A multi-objective collaborative optimization method for turning process
By employing a multi-objective collaborative optimization method, the problem of multi-objective conflict in the turning of SiCp/Al composite materials was solved, achieving an efficient and safe combination of machining parameters, thereby improving machining efficiency and surface quality.
Patent Information
- Application Number
- CN202611275973.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-21
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies for processing silicon carbide particle-reinforced aluminum matrix composites lack a systematic approach to balancing the complex conflicting relationships between cutting temperature, tool wear, surface integrity, and processing efficiency. Furthermore, they fail to scientifically determine the optimal combination of processing parameters, resulting in low processing efficiency, high costs, and poor surface quality.
A multi-objective collaborative optimization method is adopted. Through experimental design, surrogate model construction and multi-objective evolutionary algorithm, a mathematical prediction model for cutting speed, feed rate and depth of cut is established, and Pareto optimal solution set is output, providing an optimized parameter combination that takes into account machining safety, tool life, surface quality and machining efficiency.
This method enables a systematic approach to balancing the complex conflicts between multiple objectives in the turning of SiCp/Al composite materials, thereby improving machining efficiency and surface quality, reducing tool wear, and ensuring machining safety and cost control.
Smart Images

Figure CN123113732A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of precision machining technology for composite materials, and more specifically to a multi-objective collaborative optimization method for turning processes. Background Technology
[0002] Silicon carbide particle-reinforced aluminum matrix composites (SiCp / Al composites) significantly improve the specific strength, specific stiffness, wear resistance, and corrosion resistance of aluminum alloys by introducing high-hardness, high-melting-point silicon carbide particles into the aluminum alloy matrix, while maintaining the lightweight properties of aluminum alloys. They are currently widely used in high-end equipment fields such as aerospace, shipbuilding, automotive, and electronics. As the application of this material expands to complex structural components such as large-size cylinders and compartments, higher requirements are being placed on its machining accuracy and surface quality.
[0003] However, this type of material presents significant common technical challenges in turning. The silicon carbide reinforcing phase has a Mohs hardness as high as 9.5, approaching that of diamond, and the material itself has a relatively low thermal conductivity, resulting in extremely high temperatures and intense alternating mechanical impact loads in the turning zone during machining. Even using polycrystalline diamond tools, currently considered the most suitable for machining this material, the tool flank wear increases rapidly under normal machining parameters, leading to a short overall tool life and severely restricting machining efficiency and cost control. Simultaneously, excessively high cutting temperatures not only accelerate thermal damage and adhesive wear of the tool but may also introduce surface defects such as microcracks, pull-out or breakage of reinforcing particles, thereby reducing component fatigue performance and service life.
[0004] Currently, the selection of machining parameters for SiCp / Al composite materials in practical engineering mainly relies on the operator's personal experience, general recommendations from tool suppliers, or turning manuals for traditional metal materials. While a few studies mention parameter optimization through orthogonal experiments or single-factor experiments, these methods often only focus on maximizing material removal or minimizing a single turning force as a single optimization objective, failing to systematically consider the complex conflicting relationships between cutting temperature, tool wear, surface integrity, and machining efficiency. More importantly, there is a significant coupling interaction effect among the three core parameters: cutting speed, feed rate, and depth of cut. Existing methods lack the technical means to mathematically express these coupling constraints and incorporate them into the optimization model. Furthermore, current techniques fail to treat the limiting thermal damage temperature threshold of polycrystalline diamond tools as a rigid constraint that must be met during the turning process, and also fail to fully consider the nonlinear effects of significant variations in the silicon carbide particle volume fraction between 20% and 70% on the optimal machining parameter range. Therefore, there is an urgent need for a collaborative optimization method that can scientifically and systematically determine the optimal combination of processing parameters for SiCp / Al composite materials, in order to overcome the shortcomings of existing empirical methods, such as arbitrariness, single-objective optimization that neglects one aspect while focusing on another, and ignoring hard constraints of the process. Summary of the Invention
[0005] The purpose of this invention is to address the aforementioned problems by providing a multi-objective collaborative optimization method for turning processes. This method overcomes the technical deficiency of existing technologies in lacking a mathematical expression of multi-parameter coupling constraints. By using the mathematical prediction models of each response objective as objective functions and employing a multi-objective evolutionary algorithm to solve the problem, this method overcomes the technical deficiency of existing methods that optimize only a single objective and cannot systematically consider the conflicts between multiple objectives. The final output non-dominated solution set provides engineers with an optimal combination of machining parameters that balances machining safety, tool life, surface quality, and machining efficiency.
[0006] The technical solution adopted in this invention is as follows: A multi-objective collaborative optimization method for turning processes includes the following steps: Experimental design and data acquisition steps: Obtain the material properties of the workpiece to be processed, and determine the value range of the three machining parameters, namely cutting speed, feed rate, and depth of cut, based on the material properties; encode the upper limit of each machining parameter within its respective value range as +1, the lower limit as -1, and the center value as 0; construct 12 sets of edge center point tests by combining any two machining parameters with their extreme values and the remaining machining parameter with its center value, and construct center point tests by combining all machining parameters with their center values. The edge center point tests and center point tests together constitute the turning test scheme; execute each set of turning tests sequentially according to the turning test scheme. After each set of turning tests is completed, collect and record the four response targets under that set of tests, namely cutting temperature, flank wear, surface roughness, and material removal rate. The proxy model construction steps are as follows: For each of the four response targets, a second-order polynomial response surface model is constructed, which includes a constant term, a linear term of each processing parameter, a quadratic term of each processing parameter, and an interaction term between each processing parameter. The second-order polynomial response surface model is then fitted by regression based on the collected experimental data to obtain the mathematical prediction model for each response target. Multi-objective optimization and constraint handling steps: The mathematical prediction models of each response objective are used as objective functions. Minimizing cutting temperature, minimizing tool wear, minimizing surface roughness, and maximizing material removal rate are set as optimization objectives to form a multi-objective optimization model. A multi-objective evolutionary algorithm is used to predict and iterate the multi-objective optimization model, converging towards the Pareto front until the preset maximum number of generations or the population converges to a stable Pareto front is reached. The output is a Pareto optimal solution set containing several non-dominated solutions, where each solution represents a feasible combination of machining parameters.
[0007] By employing the aforementioned technical solution, this method first changes the traditional reliance on operator experience for selecting machining parameters through a systematic experimental design process. Specifically, by introducing a coded experimental design concept, the upper limit of each machining parameter within its respective range is encoded as +1, the lower limit as -1, and the center value as 0. This constructs 12 sets of edge-center point experiments and center-point experiments. This turning experiment scheme covers the range where parameter interactions are most significant while avoiding the destructive experimental risk of tool breakage due to extreme corner points (such as when all three parameters are simultaneously at their maximum), making it particularly suitable for difficult-to-machine materials like SiCp / Al. Simultaneously, by systematically collecting data on four key response targets—cutting temperature, flank wear, surface roughness, and material removal rate—a sufficient and structured data foundation is provided for the subsequent establishment of the proxy model, avoiding optimization biases caused by data gaps or insufficient samples in traditional methods. Secondly, this method constructs second-order polynomial response surface models for each of the four response objectives through a surrogate model construction step. Based on the collected experimental data, regression fitting is performed to mathematically express the nonlinear relationship between the three core parameters—cutting speed, feed rate, and depth of cut—as well as the coupling and interaction effects between these parameters. This not only characterizes the independent influence of each machining parameter on cutting temperature but also quantitatively describes the complex interactions such as synergistic enhancement or mutual inhibition between pairs of machining parameters, thus overcoming the technical deficiency in existing technologies that lack a mathematical expression of multi-parameter coupling constraints. Furthermore, through multi-objective optimization and constraint processing steps, this method uses the mathematical prediction models of each response objective as objective functions, employs a multi-objective evolutionary algorithm for prediction iteration, and converges towards the Pareto front. Finally, it outputs a Pareto optimal solution set containing several non-dominated solutions, thus overcoming the technical deficiency of existing orthogonal or single-factor experimental methods that only focus on maximizing material removal rate or minimizing a single turning force as a single optimization objective, failing to systematically consider the complex conflict relationships between multiple objectives.
[0008] Furthermore, the process includes engineering decision-making and output steps: An initial decision matrix is constructed by treating the non-dominated solutions in the Pareto optimal solution set as rows and the four optimization objectives as columns; the decision matrix is standardized using vector normalization to obtain a normalized decision matrix; the positive and negative ideal solutions for the four optimization objectives are determined column-wise, with the positive ideal solution including the minimum values of cutting temperature, tool wear, and surface roughness, and the maximum value of material removal rate; the negative ideal solution includes the maximum values of cutting temperature, tool wear, and surface roughness, and the minimum value of material removal rate. The Euclidean distance between each machining parameter combination in the normalized decision matrix and the positive and negative ideal solutions are calculated sequentially, followed by the calculation of the relative proximity of each machining parameter combination; the proximity value ranges from 0 to 1, with a larger value indicating that the machining parameter combination is closer to the positive ideal solution and further away from the negative ideal solution; all machining parameter combinations are sorted in descending order of relative proximity, and at least one of the top-ranked machining parameter combinations is selected as a recommended solution and included in the recommended parameter library for user selection.
[0009] By employing the aforementioned technical solution, the engineering decision-making and output steps construct an initial decision matrix by using the non-dominated solutions in the Pareto optimal solution set as rows and the four optimization objectives as columns. Vector normalization is then used for standardization, eliminating the numerical scale differences caused by different dimensions among the four objectives—cutting temperature, flank wear, surface roughness, and material removal rate—ensuring fair comparison of all objectives on the same scale. Based on this, by separately determining the positive and negative ideal solutions and calculating the Euclidean distances between each machining parameter combination and the positive and negative ideal solutions, a relative proximity score ranging from 0 to 1 is obtained. This proximity score, as a single quantitative index, comprehensively reflects the overall superiority or inferiority of each candidate solution relative to the optimal and worst-case boundaries, solving the engineering decision-making problem where multiple non-dominated solutions in the Pareto optimal solution set cannot be directly ranked. Finally, the machining parameter combinations are sorted in descending order of relative proximity score, and at least one of the top-ranked combinations is included in the recommended parameter library. This provides operators with clear and operable machining parameter guidance, effectively transforming multi-objective optimization results into actual production.
[0010] Furthermore, when calculating the Euclidean distance between each combination of processing parameters and the positive and negative ideal solutions, the corresponding weight coefficients are multiplied to adjust the influence of each optimization objective in the ranking process. The sum of the weight coefficients of all optimization objectives is 1.
[0011] By employing the aforementioned technical solution, this method allows users to flexibly adjust the weighting of each optimization objective in the ranking process based on the specific priority requirements of the current processing task. This is achieved by multiplying the Euclidean distance between each combination of processing parameters and the positive and negative ideal solutions by corresponding weighting coefficients. For example, when the processing task has stringent requirements for surface quality, the weighting coefficient corresponding to surface roughness can be increased, tilting the recommended solution towards a better surface quality. When the task deadline is tight, the weighting coefficient corresponding to material removal rate can be increased, making the recommended solution more focused on processing efficiency. This weighted adjustment mechanism enables the decision-making process to organically integrate engineering experience and operational preferences, maintaining the objectivity of the decision-making method while taking into account the differentiated needs under different production scenarios, significantly enhancing the method's flexibility in adapting to different processing task configurations.
[0012] Furthermore, it also includes verification and correction steps: selecting a scheme to be verified from the Pareto optimal solution set, conducting a verification turning experiment on a CNC machine tool, comparing the measured response value with the predicted value output by the corresponding mathematical prediction model; if the relative error is within the preset threshold range, the verification is passed, and the corresponding scheme is included in the final recommended parameter library; if the relative error exceeds the preset threshold range, the verification data is added to the original turning test data as new sample points, and the proxy model construction step and the multi-objective optimization and constraint processing step are repeated for iterative correction until the verification is passed.
[0013] By employing the aforementioned technical solution, this method selects a solution to be verified from the Pareto optimal solution set and conducts verification turning experiments on a CNC machine tool. The measured response values are compared with the predicted values output by the corresponding mathematical prediction model, establishing a verification channel between the optimization results and the actual machining environment. When the relative error is within a preset threshold range, the verification is deemed successful, and the corresponding solution is included in the final recommended parameter library, ensuring the actual effectiveness and reliability of the recommended solution. When the relative error exceeds the preset threshold range, the verification data is added as new sample points to the original turning test data, and the proxy model construction steps and multi-objective optimization and constraint processing steps are repeatedly executed for iterative correction until the verification is successful. This closed-loop iterative correction mechanism enables the mathematical prediction model to continuously absorb actual machining feedback information for self-optimization, effectively suppressing model prediction deviations caused by engineering factors such as material batch fluctuations, tool condition differences, and machine tool thermal drift, significantly improving the robustness and engineering applicability of the optimization method in the actual machining environment.
[0014] Furthermore, in the surrogate model construction step, the second-order polynomial response surface model is shown in the following equation:
[0015] In the formula, This represents the measured value of the response target, where n is an integer from 1 to 4. The measured value representing the cutting temperature. The measured value representing the amount of tool wear. The measured value representing surface roughness. The measured value representing the material removal rate; The coded value representing the cutting speed. Code indicating feed rate The encoded value representing the turning depth, where The model intercept represents the baseline response value. , and These are the linear main effect coefficients for cutting speed, feed rate, and depth of cut, respectively, characterizing the degree of direct influence of each machining parameter on the target response. , and The coefficients of the quadratic term represent the nonlinear curvature effect of each processing parameter on the target response. , and The coefficients represent the pairwise interaction effects, characterizing the synergistic enhancement or inhibition between processing parameters.
[0016] Furthermore, the turning test scheme includes N sets of turning tests, where N is an integer greater than or equal to 13; the encoded value of each set of turning tests is substituted into a second-order polynomial response surface model to establish an N-row, 10-column X matrix; based on the measured value of each set of turning tests with respect to a certain response target, a corresponding N-row, 1-column optimization vector Y is established, and matrix operation formulas are used... Solve for the regression coefficient vector, the regression coefficient vector is For each response target, the matrix operation is performed and the corresponding regression coefficients are output. In order to obtain mathematical prediction models for each response target.
[0017] By employing the aforementioned technical solution, this method establishes an N-row, 10-column X matrix by substituting the coded values of each turning test group into a second-order polynomial response surface model. Based on the measured values of each turning test group for a specific response objective, an corresponding N-row, 1-column optimization vector Y is established. The regression coefficient vector is then solved using matrix operation formulas, achieving efficient and accurate calculation of the regression coefficients. The above matrix operations are performed on the four response objectives—cutting temperature, flank wear, surface roughness, and material removal rate—and the corresponding regression coefficients are output, thus obtaining independent mathematical prediction models for each response objective. Since N is an integer greater than or equal to 13, the number of test groups exceeds the number of regression coefficients, providing sufficient residual degrees of freedom for subsequent fit testing and model accuracy evaluation, ensuring the statistical reliability of each mathematical prediction model.
[0018] Further, the significance probability value corresponding to each regression coefficient is calculated, and the significance probability value is compared with a preset significance level. If the significance probability value of a certain regression coefficient is greater than or equal to the preset significance level, the term corresponding to that regression coefficient is determined to be an insignificant term. Each time, the term with the largest current significance probability value is removed. After removal, the remaining terms are refitted, and the significance probability values of the remaining regression coefficients are recalculated. The above removal and refitting process is repeated until the significance probability values of all terms retained in the second-order polynomial response surface model are less than the preset significance level, thus obtaining the optimized mathematical prediction model of each response target. During the removal process, if the square term or interaction term of a certain processing parameter is retained, the linear term corresponding to that processing parameter, even if its significance probability value is greater than or equal to the preset significance level, is still forcibly retained in the second-order polynomial response surface model and does not participate in the removal.
[0019] By employing the aforementioned technical solution, the model optimization step, through calculating the significance probability values corresponding to each regression coefficient and comparing them with a preset significance level, can identify redundant terms that have no significant impact on the response objective from the initially established second-order polynomial response surface model. Specifically, this method uses a successive elimination strategy, removing only the term with the highest current significance probability value each time. After elimination, the remaining terms are refitted, and the significance probability values of the remaining regression coefficients are recalculated. This process is repeated until the significance probability values of all remaining terms in the model are less than the preset significance level. This step-by-step iterative elimination strategy avoids the model distortion problem that may be caused by eliminating multiple insignificant terms at once, ensuring the robustness of the model simplification process. More importantly, this solution explicitly introduces a hierarchical principle: if the square term or interaction term of a certain processing parameter is retained during the elimination process, the linear term corresponding to that processing parameter, even if its significance probability value is greater than or equal to the preset significance level, is still forcibly retained in the model and does not participate in the elimination. This principle ensures that the contribution of each parameter is not destroyed when the encoded variables undergo scaling transformation, maintaining the physical rationality and mathematical integrity of the model. The simplified model obtained after the above optimization eliminates statistically insignificant redundant terms, reducing the risk of model overfitting.
[0020] Furthermore, in the multi-objective optimization and constraint processing steps, constraints are set for the multi-objective optimization model. These constraints include: sorting the data according to the measured temperature values from the turning test; selecting at least two sets of edge center point test data in descending order of the measured temperature values; and constructing coupled constraints between cutting speed and feed rate using linear fitting. When using a multi-objective evolutionary algorithm to predict and iterate the multi-objective optimization model, the coupled constraints are verified for each individual in the population during each generation of evolution. If the individuals in this population are located within a safe zone, and the four objective function values are preserved and calculated normally, then... If an individual in the population crosses the safety boundary, it is given a very high penalty fitness value, causing it to be eliminated in subsequent non-dominated rankings. In the formula, f represents the feed rate. Represents cutting speed. Represents cutting speed The function, whose expression is obtained by linear fitting of selected experimental data.
[0021] By adopting the above technical solution, this method introduces a coupling constraint between cutting speed and feed rate in the multi-objective optimization and constraint processing steps. The physical meaning of this coupling constraint is that the maximum feed rate allowed under a given cutting speed is limited by the tool thermal damage threshold. The introduction of this equation makes a high-temperature failure angle cut off from the original square design space, ensuring that no physically unrealizable parameter combinations will be generated in the subsequent optimization process.
[0022] Furthermore, the constraints also include machining parameter boundary constraints and response target boundary constraints. The machining parameter boundary constraints include setting the value range of cutting speed, feed rate and depth of cut. The response target boundary constraints include setting the upper limit of cutting temperature, tool wear and surface roughness, and the lower limit of material removal rate.
[0023] By employing the aforementioned technical solution, this method constructs a triple hard constraint system, including boundary constraints for machining parameters, boundary constraints for response targets, and coupling constraints between cutting speed and feed rate, which together constitute a complete constraint framework for the multi-objective optimization model. Specifically, the boundary constraints for machining parameters, by setting the ranges for cutting speed, feed rate, and depth of cut, ensure that candidate solutions during the optimization process always fall within the physical range allowed by machine tool performance and validated by previous single-factor experiments, avoiding the algorithm from searching for theoretically feasible but practically unfeasible extreme parameter combinations on the machine tool. The boundary constraints for response targets explicitly embed engineering safety baselines such as upper limits for cutting temperature, tool wear, surface roughness, and material removal rate, as well as the critical thermal damage temperature of the tool (e.g., typically 700℃ for polycrystalline diamond tools) and wear threshold (typically 0.3mm) into the optimization model, ensuring that all candidate solutions meet the basic requirements for machining safety and quality compliance. The triple constraints, from different dimensions, jointly limit the feasible solution space of the multi-objective optimization model, constrain the search direction of the multi-objective evolutionary algorithm, and ensure that each solution in the final Pareto optimal solution set has engineering feasibility, safety and reliability, and quality compliance.
[0024] Furthermore, the number of center point experiments is at least 3 sets; the method also includes a misfit test step: calculating the pure error square and pure error degrees of freedom based on the experimental data of at least 3 sets of center point experiments, calculating the total residual sum of squares and total residual degrees of freedom based on the experimental data of all turning experiments in the turning test scheme, and calculating the misfit F value using the pure error sum of squares, pure error degrees of freedom, total residual sum of squares and total residual degrees of freedom; if the misfit F value is less than the critical value of the F distribution under the preset significance level, it is determined that the second-order polynomial response surface model is not significantly misfitted, and the multi-objective optimization and constraint processing step is entered; if the misfit F value is greater than or equal to the critical value of the F distribution, it is determined that the second-order polynomial response surface model is significantly misfitted, and the value range of at least one of the cutting speed, feed rate and turning depth is adjusted, and the experimental design and data acquisition steps and the surrogate model construction steps are re-executed.
[0025] By employing the aforementioned technical solution, this method establishes a statistical discrimination mechanism for the effectiveness of the second-order polynomial response surface model by setting the number of center point tests to at least three sets and introducing a lack-of-fit test step. Specifically, the three sets of center point tests provide two degrees of freedom for calculating the sum of squared pure errors, which reflect the response fluctuations caused by random factors such as material inhomogeneity, machine tool vibration, and measurement noise under identical turning conditions. The total sum of squared residuals is calculated based on the test data from all turning tests. The total residuals minus the pure errors yield the lack-of-fit error, which reflects the systematic biases that the model structure itself failed to capture. By calculating the lack-of-fit F-value and comparing it with the critical value of the F-distribution at a preset significance level, it is possible to quantitatively determine, from a statistical perspective, whether the second-order polynomial response surface model has significant structural defects. If the F-value of the misfit is less than the critical value of the F-distribution, the model is considered not significantly misfitted, indicating that the current second-order polynomial response surface model has fully captured the physical mapping relationship between the machining parameters and the response target, and the model is effective. If the F-value of the misfit is greater than or equal to the critical value of the F-distribution, the model is considered significantly misfitted, indicating that the current model has structural biases. In this case, the model is corrected by adjusting the range of at least one of the cutting speed, feed rate, and depth of cut, and by re-executing the experimental design and modeling steps. This misfit detection mechanism ensures that the mathematical prediction model on which subsequent multi-objective optimization depends is statistically reliable, avoiding the risk of optimization results failing due to model structural errors. Secondly, in the aforementioned technical document, the regression coefficient to be determined is 10, which limits the center point experiment to at least 3 repetitions, i.e., the minimum number of experiments is 15, thus enabling the detection of whether there is a serious misfit phenomenon in the model, thereby ensuring that the fitted response surface equation is reliable. 15 experiments are the minimum and most economical configuration to ensure that the model has statistical testing capability, maintaining good mathematical interpretability and computational efficiency while ensuring high prediction accuracy.
[0026] In summary, due to the adoption of the above technical solutions, the beneficial effects of this invention are as follows: This application, through systematic experimental design, employs a combination of edge-center point experiments and center-point experiments to form a complete turning test scheme. This comprehensively covers the influence of various machining parameters and their combinations on the machining results with fewer experiments, significantly reducing experimental costs. Furthermore, by constructing a second-order polynomial response surface model and performing regression fitting based on the collected experimental data, the nonlinear relationship between the three core parameters—cutting speed, feed rate, and depth of cut—as well as the coupling interaction effects between parameters, are mathematically expressed, overcoming the technical deficiency in existing technologies that lack a mathematical expression of multi-parameter coupling constraints. Moreover, by using the mathematical prediction models of each response objective as objective functions, conflicting optimization objectives such as minimizing cutting temperature, minimizing tool wear, and minimizing surface roughness are incorporated into a unified multi-objective optimization model. A multi-objective evolutionary algorithm is used to converge to the Pareto front, overcoming the technical deficiency of existing methods that optimize only a single objective and cannot systematically consider the complex conflicting relationships between multiple objectives. Simultaneously, during the multi-objective optimization process, a coupling constraint between cutting speed and feed rate is introduced, which selects at least two sets of center point test data in descending order of measured temperature values and constructs a linear fitting model. This transforms the critical thermal damage temperature threshold of the tool into a safety zone boundary line on the speed-feed parameter plane. Together with the machining parameter boundary constraint and the response target boundary constraint, this forms a complete triple hard constraint system, ensuring that every solution in the final Pareto optimal solution set meets the basic requirements of machining safety and quality compliance. Furthermore, by setting the number of center point tests to at least three sets and introducing a fit-out test step, it is possible to quantitatively determine from a statistical perspective whether there are significant structural defects in the second-order polynomial response surface model, ensuring that the mathematical prediction model entering the multi-objective optimization process is statistically reliable. Attached Figure Description
[0027] Figure 1 This is a schematic diagram of an application scenario provided by the present invention; Figure 2 This is a flowchart of the actual method of the present invention. Detailed Implementation
[0028] The present invention will now be described in detail with reference to the accompanying drawings.
[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0030] Example 1 A multi-objective collaborative optimization method for turning processes, such as Figure 1 , Figure 2 As shown, it includes the following steps: Experimental design and data acquisition steps: Obtain the material properties of the workpiece to be processed, and determine the value range of the three machining parameters, namely cutting speed, feed rate, and depth of cut, based on the material properties; encode the upper limit of each machining parameter within its respective value range as +1, the lower limit as -1, and the center value as 0; construct 12 sets of edge center point tests by combining any two machining parameters with their extreme values and the remaining machining parameter with its center value, and construct center point tests by combining all machining parameters with their center values. The edge center point tests and center point tests together constitute the turning test scheme; execute each set of turning tests sequentially according to the turning test scheme. After each set of turning tests is completed, collect and record the four response targets under that set of tests, namely cutting temperature, flank wear, surface roughness, and material removal rate.
[0031] The proxy model construction steps are as follows: For each of the four response targets, a second-order polynomial response surface model is constructed, which includes a constant term, a linear term of each processing parameter, a quadratic term of each processing parameter, and an interaction term between each processing parameter. The second-order polynomial response surface model is then fitted by regression based on the collected experimental data to obtain the mathematical prediction model for each response target.
[0032] Multi-objective optimization and constraint handling steps: The mathematical prediction models of each response objective are used as objective functions. Minimizing cutting temperature, minimizing tool wear, minimizing surface roughness, and maximizing material removal rate are set as optimization objectives to form a multi-objective optimization model. A multi-objective evolutionary algorithm is used to predict and iterate the multi-objective optimization model, converging towards the Pareto front until the preset maximum number of generations or the population converges to a stable Pareto front is reached. The output is a Pareto optimal solution set containing several non-dominated solutions, where each solution represents a feasible combination of machining parameters.
[0033] Specifically, this method introduces a coded experimental design, encoding the upper limit of each machining parameter within its respective range as +1, the lower limit as -1, and the center value as 0. Twelve sets of edge-center point and center-point tests are constructed accordingly. This turning test scheme covers the range where parameter interactions are most significant while avoiding the destructive test risk of tool breakage due to extreme corner points, making it particularly suitable for difficult-to-machine materials like SiCp / Al. Simultaneously, by systematically collecting four key response targets, a sufficient and structured data foundation is provided for the subsequent establishment of the proxy model. Secondly, second-order polynomial response surface models are constructed for each of the four response targets, mathematically expressing the nonlinear relationship between the three core parameters—cutting speed, feed rate, and depth of cut—and the coupling interaction effects between parameters. This not only characterizes the independent influence of each machining parameter on cutting temperature but also quantitatively describes the complex interactions such as synergistic enhancement or mutual inhibition between pairs of machining parameters. Based on this, the mathematical prediction models of each response objective are used as objective functions, and a multi-objective evolutionary algorithm is used for prediction iteration and convergence to the Pareto front. Finally, a Pareto optimal solution set containing several non-dominated solutions is output, which overcomes the technical defect of existing methods that optimize only a single objective and cannot systematically take into account the complex conflict relationship between multiple objectives.
[0034] Understandably, repeating the center point experiment not only provides the degree of freedom for pure error estimation in subsequent fit tests, but also allows for real-time monitoring of whether the system is in a statistically controlled state during the experiment. That is, if the fluctuation of the center point response value exceeds the expected range, it indicates the presence of uncontrolled noise factors, and the cause of the anomaly needs to be investigated and eliminated before continuing the experiment.
[0035] Engineering Decision and Output Steps: An initial decision matrix is constructed by treating the non-dominated solutions in the Pareto optimal solution set as rows and the four optimization objectives as columns. The decision matrix is then standardized using vector normalization to obtain a normalized decision matrix. Positive and negative ideal solutions for each of the four optimization objectives are determined column-wise. The positive ideal solution includes the minimum values of cutting temperature, tool wear, and surface roughness, and the maximum value of material removal rate. The negative ideal solution includes the maximum values of cutting temperature, tool wear, and surface roughness, and the minimum value of material removal rate. For each machining parameter combination in the normalized decision matrix, the Euclidean distance between it and the positive and negative ideal solutions are calculated. Then, the relative proximity of each machining parameter combination is calculated. This proximity value ranges from 0 to 1; a higher value indicates that the machining parameter combination is closer to the positive ideal solution and further away from the negative ideal solution. All machining parameter combinations are sorted in descending order of relative proximity, and at least one of the top-ranked combinations is selected as a recommended solution and added to the recommended parameter library for user selection.
[0036] Specifically, this engineering decision-making and output process constructs an initial decision matrix by using the non-dominated solutions in the Pareto optimal solution set as rows and the four optimization objectives as columns. Vector normalization is then used for standardization, eliminating the numerical scale differences caused by the different dimensions among the four objectives—cutting temperature, flank wear, surface roughness, and material removal rate—ensuring fair comparison of each objective on the same scale. Based on this, by separately determining the positive and negative ideal solutions and calculating the Euclidean distances between each machining parameter combination and the positive and negative ideal solutions, a relative closeness value ranging from 0 to 1 is obtained. This closeness, as a single quantitative index, comprehensively reflects the overall superiority or inferiority of each candidate solution relative to the optimal and worst-case boundaries, solving the engineering decision-making problem where multiple non-dominated solutions in the Pareto optimal solution set cannot be directly ranked. This ranking result can be directly retrieved by engineers from the recommended parameter library according to specific production tasks, without the need for repeated experiments or calculations.
[0037] When calculating the Euclidean distance between each combination of processing parameters and the positive and negative ideal solutions, the corresponding weight coefficients are multiplied to adjust the influence of each optimization objective in the ranking process. The sum of the weight coefficients of all optimization objectives is 1.
[0038] Specifically, this method allows users to flexibly adjust the weighting of each optimization objective in the ranking process based on the specific priorities of the current processing task by multiplying the Euclidean distance between each combination of processing parameters and the positive and negative ideal solutions by corresponding weighting coefficients. For example, when the processing task has stringent requirements for surface quality, the weighting coefficient corresponding to surface roughness can be increased, tilting the recommended solution towards a better surface quality; when the task deadline is tight, the weighting coefficient corresponding to material removal rate can be increased, making the recommended solution more focused on processing efficiency. This weighted adjustment mechanism enables the decision-making process to organically integrate engineering experience and operational preferences, maintaining the objectivity of the decision-making method while taking into account the differentiated needs under different production scenarios. It is understandable that the setting of the weighting coefficients reflects the decision-maker's subjective judgment on the importance of the four optimization objectives, and the sum of the weighting coefficients being 1 ensures the scale invariance of the decision matrix before and after weighting. In practical applications, if there is a lack of clear prior preferences, an equal weighting configuration (each weight being 0.25) can be used to make the decision result completely data-driven; if there are clear process constraints or customer requirements, the weight of the corresponding objective can be increased to tilt the recommended solution in a specific direction.
[0039] Verification and correction steps: Select the scheme to be verified from the Pareto optimal solution set, conduct a verification turning experiment on a CNC machine tool, and compare the measured response value with the predicted value output by the corresponding mathematical prediction model; if the relative error is within the preset threshold range, the verification is passed, and the corresponding scheme is included in the final recommended parameter library; if the relative error exceeds the preset threshold range, the verification data is added to the original turning test data as new sample points, and the surrogate model construction step and multi-objective optimization and constraint processing step are repeated for iterative correction until the verification is passed.
[0040] Specifically, this method establishes a verification channel between the optimization results and the real machining environment by selecting a solution to be verified from the Pareto optimal solution set and conducting verification turning experiments on a CNC machine tool. When the relative error is within a preset threshold range, the verification is deemed successful, and the corresponding solution is included in the final recommended parameter library, ensuring the actual effectiveness and reliability of the recommended solution. When the relative error exceeds the preset threshold range, the verification data is added to the original turning test data as new sample points, and the surrogate model construction step and the multi-objective optimization and constraint processing step are repeatedly executed for iterative correction until the verification is successful. This closed-loop iterative correction mechanism enables the mathematical prediction model to continuously absorb actual machining feedback information for self-optimization.
[0041] Understandably, the verification turning experiment should be conducted under the same machine tool, cutting tool, and material batch conditions as the modeling experiment to eliminate the interference of system deviations on the verification results. This closed-loop iterative correction mechanism incorporates the verification data into the new round of modeling, allowing the model to approach the real physical system after each iteration until it reaches a preset accuracy threshold and then terminates, ensuring the engineering reliability of the recommended scheme.
[0042] In other embodiments, the selection strategy for the verification scheme is not limited to selecting from the Pareto optimal solution set. Alternatively, an extreme scheme can be selected at each end of the Pareto front (such as the lowest temperature end and the highest efficiency end) as a control for destructive testing to verify whether the model's predictive ability under boundary conditions remains effective.
[0043] In the proxy model construction step
[0044] In the formula, This represents the measured value of the response target, where n is an integer from 1 to 4. The measured value representing the cutting temperature. The measured value representing the amount of tool wear. The measured value representing surface roughness. The measured value representing the material removal rate; The coded value representing the cutting speed. Code indicating feed rate The encoded value representing the turning depth, where The model intercept represents the baseline response value. , and These are the linear main effect coefficients for cutting speed, feed rate, and depth of cut, respectively, characterizing the degree of direct influence of each machining parameter on the target response. , and The coefficients of the quadratic term represent the nonlinear curvature effect of each processing parameter on the target response. , and The coefficients represent the pairwise interaction effects, characterizing the synergistic enhancement or inhibition between processing parameters.
[0045] The turning test scheme includes N sets of turning tests, where N is an integer greater than or equal to 13. The encoded values of each set of turning tests are substituted into a second-order polynomial response surface model to establish an N-row, 10-column X matrix. Based on the measured values of each set of turning tests with respect to a certain response target, a corresponding N-row, 1-column optimization vector Y is established. Matrix operation formulas are then used... Solve for the regression coefficient vector, the regression coefficient vector is For each response target, the matrix operation is performed and the corresponding regression coefficients are output. In order to obtain mathematical prediction models for each response target.
[0046] Specifically, this method establishes an N-row, 10-column X matrix by substituting the coded values of each turning test into a second-order polynomial response surface model. Based on the measured values of each turning test for a specific response objective, an N-row, 1-column optimization vector Y is established. The regression coefficient vector is then solved using matrix operation formulas, achieving efficient and accurate calculation of the regression coefficients. The above matrix operations are performed for each of the four response objectives, and the corresponding regression coefficients are output, thus obtaining independent mathematical prediction models for each response objective. Since N is an integer greater than or equal to 13, the number of test groups is ensured to exceed the number of regression coefficients, providing sufficient residual degrees of freedom for subsequent fit testing and model accuracy evaluation.
[0047] It is understandable that matrix operation formulas Derived from least squares estimation theory, its geometric meaning is to find the regression coefficient vector that minimizes the sum of squared residuals in the 10-dimensional space spanned by the X matrix. When there is multicollinearity among the columns of the X matrix... Numerical calculations may be unstable, but the orthogonality in the design of the experimental scheme ensures that the columns are approximately orthogonal, thus ensuring the numerical stability of matrix inversion.
[0048] In other embodiments, the regression coefficients are not limited to the least squares method; ridge regression or lasso regression can also be used as alternatives. However, the above methods all introduce additional regularization parameters that need to be determined through cross-validation, increasing computational complexity. Therefore, this embodiment prefers the least squares method.
[0049] Calculate the significance probability value corresponding to each regression coefficient and compare the significance probability value with a preset significance level. If the significance probability value of a regression coefficient is greater than or equal to the preset significance level, the term corresponding to that regression coefficient is determined to be insignificant. Each time, the term with the largest current significance probability value is removed. After removal, the remaining terms are refitted, and the significance probability values of the remaining regression coefficients are recalculated. The above removal and refitting process is repeated until the significance probability values of all terms retained in the second-order polynomial response surface model are less than the preset significance level, thus obtaining the optimized mathematical prediction model for each response target. During the removal process, if the square term or interaction term of a certain processing parameter is retained, the linear term corresponding to that processing parameter, even if its significance probability value is greater than or equal to the preset significance level, is still forcibly retained in the second-order polynomial response surface model and does not participate in the removal.
[0050] Specifically, the model optimization step identifies redundant terms with insignificant impact on the response objective from the initially established second-order polynomial response surface model by calculating the significance probability values of each regression coefficient and comparing them with a preset significance level. This method employs a successive elimination strategy, removing only the term with the highest current significance probability value each time. After elimination, the remaining terms are refitted, and the significance probability values of the remaining regression coefficients are recalculated. This process is repeated until the significance probability values of all remaining terms in the model are less than the preset significance level. This iterative elimination strategy avoids model distortion that might result from removing multiple insignificant terms at once, ensuring the robustness of the model simplification process. More importantly, this scheme explicitly introduces a hierarchical principle: if the squared term or interaction term of a certain processing parameter is retained during the elimination process, the corresponding linear term, even if its significance probability value is greater than or equal to the preset significance level, is forcibly retained in the model and does not participate in the elimination, ensuring that the contribution of each parameter is not destroyed when the encoded variables undergo scaling transformation.
[0051] Understandably, the reason for eliminating variables one by one rather than all at once is that the estimated value of the coefficients of the remaining variables will change after one variable is eliminated, and the corresponding standard error and p-value will also change. Therefore, after each elimination, the p-value of all remaining coefficients needs to be recalculated to ensure the timeliness of the criterion.
[0052] In the multi-objective optimization and constraint processing steps, constraints are set for the multi-objective optimization model. These constraints include: sorting the data according to the measured temperature values from the turning test; selecting at least two sets of edge center point test data in descending order of the measured temperature values; and constructing coupled constraints between cutting speed and feed rate using linear fitting. When using a multi-objective evolutionary algorithm to predict and iterate the multi-objective optimization model, the coupled constraints are verified for each individual in the population during each generation of evolution. If the individuals in this population are located within a safe zone, and the four objective function values are preserved and calculated normally, then... If an individual in the population crosses the safety boundary, it is given a very high penalty fitness value, causing it to be eliminated in subsequent non-dominated rankings. In the formula, f represents the feed rate. Represents cutting speed. Represents cutting speed The function, whose expression is obtained by linear fitting of selected experimental data.
[0053] The constraints also include machining parameter boundary constraints and response target boundary constraints. The machining parameter boundary constraints include setting the range of values for cutting speed, feed rate, and depth of cut. The response target boundary constraints include setting the upper limit of cutting temperature, tool wear, and surface roughness, as well as the lower limit of material removal rate.
[0054] Specifically, this method introduces a coupling constraint between cutting speed and feed rate in the multi-objective optimization and constraint processing steps. The physical meaning of this coupling constraint is that the maximum allowable feed rate at a given cutting speed is limited by the tool's thermal damage threshold. The introduction of this equation removes a high-temperature failure angle from the originally square design space, ensuring that physically unrealizable parameter combinations are not generated during subsequent optimization. This method also constructs a triple hard constraint system including machining parameter boundary constraints, response target boundary constraints, and coupling constraints. Machining parameter boundary constraints, by setting the value range of each parameter, ensure that candidate solutions during optimization are always within the physical range allowed by machine tool performance and valid in previous single-factor experiments. Response target boundary constraints, by setting upper limits for cutting temperature, tool wear, surface roughness, and material removal rate, explicitly embed engineering safety baselines such as the tool's critical thermal damage temperature and wear threshold into the optimization model. It can be understood that the linear fitting of the coupling constraints is based on the measured temperature values approximating the extreme operating points of the tool's critical thermal damage temperature. These boundary points are located at the intersection of the safe and dangerous regions on the speed-feed plane. By performing coupling constraint verification on individuals in each generation of evolution, individuals that violate the safety boundary are given a large penalty fitness value so that they are naturally eliminated in the non-dominated sorting, ensuring that the final Pareto optimal solution set all fall within the safety region.
[0055] In other embodiments, the construction of coupling constraints is not limited to linear fitting; quadratic polynomials can also be used to more accurately match the actual distribution of boundary points. When boundary points exhibit obvious curvature characteristics, quadratic polynomial fitting can provide higher boundary delineation accuracy, but at least three sets of boundary points are required to determine the quadratic term coefficients. Furthermore, the temperature threshold is not limited to 700°C and can be adjusted according to the actual thermal damage temperature of the tool material.
[0056] The number of center point experiments is at least 3 sets; the method also includes a misfit test step: calculating the pure error square and pure error degrees of freedom based on the experimental data of at least 3 sets of center point experiments, calculating the total residual sum of squares and total residual degrees of freedom based on the experimental data of all turning experiments in the turning test scheme, and calculating the misfit F value using the pure error sum of squares, pure error degrees of freedom, total residual sum of squares and total residual degrees of freedom; if the misfit F value is less than the critical value of the F distribution under the preset significance level, it is determined that the second-order polynomial response surface model is not significantly misfitted, and the multi-objective optimization and constraint processing step is entered; if the misfit F value is greater than or equal to the critical value of the F distribution, it is determined that the second-order polynomial response surface model is significantly misfitted, and the value range of at least one of the cutting speed, feed rate and turning depth is adjusted, and the experimental design and data acquisition steps and the surrogate model construction steps are re-executed.
[0057] Specifically, this method establishes a statistical discrimination mechanism for the effectiveness of the second-order polynomial response surface model by setting the number of center point experiments to at least three sets and introducing a lack-of-fit test step. The three sets of center point experiments provide two degrees of freedom for calculating the sum of squared pure errors, which reflect response fluctuations caused by random factors such as material inhomogeneity, machine tool vibration, and measurement noise under identical turning conditions. The total sum of squared residuals is calculated based on the experimental data from all turning experiments. The total residuals minus the pure errors yield the lack-of-fit error, which reflects systematic deviations that the model structure itself failed to capture. By calculating the lack-of-fit F-value and comparing it with the critical value of the F-distribution at a preset significance level, the existence of significant structural defects in the second-order polynomial response surface model can be quantitatively determined statistically. If the lack-of-fit F-value is less than the critical value of the F-distribution, it indicates that the current model has sufficiently captured the physical mapping relationship between the machining parameters and the response target; if the lack-of-fit F-value is greater than or equal to the critical value of the F-distribution, the model is corrected by adjusting the range of machining parameters and re-executing the experimental design and modeling steps.
[0058] Understandably, since there are 10 regression coefficients to be determined, the center point trial must be repeated at least 3 times, with a minimum of 15 trials. The total residual degrees of freedom are 15-10=5, of which pure error accounts for 2 degrees of freedom and misfit accounts for 3 degrees of freedom. This allows us to test whether the model has serious misfit. 15 trials is the minimum and most economical configuration to ensure that the model has statistical testing capability.
[0059] Example 2 Example 2 is a specific implementation of Example 1; the material properties of the workpiece to be processed are set as follows: SiCp volume fraction 45%, matrix is 6061 aluminum alloy, and material condition is T6 heat treated state. Based on the results of the previous single-factor pre-experiment and the performance limitations of the machine tool spindle, the value ranges of the three processing parameters are determined as follows: cutting speed The feed rate is 30-120 m / min. The turning depth is 0.06-0.2 mm / r. The range is 0.2-2mm. The three processing parameters mentioned above are coded according to Table 1, as shown in Table 1.
[0060] Table 1 Processing Parameter Coding Table
[0061] A turning test scheme was designed, including 12 sets of edge center point tests and 3 sets of center point tests, totaling 15 sets of tests. Each set of turning tests was executed. To eliminate systematic errors such as machine tool thermal drift and tool batch differences, the execution order of the 15 sets of tests was randomized. The 3 sets of center point tests were interspersed at the beginning, middle, and end of the test sequence, respectively. The coding scheme and measured response data for each set of tests are shown in Table 2.
[0062] Table 2 Turning test plan and measured response data
[0063] All 15 turning tests were performed on a precision CNC lathe using polycrystalline diamond (PCD) tools. A new tool tip was used for each test to ensure consistent initial tool conditions. Cutting temperature was measured using an infrared thermal imager focused on the tool tip area, recording the average temperature during the stable cutting phase. Rake face wear was measured using a depth-of-field microscope, surface roughness was measured using a surface profilometer, and material removal rate was calculated using the formula... Obtained through calculation.
[0064] Understandably, the three sets of center point tests (numbers 13, 14, and 15) were repeated under the same cutting parameters (speed 75 m / min, feed rate 0.13 mm / r, depth of cut 1.1 mm), and their temperature response values were 615℃, 618℃, and 613℃, respectively, with a fluctuation range of only 5℃. This indicates that the test system has good repeatability and stability at this center point. If the relative deviation of the three sets of center point response values exceeds 5%, it suggests an abnormality in the machine tool's thermal stability, tool consistency, or material batch uniformity, requiring investigation of the cause and re-performing all 15 sets of tests.
[0065] Cutting temperature For example, the coded values of the 15 groups of experiments in Table 2 Substituting the measured temperature values into the following formula, a second-order polynomial response surface model with respect to cutting temperature,
[0066] Establish a 15x10 matrix X. Based on the measured values of each turning test for a certain response target, establish a corresponding 15x1 optimization vector Y. Then, use matrix operation formulas... To solve for the regression coefficient vector, in this embodiment, the Minitab statistical software is used to perform the above matrix operations, and the regression coefficients of the temperature response surface model are output as follows:
[0067] Similarly, second-order polynomial response surface models are established for the back face wear amount VB, surface roughness Ra, and material removal rate MRR, respectively.
[0068] The significance probability value (P-value) for each regression coefficient is calculated based on the t-test, with a significance level set. For the cutting temperature model, and The p-values were 0.285 and 0.395, respectively, both greater than 0.05, and were therefore considered insignificant. A successive backward elimination method was used, removing the term with the largest p-value in each round: the first round of elimination... After refitting, The p-value was 0.268. This was still not significant, therefore a second round of elimination was performed. After refitting, the p-values of all remaining terms were less than 0.05. The final optimized simplified cutting temperature model is as follows:
[0069] Based on the measured temperature values of 615℃, 618℃, and 613℃ from the three groups of center point tests in Table 2, the average measured temperature of the center point tests was calculated as follows: The sum of squared errors is:
[0070] In the formula, The measured cutting temperature at a single center point is given, with pure error degrees of freedom. The pure error mean square is Based on the measured temperatures and simplified model predictions from all 15 sets of experiments, the total residual sum of squares was calculated. Total residual degrees of freedom The sum of squared errors of the misfit is - =69.63, the unfit degrees of freedom are The misfit mean square is The F-value for the lack of fit is calculated as follows: =2.198 Set significance level Consult the F-distribution table (numerator: 5 degrees of freedom, denominator: 2 degrees of freedom) for the critical value. Since 2.198 < 19.30, the lack-of-fit test is not significant. Therefore, the current second-order polynomial response surface model is determined to have no structural defects and is valid, allowing it to proceed to the multi-objective optimization step. In other embodiments, if the lack-of-fit F-value is greater than or equal to the critical value, the range of cutting parameters can be adjusted, and all 15 sets of experiments can be re-executed.
[0071] Based on the measured temperature data in Table 2, the extreme operating conditions approaching the critical thermal damage temperature (700℃) of the PCD tool were selected by sorting the temperature values from highest to lowest. The test data for the edge center point with the highest temperature was selected from Table 2: Group 4. ), and operating points where the temperature approaches the critical value: Group 8 Group 12 The coded values of the above three sets of experimental data were converted into physical coordinates according to Table 1, yielding key data points regarding feed rate and cutting temperature: point 4 (0.20, 712), point 8 (0.13, 685), and point 12 (0.2, 672). Since the temperature value of the fourth set of experimental data exceeds the critical thermal damage temperature of the tool, this point itself cannot be directly used as a valid point on the safety boundary for fitting. Therefore, using linear interpolation between points 4 (0.20, 712) and 8 (0.13, 685) at the same cutting speed (120 m / min), the critical feed rate at a cutting speed of 120 m / min and a temperature of 700℃ was calculated. The value is 0.169 mm / r; simultaneously, using point 12 and combining the experimental data from the 10th group at the same speed, the critical feed rate corresponding to 700℃ at a cutting speed of 75 m / min is extrapolated. With a velocity of 0.215 mm / r, the two critical points (75, 0.215) and (120, 0.169) are connected on the velocity-feed plane to obtain the coupled constraint boundary line. Based on the principle of engineering conservatism, the maximum allowable feed rate is further tightened at the same cutting speed, thereby reserving a safety margin for uncertainties such as material batch fluctuations and changes in tool wear. The intercept can also be finely adjusted downward from 0.29, for example, to 0.28.
[0072] The optimization objectives are set as follows: minimize T, minimize VB, minimize Ra, and maximize MRR. The constraints are set as follows: (1) Boundary constraints of cutting parameters: , , .
[0073] (2) Response to target boundary constraints: , , , .
[0074] (3) Coupling constraints: .
[0075] This embodiment uses the NSGA-II multi-objective evolutionary algorithm for iterative solution. The algorithm parameters are configured as follows: population size 150, maximum number of generations 300, simulated binary crossover probability 0.90, and polynomial mutation probability 0.10. In each generation, the algorithm sequentially performs coupling constraint verification on each individual in the population. If the individuals in this population are located within a safe zone, and the four objective function values are preserved and calculated normally, then... When an individual in the population crosses the safety boundary, it is given a large penalty fitness value, causing it to be eliminated in subsequent non-dominated sorting. After 300 generations of iteration, the population converges to a stable Pareto front, outputting a Pareto optimal solution set containing non-dominated solutions. Partial solution sets are shown in Table 3.
[0076] Table 3. Representative solutions in the Pareto optimal solution set
[0077] An initial decision matrix is constructed by taking the non-dominated solutions in the Pareto optimal solution set as rows and the four optimization objectives as columns. The decision matrix is then standardized using vector normalization to obtain a normalized decision matrix. The positive and negative ideal solutions for the four optimization objectives are determined column-wise. Positive Ideal Solution : , , ,
[0078] Negative ideal solution : , , ,
[0079] Calculate the Euclidean distance between each scheme and the positive and negative ideal solutions, and then calculate the relative proximity. This embodiment uses a balanced weight configuration (each objective weight is 0.25). The top three schemes in the final ranking are Scheme B, Scheme C, and Scheme D. If the current processing task is the first trial production of a critical cylinder with high requirements for surface quality and tool life, Scheme B can be selected. If it is for mass production and a balance between efficiency and quality is pursued, Scheme C can be selected. If the time limit is tight and output needs to be maximized, Scheme D can be selected.
[0080] This document uses specific embodiments to illustrate the principles and implementation methods of the present invention. The descriptions of these embodiments are only intended to aid in understanding the method and core ideas of the present invention. It should be noted that those skilled in the art can make various improvements and modifications to the present invention without departing from its principles, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.
[0081] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the product of this invention is usually placed during use. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.
[0082] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.
Claims
1. A multi-objective collaborative optimization method for turning processes, characterized in that, Includes the following steps: Experimental design and data acquisition steps: Obtain the material properties of the workpiece to be processed, and determine the value range of the three machining parameters, namely cutting speed, feed rate, and depth of cut, based on the material properties; encode the upper limit of each machining parameter within its respective value range as +1, the lower limit as -1, and the center value as 0; construct 12 sets of edge center point tests by combining any two machining parameters with their extreme values and the remaining machining parameter with its center value, and construct center point tests by combining all machining parameters with their center values. The edge center point tests and center point tests together constitute the turning test scheme; execute each set of turning tests sequentially according to the turning test scheme. After each set of turning tests is completed, collect and record the four response targets under that set of tests, namely cutting temperature, flank wear, surface roughness, and material removal rate. The proxy model construction steps are as follows: For each of the four response targets, a second-order polynomial response surface model is constructed, which includes a constant term, a linear term of each processing parameter, a quadratic term of each processing parameter, and an interaction term between each processing parameter. The second-order polynomial response surface model is then fitted by regression based on the collected experimental data to obtain the mathematical prediction model for each response target. Multi-objective optimization and constraint handling steps: The mathematical prediction models of each response objective are used as objective functions. Minimizing cutting temperature, minimizing tool wear, minimizing surface roughness, and maximizing material removal rate are set as optimization objectives to form a multi-objective optimization model. A multi-objective evolutionary algorithm is used to predict and iterate the multi-objective optimization model, converging towards the Pareto front until the preset maximum number of generations or the population converges to a stable Pareto front is reached. The output is a Pareto optimal solution set containing several non-dominated solutions, where each solution represents a feasible combination of machining parameters.
2. The multi-objective collaborative optimization method for turning processes as described in claim 1, characterized in that, Also includes Engineering decision-making and output steps: Take the non-dominated solutions in the Pareto optimal solution set as rows and the four optimization objectives as columns to form an initial decision matrix; The decision matrix is standardized using vector normalization to obtain a normalized decision matrix. Positive and negative ideal solutions for the four optimization objectives are determined column-wise. The positive ideal solution includes the minimum values of cutting temperature, tool wear, and surface roughness, and the maximum value of material removal rate. The negative ideal solution includes the maximum values of cutting temperature, tool wear, and surface roughness, and the minimum value of material removal rate. For each machining parameter combination in the normalized decision matrix, the Euclidean distance between it and the positive and negative ideal solutions are calculated. Then, the relative proximity of each machining parameter combination is calculated. The proximity value ranges from 0 to 1; a larger value indicates that the machining parameter combination is closer to the positive ideal solution and further away from the negative ideal solution. All machining parameter combinations are sorted in descending order of relative proximity, and at least one of the top-ranked combinations is selected as a recommended solution and included in the recommended parameter library for users to choose from.
3. The multi-objective collaborative optimization method for turning processes as described in claim 2, characterized in that, When calculating the Euclidean distance between each combination of processing parameters and the positive and negative ideal solutions, the corresponding weight coefficients are multiplied to adjust the influence of each optimization objective in the ranking process. The sum of the weight coefficients of all optimization objectives is 1.
4. The multi-objective collaborative optimization method for turning processes as described in any one of claims 1-3, characterized in that, Also includes Verification and correction steps: Select the scheme to be verified from the Pareto optimal solution set, conduct a verification turning experiment on a CNC machine tool, and compare the measured response value with the predicted value output by the corresponding mathematical prediction model. If the relative error is within the preset threshold range, the verification is successful, and the corresponding scheme is included in the final recommended parameter library; if the relative error exceeds the preset threshold range, the verification data is added as new sample points to the original turning test data, and the proxy model construction steps and multi-objective optimization and constraint processing steps are repeated for iterative correction until the verification is successful.
5. The multi-objective collaborative optimization method for turning processes as described in claim 1, characterized in that, In the proxy model construction step, the second-order polynomial response surface model is shown in the following equation. In the formula, This represents the measured value of the response target, where n is an integer from 1 to 4. The measured value representing the cutting temperature. The measured value representing the amount of tool wear. The measured value representing surface roughness. The measured value representing the material removal rate; The coded value representing the cutting speed. Code indicating feed rate The encoded value representing the turning depth, where The model intercept represents the baseline response value. , and These are the linear main effect coefficients for cutting speed, feed rate, and depth of cut, respectively, characterizing the degree of direct influence of each machining parameter on the target response. , and The coefficients of the quadratic term represent the nonlinear curvature effect of each processing parameter on the target response. , and The coefficients represent the pairwise interaction effects, characterizing the synergistic enhancement or inhibition between processing parameters.
6. The multi-objective collaborative optimization method for turning processes as described in claim 5, characterized in that, The turning test scheme includes N sets of turning tests, where N is an integer greater than or equal to 13. The encoded values of each set of turning tests are substituted into a second-order polynomial response surface model to establish an N-row, 10-column X matrix. Based on the measured values of each set of turning tests with respect to a certain response target, a corresponding N-row, 1-column optimization vector Y is established. Matrix operation formulas are then used... Solve for the regression coefficient vector, the regression coefficient vector is For each response target, perform matrix operations and output the corresponding regression coefficients. In order to obtain mathematical prediction models for each response target.
7. The multi-objective collaborative optimization method for turning process as described in claim 6, characterized in that, Calculate the significance probability value corresponding to each regression coefficient and compare the significance probability value with a preset significance level. If the significance probability value of a regression coefficient is greater than or equal to the preset significance level, the term corresponding to that regression coefficient is determined to be insignificant. Each time, the term with the largest current significance probability value is removed. After removal, the remaining terms are refitted, and the significance probability values of the remaining regression coefficients are recalculated. The above removal and refitting process is repeated until the significance probability values of all terms retained in the second-order polynomial response surface model are less than the preset significance level, thus obtaining the optimized mathematical prediction model for each response target. During the removal process, if the square term or interaction term of a certain processing parameter is retained, the linear term corresponding to that processing parameter, even if its significance probability value is greater than or equal to the preset significance level, is still forcibly retained in the second-order polynomial response surface model and does not participate in the removal.
8. The multi-objective collaborative optimization method for turning processes as described in claim 1, characterized in that, In the multi-objective optimization and constraint processing steps, constraints are set for the multi-objective optimization model. These constraints include: sorting the data according to the measured temperature values from the turning test; selecting at least two sets of edge center point test data in descending order of the measured temperature values; and constructing coupled constraints between cutting speed and feed rate using linear fitting. When using a multi-objective evolutionary algorithm to predict and iterate the multi-objective optimization model, the coupled constraints are verified for each individual in the population during each generation of evolution. If the individuals in this population are located within a safe zone, and the four objective function values are preserved and calculated normally, then... If an individual in the population crosses the safety boundary, it is given a very high penalty fitness value, causing it to be eliminated in subsequent non-dominated rankings. In the formula, f represents the feed rate. Represents cutting speed. Represents cutting speed The function, whose expression is obtained by linear fitting of selected experimental data.
9. The multi-objective collaborative optimization method for turning processes as described in claim 8, characterized in that, The constraints also include machining parameter boundary constraints and response target boundary constraints. The machining parameter boundary constraints include setting the range of values for cutting speed, feed rate, and depth of cut. The response target boundary constraints include setting the upper limit of cutting temperature, tool wear, and surface roughness, as well as the lower limit of material removal rate.
10. The multi-objective collaborative optimization method for turning processes as described in claim 1, characterized in that, The number of center point experiments is at least 3 sets; the method also includes a misfit test step: calculating the pure error sum of squares and pure error degrees of freedom based on the experimental data of at least 3 sets of center point experiments, calculating the total residual sum of squares and total residual degrees of freedom based on the experimental data of all turning experiments in the turning test scheme, and calculating the misfit F-value using the pure error sum of squares, pure error degrees of freedom, total residual sum of squares and total residual degrees of freedom; if the misfit F-value is less than the critical value of the F distribution under the preset significance level, it is determined that the second-order polynomial response surface model is not significantly misfitted, and the multi-objective optimization and constraint processing step is entered; if the misfit F-value is greater than or equal to the critical value of the F distribution, it is determined that the second-order polynomial response surface model is significantly misfitted, and the value range of at least one of the cutting speed, feed rate and turning depth is adjusted, and the experimental design and data acquisition steps and the surrogate model construction steps are re-executed.
