Hydrodynamic transmission power chain optimization method and system

Abstract

The invention relates to the technical field of torque converters, in particular to a hydrodynamic transmission power chain optimization method and system.The method comprises the steps that a hydrodynamic torque converter parameterized geometric model is built, and a pump impeller blade outflow angle, a guide wheel runner input / output angle, turbine blade geometric parameters and the relative position of a pump impeller and a turbine are set as design variables; based on CFD three-dimensional flow field simulation, a k-epsilon model is adopted for rapid analysis, an SST model is adopted for transient simulation, and simulation data under different variable combinations are obtained; training a BP neural network by using the simulation data, inputting each design variable, and constructing a multi-target optimization model with maximization of efficiency and minimization of noise as targets; a design variable combination is selected in the Pareto optimal solution set, a model is reconstructed, and an optimized hydraulic torque converter is obtained; according to the method, the problems that the design optimization efficiency of the hydraulic torque converter is low, the parameter adjusting period is long, and good compromise between the transmission efficiency and noise suppression is difficult to obtain can be solved.

Description

Technical Field

[0001] The present invention relates to the technical field of torque converters, and in particular to a method and system for optimizing a hydraulic transmission power chain. Background Art

[0002] The torque converter is a core component in the hydraulic transmission system of construction machinery. Its performance directly impacts the vehicle's dynamic response, fuel economy, and noise level. Traditionally, the torque converter's geometric parameters have been adjusted primarily based on engineering experience and limited simulation samples. This results in low design optimization efficiency, long parameter adjustment cycles, and difficulty achieving a good compromise between transmission efficiency and noise suppression. Summary of the Invention

[0003] In view of the above technical problems, the present invention provides a hydraulic transmission power chain optimization method and system to solve the problems of low design optimization efficiency of the hydraulic torque converter, long parameter adjustment cycle, and difficulty in achieving a good compromise between transmission efficiency and noise suppression.

[0004] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by practice of the present disclosure.

[0005] According to one aspect of the present invention, a method for optimizing a hydraulic transmission power train is proposed, the method comprising: A parametric geometric model of the torque converter is constructed, in which the impeller blade outflow angle, the guide wheel flow channel input angle, the guide wheel flow channel output angle, the turbine blade geometric parameters, and the relative position of the impeller and turbine are used as design variables; Performing a three-dimensional flow field simulation of the torque converter based on computational fluid dynamics, using a k-ε turbulence model for rapid analysis of an initial solution, and using a shear stress transmission turbulence model for transient simulation, to obtain simulation data of the torque converter under different combinations of the design variables, the simulation data including flow field characteristics, pressure distribution, velocity field distribution, turbulence intensity, transmission efficiency, and corresponding noise values; A training set is established using the simulation data to train a back-propagation neural network model; wherein the input layer nodes of the back-propagation neural network model are the design variables of the torque converter, the hidden layer uses at least one layer, each layer contains 8 neurons and uses a hyperbolic tangent sigmoid activation function, the output layer nodes are the transmission efficiency and noise prediction values ​​of the torque converter, the output layer uses a linear activation function, and the back-propagation neural network model uses a Levenberg-Marquardt algorithm for back-propagation training; Constructing a multi-objective optimization model, wherein the multi-objective optimization model takes maximizing the transmission efficiency of the torque converter and minimizing the noise level as objective functions, and sets constraints on the design variables, wherein the constraints include limited value ranges of the impeller blade outflow angle, the guide wheel flow channel input and output angles, and the turbine blade geometric parameters; Solving the multi-objective optimization model using a non-dominated sorting genetic algorithm, using the trained back propagation neural network model as a fitness function to evaluate the objective value of each candidate design, and obtaining a Pareto optimal solution set that includes a trade-off relationship between transmission efficiency and noise performance; Based on the combination of torque converter design variables obtained from the Pareto optimal solution set, a solution is determined and a model is reconstructed to obtain the optimized torque converter.

[0006] Furthermore, the design variable value ranges of the parameterized geometric model of the torque converter specifically include: the impeller blade outflow angle is 15°~45°, the guide wheel flow channel input angle is 20°~60°, the guide wheel flow channel output angle is 5°~35°, and the turbine blade geometric parameters include blade thickness and flow channel curvature radius.

[0007] Furthermore, after obtaining the optimized torque converter, the three-dimensional flow field simulation of the torque converter is performed again based on computational fluid dynamics. During the simulation, the Taguchi orthogonal test method is used to generate initial sample points, and random sampling is combined to supplement additional sample points to form a comprehensive sample set covering the entire design space. The comprehensive sample set is used to construct training data for the back propagation neural network model.

[0008] Furthermore, the genetic operation of the non-dominated sorting genetic algorithm includes: after the initial population is randomly generated, the priority of the population individuals is determined by non-dominated sorting, the elite individuals are screened by using the crowding distance, and a new generation of population is generated for the selected individuals through intermediate crossover and Gaussian mutation strategies, and the maximum number of genetic iterations is set to 300 generations.

[0009] Furthermore, when solving the multi-objective optimization model using a non-dominated sorting genetic algorithm, the method specifically includes: The trained back-propagation neural network model is used as a fitness evaluation tool to perform population initialization, non-dominated sorting, crowding calculation, crossover and mutation genetic operations during the solution. During the operation, the population size is set to 100, the crossover probability is 0.8, and the Gaussian mutation strategy is adopted. The iterative calculation is performed until the termination condition is met to generate the Pareto optimal solution set that includes the trade-off relationship between transmission efficiency and noise performance.

[0010] Furthermore, in the process of solving the multi-objective optimization model, the objective function is weighted and synthesized using the weighted summation method to form an initial optimization population, wherein the objective function based on the weighted summation method is expressed as: ; in, is the weighted comprehensive objective function, represents the transmission efficiency objective function, represents the noise objective function, 、 are the weight coefficients of transmission efficiency and noise level, respectively, and satisfy the relationship + =1; The optimization result obtained by the weighted summation method is input as the initial population into the non-dominated sorting genetic algorithm for secondary optimization to obtain the Pareto optimal solution set.

[0011] Furthermore, in the three-dimensional flow field simulation of the torque converter based on computational fluid dynamics, the obtained multi-dimensional flow field output data is subjected to tensor preprocessing to reduce the data dimension and extract key features. The preprocessing includes normalizing the pressure distribution and the velocity field, filtering out numerical noise, and calculating field parameters that characterize the overall performance; the field parameters include the flow velocity ratio obtained by area averaging the velocity field distribution, and the flow velocity ratio is used to represent the overall velocity gain effect of the flow field.

[0012] Furthermore, when solving the multi-objective optimization model, it also includes: An optimized coupling mechanism for hydraulic transmission parameters is established by deriving a mathematical relationship between the specific speed ratio and the torque ratio of the torque converter, wherein the specific speed ratio is defined as the ratio of the rotational speed of the torque converter turbine to the rotational speed of the pump impeller, and the torque ratio is defined as the ratio of the output torque of the turbine to the input torque of the pump impeller; A performance characteristic curve of the torque converter is constructed based on a relationship between the specific speed ratio and the torque ratio, and the hydraulic transmission parameters are combined with an objective function in the multi-objective optimization model.

[0013] According to a second aspect of the present disclosure, a hydraulic transmission power chain optimization chain system is provided, the system comprising: A 3D modeling module is used to construct a parametric geometric model of the torque converter, in which the impeller blade outflow angle, the guide wheel flow channel input angle, the guide wheel flow channel output angle, the turbine blade geometry parameters, and the relative position of the impeller and turbine are used as design variables; a simulation module for performing a three-dimensional flow field simulation of the torque converter based on computational fluid dynamics, using a k-ε turbulence model for rapid analysis of an initial solution, and using a shear stress transmission turbulence model for transient simulation, to obtain simulation data of the torque converter under different combinations of the design variables, the simulation data including flow field characteristics, pressure distribution, velocity field distribution, turbulence intensity, transmission efficiency, and corresponding noise values; a machine learning module, configured to establish a training set using the simulation data and train a back-propagation neural network model; wherein the input layer nodes of the back-propagation neural network model are the design variables of the torque converter, the hidden layer has at least one layer, each layer includes 8 neurons and uses a hyperbolic tangent sigmoid activation function, the output layer nodes are the transmission efficiency and noise prediction values ​​of the torque converter, the output layer uses a linear activation function, and the back-propagation neural network model uses a Levenberg-Marquardt algorithm for back-propagation training; a mathematical modeling module for constructing a multi-objective optimization model, wherein the multi-objective optimization model takes maximizing the transmission efficiency of the torque converter and minimizing the noise level as objective functions, and sets constraints on the design variables, wherein the constraints include limited value ranges of the impeller blade outflow angle, the guide wheel flow channel input and output angles, and the turbine blade geometric parameters; a calculation module, configured to solve the multi-objective optimization model using a non-dominated sorting genetic algorithm, and use the trained back propagation neural network model as a fitness function to evaluate the objective value of each candidate design, thereby obtaining a Pareto optimal solution set that includes a trade-off relationship between transmission efficiency and noise performance; A reconstruction module is used to determine a solution and reconstruct a model based on the combination of torque converter design variables obtained from the Pareto optimal solution set to obtain the optimized torque converter.

[0014] The technical solution disclosed in this disclosure has the following beneficial effects: By introducing a back-propagation neural network to model and predict the transmission efficiency and noise level of the torque converter under different design parameters, the high-cost repeated fluid dynamics simulation calculations during the optimization iteration process are avoided, the computing resource consumption is significantly reduced, and the design efficiency is improved. The non-dominated sorting genetic algorithm is used for multi-objective search, which can simultaneously optimize the two performance indicators of transmission efficiency and noise control, obtain a set of balanced Pareto optimal solutions, and break through the technical bottleneck of traditional methods that are difficult to compromise between efficiency and noise. The training samples are constructed by combining Taguchi orthogonal experiments with random sampling to cover the design variable space, which improves the generalization ability of the neural network model, can adapt to the complex and changeable torque converter design requirements, and output a more applicable optimal parameter combination. By setting the weight factor, fitness function structure and iterative control parameters, the optimization target proportion can be flexibly adjusted according to the performance focus of the vehicle, realizing customized parameter optimization and improving the overall performance of the powertrain system and the level of design intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flow chart of a method for optimizing a hydraulic transmission power chain according to an embodiment of this specification; Figure 2 This is a schematic diagram of key components of a hydraulic transmission chain in an embodiment of this specification; Figure 3 is a schematic diagram of the internal flow field of the torque converter in the embodiment of this specification; Figure 4 is a pressure distribution diagram of the internal flow field of the torque converter in the embodiment of this specification; Figure 5 is a velocity distribution diagram of the internal flow field of the torque converter in the embodiment of this specification; Figure 6 is a turbulent viscosity distribution diagram of the flow field in the torque converter in the embodiment of this specification; Figure 7 This is a structural block diagram of a hydraulic transmission power chain optimization chain system in an embodiment of this specification; Figure 8 It is a terminal device for implementing a method for optimizing a hydraulic transmission power chain in an embodiment of this specification; Figure 9 It is a computer-readable storage medium storing a method for optimizing a hydraulic transmission power chain in an embodiment of this specification. DETAILED DESCRIPTION

[0016] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the present disclosure will be more comprehensive and complete and will fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the specific details, or that other methods, components, devices, steps, etc. may be employed. In other cases, well-known technical solutions are not shown or described in detail to avoid obscuring various aspects of the present disclosure.

[0017] The accompanying drawings are merely schematic illustrations of the present disclosure. Identical reference numerals in the drawings denote identical or similar components, and thus their repeated description will be omitted. Some of the blocks shown in the accompanying drawings represent functional entities that do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0018] The present invention provides a method for optimizing the hydraulic transmission power chain of a product. Figure 1 FIG. 1 is a flow chart of a method for optimizing a hydraulic transmission power train according to an embodiment of the present invention. The method can be applied to electronic devices such as personal computers and servers. The method can be performed by a device that can be implemented by software and / or hardware. The method can specifically include the following steps S101 to S106: In step S101, a parameterized geometric model of the torque converter is constructed, in which the outflow angle of the impeller blade, the input angle of the guide wheel flow channel, the output angle of the guide wheel flow channel, the geometric parameters of the turbine blades, and the relative positions of the impeller and the turbine are used as design variables.

[0019] Among them, the torque converter is mainly composed of a pump wheel, a turbine and a guide wheel. The basic structure is as follows: Figure 2As shown in the figure. When the torque converter's circulation chamber is filled with working medium, the engine rotates the impeller. The working medium, propelled by the impeller blades, gains kinetic energy and velocity, flowing along its flow path toward the outlet. Energy is transferred from the engine to the impeller, where it is then transferred to the working medium by the impeller blades, converting mechanical energy into fluid kinetic energy. After exiting the impeller, the working medium enters the turbine area, where it propels the turbine blades, causing the turbine to rotate. This transfer of energy from the working medium to the turbine converts fluid kinetic energy into mechanical energy. The turbine's rotation drives the drive shaft, transferring energy to the transmission. The energy input from the engine is now output to the transmission through this flexible connection. After the working medium propels the turbine blades, generating torque, it flows out of the turbine area along the blades and impacts the stator blades. Because the stator is stationary, the velocity and direction of the working medium flowing through the stator area change, causing a corresponding change in angular momentum, thus achieving torque conversion.

[0020] In traditional torque converter design, geometric parameters such as the impeller blade outflow angle, the guide wheel flow channel input and output angles, the turbine blade geometry, and the relative position between the impeller and turbine have a direct impact on the overall transmission efficiency, torque conversion characteristics, and operating noise. However, in existing technologies, these parameters often rely on empirical experience or manual adjustment of single variables, making it difficult to achieve an ideal balance between efficiency and noise suppression. To systematically perform multi-dimensional optimization, it is necessary to extract these key geometric features as adjustable design variables and construct a parameterized geometric model for unified iteration and screening in simulation and optimization algorithms.

[0021] The parametric geometric model described above represents the impeller blade outflow angle, the stator flow channel input and output angles, the turbine blade thickness and curvature distribution, and the relative axial or radial arrangement of the impeller and turbine as numerical variables. These variables then automatically modify the blade profile or component position within the CAD or modeling software based on these variables. For example, the impeller blade outflow angle describes the inclination of the liquid relative to the tangential direction as it leaves the impeller. Different outflow angles alter the kinetic energy distribution of the liquid as it is ejected from the impeller outlet. Larger outflow angles increase the radial velocity component of the liquid, but also increase the likelihood of blade exit separation and localized pressure fluctuations, leading to energy loss or increased noise. However, smaller outflow angles reduce kinetic energy loss, but the exit velocity direction is too tangential, resulting in insufficient torque extraction. Therefore, by setting this angle as an adjustable variable during the optimization process, the algorithm automatically balances kinetic energy conversion efficiency with flow channel stability and noise.

[0022] The stator flow channel input and output angles are equally important. The stator input angle determines the degree of alignment of the angle of attack of the fluid exiting the turbine as it enters the stator. If the input angle is inconsistent with the incoming flow direction, localized separation vortices will form at the stator inlet, resulting in energy loss and increased noise. The output angle determines the direction of the fluid passing through the stator and whether the velocity vector direction of the fluid returning to the impeller inlet matches the inlet angle of the impeller blades. Otherwise, impact or backflow will occur at the impeller inlet, reducing efficiency and generating noise. Setting both the stator input and output angles as design variables allows the simulation and optimization process to automatically adjust the entire fluid cycle from the impeller to the turbine and back to the stator as smoothly and efficiently as possible. Turbine blade geometry includes blade thickness, inlet and outlet radii of curvature, and, if necessary, twist distribution. Turbine blades convert the kinetic energy of the fluid ejected from the impeller into mechanical torque. Their geometry directly determines the separation location of the boundary layer on the blade surface and the degree of vortex dissipation. For example, if the blade radius of curvature is too large, a sharp turn may occur at the inlet, causing fluid separation and noise. If the blade is too thin or has too little curvature, it is prone to strong vibration or fatigue failure under high loads, and the performance optimization space is limited. By parameterizing the blade thickness and curvature distribution, new blade geometries can be continuously generated in simulation. Combined with a neural network model, performance is predicted, and a genetic algorithm is used to select the optimal geometry that both improves torque output and suppresses noise. Furthermore, the relative position between the impeller and turbine also affects the flow path length and fluid flow path. When the axial spacing between the two components is small, the fluid travels a short path within the cavity, minimizing dynamic pressure losses, but the local flow velocity in the channel varies dramatically, which can easily generate noise. If the axial spacing is too large, the fluid travels a longer path, resulting in more gradual dynamic pressure recovery, but increasing friction losses and reducing transmission efficiency. Similarly, radial eccentricity can cause asymmetry in the overall flow field, resulting in increased turbulence in certain areas or changes in the flow cross-section. By setting relative position as a design variable, the optimization algorithm can analyze the transmission efficiency and noise performance under different position combinations and find a balance.

[0023] In summary, by using a parameterized geometric model as six adjustable design variables—the impeller blade outflow angle, the guide channel input angle, the guide channel output angle, the turbine blade geometry, and the relative position of the impeller and turbine—the geometric design process of the torque converter can be transformed into a high-dimensional numerical optimization problem. This approach enables comprehensive and automatic generation of corresponding flow field results in subsequent CFD (fluid dynamics) simulations. This is then rapidly predicted using a back-propagation neural network, combined with a multi-objective genetic algorithm for global search, to efficiently obtain the optimal design solution that balances transmission efficiency and noise suppression.

[0024] In step S102, a three-dimensional flow field simulation of the torque converter is performed based on computational fluid dynamics, a k-ε turbulence model is used to quickly analyze the initial solution, and a shear stress transmission turbulence model is used to perform transient simulation to obtain simulation data of the torque converter under different combinations of the design variables, wherein the simulation data includes flow field characteristics, pressure distribution, velocity field distribution, turbulence intensity, transmission efficiency, and corresponding noise values.

[0025] During operation, the internal flow field of a torque converter is complex and turbulent, resulting in high testing costs. To provide theoretical guidance for torque converter performance testing and thereby shorten the development cycle, a CFD-based flow field analysis model was used to study the performance of a torque converter. To ensure that the flow field model remains reasonably complex and can be solved within a reasonable timeframe, while closely resembling the actual torque converter operating conditions, the following assumptions were made for the torque converter flow field analysis: (1) The working fluid of the torque converter is a homogeneous viscous fluid, free of impurities and voids, and its physical and chemical properties remain unchanged; (2) The influence of temperature change on the working condition of the torque converter is not considered, the heat conversion and heat transfer in various parts of the torque converter are ignored, and the energy equation is not used in the flow field model; (3) The inner and outer ring shells and blades of the torque converter are all rigid bodies, and the fluid-solid coupling between the working oil and the torque converter cavity is ignored; (4) There is no leakage of working oil at the junction between the impellers, and the flow channel inside the torque converter is a completely closed annular flow channel.

[0026] The most fundamental governing equations for solving the flow field within a torque converter are the mass and momentum equations. The mass equation, also known as the continuity equation, is used to determine the mass increment caused by fluid entering the control volume per unit time. The mass equation shows that the mass of fluid flowing out of the control volume through the control surface per unit time is equal to the mass of fluid flowing out of the control volume through the control surface during the same period of time. The mass equation can be expressed in integral form as follows: ; Where t represents time; ρ is the density of the fluid; υ is the velocity of the fluid; n is the unit vector normal to the control surface; and A is the control surface. The first term is the increment of the control volume's mass per unit time, and the second term is the net mass of the fluid flowing out of the control volume through the control surface. When the fluid density remains constant, the continuity equation can be rewritten in differential form in rectangular coordinates: ; Where u, v, and w are the velocity components of the fluid velocity υ in the x, y, and z directions, respectively. From the above equation, we can see that the divergence of the velocity vector field of the fluid in the control volume is zero.

[0027] When solving the torque converter flow field using CFD, in addition to solving the mass equation, the momentum equation, also known as the Navier-Stokes (NS) equations, is required. The NS equations describe the law of conservation of momentum in fluid motion, stating that the sum of the external forces acting on a control volume is equal to the rate of change of the momentum of the fluid within the control volume with respect to time. When the density and viscosity of the fluid are constant, the NS equations can be expressed in a rectangular coordinate system as: ； ； ; Where u, v, w are the velocity components of the fluid velocity υ in the x, y, and z directions respectively; t represents time; ρ is the density of the fluid; fx, fy, fz are the external forces acting on the control body in the x, y, and z directions respectively; p is the pressure of the fluid; μ is the dynamic viscosity of the fluid; and ∇ is the Hamiltonian operator.

[0028] During operation, the impeller speeds of a torque converter vary, and the operating oil flows turbulently, creating complex flow conditions within the converter. Torque converter flow analysis models must capture these turbulent flows and more closely examine factors such as eddy current diffusion to fully investigate torque converter performance. While engineering doesn't require detailed understanding of turbulent flow, it is crucial to assess the overall impact of turbulence on the internal flow field. Therefore, a specific turbulence model is necessary to simplify the simulation process.

[0029] During simulation, using an existing torque converter as an example, the boundary conditions are first defined. These define the mathematical and physical conditions that flow field variables must satisfy at the boundaries of the computational domain, and their definition is crucial to the success of fluid dynamics calculations. For the numerical calculation of a torque converter, this process simulates the internal flow of oil within the enclosed cavities of the impeller, turbine, and stator. Therefore, the circulation cavity of the impeller, turbine, and stator is defined as the flow region, and the contact surfaces from the impeller outlet to the turbine inlet, the turbine outlet to the stator inlet, and the stator outlet to the impeller inlet are defined as the interfaces. Because the flow and impact of the oil within the cavity do not change the stiffness of the torque converter blades and outer wall, the blades and outer wall are defined as rigid bodies.

[0030] Then, parameter settings are performed. The pump speed is set to 1800 r / min, the initial turbine speed is set to 0, and the stator speed is constant at 0 because the stator is fixed. Assuming that the torque converter flow field is closed and ignoring the effect of medium temperature rise on the numerical simulation, the detailed simulation parameters of the torque converter are as follows:

[0031] Then the flow field characteristics of the torque converter are analyzed. When the torque converter is in the starting condition, the impeller speed ratio is the largest, the torque on the impeller is the largest, and the internal flow field is the most complex. Taking this flow field state as an example, the flow condition inside the torque converter is as follows: Figure 3 The upper part of the figure, dominated by yellow-green streamlines, represents the internal flow field of the turbine, while the lower part, dominated by blue streamlines, represents the internal flow field of the impeller. The working oil is accelerated by the impeller to impact the turbine, and then changes direction through the guide impeller and returns to the impeller.

[0032] like Figure 3 As shown in the diagram, during the working oil's circulation, the fluid enters the impeller at a low velocity. After being accelerated by the impeller, it impacts the turbine blades. The inlet of the turbine flow passage is the point of maximum oil velocity within the torque converter. As the oil flows through the turbine blades, its velocity gradually decreases, and its kinetic energy is converted back into mechanical energy for turbine rotation. The oil then passes through the guide wheel and reaches the impeller flow passage inlet, where it is again accelerated and ejected by the impeller.

[0033] When the oil enters the turbine from the impeller, there are two high-speed sections: the high-speed section in the gap between the two impellers when the oil is just thrown out by the impeller, and the high-speed section formed after the oil completely enters the turbine blade flow channel. The velocity direction of the liquid in the two high-speed sections of the liquid flow is different. The shape of the turbine blades causes the liquid flow to change direction, and the velocity of the fluid temporarily drops due to the thickness of the blades and the shape of the blade edges. When the outlet angle of the impeller blades changes relative to the inlet angle of the turbine blades, the flow field conditions at the inlet of the turbine flow channel are changed accordingly. The design of these two angles has a great influence on the local energy loss of the fluid at the junction. The flow in the impeller is very turbulent. There is a deviation between the velocity direction of the liquid entering the impeller and the inlet angle of the impeller blades, which causes part of the liquid to impact back and forth between the two blades of the impeller flow channel, consuming kinetic energy.

[0034] First, the pressure field of the torque converter is analyzed. Under starting conditions, the overall pressure of the flow field in the torque converter is relatively high, such as Figure 4 An example is shown. In this example, the pressure distribution within the impeller flow path is relatively uniform, gradually increasing from the inlet to the outlet, with a slight jump at the trailing blade ends. The pressure variations within the turbine flow path are more pronounced. First, a significant localized high-pressure zone exists at the turbine blade inlet, which is detrimental to the blade structure. The magnitude of the pressure in this high-pressure zone is related not only to the turbine blade inlet angle but also to the blade thickness and edge shape. The turbine blades force the oil to change direction, causing the pressure in the flow field behind the blades to rise again, resulting in an increase in turbine torque that exceeds the impeller torque. After passing through the stator, the oil pressure drops to the pressure near the impeller inlet.

[0035] Further Figure 5As shown in the figure, the overall flow rate in the impeller flow passage of this example is low, and the oil collides back and forth between the blades, resulting in significant energy loss and detrimental to hydraulic transmission. In the turbine flow passage, the oil separates at the very front end of the blade. Some oil adheres to the pressure surface, converting the fluid's kinetic energy into mechanical energy for the impeller. The other part rushes at high speed toward the next blade on the suction side of the current blade, causing a very significant vortex flow, which slows the flow rate until the vortex is essentially dissipated at the blade's exit.

[0036] Since the flow inside the torque converter is extremely complex and the entire internal flow field is basically in a turbulent flow state, it is necessary to perform flow field analysis on the turbulent flow of the torque converter. Using turbulent viscosity as a quantitative index, the turbulent viscosity distribution in the blade flow field is calculated, such as Figure 6 As shown in the figure, turbulent viscosity differs from the dynamic viscosity of an oil. Its essence is to reflect the dissipation of eddies. As the working oil circulates within cavities of varying shapes, its flow state changes dramatically, generating numerous eddies in the flow field. The governing equations within a turbulence model essentially account for the motion and dissipation of eddies, so the turbulent viscosity distribution diagram is a graphical representation of the turbulence model's solution.

[0037] from Figure 6 As can be seen from the diagram, the turbulent viscosity in this example is higher on both sides of the turbine blade, while the turbulent viscosity is lower near the blade wall. The blade wall is a no-slip surface. Due to viscous forces, the fluid velocity near the wall approaches the wall velocity, resulting in a low Reynolds number and low turbulent viscosity. Away from the wall, the fluid begins to flow freely, with a high Reynolds number and correspondingly higher turbulent viscosity. Velocity analysis shows that significant vortices are generated at the turbine blade inlet and continuously dissipate in the flow. Consequently, a distinct region of high turbulent viscosity appears at and near the turning point of the turbine blade, resulting in significant loss of fluid kinetic energy. As the vortices gradually disappear downstream, the turbulent viscosity decreases.

[0038] The above analysis of the flow field characteristics of the torque converter reveals the flow characteristics of the flow field inside the torque converter and various conditions of the flow field, so it is necessary to optimize the design parameters of the torque converter.

[0039] In step S103, a training set is established using the simulation data to train a back-propagation neural network model; wherein, the input layer nodes of the back-propagation neural network model are the various design variables of the torque converter, its hidden layer adopts at least one layer, each layer contains 8 neurons and adopts a hyperbolic tangent S-type activation function, the output layer nodes are the transmission efficiency and noise prediction values ​​of the torque converter, the output layer adopts a linear activation function, and the back-propagation neural network model adopts the Levenberg-Marquardt algorithm for back-propagation training.

[0040] Among them, after using CFD simulation to obtain the flow field characteristics, pressure distribution, velocity field, and corresponding transmission efficiency and noise values, these simulation data are integrated into a tabular training set containing multiple sets of "design variable-performance index" correspondences. Specifically, each row of the table corresponds to a design scheme formed by weighted parameters such as the impeller blade outflow angle, the guide wheel flow channel input angle, the guide wheel flow channel output angle, the turbine blade geometric parameters, and the relative position of the impeller and the turbine. It also includes the simulation output under this scheme: transmission efficiency value and noise decibel value. By performing CFD simulation on a large number of discrete parameter combinations, it is possible to construct a training data set with high-dimensional features and detailed mapping of target variables that can cover the key change trends in the design space. This approach not only ensures that the nonlinear relationship between design variables and performance indicators is fully sampled, but also provides the necessary sample depth and breadth for the subsequent training of the neural network model.

[0041] Based on the aforementioned training set, a multi-layer feedforward back-propagation neural network (BPNN) was introduced. Five or more design variables—the impeller blade outflow angle, the stator flow channel input angle, the stator flow channel output angle, the turbine blade geometry, and the relative position of the impeller and turbine—were used as input layer nodes. Each node received the value of the corresponding parameter. The hidden layer consisted of at least one layer, each containing eight neurons, and employed a hyperbolic tangent (tanh) sigmoid activation function. The choice of eight neurons was based on engineering experience and cross-validation considerations: too few nodes would result in insufficient expressive power and an inability to approximate complex nonlinear mappings; too many nodes would easily lead to overfitting and increase computational complexity. The hyperbolic tangent sigmoid activation function maps the input to the range [-1, 1], facilitating stable gradient propagation during the optimization process. Compared to the previously commonly used sigmoid function, its zero-centering and gradient decay properties are less pronounced, resulting in improved numerical stability and convergence speed for the hidden layer when learning complex flow field and performance mappings. The output layer has two nodes, which are used for regression prediction of "transmission efficiency" and "noise prediction value" respectively. A linear activation function is used to ensure that the output value can be freely mapped within any real number range, thereby matching the continuous value obtained by simulation.

[0042] During the network training phase, the Levenberg–Marquardt algorithm is used to iteratively update the error through backpropagation. This algorithm combines the fast convergence characteristics of the Gauss-Newton method with the stability of the gradient descent method. By dynamically balancing the Jacobian matrix and the Hessian matrix approximation, it can achieve a more rapid reduction in training error and a more sensitive search for complex error surfaces when processing medium-sized network parameters. In the specific process, each iteration first calculates the mean squared error (MSE) between the network output and the true simulation value under the current weights, and then fine-tunes the weights and biases of each layer in the network using the Levenberg–Marquardt corrected step size strategy. As the number of iterations increases, the training error continues to converge. When the error change rate is lower than the preset threshold or reaches the maximum number of iterations, it can be considered that the network has fully learned the nonlinear mapping relationship between the design variables and performance indicators, and can be used for rapid prediction and evaluation in subsequent multi-objective optimization.

[0043] In step S104, a multi-objective optimization model is constructed. The multi-objective optimization model takes maximizing the transmission efficiency of the torque converter and minimizing the noise level as objective functions, and sets constraints on the design variables. The constraints include the impeller blade outflow angle, the guide wheel flow channel input and output angles, and the limited value range of the turbine blade geometric parameters.

[0044] In multi-objective optimization, both transmission efficiency and noise level performance indicators must be considered simultaneously. First, the transmission efficiency of the torque converter is represented by the symbol η_eff(x), which depends on the design variables x (i.e., the impeller blade outflow angle, the stator flow channel input angle, the stator flow channel output angle, the turbine blade geometry, and the relative position of the impeller and turbine). The noise level is represented by the symbol N_noise(x), which also depends on the same design variable vector x. Thus, multi-objective optimization can be formalized as: MAX η_eff(x); MIN N_noise(x); In this expression, x is a vector containing several components, for example, x = [θ_pump_outlet, θ_stator_inlet, θ_stator_outlet, Geom_turbine, Pos_rel], where θ_pump_outlet represents the outlet angle of the impeller blades, θ_stator_inlet and θ_stator_outlet represent the inlet and outlet angles of the guide channel, respectively, Geom_turbine represents a set of geometric parameters such as the thickness and camber of the turbine blades, and Pos_rel represents the relative position of the impeller and turbine. η_eff(x) is calculated using a pre-trained BPNN model or CFD numerical calculations and reflects the efficiency of the torque converter's kinetic energy transfer from the engine to the transmission for this geometric configuration. N_noise(x) represents the acoustic radiation intensity generated for the same configuration, which can be expressed in either dB or power.

[0045] To ensure the rationality and feasibility of the optimization results, the range of design variables must be constrained. First, while ensuring blade manufacturability and material strength, the impeller blade outflow angle, θ_pump_outlet, is limited to between 15° and 45°. If θ_pump_outlet is less than 15°, the radial component of the liquid ejected from the blade outlet is insufficient, resulting in reduced kinetic energy transfer efficiency. If it is greater than 45°, the tangential velocity component at the blade outlet is too low, easily causing blade trailing edge separation and significantly increasing noise. Second, the stator flow channel inlet angle, θ_stator_inlet, is limited to between 20° and 60°. When the inlet angle is less than 20°, the velocity direction of the liquid entering the stator does not match the blade angle of attack, forming vortices. When the inlet angle exceeds 60°, the inlet angle of attack is too large, causing excessive impact on the blade leading edge and generating high noise. Third, the stator flow channel outlet angle, θ_stator_outlet, must be controlled between 5° and 35° to ensure that the stator evenly guides the liquid back to the impeller inlet. If the outlet angle is less than 5°, the outflow direction deviates tangentially, causing impeller inlet shock. If it exceeds 35°, fluid separation and backflow losses increase, also reducing efficiency and increasing noise. Fourthly, turbine blade geometric parameters, such as the thickness t_turbine, should be maintained between 0.2 mm and 0.6 mm, and the radius of curvature R_bend should range from 10 mm to 30 mm. Blades that are too thin are prone to instability or vibration under high loads, while blades that are too thick will increase flow field damping, affecting overall efficiency. Furthermore, the relative position of the impeller and turbine (Pos_rel) (axial spacing and radial eccentricity) also requires a set range. The axial spacing D_axial is limited to 5 mm to 20 mm to ensure that the fluid circuit is neither too crowded (to avoid increased turbulence) nor too sparse (to avoid energy waste). The radial eccentricity D_radial is limited to 0 to 1 mm, which can reduce eccentric vortices to a certain extent. However, if it exceeds 1 mm, the energy loss caused by flow field asymmetry begins to outweigh the improvement effect.

[0046] By using these constraints, the design variables are confined to a reasonable range that ensures mechanical structural strength, blade manufacturing feasibility, and controllable turbulence and pressure recovery in the flow field. This multi-objective optimization model not only mathematically pursues maximizing η_eff(x) and minimizing N_noise(x) but also, by limiting the range of x values, avoids invalid or even unfeasible design solutions. Ultimately, an optimization algorithm (such as a non-dominated sorting genetic algorithm) searches for and evolves the parameter vector x that satisfies the constraints. Subsequently, through non-dominated sorting algorithms and congestion screening, a set of Pareto-optimal solutions that strike a good compromise between transmission efficiency and noise level can be obtained.

[0047] In step S105, a non-dominated sorting genetic algorithm (NSGA-II) is used to solve the multi-objective optimization model, and the trained back propagation neural network model is used as a fitness function to evaluate the objective value of each candidate design to obtain a Pareto optimal solution set that includes a trade-off relationship between transmission efficiency and noise performance.

[0048] When solving the multi-objective optimization model for a torque converter, the first step is to evaluate the fitness of each generation of candidate designs. Specifically, a trained back-propagation neural network model is used as the fitness function. This involves inputting the five or more design variables corresponding to the current candidate design: "impeller blade outflow angle, guide channel input angle, guide channel output angle, turbine blade geometry, and the relative position of the impeller and turbine." The neural network quickly outputs the corresponding transmission efficiency and noise predictions. This step significantly reduces the time required for precise calculations based on CFD simulations, allowing performance evaluation of hundreds or even thousands of candidate individuals in a short period of time.

[0049] Specifically, the genetic operations of the non-dominated sorting genetic algorithm include: after the initial population is randomly generated, the priority of the population individuals is determined by non-dominated sorting, the elite individuals are screened by using the crowding distance, and a new generation of population is generated for the selected individuals through intermediate crossover and Gaussian mutation strategies, and the maximum number of genetic iterations is set to 300 generations.

[0050] In each iteration of the genetic algorithm, elite solutions are first randomly generated or retained from the previous generation's population. These solutions are then encoded according to specific rules to form a set of chromosomes representing the design variables. For each chromosome, a forward propagation operation is directly invoked on the neural network model to obtain the corresponding metrics: "transmission efficiency η_eff" and "noise level N_noise." These two sets of values ​​constitute the target vector 〈η_eff, N_noise〉 required by NSGA-II. NSGA-II then performs a non-dominated sort on the entire population, assigning those solutions that are not simultaneously surpassed by any other individual in terms of "maximum efficiency / minimum noise" to the frontier. The remaining individuals are then stratified according to dominance. This stratification ensures that the algorithm simultaneously considers trade-offs between multiple objectives, rather than simply combining them into a single metric, thereby better representing the balance between these objectives.

[0051] After completing the non-dominated hierarchy, NSGA-II also needs to calculate the crowding distance (CrowdingDistance) of each individual, which is used to measure the distance between the individual and its surrounding neighbors in the target space, to help the subsequent selection process maintain the diversity of the solution set. Within the same hierarchy, if two individuals have the same level, the one with the greater crowding distance is retained first to avoid "clustering" in the solution set, thereby ensuring that the Pareto optimal frontier can cover a wider area of ​​the design space. Through the above two steps (non-dominated sorting + crowding calculation), NSGA-II will select a portion of elite individuals in the current population and merge them with the offspring individuals generated by crossover and mutation operations to form the parent candidate set of the next generation population.

[0052] During population update, the genetic operator performs binary or simulated binary crossover on selected parents, proportionally exchanging the values ​​of their design variables. Simultaneously, Gaussian mutation is performed on some genes within certain chromosomes with a small probability, breaking local optima and introducing new solutions. Each round of genetic recombination and mutation yields a new batch of candidate individuals, which are again evaluated for fitness using the neural network model. A new non-dominated sorting and crowding screening process is then performed, iteratively repeating until a pre-defined termination condition (such as reaching the maximum number of generations or the overall population improvement rate falling below a threshold) is met. Throughout the evolutionary process, all individuals identified as the first frontier in any generation's non-dominated sorting accumulate to form a set of solutions that continuously evolves and increasingly approaches the true Pareto optimal frontier.

[0053] When using the non-dominated sorting genetic algorithm to solve the multi-objective optimization model, it specifically includes: and using the trained back-propagation neural network model as a fitness evaluation tool, performing population initialization, non-dominated sorting, congestion calculation, crossover and mutation genetic operations during the solution, and setting the population size to 100, the crossover probability to 0.8, and the Gaussian mutation strategy during the operation, iterating the calculation until the termination condition is met, and generating the Pareto optimal solution set including the trade-off relationship between transmission efficiency and noise performance.

[0054] At the end of the algorithm iteration, the design variable combinations corresponding to the individuals marked as Pareto-optimal represent the optimal compromise between transmission efficiency and noise performance, where one cannot be improved without sacrificing the other. Designers can then select a specific solution that prioritizes efficiency or low noise based on their actual needs, or use a further decision matrix to score and rank these candidate designs. Because the backpropagation neural network has learned the mapping between CFD simulations and actual performance during the training phase, it can quickly and accurately provide performance predictions during the NSGA-II iterative search process, ensuring the authenticity of the solution set while significantly reducing computational time throughout the multi-objective optimization process.

[0055] In step S106, based on the combination of torque converter design variables obtained from the Pareto optimal solution set, a solution is determined and a model is reconstructed to obtain the optimized torque converter.

[0056] After completing the multi-objective genetic algorithm iterations and obtaining the Pareto-optimal solution set, one or more representative combinations of design variables need to be selected. Based on this, the selected design variable values ​​(i.e., the impeller blade outflow angle, the guide channel input angle, the guide channel output angle, the turbine blade thickness and curvature, and the relative position of the impeller and turbine) are individually inserted into the original geometric parameterized model to regenerate a detailed 3D CAD model. This model is constructed entirely according to the numerical parameters in the optimal solution set, so that its geometry, blade profile, and component layout correspond to the point or points determined during the optimization process to provide the best compromise between transmission efficiency and noise performance.

[0057] The reconstructed geometric model is imported into the meshing module, using the same mesh generation strategy as before (for example, preserving the boundary layer mesh thickness and ensuring that the mesh density at the leading and trailing edges of the blades meets the requirement of y+<1) to prepare for subsequent verification simulations. A high-precision transient simulation of the reconstructed model is then performed in the CFD solver, using the SST turbulence model to capture detailed features such as blade boundary layer separation and corner return vortices. Boundary conditions and solution parameters (such as inlet velocity, rotational speed, simulation time step, convergence residual, etc.) are set consistent with the training data. Through this relatively complete CFD simulation, the actual transmission efficiency η_eff_opt and noise level N_noise_opt for this combination can be obtained to verify whether they are consistent with the values ​​predicted by the BPNN model in the genetic algorithm.

[0058] If the verification results show that the prediction errors are within acceptable limits (e.g., efficiency deviation <2% and noise deviation <1dB), the design combination is confirmed as the true optimal solution. Otherwise, adjacent solutions can be selected from nearby Pareto frontier points for re-simulation, or the BPNN model can be fine-tuned and searched again. When the error reaches the preset standard, the optimized torque converter geometry is finalized. At this point, the torque converter's impeller blade outlet angle, stator inlet / outlet angles, turbine blade geometry, and the relative impeller-turbine position have all been quantified and packaged into a complete CAD file.

[0059] At this point, the entire "solution determination and model reconstruction" process has completed the conversion and verification from the Pareto optimal solution to the actual geometric model. The optimized torque converter not only retains the structural strength and assembly space required by the original design, but also achieves the dual goals of maximizing transmission efficiency and minimizing noise levels without compromising reliability through joint verification through numerical simulation and machine learning. The final CAD file can then be used in downstream manufacturing processes such as mold design, blade stamping, or CNC milling programming for mass production and vehicle installation verification. In this way, the feasibility and engineering feasibility of the adaptive multi-objective optimization results are guaranteed through the "model reconstruction-high-precision verification" step.

[0060] In another embodiment, different from the above, after obtaining the optimized torque converter, the three-dimensional flow field simulation of the torque converter is performed again based on computational fluid dynamics. During the simulation, the Taguchi orthogonal test method is used to generate initial sample points, and random sampling is combined to supplement additional sample points to form a comprehensive sample set covering the entire design space. The comprehensive sample set is used to construct training data for the back propagation neural network model.

[0061] After obtaining the optimized design parameter combination for the torque converter, the final geometric scheme is subjected to another 3D flow field simulation to verify the new model's performance under real-world operating conditions and to expand the training dataset. Specifically, the impeller blade outflow angle, guide channel inlet / outlet angle, turbine blade geometry, and the relative position of the impeller and turbine are each constrained to a number of discrete levels based on the optimized design variable range and adjacent space. A representative set of initial sample points is then selected using the Taguchi orthogonal array method. For example, if three values ​​are taken for each parameter level, an orthogonal array containing 27 test combinations can be constructed. Next, to prevent the sample distribution from being too regular due to reliance solely on the orthogonal scheme, additional sample points are generated through random sampling within the same parameter space. This ensures that the tested design points cover not only the typical high, low, and medium range combinations in the orthogonal array, but also some random sampling between the two ranges, resulting in a comprehensive sample set totaling tens to hundreds of points. For each sample combination, the corresponding 3D geometry is reconstructed and meshed in a CFD solver, maintaining the same boundary conditions and solution accuracy as in the previous validation phase (e.g., using the SST turbulence model, time step, and convergence threshold). This allows for more accurate flow field characteristics, including local pressure distribution, velocity vector field, turbulent viscosity distribution, and ultimately, transmission efficiency and noise levels. Simulation results for all samples are tabulated to form a training set containing a mapping of design variables to transmission efficiency and noise. Based on this training set, the back-propagation neural network (BPNN) is retrained or fine-tuned to achieve higher prediction accuracy on the newly obtained sample distribution. The orthogonal design ensures systematic sampling of parameter main and interaction effects, while random sampling enhances the model's sensitivity to potential nonlinearities at the boundaries and intermediate regions. The resulting comprehensive sample set captures both local variations near the optimal design and trends across the entire design space. This provides the BPNN model with richer and more uniform training data, further improving its overall accuracy and robustness in subsequent prediction and optimization iterations.

[0062] In one embodiment, during the solution of the multi-objective optimization model, the objective function is weighted and synthesized using a weighted summation method to form an initial optimization population, wherein the objective function based on the weighted summation method is expressed as: ; in, is the weighted comprehensive objective function, represents the transmission efficiency objective function, i.e. the above-mentioned η_eff(x), Represents the noise objective function, that is, N_noise(x) mentioned above, 、 are the weight coefficients of transmission efficiency and noise level, respectively, and satisfy the relationship + =1; The optimization result obtained by the weighted summation method is input as the initial population into the non-dominated sorting genetic algorithm for secondary optimization to obtain the Pareto optimal solution set.

[0063] In this formula, by introducing the noise level as a negative term into the comprehensive objective, the dual goals of "improving efficiency and reducing noise" are achieved in a single function. If you want to measure the noise suppression effect with a positive value, you can also first normalize the noise level and take the opposite number, and then use it directly in the weighted summation formula. , when η_eff(x) increases (efficiency improves), F(x) increases; when N_noise(x) decreases, −N_noise(x) increases, and thus F(x) also increases; the needs of improving transmission efficiency and reducing noise can be simultaneously considered in the process of maximizing F(x).

[0064] Next, according to the pre-set weight coefficient and Within the constraints of the design variables (for example, the outflow angle of the impeller blade is between 15° and 45°, the inlet angle of the guide wheel flow channel is between 20° and 60°, the outlet angle of the guide wheel flow channel is between 5° and 35°, the thickness and curvature of the turbine blade are within a reasonable range, and the relative position is within the limited axial spacing and radial eccentricity range), a number of "design variable vectors" are generated randomly or according to rules. "The candidate solution can be calculated by corresponding to the gridded CFD geometric model or directly calling the trained BPNN model. After substituting it into the above weighted summation formula, the comprehensive target value F(x) of the candidate solution can be obtained.

[0065] By simultaneously calculating F(x) for a large number of candidate solutions, all individuals can be ranked from largest to smallest based on numerical magnitude, and a certain percentage of the top-ranked solutions (such as the top 30% or top 50%) can be selected as the initial parent population. This ensures that the next stage of genetic iteration will begin searching for solutions that are relatively balanced in terms of efficiency and noise. This avoids both the uneven quality of the initial population caused by relying entirely on random generation and the problem of uncontrolled noise caused by directly focusing on a single goal (such as maximizing efficiency).

[0066] After obtaining this initial population, based on the weighted summation results and exhibiting a first-round trade-off, it is fed into the non-dominated sorting genetic algorithm (NSGA-II) for secondary optimization. Within NSGA-II, the fitness evaluation of each individual still relies on the original values ​​of the two objectives, transmission efficiency and noise level (which can be quickly predicted by the BPNN model), rather than the merged F(x) value. NSGA-II first performs a non-dominated hierarchical classification on the initial population, classifying solutions that cannot be simultaneously surpassed by other individuals in terms of "increased efficiency without increasing noise" or "reduced noise without decreasing efficiency" as the first frontier. The remaining individuals are then sequentially stratified. Next, individuals with high diversity within the same hierarchical level are selected based on crowding calculations, ensuring that the next generation of the selected population is evenly distributed in the target space. Through multiple iterations of crossover, mutation, non-dominated sorting, and crowding selection, NSGA-II continuously expands and approaches the true Pareto frontier optimal solution.

[0067] Ultimately, the set of individuals on the Pareto frontier output by NSGA-II constitutes the set of solutions that achieve the optimal trade-off between transmission efficiency and noise level. This demonstrates that the weighted summation method provides the genetic algorithm with a set of initial, high-quality compromise solutions that balance efficiency and noise while guiding NSGA-II to conduct a global search at more representative locations, effectively improving the algorithm's convergence speed and the diversity of multi-objective solutions.

[0068] In one embodiment, in the three-dimensional flow field simulation of the torque converter based on computational fluid dynamics, the obtained multi-dimensional flow field output data is subjected to tensor preprocessing to reduce the data dimension and extract key features. The preprocessing includes normalizing the pressure distribution and the velocity field, filtering out numerical noise, and calculating field parameters that characterize the overall performance; the field parameters include a flow velocity ratio obtained by area averaging the velocity field distribution, and the flow velocity ratio is used to represent the overall velocity gain effect of the flow field.

[0069] After completing a three-dimensional flow field simulation, a tensor is generated, consisting of multidimensional information such as pressure, velocity components, and turbulent viscosity at countless grid nodes. Directly using this high-dimensional, massive grid data for subsequent machine learning or optimization is not only computationally intensive but also prone to introducing redundancy and noise. Therefore, preprocessing is necessary to reduce the data dimensionality while extracting the key features that best reflect overall performance. First, the pressure distribution and velocity field are normalized. Normalization maps the pressure or velocity value at each grid node to a fixed numerical range (for example, 0 to 1 or -1 to 1). A common approach is to subtract the minimum value of the variable in the entire field from the original value and then divide it by the difference between the maximum and minimum values. This approach eliminates the effects of different physical quantities (for example, pressure may range from tens of megapascals, while velocity may range from ten meters per second). This prevents machine learning algorithms from biasing the processing of individual features due to scale differences. It also ensures numerical stability in subsequent filtering operations, preventing overflow or gradient explosion caused by extreme values. After normalization, the inevitable numerical noise in the simulation data must be filtered out. Numerical noise typically arises from factors such as uneven meshing, suboptimal iterative convergence, and solver roundoff errors, resulting in localized spikes or random jitter that do not conform to physical laws. Directly feeding this noise into the neural network will slow down model training convergence and reduce prediction accuracy. Therefore, a low-pass filter or local smoothing algorithm, such as a moving average, two-dimensional Gaussian blur, or a median filter based on a neighborhood mean, can be applied to each physical field (e.g., pressure field, velocity field). This method can suppress local outliers while preserving large-scale flow characteristics (such as the mainstream velocity distribution and pressure gradient direction), making subsequent feature extraction more stable and reliable. After completing the above normalization and noise filtering, the entire three-dimensional field tensor still needs to be further reduced in dimensionality, condensing millions of grid data points into a few scalars or low-dimensional vectors that directly reflect the overall performance. One effective approach is to calculate the flow velocity ratio, a field parameter that represents the "overall velocity gain effect" of the entire flow field. This parameter is defined as: first, taking the area-weighted average of the velocity vectors of all grid cells or sections of interest (such as the impeller blade outlet or turbine inlet) to obtain an average velocity value; then, the ratio of this average velocity to the inlet flow velocity (the wind speed or oil flow velocity preset in the simulation boundary conditions) is used as the final "flow velocity ratio" indicator. Its calculation formula can be expressed as: ; Where v(x,y,z) represents the vector velocity at any point on a certain cross section in the flow field; S is the area of ​​the cross section, dS is the area element; As is the total area of ​​the cross section; The nominal value of the inlet flow velocity. This indicator reflects whether the overall flow velocity in the pump to turbine or guide wheel to pump wheel circuit has been improved (when When >1, it means that the overall velocity of the fluid increases after being pressurized or directed backflow. <1, indicating energy loss during the flow, resulting in a decrease in velocity. Using the velocity ratio as a scalar field instead of the original three-dimensional velocity tensor not only significantly reduces the dimensionality of subsequent machine learning input features but also makes it easier for the neural network to learn "which geometric parameter combination achieves the greatest kinetic energy increase for the entire flow field," thereby further guiding the optimization algorithm to prioritize design solutions with high velocity ratios, indicating potential increases in transmission efficiency.

[0070] After the aforementioned preprocessing steps, the originally lengthy multidimensional field data is compressed into a few dimensionless key features that directly reflect "overall performance," such as flow velocity ratios, normalized mean pressure values, or scalar indicators representing turbulence intensity. These features form the input vectors for the subsequent neural network, while the corresponding outputs are transmission efficiency and noise levels. This approach preserves the most important physical information in the simulation results while significantly reducing the computational effort required for neural network training and prediction, improving the model's stability and convergence speed.

[0071] In one embodiment, when solving the multi-objective optimization model, the method further includes: An optimized coupling mechanism for hydraulic transmission parameters is established by deriving a mathematical relationship between the specific speed ratio and the torque ratio of the torque converter, wherein the specific speed ratio is defined as the ratio of the rotational speed of the torque converter turbine to the rotational speed of the pump impeller, and the torque ratio is defined as the ratio of the output torque of the turbine to the input torque of the pump impeller; A performance characteristic curve of the torque converter is constructed based on a relationship between the specific speed ratio and the torque ratio, and the hydraulic transmission parameters are combined with an objective function in the multi-objective optimization model.

[0072] Specifically, by analyzing the internal fluid mechanics of the torque converter, the mathematical relationship between the speed ratio and the torque ratio can be derived and integrated into the multi-objective optimization model to achieve the coupling coordination of the hydraulic transmission parameters with the efficiency and noise objective functions. First, the speed ratio SR is defined as the turbine speed. and pump impeller speed Ratio: ; The ratio of turbine speed to pump speed directly reflects the switching between "unloading" and "loading" states during the fluid circulation inside the torque converter, and has a decisive influence on the torque output and flow field stability. TR is the turbine output torque and the pump input torque Ratio: ; This ratio represents the gain in energy transferred between the impeller and turbine—that is, the ratio of mechanical torque received by the turbine to the input torque at the impeller. When TR > 1, it indicates additional torque gain during the hydraulic cycle; if TR < 1, the system is slipping or dissipating energy.

[0073] Based on the above definition, the functional relationship between the specific speed ratio and the torque ratio can be derived from the theoretical mechanism, which can generally be expressed as: ; in, is the relative geometric parameters of the pump and turbine blades (such as the difference between the blade outlet outflow angle and the inlet angle of attack), is the Reynolds number of the fluid or the shape parameter of the flow channel section. This functional relationship is obtained through theoretical analysis or CFD discrete sample fitting, and a "performance characteristic curve" representing the overall performance characteristics can be drawn under different speed and pressure conditions. Taking the typical output torque-speed curve as an example, when the pump wheel speed is When fixed, it can be varied within a certain range, and the corresponding TR value can be determined through simulation or experimental measurement, thereby plotting an "SR-to-TR" curve. This curve typically exhibits a peak characteristic that first increases and then decreases. The SR at the peak corresponds to the torque converter's optimal torque-increasing condition. When constructing a multi-objective optimization model, the aforementioned "speed ratio-torque ratio" relationship is introduced into the objective function or as an additional constraint to achieve coupling between hydraulic transmission performance and other geometric parameters. For example, a preference expression for TR can be added to the loss function or fitness function: ; in, Indicates the maximum torque ratio acceptable under design conditions, is its weight coefficient. When the individual SR makes TR close to When , the additional term approaches zero, otherwise it will lower the comprehensive target value, prompting the algorithm to give priority to geometric solutions that can obtain high torque-increasing performance. Alternatively, SR and TR can be directly used as constraints: ; This ensures that the optimized design will not fall into the invalid sliding zone and can also ensure sufficient torque-increasing capacity under most working conditions. By taking the "speed ratio-torque ratio" coupling characteristics and geometric parameters as the objective function or constraint conditions, the optimization algorithm can fully consider the inherent relationship between hydraulic efficiency, noise performance and torque-increasing capacity during the search process. For example, in the fitness evaluation link of NSGA-II, in addition to the original and , TR(x) is also predicted (with the help of BPNN model or simplified mathematical formula) to generate a three-dimensional target vector This allows torque-increasing capability to be factored into subsequent non-dominated sorting and congestion screening, resulting in a Pareto solution that balances transmission efficiency and low noise while meeting the required torque-increasing level. Ultimately, the selected optimization solution corresponds to a geometric parameter combination that achieves good turbine and impeller speed matching, significantly improved torque output, and suppressed noise, significantly improving the torque converter's overall performance under actual operating conditions.

[0074] Based on the same idea, Figure 7 The figure shows a block diagram of a hydraulic transmission power chain optimization chain system according to an embodiment of the present invention. The system comprises: a three-dimensional modeling module 201 for constructing a parameterized geometric model of a hydraulic torque converter, wherein the impeller blade outflow angle, the guide wheel flow channel input angle, the guide wheel flow channel output angle, the turbine blade geometric parameters, and the relative positions of the impeller and the turbine are used as design variables; a simulation module 202 for performing a three-dimensional flow field simulation of the hydraulic torque converter based on computational fluid dynamics, using the k-ε turbulence model for rapid analysis of the initial solution, and using the shear stress transmission turbulence model for transient simulation, to obtain simulation data of the hydraulic torque converter under different combinations of the design variables, wherein the simulation data includes flow field characteristics, pressure distribution, velocity field distribution, turbulence intensity, transmission efficiency, and corresponding noise values; The machine learning module 203 is used to establish a training set using the simulation data and train a back propagation neural network model; wherein, the input layer nodes of the back propagation neural network model are the various design variables of the torque converter, the hidden layer adopts at least one layer, each layer contains 8 neurons and adopts a hyperbolic tangent S-type activation function, the output layer nodes are the transmission efficiency and noise prediction value of the torque converter, the output layer adopts a linear activation function, and the back propagation neural network model adopts the Levenberg-Marquardt algorithm for back propagation training; the mathematical modeling module 204 is used to construct a multi-objective optimization model, and the multi-objective optimization model is used to maximize the transmission efficiency of the torque converter. rate and minimized noise level as the objective function, and setting the constraints of the design variables, wherein the constraints include the outflow angle of the impeller blade, the input and output angles of the guide wheel flow channel, and the limited value range of the turbine blade geometric parameters; a calculation module 205 is used to solve the multi-objective optimization model using a non-dominated sorting genetic algorithm, and use the trained back propagation neural network model as a fitness function to evaluate the target value of each candidate design to obtain a Pareto optimal solution set that includes a trade-off relationship between transmission efficiency and noise performance; a reconstruction module 206 is used to determine a scheme and reconstruct a model based on the combination of torque converter design variables obtained in the Pareto optimal solution set to obtain the optimized torque converter.

[0075] The specific details of the above system have been described in detail in the implementation method part. For undisclosed details, please refer to the implementation method part, and thus will not be repeated here.

[0076] This system introduces a back-propagation neural network to model and predict the transmission efficiency and noise level of the torque converter under different design parameters, avoiding the repeated high-cost fluid dynamics simulation calculations during the optimization iteration process, significantly reducing computing resource consumption, and improving design efficiency; using a non-dominated sorting genetic algorithm for multi-objective search, it can simultaneously optimize the two performance indicators of transmission efficiency and noise control, obtain a set of balanced Pareto optimal solutions, and break through the technical bottleneck of traditional methods that are difficult to compromise between efficiency and noise; a combination of Taguchi orthogonal experiments and random sampling is used to construct training samples, covering the design variable space, improving the generalization ability of the neural network model, and being able to adapt to the complex and changeable torque converter design requirements and output a more applicable optimal parameter combination; by setting weight factors, fitness function structures and iterative control parameters, the optimization target proportion can be flexibly adjusted according to the performance focus of the vehicle, realizing customized parameter optimization, and improving the overall performance of the power chain system and the level of design intelligence.

[0077] Based on the same idea, the embodiment of this specification also provides a hydraulic transmission power chain optimization chain device, such as Figure 8 shown.

[0078] The hydraulic transmission power chain optimization chain device may be the terminal device or server provided in the above embodiment.

[0079] A hydraulic transmission powertrain optimization device can vary significantly depending on its configuration or performance. It may include one or more processors 301, memory 302, and a bus. Memory 302 may store one or more applications or data. Memory 302 may include readable media in the form of volatile storage units, such as random access memory (RAM) and / or cache memory, such as plug-in removable hard drives, Smart Media Cards (SMCs), Secure Digital (SD) cards, or Flash Cards found in electronic devices. It may also include read-only storage units. Applications stored in memory 302 may include one or more program modules (not shown). Such program modules include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include the implementation of a network environment. Furthermore, processor 301 may be configured to communicate with memory 302 to execute a series of computer-executable instructions stored in memory 302 on the hydraulic transmission powertrain optimization device. The hydraulic transmission power chain optimization chain device may also include one or more power supplies 303, one or more wired or wireless network interfaces 304, one or more I / O interfaces (input and output interfaces) 305, and one or more external devices 306 (e.g., a keyboard). It may also communicate with one or more devices that enable a user to interact with the device, and / or with any device that enables the device to communicate with one or more other computing devices (e.g., a router, a network switch, etc.). Such communication may be performed via the I / O interface 305. Furthermore, the device may also communicate with one or more networks (e.g., a local area network (LAN)) via the wired or wireless interface 304.

[0080] In some embodiments, the processor 301 may be comprised of an integrated circuit, such as a single packaged integrated circuit or multiple packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips. The processor 301 is the control core (Control Unit) of the electronic device, connecting the various components of the electronic device using various interfaces and circuits. It executes programs or modules stored in the memory 11 (e.g., a program for optimizing the hydraulic transmission power train), and accesses data stored in the memory, thereby performing various functions of the processing device and processing data.

[0081] The bus may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The bus is configured to enable connection and communication between the memory 302 and at least one processor 301, etc.

[0082] The power supply can be logically connected to the at least one processor 301 via a power management device, thereby implementing functions such as charging management, discharging management, and power consumption management through the power management device. The power supply can also include one or more DC or AC power supplies, a recharging device, a power failure detection circuit, a power converter or inverter, a power status indicator, and other arbitrary components. The electronic device 1 can also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which are not further described here.

[0083] Optionally, the processing device may further include a user interface, which may be a display. Optionally, the user interface may also be a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touch screen. The display may also be appropriately referred to as a display screen or a display unit, which is used to display information processed in the electronic device 1 and to display a visual user interface. Figure 8 Only the hydraulic transmission power chain optimization chain device with components is shown, and it can be understood by those skilled in the art that Figure 8The structure shown does not constitute a limitation on the hydraulic transmission power chain optimization chain device, and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0084] Specifically in this embodiment, the hydraulic transmission power chain optimization device includes a memory and one or more programs, wherein the one or more programs are stored in the memory, and the one or more programs may include one or more modules, and each module may include a series of computer-executable instructions for the hydraulic transmission power chain optimization device, and the one or more programs are configured to be executed by one or more processors, including computer-executable instructions for performing the following: A parametric geometric model of the torque converter is constructed, in which the impeller blade outflow angle, the guide wheel flow channel input angle, the guide wheel flow channel output angle, the turbine blade geometric parameters, and the relative position of the impeller and turbine are used as design variables; Performing a three-dimensional flow field simulation of the torque converter based on computational fluid dynamics, using a k-ε turbulence model for rapid analysis of an initial solution, and using a shear stress transmission turbulence model for transient simulation, to obtain simulation data of the torque converter under different combinations of the design variables, the simulation data including flow field characteristics, pressure distribution, velocity field distribution, turbulence intensity, transmission efficiency, and corresponding noise values; A training set is established using the simulation data to train a back-propagation neural network model; wherein the input layer nodes of the back-propagation neural network model are the design variables of the torque converter, the hidden layer uses at least one layer, each layer contains 8 neurons and uses a hyperbolic tangent sigmoid activation function, the output layer nodes are the transmission efficiency and noise prediction values ​​of the torque converter, the output layer uses a linear activation function, and the back-propagation neural network model uses a Levenberg-Marquardt algorithm for back-propagation training; Constructing a multi-objective optimization model, wherein the multi-objective optimization model takes maximizing the transmission efficiency of the torque converter and minimizing the noise level as objective functions, and sets constraints on the design variables, wherein the constraints include limited value ranges of the impeller blade outflow angle, the guide wheel flow channel input and output angles, and the turbine blade geometric parameters; Solving the multi-objective optimization model using a non-dominated sorting genetic algorithm, using the trained back propagation neural network model as a fitness function to evaluate the objective value of each candidate design, and obtaining a Pareto optimal solution set that includes a trade-off relationship between transmission efficiency and noise performance; Based on the combination of torque converter design variables obtained from the Pareto optimal solution set, a solution is determined and a model is reconstructed to obtain the optimized torque converter.

[0085] Based on the same idea, the exemplary embodiments of the present invention further provide a computer-readable storage medium storing a program product capable of implementing the methods described above. In some possible implementations, various aspects of the present disclosure may also be implemented in the form of a program product comprising program code. When the program product is executed on a terminal device, the program code is configured to cause the terminal device to execute the steps described in the "Exemplary Methods" section above according to various exemplary embodiments of the present disclosure.

[0086] refer to Figure 9 As shown, a program 400 for implementing the above method according to an exemplary embodiment of the present disclosure is described. The program 400 may be a portable compact disc read-only memory (CD-ROM) and include program code, and may be run on a terminal device, such as a personal computer. However, the program product of the present disclosure is not limited thereto. In this document, a readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0087] The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0088] A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0089] Program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, CSS, HTML, and the like, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a standalone software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0090] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solution according to the embodiments of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the exemplary embodiment of the present disclosure.

[0091] Furthermore, the figures above are merely illustrative of the processes included in the methods according to exemplary embodiments of the present disclosure and are not intended to be limiting. It is readily understood that the processes illustrated in the figures above do not indicate or limit the temporal order of these processes. Furthermore, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0092] It should be noted that although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the exemplary embodiments of the present disclosure, the features and functions of two or more modules or units described above can be concretized in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into multiple modules or units to be concretized.

[0093] Other embodiments of the present disclosure will readily occur to those skilled in the art after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art not disclosed herein. The description and embodiments are to be considered as exemplary only, with the true scope and spirit of the present disclosure being indicated by the claims.

[0094] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes can be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A hydraulic transmission power chain optimization method, characterized in that: The method comprises: A parametric geometric model of the torque converter is constructed, in which the impeller blade outflow angle, the guide wheel flow channel input angle, the guide wheel flow channel output angle, the turbine blade geometric parameters, and the relative position of the impeller and turbine are used as design variables; Performing a three-dimensional flow field simulation of the torque converter based on computational fluid dynamics, using a k-ε turbulence model for rapid analysis of an initial solution, and using a shear stress transmission turbulence model for transient simulation, to obtain simulation data of the torque converter under different combinations of the design variables, the simulation data including flow field characteristics, pressure distribution, velocity field distribution, turbulence intensity, transmission efficiency, and corresponding noise values; A training set is established using the simulation data to train a back-propagation neural network model; wherein the input layer nodes of the back-propagation neural network model are the design variables of the torque converter, the hidden layer uses at least one layer, each layer contains 8 neurons and uses a hyperbolic tangent sigmoid activation function, the output layer nodes are the transmission efficiency and noise prediction values ​​of the torque converter, the output layer uses a linear activation function, and the back-propagation neural network model uses a Levenberg-Marquardt algorithm for back-propagation training; Constructing a multi-objective optimization model, wherein the multi-objective optimization model takes maximizing the transmission efficiency of the torque converter and minimizing the noise level as objective functions, and sets constraints on the design variables, wherein the constraints include limited value ranges of the impeller blade outflow angle, the guide wheel flow channel input and output angles, and the turbine blade geometric parameters; Solving the multi-objective optimization model using a non-dominated sorting genetic algorithm, using the trained back propagation neural network model as a fitness function to evaluate the objective value of each candidate design, and obtaining a Pareto optimal solution set that includes a trade-off relationship between transmission efficiency and noise performance; Based on the combination of torque converter design variables obtained from the Pareto optimal solution set, a solution is determined and a model is reconstructed to obtain the optimized torque converter.

2. The hydraulic transmission power chain optimization method according to claim 1, characterized in that: The design variable value range of the parameterized geometric model of the torque converter specifically includes: the outflow angle of the impeller blade is 15°~45°, the input angle of the guide wheel flow channel is 20°~60°, the output angle of the guide wheel flow channel is 5°~35°, and the turbine blade geometric parameters include blade thickness and flow channel curvature radius.

3. The hydraulic transmission power chain optimization method according to claim 1, characterized in that: After obtaining the optimized torque converter, the three-dimensional flow field simulation of the torque converter is performed again based on computational fluid dynamics. During the simulation, the Taguchi orthogonal test method is used to generate initial sample points, and additional sample points are supplemented by random sampling to form a comprehensive sample set covering the entire design space. The comprehensive sample set is used to construct training data for the back propagation neural network model.

4. The hydraulic transmission power chain optimization method according to claim 1, characterized in that: The genetic operation of the non-dominated sorting genetic algorithm includes: after the initial population is randomly generated, the priority of the population individuals is determined by non-dominated sorting, the elite individuals are screened by using the crowding distance, and the new generation population is generated for the selected individuals through the intermediate crossover and Gaussian mutation strategies, and the maximum number of genetic iterations is set to 300 generations.

5. The hydraulic transmission power chain optimization method according to claim 1, characterized in that: When solving the multi-objective optimization model using a non-dominated sorting genetic algorithm, the method specifically includes: The trained back-propagation neural network model is used as a fitness evaluation tool to perform population initialization, non-dominated sorting, crowding calculation, crossover and mutation genetic operations during the solution. During the operation, the population size is set to 100, the crossover probability is 0.8, and the Gaussian mutation strategy is adopted. The iterative calculation is performed until the termination condition is met to generate the Pareto optimal solution set that includes the trade-off relationship between transmission efficiency and noise performance.

6. The hydraulic transmission power chain optimization method according to claim 5, characterized in that: In the process of solving the multi-objective optimization model, the objective function is weighted and synthesized using the weighted summation method to form an initial optimization population, wherein the objective function based on the weighted summation method is expressed as: ； in, is the weighted comprehensive objective function, represents the transmission efficiency objective function, represents the noise objective function, 、 are the weight coefficients of transmission efficiency and noise level, respectively, and satisfy the relationship + =1; The optimization result obtained by the weighted summation method is input as the initial population into the non-dominated sorting genetic algorithm for secondary optimization to obtain the Pareto optimal solution set.

7. The hydraulic transmission power chain optimization method according to claim 1, characterized in that: In a computational fluid dynamics-based three-dimensional flow field simulation of the torque converter, tensor preprocessing is performed on the obtained multi-dimensional flow field output data to reduce the data dimension and extract key features. The preprocessing includes normalizing the pressure distribution and the velocity field, filtering out numerical noise, and calculating field parameters that characterize the overall performance. The field parameters include a flow velocity ratio obtained by performing area averaging on the velocity field distribution, and the flow velocity ratio is used to represent the velocity gain effect of the entire flow field.

8. The hydraulic transmission power chain optimization method according to claim 1, characterized in that: When solving the multi-objective optimization model, the method further includes: An optimized coupling mechanism for hydraulic transmission parameters is established by deriving a mathematical relationship between the specific speed ratio and the torque ratio of the torque converter, wherein the specific speed ratio is defined as the ratio of the rotational speed of the torque converter turbine to the rotational speed of the pump impeller, and the torque ratio is defined as the ratio of the output torque of the turbine to the input torque of the pump impeller; A performance characteristic curve of the torque converter is constructed based on a relationship between the specific speed ratio and the torque ratio, and the hydraulic transmission parameters are combined with an objective function in the multi-objective optimization model.

9. A hydraulic transmission power chain optimization chain system, the system comprising: A 3D modeling module is used to construct a parametric geometric model of the torque converter, in which the impeller blade outflow angle, the guide wheel flow channel input angle, the guide wheel flow channel output angle, the turbine blade geometry parameters, and the relative position of the impeller and turbine are used as design variables; a simulation module for performing a three-dimensional flow field simulation of the torque converter based on computational fluid dynamics, using a k-ε turbulence model for rapid analysis of an initial solution, and using a shear stress transmission turbulence model for transient simulation, to obtain simulation data of the torque converter under different combinations of the design variables, the simulation data including flow field characteristics, pressure distribution, velocity field distribution, turbulence intensity, transmission efficiency, and corresponding noise values; a machine learning module, configured to establish a training set using the simulation data and train a back-propagation neural network model; wherein the input layer nodes of the back-propagation neural network model are the design variables of the torque converter, the hidden layer has at least one layer, each layer includes 8 neurons and uses a hyperbolic tangent sigmoid activation function, the output layer nodes are the transmission efficiency and noise prediction values ​​of the torque converter, the output layer uses a linear activation function, and the back-propagation neural network model uses a Levenberg-Marquardt algorithm for back-propagation training; a mathematical modeling module for constructing a multi-objective optimization model, wherein the multi-objective optimization model takes maximizing the transmission efficiency of the torque converter and minimizing the noise level as objective functions, and sets constraints on the design variables, wherein the constraints include limited value ranges of the impeller blade outflow angle, the guide wheel flow channel input and output angles, and the turbine blade geometric parameters; a calculation module, configured to solve the multi-objective optimization model using a non-dominated sorting genetic algorithm, and use the trained back propagation neural network model as a fitness function to evaluate the objective value of each candidate design, thereby obtaining a Pareto optimal solution set that includes a trade-off relationship between transmission efficiency and noise performance; A reconstruction module is used to determine a solution and reconstruct a model based on the combination of torque converter design variables obtained from the Pareto optimal solution set to obtain the optimized torque converter.

Images