the training set to avoid overfitting the model to specific data.

[0060] Preferably, in step 7, the accuracy verification index adopts one of the determination coefficient, mean square error, root mean square error and mean absolute error, preferably the determination coefficient R 2 As an indicator of the accuracy of the predicted value, R 2 It reflects the proportion of all changes in the dependent variable explained by the independent variable through the regression relationship, that is, using the average value of the data as a benchmark, observing the degree of deviation between the prediction error and the average reference error.

[0061] Preferably, in step 7, adjusting the sampling method of step 5 includes increasing the number of profile samples and / or reducing the sampling interval. Preferably, the number of profile samples is increased to 1.3 to 2 times the number of profile samples in step 5. In this embodiment, the number is increased from 30 groups to 50 groups, thereby improving the diversity and representativeness of the training data. The sampling interval is reduced to 0.5 to 0.8 times the sampling interval in step 5. In this embodiment, the number is reduced from displacement × 50% to displacement × 150% to displacement × 70% to displacement × 130%, thereby capturing more possible variations.

[0062] Step 8: Based on the machine learning agent model obtained in step 7, the contact pressure and stress of the skirt top contact are used as optimization indicators, the position and indentation of the key deformation points selected in the radial profile of the skirt top contact, and the eigenvalues ​​of the fitting function of the skirt top contact circumference are used as optimization target parameters, and the optimization indicators are continuously iterated and solved through a multi-objective optimization algorithm until the set constraint value or the maximum number of iterations is reached, thereby obtaining the optimal target parameters;

[0063] Preferably, in step 8, the multi-objective optimization algorithm adopts an improved particle swarm algorithm or a genetic algorithm.

[0064] The improved particle swarm optimization algorithm (PSO) is an optimization algorithm based on swarm intelligence. It simulates the foraging behavior of bird flocks and seeks the optimal solution through information sharing among individuals in the flock. The improved PSO combines local and global optima, designing a hybrid update strategy. This strategy enhances local search capabilities while maintaining global search capabilities, thus reducing the risk of being stuck in local optima. The improved PSO is simple to implement, requires few parameters, and is suitable for continuous problems with rapid convergence. However, it is prone to being stuck in local optima and has limited search capabilities for complex multimodal problems.

[0065] A genetic algorithm is an optimization algorithm based on natural selection and heredity, simulating the process of biological evolution to find the optimal solution. Its elite retention strategy involves retaining the best individuals in each generation to the next, preventing the optimal solution from being lost during crossover and mutation. This ensures the persistence of the optimal solution, improving the algorithm's stability and convergence speed. In contrast, genetic algorithms are applicable to both discrete and continuous problems, have strong global search capabilities, and maintain population diversity through crossover and mutation. However, they have more parameters, are more complex to adjust, and converge more slowly than particle swarm optimization algorithms.

[0066] Preferably, in step 8, the contact pressure and stress of the skirt top contact are specifically: the optimization indicators are the maximum contact pressure of the piston skirt top contact surface, the maximum Mises stress of the piston skirt top contact surface, the maximum Mises stress at the locating pin hole 3 and the maximum Mises stress at the bolt hole 4.

[0067] Step 9: Refit the optimal target parameters obtained in step 8 to obtain the optimal radial and circumferential profiles of the skirt top contact, respectively. Then, linearly superimpose the optimal radial and circumferential profiles to form a three-dimensional curved surface as the optimal profile at the contact surface of the piston top 1 or the piston skirt 2 (preferably the piston skirt 2).

[0068] Preferably, the method also includes: step 10, result verification: the optimal surface obtained in step 9 is adjusted by the adjustment method of step 6, and loaded into the finite element simulation model of step 1 to perform simulation calculation under preset simulation conditions (i.e., simulation verification) to obtain simulation results of contact pressure and stress; then the simulation results are compared with the simulation results of step 6 to evaluate the optimization effect.

[0069] Preferably, in step 10, the comparison includes contact pressure and stress. In this embodiment, the simulation results show that the contact pressure distribution is as follows: the middle of the skirt contact surface is the primary contact pressure bearing area and is evenly distributed. The peak contact pressure is significantly improved compared to the non-optimized profile solution, confirming that the skirt contact profile has high reliability. The stress distribution shows that the Mises stress values ​​of the structure at the piston skirt contact surface do not reach or are far below the yield limit of the piston material.

[0070] Any matters not described in the present invention are applicable to the prior art.

Claims

1. A method for designing a bolt-connected combined piston skirt top contact profile, characterized in that: The method comprises the following steps: Step 1: Establish a finite element simulation model of the piston; Step 2: Under preset simulation conditions, perform a thermal-mechanical coupling simulation on the finite element simulation model of step 1 to obtain the radial and circumferential surface deformations of the piston skirt top contact; Step 3: fitting the radial and circumferential surface deformations of the piston skirt top contact obtained in step 2 to obtain initial radial and circumferential profiles of the skirt top contact, respectively; wherein the initial circumferential profile of the skirt top contact is characterized by a fitting function that is most similar to the surface deformation state of the skirt top contact; Step 4: In the radial direction of the skirt top contact, the positions and indentations of the key deformation points in the initial radial profile of the skirt top contact obtained in Step 3 are selected as the target parameters for optimization; in the circumferential direction of the skirt top contact, the eigenvalues ​​of the fitting function of the initial circumferential profile obtained in Step 3 are selected as the target parameters for optimization; Step 5: Sampling the radial and circumferential target parameters of the skirt top contact according to the sampling sample interval to obtain n1 groups of profile line samples; Step 6: Based on the n1 groups of profile samples from step 5, adjust the finite element simulation model from step 1 to obtain finite element simulation models of n1 groups of different profiles of the same scale; then perform finite element simulation on each finite element simulation model to obtain simulation results as output values ​​for training the machine learning proxy model, and use the corresponding profile samples as input values ​​for training the machine learning proxy model to obtain a trained machine learning proxy model; the simulation results include contact pressure and stress; Step 7: Repeat steps 5 and 6 to extract n2 groups of profile samples, and then perform finite element simulation according to step 6. The obtained simulation results and their corresponding profile samples are used as the validation set of the trained machine learning proxy model obtained in step 6. The accuracy of the trained machine learning proxy model is verified by selecting the corresponding accuracy verification index. If the accuracy does not meet the requirements, adjust the sampling method of step 5, retrain the machine learning proxy model until the accuracy meets the requirements, and save the trained machine learning proxy model that meets the accuracy requirements to replace the finite element simulation process. Step 8: Based on the machine learning agent model obtained in step 7, the contact pressure and stress of the skirt top contact are used as optimization indicators, the position and indentation of the key deformation points selected in the radial profile of the skirt top contact, and the eigenvalues ​​of the fitting function of the circumferential direction of the skirt top contact are used as optimization target parameters, and the optimization indicators are continuously iteratively solved through a multi-objective optimization algorithm to obtain the optimal target parameters; Step 9: Refit the optimal target parameters obtained in step 8 to obtain the optimal radial and circumferential profiles of the skirt top contact, respectively. Then, the three-dimensional surface formed by linearly superimposing the optimal radial and circumferential profiles is used as the optimal profile at the contact surface of the piston top or piston skirt.

2. The method for designing a bolted combined piston skirt top contact profile according to claim 1, characterized in that: In step 1, the finite element simulation model includes the piston body, piston pin, cylinder liner and connecting rod; During the modeling process, the contact relationship of the skirt top is established in a point-to-point manner. The subsequent contact surface is obtained through mapping, and the contact surface nodes are copied one by one.

3. The method for designing a bolted combined piston skirt top contact profile according to claim 1, characterized in that: In step 2, on the load boundary, a preset simulation condition is established to take into account the thermal and mechanical loads that the piston is subjected to during actual operation. The thermal load is the thermal load generated by the piston under a steady-state temperature field, and the mechanical load is the multi-source mechanical load borne by the piston under maximum explosion pressure, which is the most dangerous working condition for the piston. On the constraint boundary, the preset simulation working condition fully constrains the piston and cylinder sleeve part, and constrains the connecting rod body direction and the vertical connecting rod swing direction of the piston connecting rod big end part.

4. The method for designing a bolt-connected assembled piston skirt top contact profile according to claim 3, characterized in that: In step 2, for the piston steady-state temperature field, if there is no piston test measured temperature, the piston steady-state temperature field is set according to the maximum temperature of the piston temperature design limit; if there is a piston test measured temperature, the final steady-state temperature field is compared with the test measured temperature, so that the error value between the simulated steady-state temperature field and the test measured temperature is controlled within 10%; The multi-source mechanical loads include bolt preload, gas pressure generated at maximum explosion pressure, inertia force and lateral force given by the cylinder liner.

5. The method for designing a bolted combined piston skirt top contact profile according to claim 1, characterized in that: In step 4, the initial radial profile of the skirt top contact is a three-segment profile, which is a structural form of a straight line segment in the middle and curved segments on both sides. The key deformation points are the endpoints on both sides of the profile and the inflection point in the middle.

6. The method for designing a bolt-connected assembled piston skirt top contact profile according to claim 1, characterized in that: In step 5, the sampling interval is from displacement × 50% to displacement × 150%; Sampling was done using Latin hypercube sampling.

7. The method for designing a bolt-connected assembled piston skirt top contact profile according to claim 1, characterized in that: In step 6, the adjustment method includes a grid mapping method, a method for adjusting the contact gap, and a method for adjusting the coordinates of the grid nodes on the surface of the contact part in the finite element simulation model; The machine learning agent model is a response surface model, a Kriging model, or a neural network model; The contact pressure is the maximum contact pressure of the piston skirt top contact surface, and the stress includes the maximum Mises stress of the piston skirt top contact surface, the maximum Mises stress at the locating pin hole, and the maximum Mises stress at the bolt hole.

8. The method for designing a bolt-connected assembled piston skirt top contact profile according to claim 1, characterized in that: In step 7, the accuracy verification index adopts one of the determination coefficient, mean square error, root mean square error and mean absolute error indicators; Adjusting the sampling method of step 5 includes increasing the number of profile samples and / or reducing the sampling sample interval; wherein, increasing the number of profile samples is 1.3 to 2 times the number of profile samples in step 5; reducing the sampling sample interval is 0.5 to 0.8 times the sampling sample interval in step 5.

9. The method for designing a bolt-connected assembled piston skirt top contact profile according to claim 1, characterized in that: In step 8, the multi-objective optimization algorithm adopts an improved particle swarm algorithm or a genetic algorithm.

10. The method for designing a bolt-connected assembled piston skirt top contact profile according to claim 1, characterized in that: The method also includes: step 10, result verification: the optimal surface obtained in step 9 is adjusted by the adjustment method of step 6, and loaded into the finite element simulation model of step 1 to perform simulation calculations under preset simulation conditions to obtain simulation results of contact pressure and stress; the simulation results are then compared with the simulation results of step 6 to evaluate the optimization effect.

Images