A method for optimizing a hardware manufacturing process based on digital twinning

Abstract

The application belongs to the technical field of digital twinning and intelligent manufacturing, and particularly relates to a hardware manufacturing process optimization method based on digital twinning. The method comprises the following steps: constructing a high-fidelity digital twinning model integrating stamping, bending and heat treatment processes; defining a multi-objective problem with strength, accuracy, surface quality, energy consumption and cost as optimization objectives; performing mutualistic symbiosis, commensalism and parasitism operations in the process parameter space for global optimization by using a symbiotic organism search algorithm; evaluating fitness based on virtual trial simulation and maintaining a high-quality Pareto front solution set; and finally outputting an optimal process parameter scheme meeting production constraints. Through deep coupling of digital twinning and super-parameter-free optimization algorithm, the application improves global search capability, multi-objective trade-off effect and process development efficiency, reduces material and energy consumption, and enhances manufacturing flexibility and green level.

Description

Technical Field

[0001] This invention belongs to the field of digital twin and intelligent manufacturing technology, specifically relating to a method for optimizing the manufacturing process of hardware parts based on digital twin. Background Technology

[0002] With the deepening of intelligent manufacturing and Industry 4.0 concepts, digital twin technology is increasingly widely used in the manufacturing industry, especially demonstrating its advantages in modeling, monitoring, and optimizing complex processes. Hardware manufacturing, as a typical discrete manufacturing process involving multiple processes and parameters, relies heavily on the precise control of key process parameters such as stamping, bending, and heat treatment for product quality, production efficiency, and cost control. However, due to the complex nonlinear interactions between process parameters and the often conflicting optimization objectives, traditional experience-based or single-objective optimization methods are insufficient to meet the demands of highly flexible and high-quality production.

[0003] Digital twin-based process optimization methods attempt to iteratively fine-tune process parameters in a simulation environment by constructing a virtual mapping model. However, existing optimization engines generally employ classic intelligent algorithms such as genetic algorithms and particle swarm optimization. These methods are prone to getting trapped in local optima in high-dimensional parameter spaces, have slow convergence speeds, and are sensitive to their own hyperparameters, leading to instability and poor repeatability in the optimization process. Furthermore, when facing multi-objective optimization problems, the Pareto solution sets generated by traditional algorithms are unevenly distributed and incompletely covered, making it difficult for decision-makers to effectively balance performance, cost, and efficiency.

[0004] Therefore, there is an urgent need for a digital twin-driven method that integrates a novel intelligent optimization mechanism, which can efficiently explore the high-dimensional process parameter space without frequent parameter tuning and stably output a high-quality multi-objective optimization solution set, thereby achieving global, collaborative and adaptive optimization of hardware manufacturing processes. Summary of the Invention

[0005] To achieve the above objectives, the present invention provides a method for optimizing the manufacturing process of hardware parts based on digital twins, comprising the following steps: Based on the equipment operating parameters, environmental variables and historical production data of hardware parts, a high-fidelity virtual mapping model is established using multi-physics coupling simulation technology. The model synchronously simulates the stress and strain distribution during the stamping process, the springback behavior during the bending stage and the phase transformation and residual stress evolution during the heat treatment process, and updates the state at a frequency not less than a preset threshold. By taking strength, dimensional accuracy, surface quality, energy consumption and manufacturing cost as optimization objectives, and stamping speed, die clearance, bending angle, springback compensation and heat treatment temperature curve as decision variables, a high-dimensional nonlinear multi-objective optimization problem is formed. Initialize the symbiotic organism search algorithm population. In the solution space of process parameters, the initial population is generated by Latin hypercube sampling strategy. Each individual represents a complete combination of process parameters. The value range of each parameter is set according to the equipment capacity boundary and process safety threshold. The symbiotic search algorithm is iteratively optimized by performing mutualistic, symbiotic, and parasitic operations on each individual in the population. Mutualistic operations improve fitness through cooperation between two individuals, symbiotic operations benefit one individual while the other remains unaffected, and parasitic operations introduce local perturbations in the individual's neighborhood to enhance global exploration capabilities. The fitness of individuals is evaluated and the Pareto front solution set is maintained. Virtual prototype simulation is performed on each individual based on a digital twin model to obtain the quantitative results of each optimization objective. Non-dominated sorting mechanism is used to screen non-dominated solutions and dynamically maintain a uniformly distributed and comprehensive Pareto front solution set. The system outputs optimized process parameter schemes. When the convergence condition is met or the maximum number of iterations is reached, it selects the optimal compromise solution that meets the actual production constraints from the Pareto front solution set, generates an executable process parameter configuration file, and pushes it to the physical manufacturing system for execution.

[0006] Preferably, the digital twin model integrates finite element analysis, thermodynamic simulation and kinematic modeling modules. The finite element analysis module uses an implicit dynamic solver to perform millisecond-level dynamic simulation of the plastic deformation and stress concentration areas of the sheet metal during the stamping process. The mesh generation adopts an adaptive refinement strategy, which automatically refines the mesh to a unit size of 0.2 mm in areas with a curvature radius of less than 2 mm. The thermodynamic simulation module is based on the coupling of the Fourier heat conduction equation and the JMAK phase transformation kinetic model to calculate the temperature field evolution, the proportion of austenite to martensite transformation and the distribution of residual stress during the heat treatment stage. The kinematic modeling module reproduces the motion trajectory of the bending machine slider and the opening and closing logic of the mold, and combines the springback compensation algorithm to predict the geometric deviation of the formed part.

[0007] Preferably, in the multi-objective optimization problem, the strength objective is characterized by tensile strength or yield strength, the dimensional accuracy objective is measured by the absolute value of the maximum deviation of the key feature dimension, the surface quality objective is characterized by the surface roughness Ra value or defect density, the energy consumption objective is calculated by the total electrical energy consumption of a single product, and the manufacturing cost objective includes the comprehensive cost of material loss, tool wear and labor hours; each objective is normalized and then scalarized using the weighted Chebyshev paradigm to eliminate dimensional differences.

[0008] Preferably, when generating the initial population, the boundary values ​​of each decision variable are provided by the process parameter constraint library. During initialization, the legality of all individuals is verified, out-of-bounds individuals are removed, and resampling is performed until all individuals in the population satisfy the equipment capability boundary and process safety threshold constraints.

[0009] Preferably, the update formula for the mutually beneficial symbiotic operation is: ; and They are two different individuals. It is the best individual in the current population. A uniformly distributed random number in the interval [0,1). This is the scaling factor; The update formula for symbiotic operations is: ; The random number is a uniformly distributed number in the interval [-1, 1). Parasitic operations in individuals Random replacements are performed on several parameter dimensions, with the disturbance amplitude not exceeding a predetermined proportion of the parameter range.

[0010] Preferably, the maximum capacity of the Pareto front solution set is set to a preset upper limit of solution set capacity. When the solution set size is greater than the preset upper limit of solution set capacity, the crowding distance calculation mechanism is used to preferentially retain solutions in sparsely distributed regions. The crowding distance is determined by calculating the sum of the distances between adjacent individuals after sorting each target dimension.

[0011] Preferably, the convergence condition includes the rate of change of the hypervolume index of the Pareto front solution set being less than a preset rate of change threshold for several consecutive generations, or the standard deviation of all individuals in the population on each target being less than a preset threshold; the maximum number of iterations is set to a preset maximum number of iterations.

[0012] Preferably, the method also includes performing closed-loop verification of the optimized process parameter scheme in a digital twin model, generating a product quality prediction report through virtual trial production, and if the key indicators fail to meet the standards, performing local sampling centered on the current Pareto solution set to trigger a new round of optimization iteration.

[0013] Preferably, the method further includes storing historical optimization cases and corresponding actual production results in a process knowledge base. This process knowledge base contains process parameter-performance result pairs, product models, material batches, equipment numbers, and environmental conditions, which are used to guide population initialization or as training data for machine learning models during subsequent optimization processes.

[0014] Preferably, the symbiotic organism search algorithm does not require setting crossover rate, mutation rate or inertia weight during execution. Its internal operation only relies on random numbers and the state of the population itself, making it suitable for parameter-free optimization when switching between different hardware products.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. By introducing a search mechanism that simulates biological symbiosis, global exploration is achieved in the high-dimensional process parameter space, avoiding the problem of traditional genetic algorithms or particle swarm algorithms getting stuck in local optima due to premature convergence, thus improving the success rate of multi-objective optimization; the symbiotic biological search algorithm itself has very few hyperparameters, so there is no need to repeatedly adjust the algorithm configuration for different hardware products, reducing the threshold for engineering applications and significantly improving the stability of the optimization process.

[0016] 2. The generated Pareto front solution set is uniformly distributed and comprehensively covered, achieving a fine trade-off between conflicting objectives such as strength, accuracy, and cost. The solution set hypervolume index is improved compared to traditional methods. It provides a variety of feasible solutions for process engineers to choose from, supports dynamic screening based on actual production constraints, and effectively supports the needs of highly flexible manufacturing.

[0017] 3. The digital twin model is deeply integrated with the optimization engine, and the cycle of a single complete optimization is controlled within a predetermined time period, which is shorter than the traditional trial and error method. By replacing a large number of physical experiments with virtual prototyping, material waste in a single process development is significantly reduced, energy consumption is reduced, and the level of green manufacturing is improved. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention. Detailed Implementation

[0019] Example 1: Reference Figure 1 In the method for optimizing the manufacturing process of hardware parts based on digital twins, step 1, constructing a digital twin model of the hardware parts manufacturing process, specifically includes the following sub-operations: First, obtain the three-dimensional geometric model data of the hardware to be optimized. This three-dimensional geometric model data comes from the computer-aided design system and is in the STEP or IGES standard format, with a geometric accuracy of not less than 0.01 mm. Then, the physical property parameters of the materials used in the hardware are collected, including but not limited to elastic modulus, Poisson's ratio, yield strength, coefficient of thermal expansion, thermal conductivity and critical phase transition temperature. These parameters are measured by a material testing machine and a differential scanning calorimeter, and stored in the material database in the form of a structured data table. Then, the real-time operating parameters of key manufacturing equipment such as stamping machines, bending machines, and heat treatment furnaces are simultaneously accessed, including spindle speed, hydraulic pressure, servo motor position feedback, mold temperature, cooling rate, etc., with a sampling frequency of not less than 100 Hz. The data is transmitted to the edge computing node through the OPC UA protocol. At the same time, environmental variables in the workshop, such as ambient temperature, humidity, and vibration spectrum, are collected and uploaded in real time by an industrial IoT sensor array deployed around the production line. Finally, historical production data, including process parameter records, quality inspection reports, and equipment maintenance logs for similar hardware parts over the past six months, are integrated to form a time-series dataset.

[0020] After completing the above data acquisition, a high-fidelity virtual mapping model was constructed using a multiphysics coupling simulation engine. This high-fidelity virtual mapping model consists of three core modules: a finite element analysis module, a thermodynamic simulation module, and a kinematic modeling module.

[0021] The finite element analysis module uses an implicit dynamic solver to perform millisecond-level dynamic simulation of the plastic deformation, stress concentration areas and crack initiation risk of sheet metal during the stamping process. The mesh generation adopts an adaptive refinement strategy, which automatically refines the mesh to an element size of 0.2 mm in areas with a curvature radius of less than 2 mm. The thermodynamic simulation module is based on the Fourier heat conduction equation and the JMAK phase transformation kinetic model. It coupled calculations of the temperature field evolution, the proportion of austenite to martensite transformation and the residual stress distribution during the heat treatment stage, with the time step controlled within 0.5 seconds. The kinematic modeling module accurately reproduces the movement trajectory of the bending machine slider and the opening and closing logic of the mold, and combined with the springback compensation algorithm, predicts the geometric deviation of the formed part.

[0022] The three modules mentioned above achieve millisecond-level data exchange through a shared memory mechanism, ensuring strong coupling and consistency among multiple physical fields. The update frequency of the entire digital twin model is set to no less than 5 Hz, that is, to complete state synchronization once every 200 milliseconds, in order to ensure real-time synchronization with the physical production line.

[0023] In the above method, step 2, defining a multi-objective optimization problem, is specifically executed as follows: Based on the final application scenario of the hardware components, a performance requirement index system should be defined. Strength targets should be quantified using tensile strength (unit: MPa) or yield strength (unit: MPa), and their values ​​must meet the minimum threshold requirements in the product design specifications. The dimensional accuracy target is measured by the absolute value of the maximum deviation (in millimeters) of key feature dimensions (such as hole spacing, flange height, and bending angle). The deviation is calculated by performing ICP registration between the geometric point cloud output by the digital twin model and the original CAD model. Surface quality targets are characterized by surface roughness Ra (unit: micrometer) or defect density (unit: defects / cm²), which are generated by simulation from the virtual optical inspection module; The energy consumption target is calculated based on the total electrical energy consumption (unit: kilowatt-hour) of a single product from raw material input to finished product output. The data is obtained by power integration of each device in the digital twin model. The manufacturing cost target is a comprehensive cost, which includes material loss cost (calculated by multiplying the weight of scrap by the unit price), tool wear cost (accumulated depreciation based on the number of stampings and bending strokes), and labor cost (calculated based on the standard labor hour rate).

[0024] The above five optimization objectives are used to construct an objective vector. , For strength, For dimensional accuracy, For surface quality, For energy consumption, Cost. Decision variable vector. ,, for Each vector contains The adjustable process parameters include: stamping speed (unit: mm / s), ranging from 300 to 800; and die clearance (unit: mm), ranging from 8% to 12% of the sheet thickness. Bending angle (unit: degrees), ranging from 85 to 95; Springback compensation (unit: degrees), ranging from 1.0 to 3.5; The heat treatment heating rate (unit: degrees Celsius / minute) ranges from 5 to 20. Insulation temperature (unit: degrees Celsius), ranging from 850 to 950; Insulation time (unit: minutes), ranging from 30 to 90; Cooling rate (unit: degrees Celsius / minute), ranging from 10 to 50.

[0025] This leads to a high-dimensional nonlinear multi-objective optimization problem.

[0026] To eliminate scale differences between objectives with different dimensions, the mathematical expression of this high-dimensional nonlinear multi-objective optimization problem is normalized using the weighted Chebyshev normal form. Specifically, firstly, for each objective... Normalization , and These are the theoretical minimum and maximum values ​​of the objective within the feasible region, obtained through preliminary experiments or historical data statistics; subsequently, a weighted Chebyshev scalarization function is constructed: , These are the weighting coefficients. This is the reference value for the ideal point.

[0027] In the above method, step 3, initializing the symbiotic organism search algorithm population, is specifically implemented as follows: Within the solution space of the process parameters composed of the above n decision variables, an initial population is generated using a Latin hypercube sampling strategy. This Latin hypercube sampling strategy divides the value range of each parameter dimension into N equal parts (N is the preset population size, usually set to an integer between 50 and 200), and then randomly selects a sample point from each part to ensure that the samples in each dimension are evenly distributed and without repetition. The population generated by this method... Each individual represents a complete combination of process parameters.

[0028] The boundaries of each process parameter are strictly set based on the equipment capacity limits and process safety thresholds. For example, the upper limit of the stamping speed of 800 mm / s is limited by the maximum output power of the main motor of the stamping press and the response delay of the hydraulic system, while the lower limit of 300 mm / s is the minimum process requirement to ensure forming quality; the upper limit of the heat treatment holding time of 90 minutes is determined by the production cycle time constraint, while the lower limit of 30 minutes is the shortest time to ensure complete austenite formation.

[0029] All boundary values ​​are stored in the process parameter constraint library. The initialization process automatically calls the process parameter constraint library to perform legality verification, removes any out-of-bounds individuals and resamples until all individuals in the population meet the constraint conditions. While ensuring uniform coverage of the parameter space, it effectively avoids the clustering effect that may be caused by random sampling, laying a diversity foundation for subsequent global search.

[0030] In the above method, step 4 involves iteratively optimizing the symbiotic organism search algorithm, specifically by sequentially performing mutualistic, symbiotic, and parasitic operations on each individual in the population. This process is repeated in each generation until the convergence condition is met.

[0031] Mutualism simulates the behavior of two organisms cooperating to improve overall fitness. In the algorithm, two different individuals are randomly selected. and Calculate its fitness value (i.e., the multi-objective function value). Let The optimal individual in the current population (determined through hypervolume contribution or weighted sum method). The formula for updating the new position is: ; A uniformly distributed random number in the interval [0,1). This is the scaling factor, a preset scaling coefficient between 0.5 and 1.5, used to control the search step size. The operation makes... Move to a better area, while introducing Information to enhance exploration capabilities.

[0032] Symbiotic operations simulate a relationship where one organism benefits while the other remains unaffected. Two different individuals are randomly selected. and , Benefit It remains unchanged. The formula for updating the new position is: ; Generate uniformly distributed random numbers within the interval [-1, 1). The operation is performed on... and Making local disturbances along the direction of the line connecting the points helps to break out of the local flat area.

[0033] Parasitic manipulation simulates the localized invasion of a host by a parasitic organism. This is targeted at the individual. Randomly select "host" individuals within its neighborhood. (Usually individuals with similar fitness), and then Random replacement is performed on several parameter dimensions. Specifically, random selection... Dimensions ( From 1 to (integers between each selected dimension) Generate new values , For individuals In dimensions The original value on, For dimension The maximum value of the above parameter, For dimension The maximum value of the above parameter, The random perturbation coefficient is set within the interval [-0.1, 0.1], ensuring that the perturbation amplitude does not exceed 10% of the parameter range. This operation introduces local diversity to prevent premature population convergence.

[0034] The three operations described above are applied sequentially to each individual in the population to complete one iteration. Throughout the process, the algorithm does not require setting hyperparameters such as crossover rate, mutation rate, or inertia weight; its internal operations rely solely on random numbers and the population's own state, exhibiting extremely strong robustness.

[0035] In the above method, step 5, evaluating individual fitness and maintaining the Pareto front solution set, is specifically executed as follows: For each individual in the population... The corresponding process parameters are then input into the digital twin model constructed in step 1, and the virtual trial production simulation is initiated. The simulation process fully replicates the entire process from raw material input to finished product output, taking approximately 2 to 5 minutes (depending on the model complexity). After the simulation, the quantification results of the five optimization objectives are automatically extracted to form the target vector Fᵢ.

[0036] Subsequently, a fast non-dominated sorting algorithm is used to jointly sort the current population and the historical Pareto solution set. This fast non-dominated sorting algorithm divides all individuals into multiple non-dominated levels, with the first level being the current Pareto front. To maintain the distribution of the solution set, the crowding distance of each individual is calculated within the same non-dominated level: for each target dimension, individuals are sorted by target value, with the crowding distance of the first and last individuals set to infinity, and the crowding distance of the middle individuals being the sum of the distances of their two adjacent individuals on the target. The larger the crowding distance, the sparser the region where the individual is located.

[0037] The maximum capacity of the Pareto front solution set is set to a preset upper limit (typically 100 to 300). When the solution set size exceeds the preset upper limit, individuals with the smallest crowding distance are preferentially removed, i.e., solutions from sparsely distributed regions are retained to ensure that the front solution set is comprehensive and evenly distributed. This Pareto front solution set is dynamically updated after each iteration and persistently stored in the optimization engine's memory cache for subsequent selection and verification.

[0038] In the above method, step 6, outputting the optimized process parameter scheme, specifically includes the following operations: First, determine whether the algorithm meets the preset convergence conditions. The convergence conditions include two scenarios: The rate of change of the hypervolume index of the Pareto front solution set for 10 consecutive generations is less than 0.5%; The standard deviation of all individuals in the population for each objective is less than 5% of the allowable fluctuation range for each objective.

[0039] If any condition is met, or the maximum number of iterations (preset to 200) has been reached, the iteration will terminate.

[0040] Subsequently, the optimal compromise solution that meets the actual production constraints is selected from the final Pareto front solution set. The selection process is completed by process engineers through a human-computer interaction interface, which displays the performance of all non-dominated solutions on each objective in the form of a parallel coordinate graph and supports setting hard constraints (such as cost must not exceed a certain threshold, intensity must be greater than a certain value). The process parameter combination corresponding to the selected solution is encapsulated in a standard JSON format configuration file, containing the name, value, and unit of all decision variables.

[0041] The configuration file is pushed to the physical manufacturing system via the manufacturing execution system interface and automatically distributed to the controllers of the stamping machine, bending machine, and heat treatment furnace, enabling one-click parameter loading. Furthermore, this invention includes closed-loop verification of the optimized process parameter scheme in a digital twin model: virtual prototyping is run again to generate a product quality prediction report containing a strength distribution cloud map, a dimensional deviation heat map, an energy consumption decomposition pie chart, and a cost breakdown table. If any key indicator in the report fails to meet the standard (e.g., strength less than the design value, key dimensions exceeding tolerance), a new round of optimization iteration is automatically triggered. The initial population performs local sampling centered on the current Pareto solution set to accelerate convergence.

[0042] Furthermore, this invention also includes storing the current optimization case and the corresponding actual production results (obtained through an online quality inspection system) in a process knowledge base. This process knowledge base is constructed using a relational database, with each record containing a process parameter-performance result pair, product model, material batch, equipment number, and environmental conditions, accumulating to more than 10,000 records. This process knowledge base is used for population initialization guidance in subsequent optimization processes: when processing similar products, historical cases can be retrieved, and high-performance parameter combinations can be extracted as part of the initial population, significantly improving convergence speed; simultaneously, the dataset can also be used as training data for machine learning models to construct surrogate models to accelerate simulation evaluation.

[0043] To illustrate the effectiveness of this invention, the following application example is provided: A car door hinge hardware component, made of DC04 cold-rolled steel sheet, 1.5 mm thick. Its key performance requirements are: tensile strength ≥ 340 MPa, hole spacing deviation ≤ ±0.15 mm, no visible scratches on the surface, single-piece energy consumption ≤ 0.8 kWh, and comprehensive cost ≤ 12 yuan. Decision variables include stamping speed, die clearance (0.12 to 0.18 mm), bending angle (88 to 92 degrees), springback compensation (1.8 to 2.5 degrees), heat treatment heating rate (8 to 15 degrees Celsius / minute), holding temperature (880 to 920 degrees Celsius), holding time (40 to 70 minutes), and cooling rate (20 to 40 degrees Celsius / minute), totaling eight dimensions.

[0044] The initial population size was set to 100, generated using Latin hypercube sampling. The digital twin model update frequency was set to 10 Hz. The optimization objective was normalized using a weighted Chebyshev paradigm, and then the symbiotic search algorithm was run. After 156 iterations, the hypervolume change rate remained less than 0.3% for 12 consecutive iterations, indicating algorithm convergence. The final Pareto front contained 187 non-dominated solutions.

[0045] The process engineer selected a solution that met the requirements of cost ≤ 11.5 yuan and strength ≥ 350 MPa. The parameter combination was: stamping speed 620 mm / s, die clearance 0.15 mm, bending angle 90.2 degrees, springback compensation 2.1 degrees, heating rate 12 degrees Celsius / min, holding temperature 900 degrees Celsius, holding time 55 minutes, and cooling rate 30 degrees Celsius / min. Virtual trial production predicted: strength 358 MPa, hole spacing deviation ±0.12 mm, Ra = 1.6 micrometers, energy consumption 0.76 kWh, and cost 11.2 yuan.

[0046] Example 2: In another application scenario, this invention is used to optimize the deep drawing process of stainless steel sinks. The product material is SUS304, with a thickness of 0.8 mm. It has a complex shape with multiple deep cavities and narrow grooves, making it prone to cracking and wrinkling defects. The optimization objectives were adjusted to: maximum thinning rate ≤20% (indirect strength indicator), contour error ≤0.5 mm, no orange peel texture on the surface, single-piece energy consumption ≤1.2 kWh, and mold life cost ≤8 yuan / piece. The decision variables were expanded to 12 dimensions, with new variables including blank holder force (5 to 20 kN), lubricant flow rate (50 to 200 ml / min), and intermediate annealing temperature for multiple deep drawing operations (600 to 800 degrees Celsius).

[0047] In the digital twin model, the finite element analysis module employs an explicit dynamics solver to capture transient effects under high-speed stamping, with the mesh size refined to 0.1 mm. The thermodynamics module adds a recrystallization model for the annealing process. The initial population size is set to 150, using a knowledge-based guided strategy: 50 historical stainless steel deep-drawing cases are retrieved, the top 20% of high-performance solutions are extracted as 30% of the initial population, and the remaining 70% are still generated by Latin hypercube sampling.

[0048] During the operation of the symbiotic organism search algorithm, the perturbation amplitude of the parasitic operation was adjusted to 15% to cope with the complex terrain of the high-dimensional space. The convergence condition was set as a maximum of 300 iterations or a hypervolume change rate of less than 0.4% for 15 consecutive iterations. A Pareto solution set of 215 solutions was ultimately obtained. The optimized scheme achieved a maximum thinning rate of 18.7%, a contour thickness of 0.42 mm, energy consumption of 1.15 kWh, and a mold cost of 7.6 yuan.

[0049] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0050] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for optimizing the manufacturing process of hardware parts based on digital twins, characterized in that, Includes the following steps: A high-fidelity virtual mapping model is established to simultaneously simulate the stress and strain distribution during the stamping process, the springback behavior during the bending stage, and the phase transformation and residual stress evolution during the heat treatment process, and the state is updated at a frequency not less than a preset threshold. By setting optimization objectives and decision variables, a high-dimensional nonlinear multi-objective optimization problem is formed. In the solution space of process parameters, an initial population is generated by Latin hypercube sampling strategy. Each individual represents a complete combination of process parameters. The range of values ​​for each parameter is set according to the equipment capacity boundary and the process safety threshold. Each individual in the population is subjected to mutualistic, symbiotic, and parasitic operations in sequence. Mutualistic operations improve fitness through cooperation between two individuals, symbiotic operations benefit one individual while the other remains unaffected, and parasitic operations introduce local perturbations in the individual's neighborhood to enhance global exploration capabilities. Based on the digital twin model, virtual prototype simulation is performed on each individual to obtain the quantitative results of each optimization objective. The non-dominated sorting mechanism is used to screen non-dominated solutions and dynamically maintain a uniformly distributed and comprehensive Pareto front solution set. Once the convergence condition is met or the maximum number of iterations is reached, the optimal compromise solution that meets the actual production constraints is selected from the Pareto front solution set, an executable process parameter configuration file is generated, and it is pushed to the physical manufacturing system for execution. In the multi-objective optimization problem, the strength objective is characterized by tensile strength or yield strength, the dimensional accuracy objective is measured by the absolute value of the maximum deviation of the key feature dimension, the surface quality objective is characterized by the surface roughness Ra value or defect density, the energy consumption objective is calculated by the total electrical energy consumption of a single product, and the manufacturing cost objective includes the comprehensive cost of material loss, tool wear and labor hours. After normalization, each objective is scalarized using the weighted Chebyshev paradigm to eliminate dimensional differences. When generating the initial population, the boundary values ​​of each decision variable are provided by the process parameter constraint library. During initialization, the legality of all individuals is checked, out-of-bounds individuals are removed and resampled until all individuals in the population meet the equipment capability boundary and process safety threshold constraints. The update formula for the mutually beneficial symbiotic operation is: ； and They are two different individuals. The best individual in the current population. A uniformly distributed random number in the interval [0,1). This is the scaling factor; The update formula for symbiotic operations is: ； The random numbers are uniformly distributed within the interval [-1, 1). Parasitic operations in individuals Random replacements are performed on several parameter dimensions, with the disturbance amplitude not exceeding a predetermined proportion of the parameter range.

2. The method for optimizing hardware manufacturing processes based on digital twins according to claim 1, characterized in that, The digital twin model integrates finite element analysis, thermodynamic simulation and kinematic modeling modules. The finite element analysis module uses an implicit dynamic solver to perform millisecond-level dynamic simulation of the plastic deformation and stress concentration areas of the sheet metal during the stamping process. The mesh generation adopts an adaptive refinement strategy, which automatically refines the mesh to a unit size of 0.2 mm in areas with a curvature radius of less than 2 mm. The thermodynamic simulation module is based on the coupling of the Fourier heat conduction equation and the JMAK phase transformation kinetic model to calculate the temperature field evolution, the proportion of austenite to martensite transformation and the distribution of residual stress during the heat treatment stage. The kinematic modeling module reproduces the motion trajectory of the bending machine slider and the opening and closing logic of the mold, and combines the springback compensation algorithm to predict the geometric deviation of the formed part.

3. The method for optimizing hardware manufacturing processes based on digital twins according to claim 1, characterized in that, The maximum capacity of the Pareto front solution set is set to a preset upper limit. When the solution set size is greater than the preset upper limit, the crowding distance calculation mechanism is used to prioritize the retention of solutions in sparsely distributed regions. The crowding distance is determined by calculating the sum of the distances between adjacent individuals after sorting each target dimension.

4. The method for optimizing hardware manufacturing processes based on digital twins according to claim 3, characterized in that, The convergence conditions include the rate of change of the hypervolume index of the Pareto front solution set being less than a preset rate of change threshold for several consecutive generations, or the standard deviation of all individuals in the population on each target being less than a preset threshold; the maximum number of iterations is set to a preset maximum number of iterations.

5. The method for optimizing hardware manufacturing processes based on digital twins according to claim 4, characterized in that, The method also includes performing closed-loop verification of the optimized process parameter scheme in a digital twin model, generating a product quality prediction report through virtual trial production, and if the key indicators fail to meet the standards, local sampling is performed with the current Pareto solution set as the center to trigger a new round of optimization iteration.

6. The method for optimizing hardware manufacturing processes based on digital twins according to claim 5, characterized in that, The method also includes storing historical optimization cases and corresponding actual production results in a process knowledge base. This process knowledge base contains process parameter-performance result pairs, product models, material batches, equipment numbers, and environmental conditions, which are used to guide population initialization or as training data for machine learning models during subsequent optimization processes.

7. The method for optimizing hardware manufacturing processes based on digital twins according to claim 6, characterized in that, The initial population generation process does not require setting crossover rate, mutation rate, or inertia weight. The internal operations rely only on random numbers and the population's own state, making it suitable for parameter-free optimization when switching between different hardware products.

Images