Multi-pass hobbing parameter decision-making method

Abstract

The invention discloses a multi-pass gear hobbing parameter decision-making method, relates to the technical field of gear intelligent manufacturing and process optimization, and aims at solving the problems that in gear hobbing parameter optimization, a large language model is insufficient in the aspects of initial search range setting, dynamic adjustment and optimization of the center of gravity, fuzzy preference quantization and the like. The method comprises the following steps: acquiring an initial hobbing parameter to be optimized; determining a to-be-optimized initial hobbing parameter range based on the large language model; based on a multi-target artemisinin optimization algorithm and the initial gear hobbing parameter range, an optimized gear hobbing parameter solution set and a corresponding target function value are obtained; and based on a decision algorithm of a user fuzzy demand, decision making is carried out on the optimized hobbing parameter solution set and a corresponding objective function value, and a final hobbing parameter meeting the user demand is output. According to the method, a multi-target artemisinin optimization algorithm and a decision algorithm based on user fuzzy requirements are adopted, an optimized hobbing parameter solution set with better quality and wider distribution is obtained, and a final hobbing parameter scheme which most meets the user requirements is decided.

Description

Technical Field

[0001] This invention relates to a multi-pass gear hobbing parameter decision method, belonging to the field of intelligent gear manufacturing and process optimization technology. Background Technology

[0002] In multi-pass gear hobbing, the rational configuration of process parameters has a significant impact on processing energy consumption and production costs. Currently, the determination of gear hobbing process parameters mainly relies on engineers' experience and trial cutting experiments, which suffers from problems such as low parameter optimization efficiency, difficulty in balancing multiple conflicting objectives, and a lack of intelligent decision-making processes. Furthermore, most research on gear hobbing parameters focuses on single-pass processing, with a lack of research on multi-pass gear hobbing.

[0003] Current research on gear hobbing parameter optimization mainly focuses on single-objective optimization or traditional optimization algorithms, lacking a comprehensive solution that organically integrates large language model prior knowledge guidance, adaptive multi-objective optimization algorithms, and intelligent decision-making techniques. In particular, there are significant shortcomings in utilizing large language models to provide an initial search range, dynamically adjusting the optimization focus based on the convergence difficulty of the objective, and transforming user fuzzy preferences into precise weights. Summary of the Invention

[0004] The purpose of this invention is to effectively utilize a large language model to generate the initial range of hobbing parameters, employ a multi-objective artemisinin algorithm with an adaptive pressure mechanism to optimize complex multi-pass hobbing process parameters, and use a decision algorithm based on user fuzzy requirements to select the final hobbing parameter scheme that best meets user needs, thus providing intelligent decision support for the optimization of multi-pass hobbing processing parameters.

[0005] To achieve the above objectives, the present invention is implemented using the following technical solution:

[0006] In a first aspect, the present invention proposes a multi-pass gear hobbing parameter decision method, comprising:

[0007] Obtain the initial hobbing parameters to be optimized;

[0008] The initial range of hobbing parameters to be optimized is determined based on a large language model;

[0009] Based on the multi-objective artemisinin optimization algorithm and the initial range of gear hobbing parameters, the optimized gear hobbing parameter solution set and its corresponding objective function value are obtained.

[0010] The decision-making algorithm based on user fuzzy requirements makes decisions on the optimized gear hobbing parameter solution set and its corresponding objective function value, and outputs the final gear hobbing parameters that meet the user's requirements.

[0011] Furthermore, the initial hobbing parameters to be optimized include the number of machining passes r and the spindle speed for each pass. Axial feed rate for each pass Depth of cut in each pass ,in, .

[0012] Furthermore, the determination of the initial hobbing parameter range to be optimized based on the large language model includes:

[0013] Determine the basic information of the gear workpiece to be machined and the hob;

[0014] The basic information is provided to the large language model, and the initial range of the initial hobbing parameters to be optimized is determined by the large language model.

[0015] The basic information of the gear workpiece to be processed and the hob includes the model of the gear hobbing machine, the gear module m, the pressure angle a, the number of teeth Za, the tooth thickness B0, the hob diameter d, and the number of hob heads K.

[0016] Furthermore, obtaining the optimized gear hobbing parameter solution set and its corresponding objective function value based on the multi-objective artemisinin optimization algorithm and the initial gear hobbing parameter range includes:

[0017] The total energy consumption and total production cost of gear hobbing are taken as two optimization objectives that need to be minimized simultaneously, and the objective function for optimizing the gear hobbing parameters is determined. The objective function for optimizing the gear hobbing parameters is expressed as follows:

[0018] (1)

[0019] in, The objective function for optimizing gear hobbing parameters is represented by E_total and C_total, which are two optimization objectives that need to be minimized simultaneously. E_total is the total energy consumption during the gear hobbing process, and C_total is the total production cost.

[0020] The operating parameters based on the multi-objective artemisinin optimization algorithm and the process constraints of the gear hobbing process are set. The operating parameters include the maximum dimension dim_max of the variable determined by the maximum number of machining passes, the number of objectives obj_no = 2, the maximum number of fitness evaluations MaxFEs, the population size N, and the maximum capacity archive_size of the external archive used to store non-dominated solutions. The process constraints of the gear hobbing process include machine tool power constraints, workpiece surface roughness constraints, tooth profile error constraints, and constraints on the allocation rules between the total depth of cut and the depth of cut in each pass.

[0021] Using a large language model, N individuals are generated based on the initial range of gear hobbing parameters to form an initial population. Each individual is a vector of variable dimension. The vector structure of an individual with trace number r is as follows: ; Initialize all non-dominated solutions Archive_X and their corresponding objective function values ​​Archive_F in the external archive storage used to store non-dominated solutions to be empty; Initialize the energy stagnation counter stagnation_E and the cost stagnation counter stagnation_C used for the adaptive pressure mechanism to be 0;

[0022] Traverse the N individuals in the initial population, calculate the objective function value of the gear hobbing process corresponding to each individual in turn, and accumulate the fitness evaluation times FEs; perform non-dominated sorting on the current population and its objective function values, filter out all non-dominated solutions, and store them in the external archive Archive;

[0023] Based on the adaptive pressure mechanism, dynamic parameters are calculated, adaptive leader selection and population position update are performed during the iteration process until the current fitness evaluation number FEs reaches the maximum value MaxFEs.

[0024] For parameters exceeding the preset upper and lower bounds, their values ​​are set as the corresponding boundary values, and the cutting depth of each pass is summed and normalized to calculate the objective function value of the new solution, and the current fitness evaluation count (FEs) is increased accordingly. The new solution is compared with its parent in terms of Pareto dominance: if the new solution dominates its parent, it is replaced; if they do not dominate each other, they are replaced with a 50% probability. The updated population is merged with the current archive, and a new set of non-dominated solutions is selected from it to form the next generation of archive. If the archive exceeds the preset maximum capacity, it is sorted and pruned according to the crowding of solutions.

[0025] Compare the minimum energy consumption and minimum cost in the new save file with the previous save file. If a certain goal is not improved, increment the corresponding energy consumption stagnation counter stagnation_E or cost stagnation counter stagnation_C by 1; otherwise, reset it to zero.

[0026] Determine whether the current fitness evaluation count FEs has reached the maximum value MaxFEs. If it has, terminate the loop and output all non-dominated solutions Archive_X and their corresponding objective function values ​​Archive_F stored in the external archive as the final optimized hobbing parameter solution set; otherwise, continue the optimization iteration based on the adaptive pressure mechanism.

[0027] Furthermore, the calculation of dynamic parameters includes calculating two dynamic parameters based on the current fitness evaluation counts FEs: an exploration / mining balance parameter K and a variable-asynchronous long decay factor E_decay, used to control the population update strategy. The formula for calculating the exploration / mining balance parameter K is as follows:

[0028] (2)

[0029] The formula for calculating the variable asynchronous long decay factor E_decay is:

[0030] (3)

[0031] Furthermore, the adaptive leader selection includes:

[0032] Select the energy consumption leader (Leader_E), cost leader (Leader_C), and diversity leader (Leader_D) from the external archive to guide population renewal;

[0033] Get the current energy stagnation counter stagnation_E and cost stagnation counter stagnation_C;

[0034] The probability P_cost_leader of selecting the cost leader is calculated based on the energy stagnation counter stagnation_E and the cost stagnation counter stagnation_C. The formula for calculating the probability P_cost_leader is as follows:

[0035] (4)

[0036] When executing the mining strategy, a random number between [0,1] is generated. If the random number is less than P_cost_leader, the cost leader Leader_C is selected; otherwise, the energy leader Leader_E is selected. When executing the exploration strategy, the diversity leader Leader_D is selected to maintain population diversity.

[0037] Furthermore, the adaptive population update includes:

[0038] Generate a random number rand between [0,1] for each individual;

[0039] When the random number rand is less than the exploration / mining balance parameter K, the individual executes a global exploration strategy, conducting a broad search throughout the entire search space based on the large variable-asynchronous long decay factor E_decay.

[0040] When the random number rand >= the exploration / mining balance parameter K, the individual executes a local mining strategy, performing a fine search in the current dense population area based on the small variable-asynchronous long decay factor E_decay;

[0041] In the final stage of the update process, for each newly generated solution, with a 20% probability, some parameter genes of the newly generated solution are directly replaced with the corresponding genes of the selected leader.

[0042] Furthermore, the decision-making algorithm based on user fuzzy requirements makes decisions on the optimized gear hobbing parameter solution set and its corresponding objective function value, and outputs the final gear hobbing parameters that meet user requirements, including:

[0043] Obtain vague user needs;

[0044] Transform the user's vague requirements into precise weights;

[0045] Based on the precise weights of the user's fuzzy requirements and the optimized solution set of hobbing parameters, the final hobbing parameters that meet the user's requirements are determined.

[0046] Furthermore, the process of converting the user's fuzzy requirements into precise weights includes:

[0047] The system obtains the user's fuzzy language evaluation of the two optimization objectives of energy consumption and cost through an interactive interface; wherein the fuzzy language evaluation is selected from a predefined set of fuzzy language evaluations containing five importance levels.

[0048] The fuzzy language evaluation selected by the user for each optimization objective is mapped to a preset triangular fuzzy number (l, m, u), where l is the lower bound, m is the most likely value, and u is the upper bound.

[0049] The centroid method is used to defuzzify the triangular fuzzy numbers, transforming the fuzzy language evaluation selected by the user for each optimization objective into precise initial weight values. Initial weight values The calculation formula is:

[0050] (5)

[0051] in, and These correspond to the initial weight values ​​for energy consumption and cost, respectively.

[0052] Calculate the initial weight values After normalization, the final precise weight vector `weights` used for decision-making is obtained. The formula for calculating the precise weight vector `weights` is:

[0053] (6).

[0054] Furthermore, the determination of the final gear parameters that meet the user's needs based on the precise weights of the user's fuzzy requirements and the optimized gear parameter solution set includes:

[0055] Based on all non-dominated solutions Archive_X output by the multi-objective artemisinin optimization algorithm, iterate through the corresponding objective function values ​​Archive_F to find the minimum and maximum values ​​of energy consumption and cost, and construct virtual ideal and anti-ideal solutions respectively.

[0056] Based on the ideal solution as the first row, the anti-ideal solution as the second row, and the objective function value Archive_F as the subsequent rows, construct a comprehensive matrix extended_matrix that contains all alternative solutions and the optimal and worst benchmarks;

[0057] Linear normalization is performed on all solutions, including ideal and antiideal solutions, to construct the normalization matrix norm_matrix. The formula for calculating the normalization matrix norm_matrix is ​​as follows:

[0058] (7)

[0059] Multiply the normalized matrix norm_matrix by the precise weight vector weights to obtain the weighted normalized matrix weighted_matrix;

[0060] The weighted normalized matrix is ​​summed row-wise to obtain the aggregate score for each solution. This aggregate score is directly recorded as the corresponding alternative solution S_alternatives. The upward utility K_plus relative to the ideal solution and the downward utility K_minus relative to the anti-ideal solution for each alternative solution S_alternatives are calculated using the following formulas:

[0061] (8)

[0062] (9)

[0063] Where S_ideal and S_anti are the aggregate scores of the ideal solution and the antiideal solution, respectively;

[0064] By aggregating K_plus and K_minus using the comprehensive utility function f(k), the final evaluation value of each alternative solution S_alternatives is obtained. The solution with the highest evaluation value is selected as the final gear rolling parameter scheme that best meets the user's fuzzy requirements. The formula for calculating the comprehensive utility function f(k) is as follows:

[0065] (10).

[0066] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0067] 1. This invention employs a multi-objective artemisinin optimization algorithm. The algorithm uses two stall counters—"energy consumption" and "cost"—to determine the convergence difficulty of each objective in real time and dynamically allocates computational resources to the more difficult objective, achieving intelligent adjustment of the optimization focus. This mechanism, combined with a multi-leader strategy of "energy consumption leader," "cost leader," and "diversity leader," and in conjunction with the artemisinin algorithm update operator that combines global exploration and local exploitation, effectively avoids premature convergence and efficiently finds a higher-quality, more widely distributed set of optimized gear hobbing parameters.

[0068] 2. This invention combines a multi-objective artemisinin optimization algorithm with a decision-making algorithm based on user fuzzy requirements, which facilitates precise control of the desired processing results in engineering. First, this invention scientifically transforms the user's fuzzy language preferences into precise weights through a fuzzy weight transformation module. Then, using a decision-making algorithm based on user fuzzy requirements, the final gear hobbing parameter scheme that best meets the user's needs is selected from numerous optimized gear hobbing parameter solutions, thus better aligning with actual production requirements and achieving intelligent decision-making. Attached Figure Description

[0069] Figure 1 This is a schematic diagram of the overall basic process in an embodiment of the present invention;

[0070] Figure 2 This is a schematic diagram illustrating the process of optimizing gear parameters using a multi-objective artemisinin algorithm in an embodiment of the present invention. Detailed Implementation

[0071] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application are clearly and completely described below. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0072] Example 1:

[0073] This embodiment proposes a multi-pass gear hobbing parameter decision method, the overall basic flowchart of which is shown below. Figure 1 As shown, it includes:

[0074] Obtain the initial hobbing parameters to be optimized;

[0075] The initial range of hobbing parameters to be optimized is determined based on a large language model;

[0076] Based on the multi-objective artemisinin optimization algorithm and the initial range of gear hobbing parameters, the optimized gear hobbing parameter solution set and its corresponding objective function value are obtained.

[0077] The decision-making algorithm based on user fuzzy requirements makes decisions on the optimal set of hobbing parameters and their corresponding objective function values, and outputs the final hobbing parameters that meet the user's requirements.

[0078] In this embodiment, the initial hobbing parameters to be optimized include the number of machining passes r and the spindle speed for each pass. Axial feed rate for each pass Depth of cut in each pass ,in, .

[0079] In this embodiment, determining the initial range of hobbing parameters to be optimized based on the large language model includes:

[0080] Determine the basic information of the gear workpiece to be machined and the hob;

[0081] Provide basic information to deepseek, which then determines a reasonable initial range for the initial hobbing parameters to be optimized.

[0082] The basic information of the gear workpiece to be processed and the hob includes the model of the gear hobbing machine, the gear module m, the pressure angle a, the number of teeth Za, the tooth thickness B0, the hob diameter d, and the number of hob heads K.

[0083] In this embodiment, based on the multi-objective artemisinin optimization algorithm and the initial range of gear hobbing parameters, the optimized gear hobbing parameter solution set and its corresponding objective function value are obtained. The flowchart illustrating the multi-objective artemisinin optimization algorithm for optimizing gear hobbing parameters is shown below. Figure 2 As shown, it includes:

[0084] Determine the objective function for hobbing parameter optimization and initialize the algorithm parameters: Determine the total machining energy consumption E_total and the total production cost C_total as the two optimization objectives that need to be minimized simultaneously, and define the objective function hobbing_model accordingly:

[0085] (1)

[0086] Here, E_total and C_total are two optimization objectives that need to be minimized simultaneously. E_total is the total energy consumption in the gear hobbing process, measured in joules (J), including standby energy consumption, air cutting energy consumption, and cutting energy consumption. C_total is the total production cost, measured in yuan (¥), consisting of energy cost, tool cost, and time-related cost. The algorithm's operating parameters are set, including the maximum dimension dim_max (determined by the maximum number of machining passes), the number of objectives obj_no = 2 (energy consumption and cost), the maximum fitness evaluation count MaxFEs = 5000, the population size N = 100, and the maximum external archive size for storing non-dominated solutions archive_size = 100. The process constraints of the gear hobbing process are defined and handled, including machine power constraints, workpiece surface roughness constraints, tooth profile error constraints, and allocation rule constraints regarding the total depth of cut and the depth of cut in each pass. The values ​​of the gear hobbing parameters to be optimized must be within the range obtained by deepseek.

[0087] An initial population is randomly generated within the parameter range provided by the large language model: N individuals are randomly generated based on the initial gear hobbing parameter range obtained from deepseek, forming the initial population. Each individual is a vector with variable dimensions, the dimension of which is determined by the number of randomly generated tracts. Its structure is as follows: Furthermore, the cutting depth of each pass is normalized after random generation to ensure that its sum is exactly equal to the total cutting depth; all non-dominated solutions Archive_X and their corresponding objective function values ​​Archive_F in the external archive used to store non-dominated solutions are initialized to empty, where Archive_X is used to store decision variables, i.e., the combination of gear hobbing process parameters, and Archive_F is used to store objective function values; the energy stagnation counter stagnation_E and the cost stagnation counter stagnation_C used for the adaptive pressure mechanism are initialized to 0;

[0088] Calculate the initial objective function value and update the archive: Traverse the N individuals in the initial population, calculate the objective function value of the gear hobbing process corresponding to each individual in turn, and accumulate the fitness evaluation times FEs; perform non-dominated sorting on the current population and its objective function values, filter out all non-dominated solutions, and store them in the external archive Archive;

[0089] Iterative optimization based on adaptive pressure mechanism: Based on the adaptive pressure mechanism, dynamic parameters are calculated, adaptive leader selection and population position update are performed during the iteration process until the current fitness evaluation number FEs reaches the maximum value MaxFEs;

[0090] Boundary handling and archive updates are performed as follows: Boundary handling ensures that all newly generated parameters are within preset ranges, preventing invalid or infeasible solutions from exceeding these limits. For parameters exceeding preset upper and lower bounds, this includes: setting their values ​​to the corresponding boundary values ​​and summing and normalizing the cutting depths of each pass to satisfy the total cutting depth constraint; calculating the objective function value of the new solution and correspondingly increasing the current fitness evaluation count (FEs). Archive updates specifically include: comparing the Pareto dominance relationship between the newly generated solution and its parent. If the new solution dominates its parent, it is replaced; otherwise, it is replaced with a 50% probability to increase population diversity. The updated population is merged with the current archive, and new non-dominated solutions are selected to form the next generation of archive. If the archive exceeds the preset maximum capacity, solutions are sorted and pruned according to their crowding, prioritizing the elimination of solutions located in the most crowded and densest regions of the Pareto front, while retaining boundary points of the front and solutions distributed in sparse regions. This ensures the quality of the solution set while maximizing its diversity and uniformity of distribution.

[0091] Update the adaptive stress mechanism: compare the minimum energy consumption E_min and minimum cost C_min in the new archive with the previous archive. If a certain goal is not improved, increment the corresponding energy consumption stagnation counter stagnation_E or cost stagnation counter stagnation_C by 1; otherwise, reset it to zero, thus completing the feedback loop of the adaptive stress mechanism.

[0092] Determine whether the current fitness evaluation count FEs has reached the maximum value MaxFEs. If it has, terminate the loop and output all non-dominated solutions Archive_X and their corresponding objective function values ​​Archive_F stored in the external archive as the final optimized hobbing parameter solution set; otherwise, continue the optimization iteration based on the adaptive pressure mechanism.

[0093] In this embodiment, the calculation of dynamic parameters includes calculating two dynamic parameters, the exploration / mining balance parameter K and the variable-asynchronous long decay factor E_decay, based on the current fitness evaluation count FEs. The calculation formula for the exploration / mining balance parameter K is as follows:

[0094] (2)

[0095] The formula for calculating the variable asynchronous long decay factor E_decay is:

[0096] (3)

[0097] In this embodiment, adaptive leader selection includes:

[0098] Energy Leader (Leader_E), Cost Leader (Leader_C), and Diversity Leader (Leader_D) are selected from the external archive to guide population updates. Among them, Energy Leader (Leader_E) is the solution with the lowest energy consumption in the archive, used to guide the population to converge towards low energy consumption; Cost Leader (Leader_C) is the solution with the lowest cost in the archive, used to guide the population to converge towards low cost; and Diversity Leader (Leader_D) is the solution with the highest crowding in the archive, used to guide the population to explore the sparse region of the Pareto front and improve the diversity of the solution set.

[0099] Get the current energy stagnation counter stagnation_E and cost stagnation counter stagnation_C;

[0100] The probability P_cost_leader of selecting the cost leader is calculated based on the energy stagnation counter stagnation_E and the cost stagnation counter stagnation_C. The formula for calculating the probability P_cost_leader is as follows:

[0101] (4)

[0102] When executing the mining strategy, a random number between [0,1] is generated. If this random number is less than P_cost_leader, the cost leader Leader_C is selected; otherwise, the energy leader Leader_E is selected. This mechanism ensures that when the convergence of a certain objective stagnates, its corresponding stagnation counter increases. The larger the stagnation counter of an objective, the more difficult it is to optimize. Therefore, the leader (Leader_E or Leader_C) corresponding to that objective is selected with a higher probability, thereby dynamically allocating computing resources to the more difficult objective to optimize.

[0103] When implementing the exploration strategy, select a diversity leader (Leader_D) to maintain population diversity.

[0104] Adaptive population updates include:

[0105] Generate a random number rand between [0,1] for each individual;

[0106] When the random number rand is less than the exploration / mining balance parameter K, the individual executes a global exploration strategy, which performs a broad search in the entire search space based on the large variable-asynchronous long decay factor E_decay in order to escape local optima;

[0107] When the random number rand >= the exploration / mining balance parameter K, the individual executes a local mining strategy, performing a fine search in the current dense population area based on the small variable asynchronous long decay factor E_decay, in order to accelerate the convergence of the algorithm;

[0108] In the final stage of the update process, for each newly generated solution, with a 20% probability, some parameter genes of the newly generated solution are directly replaced with the corresponding genes of the selected leader to consolidate and propagate optimal information and enhance the convergence of the algorithm.

[0109] In this embodiment, the decision algorithm based on user fuzzy requirements makes a decision on the optimized gear hobbing parameter solution set and its corresponding objective function value, and outputs the final gear hobbing parameters that meet the user's requirements, including:

[0110] Obtain vague user needs;

[0111] Transform vague user needs into precise weights;

[0112] Based on the precise weights of the user's fuzzy requirements and the optimized solution set of hobbing parameters, the final hobbing parameters that meet the user's requirements are determined.

[0113] In this embodiment, converting vague user requirements into precise weights includes:

[0114] The system obtains users' fuzzy language evaluations of the two optimization objectives of energy consumption and cost through an interactive interface. The fuzzy language evaluation is selected from five predefined levels: 1: Not very important; 2: Not important; 3: Moderate; 4: Important; 5: Very important.

[0115] The fuzzy language evaluation selected by the user for each optimization objective is mapped to a preset triangular fuzzy number (l, m, u), where l is the lower bound, m is the most likely value, and u is the upper bound; for example, "not important" corresponds to the triangular fuzzy number [0.0, 0.1, 0.2]; "important" corresponds to the triangular fuzzy number [0.5, 0.7, 0.9].

[0116] The centroid method is used to defuzzify the triangular fuzzy numbers, transforming the fuzzy language evaluation selected by the user for each optimization objective into precise initial weight values. Initial weight values The calculation formula is:

[0117] (5)

[0118] Where l, m, and u are the values ​​of the triangular fuzzy numbers. and These correspond to the initial weight values ​​for energy consumption and cost, respectively.

[0119] Calculate the initial weight values After normalization, the final precise weight vector `weights` used for decision-making is obtained. The formula for calculating the precise weight vector `weights` is:

[0120] (6)

[0121] In this embodiment, based on the precise weights of the user's fuzzy requirements and the optimized set of gear hobbing parameters, the final gear hobbing parameters that meet the user's requirements are determined as follows:

[0122] Based on all non-dominated solutions Archive_X output by the multi-objective artemisinin optimization algorithm, iterate through the corresponding objective function values ​​Archive_F to find the minimum and maximum values ​​of energy consumption and cost, and construct virtual ideal and anti-ideal solutions respectively.

[0123] Based on the ideal solution as the first row, the anti-ideal solution as the second row, and the objective function value Archive_F as the subsequent rows, construct a comprehensive matrix extended_matrix that contains all alternative solutions and the optimal and worst benchmarks;

[0124] Linear normalization is applied to all solutions, including ideal and anti-ideal solutions, to construct a normalization matrix, norm_matrix. All objective values ​​are uniformly converted into a dimensionless, yield-type index that is as large as possible, where the normalization value of the ideal solution is 1. The formula for calculating the normalization matrix norm_matrix is:

[0125] (7)

[0126] The normalized matrix norm_matrix is ​​multiplied by the precise weight vector weights to obtain the weighted normalized matrix weighted_matrix, thereby incorporating user needs into the decision-making process.

[0127] The weighted normalized matrix is ​​summed row-wise to obtain the aggregate score for each solution. This aggregate score is directly recorded as the corresponding alternative solution S_alternatives. The upward utility K_plus relative to the ideal solution and the downward utility K_minus relative to the anti-ideal solution for each alternative solution S_alternatives are calculated using the following formulas:

[0128] (8)

[0129] (9)

[0130] Here, S_ideal and S_anti are the aggregate scores of the ideal solution and the antiideal solution, respectively; K_plus reflects its tendency to seek better solutions, and K_minus reflects its tendency to avoid worse solutions.

[0131] By aggregating K_plus and K_minus using the comprehensive utility function f(k), the final evaluation value of each alternative solution S_alternatives is obtained. The solution with the highest evaluation value is selected as the final gear rolling parameter scheme that best meets the user's fuzzy requirements. The formula for calculating the comprehensive utility function f(k) is as follows:

[0132] (10)

[0133] Example 2:

[0134] In this embodiment, a gear hobbing process at a gear manufacturing company is used as an example. The machine tool used is a six-axis controlled, four-axis linkage high-speed gear hobbing machine YD31125CNC6, and the software used is MATLAB. A CNC gear hobbing process example sample from the gear manufacturing workshop under this gear company is adopted. The gear type is a cylindrical helical gear, the blank material is 20 GrMnTi, the module is required to be 4mm, the number of teeth is 63, the tooth thickness is 50mm, the helix angle is 17°, the accuracy requirement is GB / T10095.1-2008 standard grade 7, and the carbide hob diameter is 112mm. The user's main optimization objectives are the number of machining passes r and the spindle speed of each pass. Axial feed rate for each pass Depth of cut in each pass ,in, .

[0135] The initial range of gear hobbing parameters is obtained using deepseek, where the number of machining passes r is... [2, 3], spindle speed [500, 2000] rpm, Z-axis feed rate For [0.1, 2.5] mm / r, depth of cut Given a radius of [0.5, 9.0] mm, a multi-objective model considering the energy consumption and processing cost of this gear hobbing operation is provided:

[0136]

[0137]

[0138]

[0139]

[0140]

[0141]

[0142]

[0143]

[0144] (11)

[0145] Among them, E total It is the total energy consumption of a multi-pass gear hobbing process, expressed in joules (J); C total This is the total cost of a multi-pass gear hobbing process, expressed in yuan (¥); t standby fz represents the standby time of the gear hobbing machine, which is a constant and is expressed in seconds. i n represents the Z-axis feed rate for each pass, in mm / r. i The spindle speed for each pass is expressed in rpm. P represents the depth of cut for each pass, in mm. standby This represents the total power consumption in standby mode. P i unload-spindle P represents the no-load power of the spindle in each pass; i unload-worktable P represents the no-load power of each pass's worktable; i unload-X P is the no-load power of the X-axis feed system. i unload-Z P is the no-load power of the Z-axis feed system. i material For material power removal, P i loss Power loss due to additional load. All energy consumption is expressed in joules (J), and all power is expressed in watts (W).

[0146] Based on the reasonable range of process parameters, the user manuals provided by tool and machine tool manufacturers, and the predictions of the variables to be optimized by DeepSeek, the following constraints are imposed on the machining process:

[0147] st (12)

[0148] Where a is the pressure angle of the gear, with a value of 20°; d is the hob diameter, with a value of 112 mm; Ra This represents the actual surface roughness value, in μm.

[0149] Table 1. Solutions to some process parameters after optimization by the multi-objective artemisinin algorithm.

[0150]

[0151] Multiple sets of optimized gear hobbing parameters were obtained through a multi-objective artemisinin optimization algorithm, and some of the optimized parameters are shown in Table 1. By setting user requirements (here, energy consumption requirement is set to medium and cost requirement to important), the user's fuzzy language requirements were transformed into precise weights using triangular fuzzy numbers and the centroid method, resulting in... Then, a decision-making algorithm based on user fuzzy requirements is used to make a decision, and the final comprehensive utility score f(k) is obtained by combining weight calculations. The solution with the highest score is the final hobbing parameter scheme that meets the user's requirements: [2, 850, 500, 0.43, 1.35, 6.3, 2.7], where the number of passes r = 2, the spindle speed n = [850, 500] rpm, the Z-axis feed rate fz = [0.43, 1.35] mm / r, and the depth of cut ap = [6.3, 2.7] mm, with a corresponding target value of [4.63×10]. 6 , 114.84 ].

[0152] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for determining parameters in multi-pass gear hobbing, characterized in that, include: Obtain the initial hobbing parameters to be optimized; The initial range of hobbing parameters to be optimized is determined based on a large language model; Based on the multi-objective artemisinin optimization algorithm and the initial range of gear hobbing parameters, the optimized gear hobbing parameter solution set and its corresponding objective function value are obtained. The decision-making algorithm based on user fuzzy requirements makes decisions on the optimized gear hobbing parameter solution set and its corresponding objective function value, and outputs the final gear hobbing parameters that meet the user's requirements.

2. The multi-pass gear hobbing parameter decision method according to claim 1, characterized in that, The initial hobbing parameters to be optimized include the number of machining passes r and the spindle speed for each pass. Axial feed rate for each pass Depth of cut in each pass ,in, .

3. The multi-pass gear hobbing parameter decision method according to claim 1, characterized in that, The range of initial hobbing parameters to be optimized based on the large language model includes: Determine the basic information of the gear workpiece to be machined and the hob; The basic information is provided to the large language model, and the initial range of the initial hobbing parameters to be optimized is determined by the large language model. The basic information of the gear workpiece to be processed and the hob includes the model of the gear hobbing machine, the gear module m, the pressure angle a, the number of teeth Za, the tooth thickness B0, the hob diameter d, and the number of hob heads K.

4. The multi-pass gear hobbing parameter decision method according to claim 2, characterized in that, The process of obtaining the optimized gear hobbing parameter solution set and its corresponding objective function value based on the multi-objective artemisinin optimization algorithm and the initial gear hobbing parameter range includes: The total energy consumption and total production cost of gear hobbing are taken as two optimization objectives that need to be minimized simultaneously, and the objective function for optimizing the gear hobbing parameters is determined. The objective function for optimizing the gear hobbing parameters is expressed as follows: （1） in, The objective function for optimizing gear hobbing parameters is represented by E_total and C_total, which are two optimization objectives that need to be minimized simultaneously. E_total is the total energy consumption during the gear hobbing process, and C_total is the total production cost. The operating parameters based on the multi-objective artemisinin optimization algorithm and the process constraints of the gear hobbing process are set. The operating parameters include the maximum dimension dim_max of the variable determined by the maximum number of machining passes, the number of objectives obj_no = 2, the maximum number of fitness evaluations MaxFEs, the population size N, and the maximum capacity archive_size of the external archive used to store non-dominated solutions. The process constraints of the gear hobbing process include machine tool power constraints, workpiece surface roughness constraints, tooth profile error constraints, and constraints on the allocation rules between the total depth of cut and the depth of cut in each pass. Using a large language model, N individuals are generated based on the initial range of gear hobbing parameters to form an initial population. Each individual is a vector of variable dimension. The vector structure of an individual with trace number r is as follows: ; Initialize all non-dominated solutions Archive_X and their corresponding objective function values ​​Archive_F in the external archive storage used to store non-dominated solutions to be empty; Initialize the energy stagnation counter stagnation_E and the cost stagnation counter stagnation_C used for the adaptive pressure mechanism to be 0; Traverse the N individuals in the initial population, calculate the objective function value of the gear hobbing process corresponding to each individual in turn, and accumulate the fitness evaluation times FEs; perform non-dominated sorting on the current population and its objective function values, filter out all non-dominated solutions, and store them in the external archive Archive; Based on the adaptive pressure mechanism, dynamic parameters are calculated, adaptive leader selection and population position update are performed during the iteration process until the current fitness evaluation number FEs reaches the maximum value MaxFEs. For parameters exceeding the preset upper and lower bounds, their values ​​are set as the corresponding boundary values, and the cutting depth of each pass is summed and normalized to calculate the objective function value of the new solution, and the current fitness evaluation count (FEs) is increased accordingly. The new solution is compared with its parent in terms of Pareto dominance: if the new solution dominates its parent, it is replaced; if they do not dominate each other, they are replaced with a 50% probability. The updated population is merged with the current archive, and a new set of non-dominated solutions is selected from it to form the next generation of archive. If the archive exceeds the preset maximum capacity, it is sorted and pruned according to the crowding of solutions. Compare the minimum energy consumption and minimum cost in the new save file with the previous save file. If a certain goal is not improved, increment the corresponding energy consumption stagnation counter stagnation_E or cost stagnation counter stagnation_C by 1; otherwise, reset it to zero. Determine whether the current fitness evaluation count FEs has reached the maximum value MaxFEs. If it has, terminate the loop and output all non-dominated solutions Archive_X and their corresponding objective function values ​​Archive_F stored in the external archive as the final optimized hobbing parameter solution set; otherwise, continue the optimization iteration based on the adaptive pressure mechanism.

5. The multi-pass gear hobbing parameter decision method according to claim 4, characterized in that, The calculation of dynamic parameters includes two dynamic parameters calculated based on the current fitness evaluation counts (FEs): an exploration / mining balance parameter K and a variable-asynchronous long decay factor E_decay, used to control the population update strategy. The formula for calculating the exploration / mining balance parameter K is as follows: （2） The formula for calculating the variable asynchronous long decay factor E_decay is: （3）。 6. The multi-pass gear hobbing parameter decision method according to claim 4, characterized in that, The adaptive leader selection includes: Select the energy consumption leader (Leader_E), cost leader (Leader_C), and diversity leader (Leader_D) from the external archive to guide population renewal; Get the current energy stagnation counter stagnation_E and cost stagnation counter stagnation_C; The probability P_cost_leader of selecting a cost leader is calculated based on the energy stagnation counter stagnation_E and the cost stagnation counter stagnation_C. The formula for calculating the probability P_cost_leader is as follows: （4） When executing the mining strategy, a random number between [0,1] is generated. If the random number is less than P_cost_leader, the cost leader Leader_C is selected; otherwise, the energy leader Leader_E is selected. When executing the exploration strategy, the diversity leader Leader_D is selected to maintain population diversity.

7. The multi-pass gear hobbing parameter decision method according to claim 6, characterized in that, The adaptive population update includes: Generate a random number rand between [0,1] for each individual; When the random number rand is less than the exploration / mining balance parameter K, the individual executes a global exploration strategy, conducting a broad search throughout the entire search space based on the large variable-asynchronous long decay factor E_decay. When the random number rand >= the exploration / mining balance parameter K, the individual executes a local mining strategy, performing a fine search in the current dense population area based on the small variable-asynchronous long decay factor E_decay; In the final stage of the update process, for each newly generated solution, with a 20% probability, some parameter genes of the newly generated solution are directly replaced with the corresponding genes of the selected leader.

8. The multi-pass gear hobbing parameter decision method according to claim 1, characterized in that, The decision algorithm based on user fuzzy requirements makes decisions on the optimized gear hobbing parameter solution set and its corresponding objective function value, and outputs the final gear hobbing parameters that meet the user's requirements, including: Obtain vague user needs; Transform the user's vague requirements into precise weights; Based on the precise weights of the user's fuzzy requirements and the optimized solution set of hobbing parameters, the final hobbing parameters that meet the user's requirements are determined.

9. The multi-pass gear hobbing parameter decision method according to claim 8, characterized in that, The process of converting vague user requirements into precise weights includes: The system obtains the user's fuzzy language evaluation of the two optimization objectives of energy consumption and cost through an interactive interface; wherein the fuzzy language evaluation is selected from a predefined set of fuzzy language evaluations containing five importance levels. The fuzzy language evaluation selected by the user for each optimization objective is mapped to a preset triangular fuzzy number (l, m, u), where l is the lower bound, m is the most likely value, and u is the upper bound. The centroid method is used to defuzzify the triangular fuzzy numbers, transforming the fuzzy language evaluation selected by the user for each optimization objective into precise initial weight values. Initial weight values The calculation formula is: （5） in, and These correspond to the initial weight values ​​for energy consumption and cost, respectively. Calculate the initial weight values After normalization, the final precise weight vector `weights` used for decision-making is obtained. The formula for calculating the precise weight vector `weights` is: （6）。 10. The multi-pass gear hobbing parameter decision method according to claim 4 or 9, characterized in that, The final gear parameters that satisfy user needs are determined based on the precise weights of the user's fuzzy requirements and the optimized gear rolling parameter solution set, including: Based on all non-dominated solutions Archive_X output by the multi-objective artemisinin optimization algorithm, iterate through the corresponding objective function values ​​Archive_F to find the minimum and maximum values ​​of energy consumption and cost, and construct virtual ideal and anti-ideal solutions respectively. Based on the ideal solution as the first row, the anti-ideal solution as the second row, and the objective function value Archive_F as the subsequent rows, construct a comprehensive matrix extended_matrix that contains all alternative solutions and the optimal and worst benchmarks; Linear normalization is performed on all solutions, including ideal and antiideal solutions, to construct the normalization matrix norm_matrix. The formula for calculating the normalization matrix norm_matrix is ​​as follows: （7） Multiply the normalized matrix norm_matrix by the precise weight vector weights to obtain the weighted normalized matrix weighted_matrix; The weighted normalized matrix is ​​summed row-wise to obtain the aggregate score for each solution. This aggregate score is directly recorded as the corresponding alternative solution S_alternatives. The upward utility K_plus relative to the ideal solution and the downward utility K_minus relative to the anti-ideal solution for each alternative solution S_alternatives are calculated using the following formulas: （8） （9） Where S_ideal and S_anti are the aggregate scores of the ideal solution and the antiideal solution, respectively; By aggregating K_plus and K_minus using the comprehensive utility function f(k), the final evaluation value of each alternative solution S_alternatives is obtained. The solution with the highest evaluation value is selected as the final gear rolling parameter scheme that best meets the user's fuzzy requirements. The formula for calculating the comprehensive utility function f(k) is as follows: （10）。

Images