Method and device for generating electric drive speed reducer scheme based on multi-target genetic algorithm

Abstract

The invention discloses a method and a device for generating an electric drive speed reducer scheme based on a multi-target genetic algorithm. The method comprises the steps of obtaining a design variable of an electric drive speed reducer of the new energy vehicle; performing iterative optimization on the design scheme of the electric-driven speed reducer by using a multi-target genetic algorithm to obtain a candidate design scheme of the electric-driven speed reducer, and in the iterative optimization process, performing grading processing on variables according to the relevance strength to obtain a candidate design scheme of the electric-driven speed reducer; a three-level fitness function based on matrix operation is used for carrying out layered sorting on constraint conditions of all levels of the fitness function, a gradient penalty coefficient is used for carrying out step-by-step reduction on a target function value, so that loss of excellent populations is prevented through a step-by-step adaptation mechanism, and the situation that an error result is output or local optimum is caused due to premature convergence is avoided; and selecting a target design scheme conforming to the Pareto optimal solution of the electric drive speed reducer from the candidate design schemes. The technical problem that the design of the electric drive speed reducer in the related technology does not have a scheme for optimizing multiple design targets is solved.

Description

Technical Field

[0001] This application relates to the field of new energy vehicle technology, and more specifically, to a method and apparatus for generating an electric drive reducer scheme based on a multi-objective genetic algorithm. Background Technology

[0002] The current development of new energy reducers is characterized by numerous design parameters and a very wide range of variables across these dimensions. Based on this, millions of gear, bearing, and spatial arrangement schemes exist to meet design requirements. Manually searching for suitable solutions only allows for a limited number of trials within a small area covered by experience, and it's impossible to iterate and arrive at the optimal target design solution in a short time. Currently, the development of new energy reducers is largely assisted by various professional simulation software. Existing simulation software typically builds models based on known, fixed parameters, and the reducer's design analysis function is divided into multiple steps or modules. Analysis and calculation of complex load spectra rely excessively on the sequential computation capabilities of computer CPUs and threads, requiring significant hardware resources and time. Effective integration is impossible in the early stages of design for specific purposes and application areas requiring optimization of multiple design objectives. Even using simple traditional algorithms to optimize a few variables cannot cover all design requirements across the entire electric drive performance dimension, often resulting in compromises. Summary of the Invention

[0003] This application provides a method and apparatus for generating electric drive reducer schemes based on multi-objective genetic algorithms, so as to at least solve the technical problem in the related art that the design of electric drive reducers does not have a scheme for multiple design objectives, full performance dimensions, and acceleration optimization.

[0004] According to one aspect of the embodiments of this application, a method for generating electric drive reducer schemes based on multi-objective genetic algorithms is provided, comprising: obtaining design variables of electric drive reducers for new energy vehicles, wherein the design variables are design parameters that meet the design requirements of electric drive reducers that meet the overall vehicle economy and power performance; using the design variables as input variables, iteratively optimizing the design schemes of the electric drive reducers using a multi-objective genetic algorithm to obtain candidate design schemes of the electric drive reducers, wherein, during the iterative optimization process, the variables are processed in a hierarchical manner according to the strength of correlation, a three-level fitness function based on matrix operations is used to hierarchically sort the constraints at each level contained in the fitness function of the multi-objective genetic algorithm, and a gradient penalty coefficient is used to progressively reduce the objective function value of the multi-objective genetic algorithm, so as to prevent the loss of excellent populations and avoid outputting incorrect results or getting trapped in local optima due to premature convergence through a gradual adaptation mechanism; and selecting a target design scheme that meets the Pareto optimal solution of the electric drive reducer from the candidate design schemes.

[0005] According to another aspect of the embodiments of this application, an apparatus for generating an electric drive reducer scheme based on a multi-objective genetic algorithm, corresponding to the above-described method for generating an electric drive reducer scheme based on a multi-objective genetic algorithm, is also provided; a computer-readable storage medium is also provided, the storage medium including a stored program, which executes the above-described method when the program is run; and an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the above-described method through the computer program.

[0006] In this embodiment, design variables for the electric drive reducer of a new energy vehicle are obtained. These design variables are design parameters that meet the design requirements of the electric drive reducer in terms of vehicle economy and power. Using these design variables as input variables, a multi-objective genetic algorithm is used to iteratively optimize the design scheme of the electric drive reducer, resulting in candidate design schemes. During the iterative optimization process, variables are processed in a hierarchical manner according to their correlation strength. A three-level fitness function based on matrix operations is used to hierarchically sort the constraints at each level contained in the fitness function of the multi-objective genetic algorithm. A gradient penalty coefficient is used to progressively reduce the objective function value of the multi-objective genetic algorithm, thereby preventing the loss of excellent populations and avoiding premature convergence leading to incorrect results or getting trapped in local optima. From the candidate design schemes, a target design scheme that meets the Pareto optimal solution of the electric drive reducer is selected, thus solving the technical problem in related technologies where the design of electric drive reducers does not address multiple design objectives, full performance dimensions, and accelerated optimization. Attached Figure Description

[0007] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0008] Figure 1 This is a flowchart of an optional method for generating an electric drive reducer scheme based on a multi-objective genetic algorithm according to an embodiment of this application;

[0009] Figure 2 This is a schematic diagram generated according to an optional electric drive reducer scheme based on a multi-objective genetic algorithm according to an embodiment of this application;

[0010] Figure 3 This is a schematic diagram of an optional multidimensional array data interaction logic according to an embodiment of this application;

[0011] Figure 4 This is a schematic diagram of an optional SN curve according to an embodiment of this application. Detailed Implementation

[0012] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0013] Due to the discrete and nonlinear nature of reducer parameters, the non-correlated nature of design variables, and the coupling of complex operating conditions, most optimization algorithms, when used alone, cannot achieve good optimization results. For example, the gradient descent method cannot handle discrete parameter problems (such as the number of teeth, speed ratio, NVH frequency avoidance, bearing selection, etc.) and multimodal functions; the Lagrange approximation method cannot handle complex problems with a large number of constraints; the Monte Carlo method requires calculating too many design schemes and cannot perform a full-dimensional search; for multimodal functions, the PSO algorithm is prone to getting trapped in local optima; due to the multidisciplinary dimensionality and manufacturing characteristics of reducer parameters, multi-objective genetic algorithms are prone to premature convergence under multiple constraints, and non-correlated variables are prone to losing excellent populations during crossover and mutation processes, failing to obtain normal solutions, etc.

[0014] Based on this, according to one aspect of the embodiments of this application, a method embodiment for generating electric drive reducer schemes based on multi-objective genetic algorithms is provided. Using an improved genetic algorithm, within the entire design parameter dimension and the entire design range, and taking the boundary conditions of electric drive design requirements, vehicle economy, and power performance as inputs, the Pareto principle for the target-oriented reducer design is sought. This algorithm uses international standards for gear and bearing design verification, combined with gear machining process standards and NVH design requirements, as the underlying logic for calculation, significantly reducing computation time while ensuring computational accuracy. To improve the convergence of the genetic algorithm, the algorithm of this invention uses a three-level fitness function based on matrix operations and gradient penalty coefficients. For each level of constraint conditions included in the fitness function of the genetic algorithm, the penalty coefficient is progressively reduced to guide the machine to establish a gradual adaptation mechanism, preventing the loss of excellent populations, avoiding premature convergence leading to incorrect output results, or getting trapped in local optima.

[0015] Figure 1 This is a flowchart of an optional method for generating an electric drive reducer scheme based on a multi-objective genetic algorithm according to an embodiment of this application, as shown below. Figure 1 As shown, the method may include the following steps:

[0016] Step S102: Obtain the design variables of the electric drive reducer for new energy vehicles. The design variables are design parameters that meet the design requirements of the electric drive reducer for vehicle economy and power.

[0017] This application uses a multi-objective genetic algorithm. The input variables are the design variables of the reducer. The coefficients involved in the underlying calculation of the optimization algorithm are stored in the environment variable file, which includes the overall scheme roadmap for electric drive development, such as oil cooling or water cooling, vehicle space dimensions, electric drive placement tilt angle, tire radius, maximum output torque required by the vehicle, several typical electric drive layouts (including parallel shaft system, planetary gear system, placement position of the parallel shaft intermediate shaft (up / down, left / right), gear helix direction, differential placement direction, distinguished by numbers, which can be directly mapped to the corresponding calculation logic when the subroutine is called), energy recovery strategy, oil pump speed control strategy, NVH design requirements, reliability coefficient, shaft and gear process route, materials used for parts, tooth surface roughness, gear accuracy grade, tooth thickness tolerance, safety clearance required for shaft and gear system layout, lubricating oil properties, reducer operating temperature, oil level and output shaft height difference, oil seal lip diameter, minimum safety factor for shaft and gear parts life, material SN curve, system iteration accuracy requirements, initial population size of the genetic algorithm, and maximum number of iterations of the genetic algorithm. Environmental coefficients are stored in a unified file that can be edited and used externally, and are repeatedly called by the main program or subroutines during the calculation process.

[0018] The parameter tables for various types of bearings and the quotations from bearing suppliers are stored in the bearing database Bearing_DataBase, which can be edited and used externally and repeatedly called by the main program or subroutines during the calculation process.

[0019] The CLTC operating point and load spectrum operating point represent the torque and speed requirements at the vehicle's output end. They can be converted to the input end based on the target speed ratio. The CLTC operating point and load spectrum operating point are stored in the CLTC operating point file and the load spectrum operating point file, respectively, and are repeatedly called by the main program or subroutines during the calculation process.

[0020] The Process_File contains the shaft and gear machining process type, tool axial movement length, system integration safety distance, axial space requirements for circlip and pressure plate installation, shaft welding space requirements, bearing material and heat treatment process requirements, bearing assembly shoulder height requirements, differential slotted shaft outer diameter, and additional process costs. The Process_File is repeatedly called by the main program or subroutines during the calculation process.

[0021] The Cost Database contains quotes from various parts suppliers. For parts that are related to materials and weight, the price is stored in units of yuan / kg.

[0022] Step S104: Using the design variables as input variables, a multi-objective genetic algorithm is used to iteratively optimize the design scheme of the electric drive reducer to obtain candidate design schemes for the electric drive reducer. During the iterative optimization process, a three-level fitness function based on matrix operations is used to hierarchically sort the constraints at each level contained in the fitness function of the multi-objective genetic algorithm, and a gradient penalty coefficient is used to gradually reduce the objective function value of the multi-objective genetic algorithm. This is to prevent the loss of excellent populations and avoid premature convergence leading to incorrect output results or getting trapped in local optima through a gradual adaptation mechanism.

[0023] 1) Using the design variables as input variables, initialize the population pop of the multi-objective genetic algorithm. The number of teeth, center distance, bearing type number and bearing model number in the population variables of the multi-objective genetic algorithm are discretized integers within the corresponding design range.

[0024] 2) Implementation of a three-level adaptation mechanism:

[0025] 2.1) In the first-level adaptation mechanism, for population pops that do not meet the population fitness conditions of the first-level fitness function, the penalty coefficient is returned in a progressively decreasing manner according to the population fitness conditions of the first-level fitness function. The second-level and third-level fitness functions are skipped, and non-dominated sorting and crowding calculation are performed directly to obtain a series of pop schemes that are relatively at the forefront of the current population Pareto. These schemes are retained as parent chromosomes and fused with offspring. The multi-objective genetic algorithm is used to perform chromosome crossover, mutation, and selection operations to carry out the next iteration in the first-level adaptation mechanism. During the iteration of the first-level adaptation mechanism, the number of teeth is used as a population variable that changes with chromosomes, while other population variables remain unchanged.

[0026] Optionally, the population fitness conditions of the first-level fitness function can be obtained; a first constraint matrix is ​​generated to represent the population fitness conditions of the first-level fitness function, wherein the population fitness conditions of the first-level fitness function include: first-level gear ratio f1, second-level gear ratio f2, overall gear ratio f3, Hunting Ratio f4, gear order frequency avoidance requirement f5; if a planetary gear structure exists, it also includes planetary gear uniformly distributed assembly condition f6 and planetary gear sideband requirement f7, for a total of seven condition elements; when determining the population pop that does not meet the population fitness conditions of the first-level fitness function using the first constraint matrix, the penalty coefficients that decrease step by step are returned according to the position of the condition elements in the first constraint matrix.

[0027] 2.2) Populations that meet the fitness conditions of the first-level fitness function are popped and enter the second-level adaptation mechanism.

[0028] 2.3) In the second-level adaptation mechanism, for population pops that do not meet the population fitness conditions of the second-level fitness function, a progressively decreasing penalty coefficient is returned according to the population fitness conditions of the second-level fitness function. Non-dominated sorting and crowding calculation are performed to obtain a series of pop schemes that are relatively at the Pareto front of the current population, and these schemes are retained as parent chromosomes and fused with offspring. The multi-objective genetic algorithm is used to perform chromosome crossover, mutation, and selection operations, thereby conducting the next iteration in the second-level adaptation mechanism. During the iteration, the number of teeth, bearing type number, and bearing model number remain unchanged, while other variables are used as variables for chromosome changes.

[0029] Optionally, the population fitness conditions of the second-level fitness function are obtained: gear design and manufacturing constraints, gear root circle diameter range, gear tip circle diameter range, NVH design requirements, maximum static torque, gear load spectrum life, and maximum input torque that the gear can withstand; a second constraint matrix is ​​generated to represent the population fitness conditions of the second-level fitness function. Taking a single-gear two-stage reduction parallel shaft system as an example, the second constraint matrix includes: the total displacement coefficient of the first-stage gear f8, the total displacement coefficient of the second-stage gear f9, the axial overlap of the first-stage gear f10, and the end face overlap of the first-stage gear. f11, Total contact ratio of the first-stage gear; f12, Axial contact ratio of the second-stage gear; f13, End face contact ratio of the second-stage gear; f14, Total contact ratio of the second-stage gear; f15, Chamfer thickness of the first-stage pinion; f16, Chamfer thickness of the first-stage wheel; f17, Chamfer thickness of the second-stage pinion; f18, Chamfer thickness of the second-stage wheel; f19, Tip circle diameter of the first-stage pinion; f20, Tip circle diameter of the first-stage wheel; f21, Tip circle diameter of the second-stage pinion; f22, Tip circle diameter of the second-stage wheel; f23, Root of the first-stage pinion. 24. Diameter of the first-stage wheel tooth root circle; 25. Diameter of the second-stage pinion tooth root circle; 26. Diameter of the second-stage wheel tooth root circle; 27. Distance between the circle where the first-stage pinion begins meshing and the base circle; 28. Distance between the circle where the second-stage pinion begins meshing and the base circle; 29. ​​Distance between the involute starting circle and the base circle of the first-stage pinion; 30. Distance between the involute starting circle and the base circle of the second-stage pinion; 31. Slip ratio at the starting point of the first-stage gear meshing; 32. Slip ratio at the ending point of the first-stage gear meshing; 33. The slip ratio at the start of meshing of the second-stage gear is f34, the slip ratio at the end of meshing of the second-stage gear is f35, the maximum torque that the gear can withstand is converted to the output end f36, the contact safety factor of the first-stage gear is f37, the bending safety factor of the first-stage gear is f38, the contact safety factor of the second-stage gear is f39, and the bending safety factor of the second-stage gear is f40, according to the load spectrum. When the second constraint matrix is ​​used to determine the population pop that does not meet the fitness conditions of the second-stage fitness function, the penalty coefficients that decrease step by step are returned according to the position of the condition element in the second constraint matrix.

[0030] 2.4) Populations that meet the fitness conditions of the second-level fitness function are popped and enter the third-level adaptation mechanism.

[0031] 2.5) In the third-level adaptation mechanism, for population pops that do not meet the population fitness conditions of the third-level fitness function, a progressively decreasing penalty coefficient is returned according to the population fitness conditions of the third-level fitness function. Non-dominated sorting and crowding calculation are performed to obtain a series of pop schemes that are relatively at the Pareto front of the current population, and these schemes are retained as parent chromosomes and fused with offspring. The multi-objective genetic algorithm is used to perform chromosome crossover, mutation, and selection operations, thereby performing the next iteration in the third-level adaptation mechanism. During the iteration, the bearing type number and bearing model number are used as variables for chromosome changes, while other variables remain unchanged.

[0032] Optionally, the population fitness conditions of the third-level fitness function are obtained: bearing size and system layout matching degree, bearing creep shoulder rate, bearing surface pressure, and bearing load spectrum life; a third constraint matrix is ​​generated to represent the population fitness conditions of the third-level fitness function. Taking a single-gear two-stage reduction parallel shaft system as an example, the third constraint matrix includes: bearing housing thinnest wall thickness f41, bearing creep shoulder rate f42, bearing surface pressure safety factor f43, ISO281 bearing damage rate f44, and ISO16281 bearing damage rate f45; when the third constraint matrix is ​​used to determine the population pop that does not meet the population fitness conditions of the third-level fitness function, the penalty coefficients that decrease step by step are returned according to the position of the condition elements in the third constraint matrix.

[0033] 2.6) For population pop that satisfies the population fitness conditions of the third-level fitness function, as the candidate design scheme, non-dominated sorting and crowding calculation are performed to obtain a series of pop schemes with the Pareto frontier of the current population, and these are retained as parent chromosomes and fused with offspring; after performing crossover, mutation, and selection operations on the number of teeth, the third-level adaptation mechanism is run again.

[0034] For example, a population pop that meets at least two of the following conditions can be selected as a candidate design: total manufacturing cost of the reducer, power loss at the CLTC operating point, life of shaft and gear components, and contact ratio.

[0035] Step S106: Select the target design scheme that meets the Pareto optimal solution of the electric drive reducer from the candidate design schemes.

[0036] Through the above steps, design variables for the electric drive reducer of new energy vehicles are obtained. These design variables are design parameters that meet the design requirements of the electric drive reducer in terms of vehicle economy and power. Using these design variables as input variables, a multi-objective genetic algorithm is used to iteratively optimize the design scheme of the electric drive reducer, obtaining candidate design schemes. During the iterative optimization process, variables are processed in hierarchical order according to their correlation strength. A three-level fitness function based on matrix operations is used to hierarchically sort the constraints at each level contained in the fitness function of the multi-objective genetic algorithm. A gradient penalty coefficient is used to progressively reduce the objective function value of the multi-objective genetic algorithm, thereby preventing the loss of excellent populations and avoiding premature convergence leading to incorrect output results or getting trapped in local optima through a gradual adaptation mechanism. The target design scheme that meets the Pareto optimal solution of the electric drive reducer is selected from the candidate design schemes, which can solve the technical problem in related technologies where the design of electric drive reducers does not have schemes optimized for multiple design objectives.

[0037] As an optional embodiment, refer to Figures 2-4 The following section describes the optimization process of this application in conjunction with the core logic:

[0038] Step 1: Use the main program to read the external environment variable file, the bearing database file Bearing_DataBase, the CLTC operating point file, the process file Process_File, the load spectrum operating point file, and the cost database Cost_Database. Convert the information in the external files into global variables, invoke the genetic algorithm main program, and pass the global variables to the next level subroutine. The environmental variables include the overall development roadmap for electric drive systems, such as oil or water cooling, vehicle dimensions, electric drive placement angle, tire radius, maximum required output torque for the vehicle, several typical electric drive layouts (including parallel shaft systems, planetary gear systems, placement of the intermediate shaft in parallel shafts (top / bottom, left / right), gear helix direction, differential placement direction, distinguished by numbers, which can be directly mapped to the corresponding calculation logic during subroutine calls), energy recovery strategies, oil pump speed control strategies, NVH design requirements, reliability coefficients, shaft and gear manufacturing processes, materials used for parts, tooth surface roughness, gear precision grade, tooth thickness tolerance, safety clearances required for shaft and gear system layout, lubricant properties, reducer operating temperature, oil level and output shaft height difference, oil seal lip diameter, minimum safety factor for shaft and gear component lifespan, material SN curve, system iteration accuracy requirements, initial population size for the genetic algorithm, maximum number of iterations for the genetic algorithm, and material SN curve reference. Figure 4 , where X is the cumulative stress spectrum and fatigue curve over the finite lifespan, N(log); Y is the stress, σ(log); and 1 is the total damage rate. The values ​​are the stress values ​​on the SN curve; σ4 and σ5 are located in the infinite life range and therefore are not included in the calculation.

[0039] Step 2: Discretize all design variables of the reducer using a genetic algorithm to pop the population for the genetic algorithm (see [link]). Figure 3 The multidimensional array data interaction logic diagram, where m is the population size, n is the chromosome size, f is the fitness function matrix, and F is the load matrix), population variables (the population variables are design variables, taking a single-gear two-stage reduction parallel shaft system as an example, including: number of gear teeth z, module mn, pressure angle α, helix angle β, displacement coefficient x, tooth width b, tooth tip height coefficient hap, first-level center distance aw_First, second-level center distance aw_Second, total center distance aw_Output, input shaft bearing span L_Bearing_Input, intermediate shaft bearing span L_Bearing_Transfert, output shaft bearing span L_Bearing_Output, bearing type number Bearing_Type, bearing model number Bearing_Nr, etc., corresponding to pop(1), pop(2), pop(3)...pop(n)), that is:

[0040]

[0041] Where n is the number of chromosomes, m is the population size; the number of teeth z, the center distance variable aw, the bearing type number, and the bearing model number are discretized integers within the design range, and pop is an m x n matrix. upper and pop lower Upper and lower bounds for population variables:

[0042] Step 3: Fitness_Function_First serves as the first-level fitness function, incorporating the tooth ratio requirement, gear order frequency avoidance requirement, and, if a planetary gear structure exists, additional fitness conditions including planetary gear uniform distribution assembly condition and planetary gear sidefrequency requirement, as population fitness conditions. Populations meeting these fitness conditions are then used to calculate the next two levels of fitness functions: Fitness_Function_Second (second-level fitness function) and Fitness_Function_Third (third-level fitness function). The result returns any 2-3 of the following reducer design variables and optimization objectives: total reducer manufacturing cost (Cost), CLTC operating point power loss (Power_Loss), shaft and gear component life (Safty), and overlap ratio.

[0043] Taking a single-gear two-stage reduction parallel shaft system as an example, f1, f2, f3, f4, and f5 constitute the constraint matrix [f1, f2, f3, f4, f5] of the first-level fitness function: f1 = u1 = z2 / z1; f2 = u2 = z4 / z3; f3 = u = u1 * u2; f4 = min(math.gcd(z2 / z1), math.gcd(z4 / z3)), which checks the coprimeness of the number of teeth and returns a logical value of 1 or 0; f5 = Gear_Order(z), which checks the frequency avoidance between each order of the gear and each order of the motor and checks the minimum percentage in the frequency avoidance matrix; if there is a planetary gear structure, it also includes f6 and f7, F6 = np%1, which checks whether the planetary gears can be evenly arranged; f7 = λ%1, which checks the phase type of the planetary gears.

[0044] The conditional discriminant coefficient for determining whether planetary gears can be uniformly distributed. The remainder of the variable np is obtained by taking the remainder of np itself.

[0045] The convolution of the planetary gear meshing frequency domain signal xm(f) and the planet carrier rotation signal xc(f) yields the following result:

[0046] X mq (f)XX c (f)*e -jπftq ,

[0047] The signal after q-modulation by the planetary gears is shifted by phase λ in the frequency domain of the first planetary gear. q :

[0048] λ q =2·pi·(q-1)·λ,

[0049] The remainder of λ is used to determine whether the planetary gears are in continuous or in phase. The engineering design ranges of f1, f2, f3, f4, and f5 serve as the upper and lower limits of the constraint matrix.

[0050] For populations that do not meet the fitness criteria, the system is hierarchically divided into levels according to f1, f2, f3, f4, and f5, returning progressively decreasing penalty coefficients to obtain values ​​deviating from the normal Target_Values. The second-level fitness function (Fitness_Function_Second) and the third-level fitness function (Fitness_Function_Third) are skipped, and non-dominated sorting and crowding calculation are performed directly. This yields a series of pop schemes that are relatively at the Pareto front of the current population, which are retained as parent chromosomes and fused with offspring. The system then returns to the genetic algorithm for chromosome crossover, mutation, and selection operations, and proceeds to the next iterative calculation. During the iteration process, the number of teeth is used as a variable for chromosome changes, while other variables remain unchanged.

[0051] The specific steps are as follows:

[0052]

[0053] The logical matrix Critical_First used to determine whether the first-level fitness is satisfied is the negation of the intersection of Critical_First_upper and Critical_First_lower, i.e.:

[0054] Critical_First=~Critical_First_upper&Critical_First_lower

[0055] If all first-level fitness conditions are met, then Critical_First is an mx5 zero matrix:

[0056]

[0057] Penalty_factor_First=Penalty_factor_First_Initial+1,

[0058] in, and These are the upper and lower bounds of the fitness matrix, respectively.

[0059] If the value returned by Penalty_factor_First is not 1, the genetic algorithm assigns Target_Values ​​to the objective function.

[0060] Penalty_factor_First means using values ​​that deviate significantly from the normal range. The purpose of using these different levels is to improve the convergence of the genetic algorithm and avoid premature convergence that leads to erroneous results.

[0061] Step 4: If all constraints of the first-level fitness function are satisfied, the genetic algorithm will start calculating the second-level fitness function. Gear design and manufacturing constraints, the range of gear root circle diameter, the range of gear tip circle diameter, NVH design requirements, maximum static torque, gear load spectrum life, and the maximum input torque the gear can withstand are used as constraints for the Fitness_Function_Second fitness function, and penalty coefficients are set sequentially in a certain order. The penalty coefficients decrease progressively to guide the machine to gradually adapt to the constraints in the fitness function, improving the convergence of the genetic algorithm.

[0062] Taking a single-gear two-stage reduction parallel shaft system as an example, the constraints of the fitness function include the following 33 matrices: total displacement coefficient of the first-stage gear f8, total displacement coefficient of the second-stage gear f9, axial contact ratio of the first-stage gear f10, end face contact ratio of the first-stage gear f11, total contact ratio of the first-stage gear f12, axial contact ratio of the second-stage gear f13, end face contact ratio of the second-stage gear f14, total contact ratio of the second-stage gear f15, chamfered tooth thickness of the first-stage pinion f16, chamfered tooth thickness of the first-stage wheel f17, chamfered tooth thickness of the second-stage pinion f18, chamfered tooth thickness of the second-stage wheel f19, addendum circle diameter of the first-stage pinion f20, addendum circle diameter of the first-stage wheel f21, addendum circle diameter of the second-stage pinion f22, addendum circle diameter of the second-stage wheel f23, dedendum circle diameter of the first-stage pinion f24, and dedendum circle diameter of the first-stage wheel. f25, diameter of the second-stage pinion root circle f26, diameter of the second-stage wheel root circle f27, distance between the first-stage pinion meshing initiation circle and the base circle f28, distance between the second-stage pinion meshing initiation circle and the base circle f29, distance between the first-stage pinion involute initiation circle and the base circle f30, distance between the second-stage pinion involute initiation circle and the base circle f31, slip ratio of the first-stage gear meshing initiation point f32, slip ratio of the first-stage gear meshing termination point f33, slip ratio of the second-stage gear meshing initiation point f34, slip ratio of the second-stage gear meshing termination point f35, maximum torque that the gear can withstand converted to the output end f36, contact safety factor of the first-stage gear checked according to the load spectrum f37, bending safety factor of the first-stage gear f38, contact safety factor of the second-stage gear f39, bending safety factor of the second-stage gear f40:

[0063] f8 = sumX1, f9 = sumX2

[0064] f16=sn_dFa1,f17=sn_dFa2,f18=sn_dFa3,f19=sn_dFa4,f20=da1,f21=da2,f22=da3,f23=da4,f2 4=df1, f25=df2, f26=df3, f27=df4, f28=dNf1-db1, f29=dNf3-db3, f30=dFf1-db1, f31=dFf3-db3,

[0065] f36=min[T_max_first_pinion,T_max_first_wheel,T_max_second_pinion,T_max_second_wheel],

[0066]

[0067] f8, f9, f10, ..., f40 constitute the constraint matrix of the second-order fitness function.

[0068] Where m is the population size for popping, and the engineering design range of f8, f9, f10, ... f40 serves as the upper and lower bounds of the constraint matrix. For populations that do not meet the fitness conditions, the system is divided into levels according to the order of f8, f9, f10, ... f40, returning progressively decreasing penalty coefficients to obtain the values ​​that deviate from the normal Target_Values. The third-level fitness function of Fitness_Function_Third is skipped, and non-dominated sorting and crowding calculation are performed directly to obtain a series of popping schemes that are relatively at the forefront of the current population's Pareto algorithm. These schemes are retained as parent chromosomes and fused with offspring chromosomes. The genetic algorithm is then used for chromosome crossover, mutation, and selection operations, followed by the next iterative calculation. During the iteration process, variables other than the number of teeth, bearing type, and bearing model are treated as variables of chromosome change, while the number of teeth, bearing type, and bearing model remain unchanged.

[0069] The specific steps are as follows:

[0070]

[0071] The logical matrix Critical_Second used to determine whether the second-order fitness is satisfied is the negation of the intersection of Critical_Second_upper and Critical_Second_lower, i.e.:

[0072] Critical_Second=~Critical_Second_upper&Critical_Second_lower,

[0073] If all second-order fitness conditions are met, then Critical_Second is an mx33 zero matrix (the other is an m-column matrix):

[0074]

[0075] Penalty_factor_Second=Penalty_factor_Second_Initial+1,

[0076] in, 与 These are the upper and lower bounds of the fitness matrix, respectively.

[0077] If the value returned by Penalty_factor_Second is not 1, the genetic algorithm assigns Target_Values ​​= Penalty_factor_Second to the objective function, that is, it uses a value that deviates significantly from the normal value. The purpose of using these different levels is to improve the convergence of the genetic algorithm and avoid premature convergence leading to erroneous results.

[0078] Step 5: If all constraints of the second-level fitness function are satisfied, the genetic algorithm will start calculating the third-level fitness function. The bearing size and system layout matching degree, bearing creep rate, bearing surface pressure, and bearing load spectrum life are used as constraints for the Fitness_Function_Third fitness function, and penalty coefficients are set sequentially in a certain order. The penalty coefficients decrease progressively to guide the machine to gradually adapt to the constraints in the fitness function, improving the convergence of the genetic algorithm.

[0079] Taking a single-gear two-stage reduction parallel shaft system as an example, the constraints of the fitness function include the following five matrices: minimum wall thickness of the bearing housing f41, bearing creep shoulder rate f42, bearing surface pressure safety factor f43, ISO281 bearing damage rate f44, and ISO16281 bearing damage rate f45.

[0080] f41=Min[aw_Input / 2-(Diameter_Bearing Input_Gear +Diameter_Bearing Transfer_Gear ) / 2,

[0081] aw_Input / 2-(Diameter_Bearing Input_Moter +Diameter_Bearing Transfer_Motor ) / 2,

[0082] aw_Output / 2-(Diameter_Bearing Transfer_Gear +Diameter_Bearing Differential_Gear ) / 2,

[0083] aw_Output / 2-(Diameter_Bearing Transfer_Motor +Diameter_Bearing Differential_Motor ) / 2];

[0084] f42 = Min[Track_Trucation] Bearing_Input_Gear Track_Trucation Bearing_Input_Motor ,

[0085] Track_TrucationBearing_Transfer_Gear Track_Trucation Bearing_Transfer_Motor ,

[0086] Track_Trucation Bearing_Differential_Gear Track_Trucation Bearing_Differential_Motor f43 = Min

[0087] [Contact_Stress_Permissible1 / Contact_Stress_max1,Contact_Stress_Permissible2 / Contact_Stress_max2,Stress_Stress_Permissible3 / Contact_Stress_max3,Contact_Stress_Permissible4 /

[0088] Contact_Stress_max4 / ,Contact_Stress_max5Contact_Stress_Permissible5,Contact_Stress_Permissible6 / Contact_Stress_max6],

[0089] f44=Max[Damage_281_1,Damage_281_2,Damage_281_3,Damage_281_4,Damage_281_5,Damage_281_6],

[0090] f45=Max[Damage_16281_1,Damage_16281_2,Damage_16281_3,Damage_16281_4,Damage_16281_5,Damage_16281_6],

[0091] f41, f42, f43, f44, and f45 constitute the constraint matrix of the second-order fitness function.

[0092] Where m is the population size of pop, the coefficient subscript is the installation position of the bearing in the shaft system, and the engineering design range of f41, f42, f43, f44, and f45 serves as the upper and lower limits of the constraint matrix.

[0093] For populations that do not meet the fitness criteria, the hierarchy is divided in the order of f41, f42, f43, f44, and f45, and the penalty coefficients are returned in progressively decreasing order to obtain the values ​​that deviate from the normal Target_Values.

[0094] Non-dominated ranking is performed on the objective function of each individual in the population to obtain population stratification and different frontiers. For two different objective function solution sets, the decision vector u = [u1, u2, ..., u...] is... P ] T With v = [v1, v2, ..., v P ] T When the following conditions are met, f(u) is in the Pareto hierarchy earlier than f(v):

[0095]

[0096] and, For each individual within the frontier, calculate the crowding degree as a subsequent selection: Where i d For congestion level, f j i+1 f represents the function value of the objective function at the (i+1)th point and the j-th objective function. j i-1 This represents the function value of the objective function at the (i-1)th point and the j-th objective function.

[0097] The obtained relatively leading strata are merged with the parent generation to form a new population. A genetic algorithm performs selection, crossover, and mutation operations. Based on non-dominance relationships and crowding, the top K optimal individuals are selected as the next generation. The iteration continues until the maximum number of iterations is reached, or the convergence error of the Pareto neighbor set exceeds a set tolerance. Otherwise, the Pareto leading set of solutions is output, and the results are summed according to the weights of different objectives, merging into a single objective. The minimum z-value is output as the optimization result, where z is:

[0098]

[0099] Where j is the number of objective functions, w i f is the weighting coefficient. j (x) represents the objective function value.

[0100] During the iteration, the bearing type and bearing model variables in the pop variable undergo crossover and mutation, while other chromosome variables remain unchanged.

[0101] The specific steps are as follows:

[0102]

[0103] The logical matrix Critical_Third used to determine whether the third-level fitness is satisfied is the negation of the intersection of Critical_Third_upper and Critical_Third_lower, i.e.:

[0104] Critical_Third=~Critical_Third_upper&Critical_Third_lower,

[0105] If all three fitness conditions are met, then Critical_Third is an mx5 zero matrix:

[0106]

[0107] Penalty_factor_Third=Penalty_factor_Third_Initial+1,

[0108] in, 与 These are the upper and lower bounds of the fitness matrix, respectively.

[0109] If the value returned by Penalty_factor_Third is not 1, the genetic algorithm assigns Target_Values ​​= Penalty_factor_Third to the objective function, that is, it uses a value that deviates significantly from the normal value. The purpose of using these different levels is to improve the convergence of the genetic algorithm and avoid premature convergence that leads to erroneous results.

[0110] Step 6: The minimum target value for reducer design optimization is the core element of the fitness function Fitness_Function_Third in the genetic algorithm. The target values ​​Target_Values ​​include any 2-3 of the following: total reducer manufacturing cost (Cost), power loss at CLTC operating point (Loss), lifespan of shaft and gear parts (Safty), and overlap ratio. Any multiple targets are converted into a single target value Target_Value through weighting coefficients.

[0111] After obtaining a local optimum while keeping all variables except bearing type and bearing model unchanged in the pop variable, before returning to the second-level fitness calculation of the genetic algorithm, crossover, mutation, and selection operations are performed on the bearing type and bearing model variables, and then steps four and five are repeated.

[0112] After returning to the second-level fitness calculation of the genetic algorithm, and obtaining a local optimum while keeping the number of teeth, bearing type, and bearing model variables in the pop variable unchanged, return to the first-level fitness calculation of the genetic algorithm, and perform crossover, mutation, and selection operations on variables other than the number of teeth, bearing type, and bearing model, and then repeat step six.

[0113] After a set number of iterations of t, the Pareto front solution set is obtained within the range of chromosome variables other than the number of teeth. After reaching the maximum number of iterations, or when the convergence error of the Pareto adjacent solution set is less than the set tolerance, a partial Pareto optimal solution is obtained.

[0114] Step 7: Return to the first-level fitness calculation of the genetic algorithm and repeat step 3. After the set number of iterations (s), obtain the Pareto front solution set for all chromosome variables. Obtain all Pareto optimal solutions after reaching the maximum number of iterations or when the convergence error of the Pareto adjacent solution set is less than the set tolerance. Then, sum the weights of the different objectives, merge them into a single objective, and output the minimum z value as the optimization result, where z is:

[0115] Where j is the number of objective functions, w i f is the weighting coefficient. j (x) represents the objective function value.

[0116] Step 8: The genetic algorithm returns the pop variable corresponding to the optimal solution to the main program. The main program processes the pop variable and converts it into a readable format file. Based on the design variables under the optimal solution, it calculates other calculation results that need to be output, such as speed vector, torque vector, reducer weight, cost, and mechanical loss Loss_Map across the entire speed and torque range.

[0117] The obtained calculation results have the following functions: directly outputting the results; serving as input conditions for other optimization algorithms to perform optimization at the electric drive system level; inputting reducer design variables to external 3D / FEA software via API to establish a multiphysics proxy model; and inputting design variables and calculated performance index data to artificial intelligence algorithms to establish an enterprise-level electric drive large model.

[0118] The specific implementation process of this invention is as follows:

[0119] 1) Main program: The main program reads the external environment variable file, the bearing database file Bearing_DataBase, the CLTC operating point file, the process file Process_File, the load spectrum operating point file, and the cost database Cost_Database. It converts the information in the external files into global variables, invokes the genetic algorithm main program, and passes the global variables to the next level of subroutines. Among them, environmental variables include the overall scheme roadmap for electric drive development, such as oil cooling or water cooling, vehicle space dimensions, electric drive placement tilt angle, tire radius, maximum output torque required by the vehicle, several typical electric drive layouts (including parallel shaft system, planetary gear system, placement position of the intermediate shaft of the parallel shaft (up / down, left / right), gear helix direction, differential placement direction, distinguished by numbers, which can be directly mapped to the corresponding calculation logic when the subroutine is called), energy recovery strategy, oil pump speed control strategy, NVH design requirements, reliability coefficient, shaft and gear process route, materials used for parts, tooth surface roughness, gear accuracy grade, tooth thickness tolerance, safety clearance required for shaft and gear system layout, lubricating oil properties, reducer operating temperature, oil level and output shaft height difference, oil seal lip diameter, minimum safety factor for shaft and gear parts life, material SN curve, system iteration accuracy requirements, initial population size of the genetic algorithm, and maximum number of iterations of the genetic algorithm.

[0120] 2) Genetic Algorithm Main Program: The genetic algorithm is used to discretize all the design variables of the reducer. Population variables are used as the population variables of the genetic algorithm. Taking a single-gear two-stage reduction parallel shaft system as an example, it includes dozens of variables such as: number of gear teeth z, module mn, pressure angle α, helix angle β, displacement coefficient x, tooth width b, tooth tip height coefficient hap, first-stage center distance aw_First, second-stage center distance aw_Second, total center distance aw_Output, input shaft bearing span L_Bearing_Input, intermediate shaft bearing span L_Bearing_Transfert, output shaft bearing span L_Bearing_Output, bearing type number Bearing_Type, bearing model number Bearing_Nr, which correspond to pop(1), pop(2), pop(3)...pop(n) respectively, that is:

[0121]

[0122] Where n is the number of chromosomes, m is the population size; the number of teeth z, the center distance variable aw, the bearing type number, and the model number of a certain type of bearing are discretized integers within the design range, and pop is an m x n matrix. upper and pop lower Upper and lower bounds for population variables:

[0123]

[0124] 3) First-level fitness function: Fitness_Function_First serves as the first-level fitness function, incorporating requirements for tooth ratio and gear order frequency avoidance. If a planetary gear structure exists, additional fitness conditions include planetary gear uniform distribution assembly conditions and planetary gear sidefrequency requirements. These are used as population fitness conditions. Populations meeting these fitness conditions are then used to calculate the next two levels of fitness functions: Fitness_Function_Second (the second-level fitness function) and Fitness_Function_Third (the third-level fitness function). The return value includes any 2-3 of the following: total reducer manufacturing cost (Cost), CLTC operating point power loss (Power_Loss), shaft and gear component life (Safty), and overlap ratio.

[0125] Taking a single-gear two-stage reduction parallel shaft system as an example, f1 to f5 constitute the constraint matrix [f1,f2,f3,f4,f5] of the first-level fitness function: f1=u1=z2 / z1;f2=u2=z4 / z3;f3=u=u1*u2;f4=min(math.gcd(z2 / z1),math.gcd(z4 / z3)), checks the coprimeness of the number of teeth, and returns a logical value of 1 or 0; f5=Gear_Order(z), checks the frequency avoidance situation between each order of the gear and each order of the motor, and checks the minimum percentage in the frequency avoidance matrix; if there is a planetary gear structure, it also includes f6 and f7; F6=np%1, checks whether the planetary gears can be evenly arranged; f7=λ%1, checks the phase type of the planetary gears; the conditional discrimination coefficient for judging whether the planetary gears can be evenly distributed:

[0126] The remainder of the variable np is taken to obtain np%1; the result is obtained by convolving the planetary gear meshing frequency domain signal xm(f) with the planet carrier rotation signal xc(f).

[0127] X mq (f)XX c (f)*e -jπftq ,

[0128] The signal after q-modulation by the planetary gears is shifted by phase λ in the frequency domain of the first planetary gear. q :

[0129] λ q =2·pi·(q-1)·λ,

[0130] The remainder of λ is used to determine whether the planetary gears are in continuous or in phase. The engineering design ranges of f1, f2, f3, f4, and f5 serve as the upper and lower limits of the constraint matrix.

[0131] For populations that do not meet the fitness criteria, the system is hierarchically divided into levels according to f1, f2, f3, f4, and f5, returning progressively decreasing penalty coefficients to obtain values ​​deviating from the normal Target_Values. The second-level fitness function (Fitness_Function_Second) and the third-level fitness function (Fitness_Function_Third) are skipped, and non-dominated sorting and crowding calculation are performed directly. This yields a series of pop schemes that are relatively at the Pareto front of the current population, which are retained as parent chromosomes and fused with offspring. The system then returns to the genetic algorithm for chromosome crossover, mutation, and selection operations, and proceeds to the next iterative calculation. During the iteration process, the number of teeth is used as a variable for chromosome changes, while other variables remain unchanged.

[0132] The specific steps are as follows:

[0133]

[0134] The logical matrix Critical_First used to determine whether the first-level fitness is satisfied is the negation of the intersection of Critical_First_upper and Critical_First_lower, i.e.:

[0135] Critical_First=~Critical_First_upper&Critical_First_lower,

[0136] If all first-level fitness conditions are met, then Critical_First is an mx5 zero matrix:

[0137]

[0138] Penalty_factor_First=Penalty_factor_First_Initial+1

[0139] in, 与 These are the upper and lower bounds of the fitness matrix, respectively.

[0140] If the value returned by Penalty_factor_First is not 1, the genetic algorithm assigns Target_Values ​​to the objective function.

[0141] Penalty_factor_First means using values ​​that deviate significantly from the normal range. The purpose of using these different levels is to improve the convergence of the genetic algorithm and avoid premature convergence that leads to erroneous results.

[0142] 4) Chromosome mutation, crossover, and selection after first-level fitness: For populations that do not meet the first-level fitness condition, skipping the second-level fitness function (Fitness_Function_Second) and the third-level fitness function (Fitness_Function_Third), directly perform non-dominated sorting and crowding calculation to obtain a series of pop schemes that are relatively at the Pareto front of the current population. These are retained as parent chromosomes and fused with offspring chromosomes. The genetic algorithm is then returned to perform chromosome crossover, mutation, and selection operations, and the next iteration is performed. During the iteration process, the number of teeth is used as a variable for chromosome changes, while other variables remain unchanged.

[0143] 5) Second-level fitness function: If all constraints of the first-level fitness function are satisfied, the genetic algorithm will start calculating the second-level fitness function. Gear design and manufacturing constraints, the range of gear root circle diameter, the range of gear tip circle diameter, NVH design requirements, maximum static torque, gear load spectrum life, and the maximum input torque the gear can withstand are used as constraints for the Fitness_Function_Second fitness function, and penalty coefficients are set sequentially in a certain order. The penalty coefficients decrease progressively to guide the machine to gradually adapt to the constraints in the fitness function, improving the convergence of the genetic algorithm.

[0144] Taking a single-gear two-stage reduction parallel shaft system as an example, the constraints of the fitness function include the following 33 matrices: total displacement coefficient of the first-stage gear f8, total displacement coefficient of the second-stage gear f9, axial contact ratio of the first-stage gear f10, end face contact ratio of the first-stage gear f11, total contact ratio of the first-stage gear f12, axial contact ratio of the second-stage gear f13, end face contact ratio of the second-stage gear f14, total contact ratio of the second-stage gear f15, chamfered tooth thickness of the first-stage pinion f16, chamfered tooth thickness of the first-stage wheel f17, chamfered tooth thickness of the second-stage pinion f18, chamfered tooth thickness of the second-stage wheel f19, addendum circle diameter of the first-stage pinion f20, addendum circle diameter of the first-stage wheel f21, addendum circle diameter of the second-stage pinion f22, addendum circle diameter of the second-stage wheel f23, dedendum circle diameter of the first-stage pinion f24, and dedendum circle diameter of the first-stage wheel. f25, diameter of the second-stage pinion root circle f26, diameter of the second-stage wheel root circle f27, distance between the first-stage pinion meshing initiation circle and the base circle f28, distance between the second-stage pinion meshing initiation circle and the base circle f29, distance between the first-stage pinion involute initiation circle and the base circle f30, distance between the second-stage pinion involute initiation circle and the base circle f31, slip ratio of the first-stage gear meshing initiation point f32, slip ratio of the first-stage gear meshing termination point f33, slip ratio of the second-stage gear meshing initiation point f34, slip ratio of the second-stage gear meshing termination point f35, maximum torque that the gear can withstand converted to the output end f36, contact safety factor of the first-stage gear checked according to the load spectrum f37, bending safety factor of the first-stage gear f38, contact safety factor of the second-stage gear f39, bending safety factor of the second-stage gear f40:

[0145] f8 = sumX1, f9 = sumX2

[0146]

[0147] f16=sn_dFa1, f17=sn_dFa2, f18=sn_dFa3, f19=sn_dFa4,

[0148] f20=da1, f21=da2, f22=da3, f23=da4, f24=df1, f25=df2, f26=df3, f27=df4

[0149] f28=dNf1-db1, f29=dNf3-db3, f30=dFf1-db1, f31=dFf3-db3

[0150]

[0151] f36=min[T_max_first_pinion,T_max_first_wheel,T_max_second_pinion,T_max_second_wheel]

[0152]

[0153] f8, f9, f10, ..., f40 constitute the constraint matrix of the second-order fitness function.

[0154] Where m is the population size for popping, and the engineering design range of f8, f9, f10, ... f40 serves as the upper and lower bounds of the constraint matrix. For populations that do not meet the fitness conditions, the system is divided into levels according to the order of f8, f9, f10, ... f40, returning progressively decreasing penalty coefficients to obtain the values ​​that deviate from the normal Target_Values. The third-level fitness function of Fitness_Function_Third is skipped, and non-dominated sorting and crowding calculation are performed directly to obtain a series of popping schemes that are relatively at the forefront of the current population's Pareto algorithm. These schemes are retained as parent chromosomes and fused with offspring chromosomes. The genetic algorithm is then used for chromosome crossover, mutation, and selection operations, followed by the next iterative calculation. During the iteration process, variables other than the number of teeth, bearing type, and bearing model are treated as variables of chromosome change, while the number of teeth, bearing type, and bearing model remain unchanged.

[0155] The specific steps are as follows:

[0156] The logical matrix Critical_Second used to determine whether the second-order fitness is satisfied is the negation of the intersection of Critical_Second_upper and Critical_Second_lower, i.e.:

[0157] Critical_Second=~Critical_Second_upper&Critical_Second_lower,

[0158] If all second-order fitness conditions are met, then Critical_Second is an mx33 zero matrix:

[0159]

[0160] Penalty_factor_Second=Penalty_factor_Second_Initial+1,

[0161] in, 与 These are the upper and lower bounds of the fitness matrix, respectively.

[0162] If the value returned by Penalty_factor_Second is not 1, the genetic algorithm assigns Target_Values ​​= Penalty_factor_Second to the objective function, that is, it uses a value that deviates significantly from the normal value. The purpose of using these different levels is to improve the convergence of the genetic algorithm and avoid premature convergence leading to erroneous results.

[0163] 6) Chromosome mutation, crossover, and selection after second-level fitness: For populations that do not meet the second-level fitness condition, skipping the third-level fitness function (Fitness_Function_Third), directly perform non-dominated sorting and crowding calculation to obtain a series of pop schemes that are relatively at the Pareto front of the current population, and retain them as parent chromosomes for fusion with offspring; return to the genetic algorithm to perform chromosome crossover, mutation, and selection operations, and proceed to the next iteration. During the iteration process, chromosomes other than tooth count, bearing type, and bearing model are used as optimization variables, while the variables of tooth count, bearing type, and bearing model remain unchanged.

[0164] 7) Third-level fitness function: If all constraints of the second-level fitness function are satisfied, the genetic algorithm will start calculating the third-level fitness function. Bearing size and system layout matching degree, bearing creep rate, bearing surface pressure, and bearing load spectrum life are used as constraints for the Fitness_Function_Third fitness function, and penalty coefficients are set sequentially in a certain order. The penalty coefficients decrease progressively, guiding the machine to gradually adapt to the constraints in the fitness function and improving the convergence of the genetic algorithm.

[0165] Taking a single-gear two-stage reduction parallel shaft system as an example, the constraints of the fitness function include the following five matrices: minimum wall thickness of the bearing housing f41, bearing creep shoulder rate f42, bearing surface pressure safety factor f43, ISO281 bearing damage rate f44, and ISO16281 bearing damage rate f45.

[0166] f41=Min[aw_Input / 2-(Diameter_Bearing Input_Gear +Diameter_Bearing Transfer_Gear ) / 2,

[0167] aw_Input / 2-(Diameter_Bearing Input_Moter +Diameter_Bearing Transfer_Motor ) / 2,

[0168] aw_Output / 2-(Diameter_Bearing Transfer_Gear +Diameter_Bearing Differential_Gear ) / 2,

[0169] aw_Output / 2-(Diameter_Bearing Transfer_Motor +Diameter_Bearing Differential_Motor ) / 2];

[0170] f42=Min[Track_Trucation Bearing_Input_Gear ,Track_Trucation Bearing_Input_Motor ,

[0171] Track_Trucation Bearing_Transfer_Gear ,Track_Trucation Bearing_Transfer_Motor ,

[0172] Track_Trucation Bearing_Differential_Gear ,Track_Trucation Bearing_Differential_Motor ]f43=Min

[0173] [Contact_Stress_Permissible1 / Contact_Stress_max1,Contact_Stress_Permissible2 / Contact_Stress_max2,Contact_Stress_Permissible3 / Contact_Stress_max3,Contact_Stress_Permissible4 / Contact_Stress_max4 / ,Contact_Stress_max5Contact_Stress_Permissible5,Contact_Stress_Permissible6 / Contact_Stress_max6]

[0174] f44=Max[Damage_281_1,Damage_281_2,Damage_281_3,Damage_281_4,Damage_281_5,Damage_281_6]

[0175] f45=Max[Damage_16281_1,Damage_16281_2,Damage_16281_3,Damage_16281_4,Damage_16281_5,Damage_16281_6]

[0176] f41, f42, f43, f44, and f45 constitute the constraint matrix of the second-order fitness function.

[0177] Where m is the population size of pop, the coefficient subscript is the installation position of the bearing in the shaft system, and the engineering design range of f41, f42, f43, f44, and f45 serves as the upper and lower limits of the constraint matrix.

[0178] For populations that do not meet the fitness criteria, the hierarchy is divided in the order of f41, f42, f43, f44, and f45, and the penalty coefficients are returned in progressively decreasing order to obtain the values ​​that deviate from the normal Target_Values.

[0179] Non-dominated ranking is performed on the objective function of each individual in the population to obtain population stratification and different frontiers. For two different objective function solution sets, the decision vector u = [u1, u2, ..., u...] is... P ] T With v = [v1, v2, ..., v P ] T When the following conditions are met, f(u) is in the Pareto hierarchy earlier than f(v):

[0180]

[0181] and, For each individual within the frontier, calculate the crowding degree as a subsequent selection: Where i d For congestion level, f j i+1 f represents the function value of the objective function at the (i+1)th point and the j-th objective function. j i-1 This represents the function value of the objective function at the (i-1)th point and the j-th objective function.

[0182] The obtained relatively leading strata are merged with the parent generation to form a new population. A genetic algorithm performs selection, crossover, and mutation operations. Based on non-dominance relationships and crowding, the top K optimal individuals are selected as the next generation. The iteration continues until the maximum number of iterations is reached, or the convergence error of the Pareto neighbor set exceeds a set tolerance. Otherwise, the Pareto leading set of solutions is output, and the results are summed according to the weights of different objectives, merging into a single objective. The minimum z-value is output as the optimization result, where z is: Where j is the number of objective functions, w i f is the weighting coefficient. j (x) represents the objective function value.

[0183] During the iteration, the bearing type and bearing model variables in the pop variable undergo crossover and mutation, while other chromosome variables remain unchanged.

[0184] The specific steps are as follows:

[0185] The logical matrix Critical_Third used to determine whether the third-level fitness is satisfied is the negation of the intersection of Critical_Third_upper and Critical_Third_lower, i.e.:

[0186] Critical_Third=~Critical_Third_upper&Critical_Third_lower,

[0187] If all three fitness conditions are met, then Critical_Third is an mx5 zero matrix:

[0188]

[0189] Penalty_factor_Third=Penalty_factor_Third_Initial+1

[0190] in, 与 These are the upper and lower bounds of the fitness matrix, respectively.

[0191] If the value returned by Penalty_factor_Third is not 1, the genetic algorithm assigns Target_Values ​​= Penalty_factor_Third to the objective function, that is, it uses a value that deviates significantly from the normal value. The purpose of using these different levels is to improve the convergence of the genetic algorithm and avoid premature convergence that leads to erroneous results.

[0192] 6) After completing the three-level fitness calculation, perform non-dominated sorting and crowding calculation to obtain a series of Pareto front solutions. Then, perform chromosome crossover, mutation, and selection. The minimum target value of reducer design optimization is the core element of the fitness function Fitness_Function_Second of the genetic algorithm. The target values ​​Target_Values ​​include any 2-3 of the following: total reducer manufacturing cost (Cost), CLTC operating point power loss (Loss), shaft and gear component life (Safty), and overlap ratio. Any multiple targets are converted into a single target value Target_Value through weight coefficients.

[0193] After obtaining a local optimum while keeping all variables except bearing type and bearing model unchanged in the pop variable, before returning to the second-level fitness calculation of the genetic algorithm, crossover, mutation, and selection operations are performed on the bearing type and bearing model variables, and then steps four and five are repeated.

[0194] After returning to the second-level fitness calculation of the genetic algorithm, and obtaining a local optimum while keeping the number of teeth, bearing type, and bearing model variables in the pop variable unchanged, return to the first-level fitness calculation of the genetic algorithm, and perform crossover, mutation, and selection operations on variables other than the number of teeth, bearing type, and bearing model, and then repeat step six.

[0195] After a set number of iterations of t, the Pareto front solution set is obtained within the range of chromosome variables other than the number of teeth. After reaching the maximum number of iterations, or when the convergence error of the Pareto adjacent solution set is less than the set tolerance, a partial Pareto optimal solution is obtained.

[0196] Return to the first-level fitness calculation of the genetic algorithm and repeat step three. After the set number of iterations (s), obtain the Pareto front solution set for all chromosome variables. Obtain all Pareto optimal solutions after reaching the maximum number of iterations or when the convergence error of the Pareto adjacent solution set is less than the set tolerance. Then, sum the weights of the different objectives, merge them into a single objective, and output the minimum z value as the optimization result, where z is:

[0197] Where j is the number of objective functions, w i f is the weighting coefficient. j (x) represents the objective function value.

[0198] 7) Gear Parameter Calculation: The pop variable is merged into the basic gear parameter variable GearBox_Par. The intermediate parameters of the gear design are calculated (including the ISO2177 gear parameter calculation process, NVH index, tooling, process attributes, etc.) and assigned to the intermediate gear parameter GearBox_Intermediate_Par. The basic gear parameter variable and the intermediate parameter variable can be repeatedly called in the main program and in each subroutine.

[0199] 8) Gear Life Calculation: The gear parameter variables converted from the pop variable in the gear parameter calculation program, along with the generated intermediate gear parameters, are imported into the gear life calculation program. Combined with the torque (Torque), speed (Speed), and number of cycles from the vehicle load spectrum, the contact safety factor (SH) and bending safety factor (SF) for each gear stage are calculated based on ISO 6336-2019-1 / -2 / -3 / -5 / -6. The calculation process involves the tooth surface load distribution coefficient k, which is related to the operating conditions. Hβ kHα k Fβ k Fβ k V The y-values ​​are processed into matrices, and then multiplied by a dot product to obtain the matrix form of the contact / bending stress σ. H and σ F Using PyTorch code as an example, let's create a two-dimensional tensor:

[0200] Import torch

[0201] σ H0 =torch.tensor(σ H01 ,σ H02 ,…σ HOi )

[0202] k Hβ =torch.tensor(k Hβ1 ,k Hβ2 ,…k Hβi )

[0203] k Hα =torch.tensor(k Hα1 ,k Hα2 ,…k Hαi )

[0204] k V =torch.tensor(k V1 ,k V2 ,…k Vi )

[0205] y = torch.sqrt(k Hβ *k Hα *k V *k A *k V )

[0206] σ H =Z*σ H0 *y where i is the number of working conditions; similarly, σ is solved. F value

[0207] σ H σ F A new matrix is ​​formed by combining the vehicle load spectrum Torque and the number of cycles Ni:

[0208] and

[0209] For σ H σ FReorder the values ​​from largest to smallest, and divide the matrix into two parts: a finite lifetime interval and an infinite lifetime interval, using the stress values ​​at the inflection points of the infinite lifetime interval as the dividing points. Then renumber the parts.

[0210]

[0211] The allowable number of cycles under the corresponding operating condition is mapped using the SN curve. and

[0212] In this calculation, the long-life range is no longer included in the load spectrum life calculation. This allows us to obtain the component damage rate under each operating condition:

[0213] and

[0214] and the total damage rate of contact and bending

[0215] Based on Miner's damage rule, the equivalent load spectrum cycle number corresponding to Damage_Cotact and Damage_Bending with the same damage level is obtained as follows:

[0216]

[0217] By N H_eqivalent and N F_eqivalent Calculate the corresponding lifetime factor Z on the SN curve. NT and Y NT And the corresponding load spectrum equivalent allowable stress value:

[0218] and

[0219] Thus, the load spectrum safety factor is obtained:

[0220]

[0221] 9) Iterative calculation of the maximum torque of the reducer to obtain Gear_Tmax: The maximum torque of the reducer, Gear_Tmax, uses a loop similar to while to try to compare the matching between the input torque and the gear load-bearing capacity. In other words, it compares the calculated stress with the maximum allowable static load stress on the SN curve.

[0222]

[0223] Gear_Tmax = (Torque – step size) / 2.5, where 2.5 is the static torsion safety factor, obtained from environmental variables, or can be defined as other values, while the step size represents the calculation accuracy.

[0224] 10) Spatial layout calculation: The spatial layout function calculates the part envelope, shaft string length, and reducer total size using known conditions such as the gear parameters generated by pop, the intermediate gear parameters obtained by calculation, the spatial layout form, bearing size, accessory size, and machining process requirements.

[0225] 11) Bearing Load Calculation: Based on the bearing type, arrangement, and bearing package dimensions extracted from the Bearing_Database, the bearing load is calculated by decomposing the gear force to each bearing position using spatial analytical geometry principles, and by iteratively calculating the loads Fr and Fa through load balance. The torque variable under operating conditions is in vector form, while the loads Fr and Fa are in matrix form, with the number of rows representing the number of operating conditions and the number of columns representing the number of bearing positions.

[0226]

[0227] 12) Bearing load spectrum damage rate calculation: Using load spectrum torque, speed, and the number of cycles corresponding to the operating point, calculate the static load safety factor, 281 corrected life damage rate, 16281 corrected life damage rate, and inner ring raceway surface pressure (elliptic integral method) based on ISO76, ISO281, and ISO16281 standards. The damage calculation needs to consider the influence of materials and processes.

[0228] DAMAGE_BEARING_LOADCASES = ISO281 calculation of damage / (heat treatment process enhancement factor * special material enhancement factor);

[0229] The calculation process involves coefficients v1 and a related to the operating conditions. ISO , k, P during L 10 L nm Process into a tensor and obtain bearing damage:

[0230] Where n i Here, represents the rotational speed under a single operating condition, and 'i' represents the number of operating conditions. Using PyTorch code as an example:

[0231] Import torch

[0232] v1 = torch.tensor(v1, v1, ..., v1) i )

[0233] a ISO =torch.tensor(a ISO1 ,a ISO2 ,…a ISOi )

[0234] k = torch.tensor(k1, k2, ..., k) i )

[0235] P during =torch.tensor(P during1 ,P during2 ,…P duringi )

[0236] L 10 =torch.tensor(L 101 ,L 102 ,…L 10i )

[0237] L nm =torch.tensor(L nm1 ,L nm2 ,…L nmi )

[0238] bearing coefficient f C X, Y, and e need to construct the Lagrange interpolation function and the Vandermonde matrix L(x) to f C For example, make:

[0239]

[0240] in:

[0241] with f C The data source is ISO281, where j is the number of datasets, obtained through the Vandermonde inverse matrix.

[0242]

[0243] Solve Thus, the Lagrange interpolation function is obtained: f C =a1 + a2*x + a3*x 2 +…+a j *x j-1 .

[0244] The other coefficients X, Y, and e are solved using the same method.

[0245] 13) Calculation of CLTC operating loss of the reducer: CLTC_Powerloss uses ISO14179-1 / -2 and commercially available bearing loss calculation methods to calculate the losses of gears, bearings, oil seals, and oil churning, and calculates the average power consumption of the reducer at the CLTC operating point. Among them:

[0246] The lubricating oil level and component immersion coefficient are evaluated through the reducer's spatial layout, calculated gear geometry parameters, and selected bearing dimensions; the dynamic viscosity of the lubricating oil is obtained by taking points from the viscosity curve at specific temperatures; the cooling method of the electric drive system determines whether there are high-speed rotating seals at the motor input; the reducer's no-load drag loss is the sum of bearing churning loss + gear churning loss + seal rotational friction loss, where the seal loss calculation is based on ISO 14179-1 with the addition of a correction factor based on actual measurement experience; the churning loss between gears and bearings is based on ISO 14179-1 with the addition of the influence of oil pump speed control strategy and opening degree within the speed range on the immersion coefficient; the loading of gears and bearings... The losses depend on the friction coefficient, CLTC speed, and the load applied at the CLTC operating point. The friction coefficient of the gears is calculated based on ISO14179-2, and can be compared and analyzed with measured data from different projects to provide corrections that differ from the standard. These corrections are not limited to the correction factor but also include adjustments to the power of the curve. The rolling resistance coefficients of each bearing are stored in the bearing database and can be directly accessed. The bearing loading loss needs to be differentiated based on the bearing selection type on different shafts. The loss assessment calculation method will differ for different types. The total loss of the reducer CLTC is the sum of gear loading loss, bearing loading loss, and no-load dragging loss, divided by the number of CLTC operating points. The duration of a single operating point is 1 second.

[0247] 14) Reducer Weight Calculation: The total weight of the reducer is estimated based on the shaft dimensions and overall reducer dimensions. Bearing weights are directly read from the database. Simplified standard parts weights are used for components such as shims, retaining rings, lock nuts, breather plugs, and bolts. The differential weight has a linear regression relationship with the vehicle's maximum output torque. The housing weight is related to the electric drive cooling method and is corrected using a correction factor based on the cooling method.

[0248]

[0249] 15) Reducer cost calculation: The total cost of the reducer is obtained by multiplying the unit price of materials by the weight of parts or the number of standard parts. The unit price of bearings is directly retrieved from the database.

[0250] GearBox_Cost = Σweight(i)*unitcost(i) + processing technology additional value, where the processing technology comes from Process_file;

[0251] The cost of the differential is also calculated using a similar weight assessment method, combined with a wheel-end torque threshold and linear regression.

[0252]

[0253]

[0254] 16) Reducer Loss Calculation: The Loss_Map calculation for the full torque range and speed range operating conditions is the same as the PowerLoss calculation for the CLTC operating conditions in terms of the underlying logic method used, but the difference lies in:

[0255] Loss_Map outputs the power loss at each point on the entire speed and torque map, while CLTC_PowerLoss calculates the average power loss at the CLTC operating point.

[0256] Where n is the number of CLTC operating conditions.

[0257] CLTC_PowerLoss performs vector operations (operating torque and speed are each column vectors), while Loss_Map uses a matrix (operating torque is a column vector, and speed is a row vector, forming a matrix) to improve calculation speed.

[0258]

[0259] Where n represents the number of torque conditions and m represents the number of speed conditions. For the transfer function, f mn Calculate coefficients for PowerLoss, including constants and complex functions.

[0260] In the technical solution of this application, an improved genetic algorithm is used to decompose the fitness function into three levels;

[0261] The algorithm input parameters consist of two parts: one part is the population variables of the genetic algorithm, which are also the design variables of the reducer; the other part is the environmental variables of the entire algorithm, which may include the overall scheme route of electric drive development, such as oil cooling or water cooling, vehicle space dimensions, electric drive placement tilt angle, tire radius, the maximum output torque required by the vehicle, several typical electric drive layouts (including parallel shaft system, planetary gear system, placement position of the parallel shaft intermediate shaft (up / down, left / right), gear helix direction, differential placement direction, distinguished by numbers, which can be directly mapped to the corresponding calculation logic when the subroutine is called), energy recovery strategy, oil pump speed control strategy, NVH design requirements, reliability coefficient, shaft and gear process route, materials used for parts, tooth surface roughness, gear accuracy grade, tooth thickness tolerance, safety clearance required for shaft and gear system layout, lubricating oil properties, reducer operating temperature, oil level and output shaft height difference, oil seal lip diameter, minimum safety factor for shaft and gear parts life, material SN curve, system iteration accuracy requirements, initial population size of the genetic algorithm, and maximum number of iterations of the genetic algorithm.

[0262] Environmental factors are stored in a unified file, which can be edited and used externally and repeatedly called by the main program or subroutines during calculations. Parameter tables for various bearing types and bearing supplier quotations are stored in the Bearing_DataBase, which can be edited and used externally and repeatedly called by the main program or subroutines during calculations. CLTC operating points and load spectrum operating points represent the vehicle's output torque, speed, and number of cycles, which can be converted to the input based on the target speed ratio. CLTC operating points and load spectrum operating points are stored in CLTC operating point files and load spectrum operating point files respectively, and are repeatedly called by the main program or subroutines during calculations. The Process_File contains the shaft and gear machining process type, tool axial movement length, system integration safety distance, axial space requirements for circlip and pressure plate installation, shaft welding space requirements, bearing material and heat treatment process requirements, bearing assembly shoulder height requirements, differential slotted shaft outer diameter, and additional process costs. The Process_File is repeatedly called by the main program or subroutines during calculations. The Cost Database contains quotes from various parts suppliers. For parts that are related to materials and weight, the price is stored as a unit: yuan / kg.

[0263] The genetic algorithm's population variables include dozens of variables such as the number of gear teeth z, module mn, pressure angle α, helix angle β, displacement coefficient x, tooth width b, addendum coefficient hap, first-order center distance aw_First, second-order center distance aw_Second, total center distance aw_Output, input shaft bearing span L_Bearing_Input, intermediate shaft bearing span L_Bearing_Transfert, output shaft bearing span L_Bearing_Output, bearing type number Bearing_Type, and bearing model number Bearing_Nr. Among these, the number of teeth z, center distance variable aw, bearing type number, and bearing model number are discretized integers within the design range.

[0264] The requirements include tooth ratio and gear order frequency avoidance. If a planetary gear structure exists, the fitness conditions additionally include planetary gear uniform distribution assembly conditions and planetary gear sidefrequency requirements, serving as first-level population fitness conditions. Populations meeting these conditions are then used to calculate the next two fitness functions: Fitness_Function_Second (second-level fitness function) and Fitness_Function_Third (third-level fitness function). The system returns any 2-3 of the following: total reducer manufacturing cost (Cost), CLTC operating point power loss (Power_Loss), shaft and gear component life (Safty), and overlap ratio.

[0265] For populations that do not meet the first-level fitness condition, the system sequentially divides the population into hierarchical levels according to the constraints, returning progressively decreasing penalty coefficients to obtain the values ​​deviating from the normal Target_Values. It then skips the second and third-level fitness functions, directly performing non-dominated sorting and crowding calculation to obtain a series of pop schemes that are relatively at the Pareto front of the current population. These are retained as parent chromosomes and merged with offspring. The system then returns to the genetic algorithm for chromosome crossover, mutation, and selection operations, and proceeds to the next iterative calculation. During the iteration process, the number of teeth is used as the variable for chromosome variation, while other variables remain unchanged.

[0266] The genetic algorithm uses gear design and manufacturing constraints, the range of gear root circle diameter, the range of gear tip circle diameter, NVH design requirements, maximum static torque, gear load spectrum life, and the maximum input torque that the gear can withstand as constraints for the second-level fitness function. Populations satisfying these fitness conditions are then used to calculate the next-level fitness function, Fitness_Function_Third (i.e., the third-level fitness function). The algorithm returns any 2-3 of the following: total reducer manufacturing cost (Cost), CLTC operating point power loss (Power_Loss), shaft and gear component life (Safty), and overlap ratio.

[0267] For populations that do not meet the second-level fitness condition, the hierarchy is divided sequentially according to the constraint conditions, returning progressively decreasing penalty coefficients to obtain the values ​​deviating from the normal Target_Values. The third-level fitness function is skipped, and non-dominated sorting and crowding calculation are performed directly. A series of pop schemes with the Pareto frontier of the current population are obtained and retained as parent chromosomes, which are then merged with the offspring. The genetic algorithm is then used for chromosome crossover, mutation, and selection operations, followed by the next iterative calculation. During the iteration process, chromosomes other than the number of teeth, bearing type, and bearing model are treated as variables, while other variables remain unchanged.

[0268] The genetic algorithm uses bearing size and system layout matching, bearing runaway shoulder rate, bearing surface pressure, and bearing load spectrum life as constraints for a three-level fitness function. The engineering design requirements for these indicators are defined as upper and lower limits of the constraint matrix values, and penalty coefficients are set sequentially according to these parameters. The penalty coefficients decrease progressively to guide the machine to gradually adapt to the constraints in the fitness function, improving the convergence of the genetic algorithm.

[0269] The optimization objectives for a reducer can be any 2-3 of the following: total manufacturing cost (Cost), power loss at the CLTC operating point (Power_Loss), lifespan of shaft and gear components (Safty), and overlap ratio. The weights of the optimization objectives can be adjusted by setting coefficients. For example, mechanical loss and manufacturing cost can be converted into a total cost relationship, and mechanical loss can be converted according to the relationship between one kilowatt-hour and battery cost. For indicators that cannot be evaluated using cost, such as NVH requirements, weight coefficients can be directly defined.

[0270] Based on the three-level fitness function, after obtaining a local optimum while keeping the parts of the pop variables other than bearing type and bearing model unchanged, the genetic algorithm returns to the second-level fitness calculation and performs crossover, mutation, and selection operations on the bearing type and bearing model variables, and then repeats the fourth and fifth steps.

[0271] After returning to the second-level fitness calculation of the genetic algorithm, and obtaining a local optimum while keeping the number of teeth, bearing type, and bearing model variables in the pop variable unchanged, return to the first-level fitness calculation of the genetic algorithm, and perform crossover, mutation, and selection operations on variables other than the number of teeth, bearing type, and bearing model, and then repeat step six.

[0272] After a set number of iterations of t, the Pareto front solution set is obtained within the range of chromosome variables other than the number of teeth. After reaching the maximum number of iterations, or when the convergence error of the Pareto adjacent solution set is less than the set tolerance, a partial Pareto optimal solution is obtained.

[0273] Return to the first-level fitness calculation of the genetic algorithm and repeat step three. After the set number of iterations (s), obtain the Pareto front solution set for all chromosome variables. Obtain all Pareto optimal solutions after reaching the maximum number of iterations or when the convergence error of the Pareto neighbor solution set is less than the set tolerance. Then, sum the weights of the different objectives, merge them into a single objective, and output the minimum z value as the optimization result.

[0274] The calculation of some variables of the reducer gear parameters is based on the ISO6336 international standard. First, a reverse iteration is performed to calculate the parameters that meet the requirement of 2.5 times the static torque of the maximum torque output at the wheel end of the whole vehicle. The second step is to perform durability verification through the vehicle load spectrum. The static torque safety factor and durability safety factor can be adjusted according to the actual engineering needs.

[0275] The speed ratio generated by the number of teeth produced by each pop during the genetic algorithm calculation can be combined with the vehicle's operating point and converted into a vehicle load spectrum.

[0276] Typical spatial layouts of reducers are clearly distinguished and can be directly identified through specific environmental input variable values. Different layouts are mapped to different calculation logics, mainly used to calculate the spatial geometric load decomposition of gears and bearings, spatial layout parameters GearBox_Layout, and the height of the lubricating oil page.

[0277] The genetic algorithm returns the pop value corresponding to the optimal solution to the main program. The main program processes the pop variable and converts it into a readable format file. Based on the design variables under the optimal solution, it calculates other required output results, such as speed vector, torque vector, reducer weight, cost, and mechanical loss Loss_Map across the entire speed and torque range. The obtained calculation results have the following functions: direct output; as input conditions for other optimization algorithms to perform electric drive system-level optimization; inputting reducer design variables to external 3D / FEA software via API to establish a multiphysics proxy model; and inputting design variables and calculated performance index data into neural network training to build an enterprise-level electric drive large-scale model.

[0278] The algorithm does not rely on any secondary development of professional transmission design software. The methods used for load spectrum calculation, CLTC operating point calculation, efficiency MAP calculation, and large-scale data processing are all based on vector or matrix operations; the array is extended to two dimensions or more, and GPUs can be used for large-scale optimization calculations. This fundamentally avoids the CPU dependence of transmission design software performing condition-by-condition calculations, resulting in millions of times faster speeds and reduced hardware resource waste compared to traditional calculation methods, thus turning the genetic algorithm into a truly feasible solution.

[0279] It can be developed using languages ​​such as Matlab, C#, PyTorch, and Tensorflow.

[0280] According to another aspect of the embodiments of this application, an apparatus for generating an electric drive reducer scheme based on a multi-objective genetic algorithm is also provided for implementing the above-described method for generating an electric drive reducer scheme based on a multi-objective genetic algorithm.

[0281] According to another aspect of the embodiments of this application, a server or terminal is also provided for implementing the above-described method for generating an electric drive reducer scheme based on a multi-objective genetic algorithm.

[0282] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0283] Embodiments of this application also provide a storage medium. Optionally, in this embodiment, the storage medium can be used to execute program code for a method of generating an electric drive reducer scheme based on a multi-objective genetic algorithm.

[0284] Optionally, in this embodiment, the storage medium may be located on at least one of the network devices in the network shown in the above embodiment.

[0285] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated here.

[0286] Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0287] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0288] If the integrated units in the above embodiments are implemented as software functional units and sold or used as independent products, they can be stored in the aforementioned computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause one or more computer devices (which may be personal computers, servers, or network devices, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0289] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0290] In the several embodiments provided in this application, it should be understood that the disclosed client can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection between units or modules, and may be electrical or other forms.

[0291] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0292] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0293] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for generating electric drive reducer schemes based on multi-objective genetic algorithms, characterized in that, include: The design variables of the electric drive reducer for new energy vehicles are obtained, wherein the design variables are design parameters that meet the design requirements of the electric drive reducer in terms of vehicle economy and power. Using the design variables as input variables, a multi-objective genetic algorithm is used to iteratively optimize the design scheme of the electric drive reducer to obtain candidate design schemes for the electric drive reducer. During the iterative optimization process, the variables are processed in a hierarchical manner according to the strength of their correlation. A three-level fitness function based on matrix operations is used to hierarchically sort the constraints at each level contained in the fitness function of the multi-objective genetic algorithm. Gradient penalty coefficients are used to gradually reduce the objective function value of the multi-objective genetic algorithm. This is to prevent the loss of excellent populations through a gradual adaptation mechanism and to avoid outputting incorrect results or getting trapped in local optima due to premature convergence. Select the target design scheme that meets the Pareto optimal solution of the electric drive reducer from the candidate design schemes.

2. The method according to claim 1, characterized in that, Using the aforementioned design variables as input variables, a multi-objective genetic algorithm is used to iteratively optimize the design scheme of the electric drive reducer, including: Using the design variables as input variables, the population pop of the multi-objective genetic algorithm is initialized to obtain the pop matrix, wherein the number of teeth, center distance, bearing type number and bearing model number in the population variables of the multi-objective genetic algorithm are discretized integers within the corresponding design range; A three-level adaptation mechanism is implemented: a three-level fitness function based on matrix operations is used to hierarchically sort the constraints at each level contained in the fitness function of the multi-objective genetic algorithm, and a gradient penalty coefficient is used to gradually reduce the objective function value of the multi-objective genetic algorithm. This step-by-step adaptation mechanism prevents the loss of excellent populations and avoids outputting incorrect results or getting trapped in local optima due to premature convergence.

3. The method according to claim 2, characterized in that, The multi-objective genetic algorithm employs a three-level fitness function based on matrix operations to hierarchically sort the constraints at each level within its fitness function, and uses gradient penalty coefficients to progressively reduce the objective function value of the multi-objective genetic algorithm, including: In the first-level adaptation mechanism, for population pop that does not meet the population fitness conditions of the first-level fitness function, the penalty coefficient is returned in a progressively decreasing manner according to the population fitness conditions of the first-level fitness function, and the second-level fitness function and the third-level fitness function are skipped. The multi-objective genetic algorithm is directly used to perform chromosome crossover, mutation and selection operations, so as to carry out the next iteration in the first-level adaptation mechanism. In the iteration process of the first-level adaptation mechanism, the number of teeth is used as the population variable for chromosome changes, while other population variables remain unchanged. Populations that meet the fitness conditions of the first-level fitness function are popped and enter the second-level adaptation mechanism. In the second-level adaptation mechanism, for population pop that does not meet the population fitness conditions of the second-level fitness function, the penalty coefficient is returned in a progressively decreasing manner according to the population fitness conditions of the second-level fitness function, and the third-level fitness function is skipped. The multi-objective genetic algorithm is directly used to perform chromosome crossover, mutation, and selection operations, so as to carry out the next iteration in the second-level adaptation mechanism. During the iteration, the number of teeth, bearing type number, and bearing model number remain unchanged, while other variables are used as variables for chromosome changes. Populations that meet the fitness conditions of the second-level fitness function are popped and enter the third-level adaptation mechanism; In the third-level adaptation mechanism, for population pop that does not meet the population fitness conditions of the third-level fitness function, a progressively decreasing penalty coefficient is returned according to the population fitness conditions of the third-level fitness function. The multi-objective genetic algorithm is used to perform chromosome crossover, mutation, and selection operations, thereby performing the next iteration in the third-level adaptation mechanism. During the iteration process, the bearing type number and bearing model number are used as variables, while other variables remain unchanged. Populations that meet the fitness conditions of the third-level fitness function are used as candidate design schemes, or the third-level adaptation mechanism is run again after crossover, mutation, and selection operations are performed on the bearing type number and bearing model number.

4. The method according to claim 3, characterized in that, Populations that satisfy the population fitness conditions of the third-level fitness function are considered as candidate design schemes, including: Populations that satisfy at least two of the following optimization objectives are selected as candidate design schemes: total manufacturing cost of reducer, power loss at CLTC operating point, life of shaft and gear components, and contact ratio.

5. The method according to claim 3, characterized in that, For population pops that do not meet the population fitness conditions of the first-level fitness function, a progressively decreasing penalty coefficient is returned according to the population fitness conditions of the first-level fitness function, including: The population fitness conditions for obtaining the first-level fitness function are: tooth ratio requirement, gear order frequency avoidance requirement. If a planetary gear structure exists, the fitness conditions additionally include planetary gear uniform distribution assembly condition and planetary gear side frequency requirement. Generate a first constraint matrix to represent the population fitness conditions of the first-level fitness function, wherein the population fitness conditions of the first-level fitness function include the first-level gear ratio f1, the second-level gear ratio f2, the total gear ratio f3, the Hunting Ratio f4, and the gear order frequency avoidance requirement f5. If a planetary gear structure exists, the population fitness conditions of the first-level fitness function also include two additional condition elements: the planetary gear uniformly distributed assembly condition f6 and the planetary gear side frequency requirement f7. When popping a population that does not meet the fitness conditions of the first-level fitness function by using the first constraint matrix, the penalty coefficients that decrease step by step are returned according to the position of the condition element in the first constraint matrix.

6. The method according to claim 3, characterized in that, For population pops that do not meet the population fitness conditions of the second-level fitness function, a progressively decreasing penalty coefficient is returned according to the population fitness conditions of the second-level fitness function, including: The population fitness conditions for obtaining the second-level fitness function are: gear design and manufacturing constraints, gear root circle diameter range, gear tip circle diameter range, NVH design requirements, maximum static torque, gear load spectrum life, and maximum input torque that the gear can withstand. A second constraint matrix is ​​generated to represent the population fitness conditions of the second-level fitness function. The second constraint matrix includes: first-level gear total displacement coefficient f8, second-level gear total displacement coefficient f9, first-level gear axial overlap ratio f10, first-level gear end face overlap ratio f11, first-level gear total overlap ratio f12, second-level gear axial overlap ratio f13, second-level gear end face overlap ratio f14, second-level gear total overlap ratio f15, first-level pinion chamfer tooth thickness f16, first-level wheel chamfer tooth thickness f17, second-level pinion chamfer tooth thickness f18, second-level wheel chamfer tooth thickness f19, first-level pinion addendum circle diameter f20, first-level wheel addendum circle diameter f21, second-level pinion addendum circle diameter f22, second-level wheel addendum circle diameter f23, first-level pinion root circle diameter f24, and first-level wheel... The following parameters are considered: I-tooth root circle diameter f25, II-stage pinion root circle diameter f26, II-stage wheel root circle diameter f27, distance between the first-stage pinion meshing start point circle and the base circle f28, distance between the second-stage pinion meshing start point circle and the base circle f29, distance between the first-stage pinion involute start circle and the base circle f30, distance between the second-stage pinion involute start circle and the base circle f31, slip ratio at the first-stage gear meshing start point f32, slip ratio at the first-stage gear meshing end point f33, slip ratio at the second-stage gear meshing start point f34, slip ratio at the second-stage gear meshing end point f35, maximum torque that the gear can withstand converted to the output end f36, contact safety factor of the first-stage gear checked according to the load spectrum f37, bending safety factor of the first-stage gear f38, contact safety factor of the second-stage gear f39, bending safety factor of the second-stage gear f40; When popping a population that does not meet the fitness conditions of the second-level fitness function by using the second constraint matrix, the penalty coefficients are returned in progressively decreasing order according to the position of the condition element in the second constraint matrix.

7. The method according to claim 3, characterized in that, For population pops that do not meet the population fitness conditions of the third-level fitness function, a progressively decreasing penalty coefficient is returned according to the population fitness conditions of the third-level fitness function, including: The population fitness conditions for obtaining the third-level fitness function are: bearing size and system layout matching degree, bearing creep shoulder rate, bearing surface pressure, and bearing load spectrum life. Generate a third constraint matrix to represent the population fitness conditions of the third-level fitness function, wherein the third constraint matrix includes: the thinnest wall thickness of the bearing housing f41, the bearing overrun shoulder rate f42, the bearing surface pressure safety factor f43, the ISO281 bearing damage rate f44, and the ISO16281 bearing damage rate f45. When popping a population that does not meet the fitness conditions of the third-level fitness function is determined using the third constraint matrix, a progressively decreasing penalty coefficient is returned according to the position of the condition element in the third constraint matrix.

8. The method according to any one of claims 1 to 7, characterized in that, The multi-objective genetic algorithm uses matrix operations for pop population, fitness function, lifetime calculation related to vehicle load spectrum and CLTC, and efficiency calculation. For externally introduced component cost data, linear regression is used to predict its relationship with weight and load-bearing capacity; For the discretization calculation coefficients of bearing parts introduced in ISO 281, a known function curve is created using the Lagrange interpolation method.

9. A device for generating an electric drive reducer scheme based on a multi-objective genetic algorithm, characterized in that, include: The acquisition unit is used to acquire the design variables of the electric drive reducer of the new energy vehicle, wherein the design variables are design parameters that meet the design requirements of the electric drive reducer that meet the overall vehicle economy and power performance. An iterative unit is used to iteratively optimize the design scheme of the electric drive reducer using the design variables as input variables and a multi-objective genetic algorithm to obtain candidate design schemes for the electric drive reducer. During the iterative optimization process, the variables are processed in a hierarchical manner according to the strength of their correlation. A three-level fitness function based on matrix operations is used to hierarchically sort the constraints at each level contained in the fitness function of the multi-objective genetic algorithm. Gradient penalty coefficients are used to gradually reduce the objective function value of the multi-objective genetic algorithm. This is to prevent the loss of excellent populations and avoid outputting incorrect results or getting trapped in local optima due to premature convergence through a gradual adaptation mechanism. The determining unit is used to select a target design scheme that meets the Pareto optimal solution of the electric drive reducer from the candidate design schemes.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the method described in any one of claims 1 to 8 through the computer program.

Images