Injection molding process parameter optimization method and system based on hybrid algorithm and model fusion

Abstract

The application relates to the technical field of artificial intelligence, in particular to a plastic injection process parameter optimization method and system based on a hybrid algorithm and model fusion, which comprises the following steps: optimizing the parameter combination of a support vector regression model through a simulated annealing algorithm, constructing a weighted fusion model based on the optimized support vector regression model and a random forest, and constructing a hybrid model of the adaptive selection weighted fusion model and the optimized support vector regression model; constructing a three-target optimization model containing warping deformation, volume shrinkage and production energy consumption; taking the hybrid model as a target value evaluation tool, and searching for a Pareto optimal solution set in a process parameter space by adopting a multi-objective genetic algorithm; and adopting a gradient descent method to locally correct key process parameters until the deviation falls within a preset threshold, so that the optimized process parameters are obtained; the application can reduce the product defect rate and reduce the production energy consumption.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to an injection molding process parameter optimization method and system based on hybrid algorithm and model fusion. BACKGROUND

[0002] In the injection molding process, product defect and energy consumption control are long-term pain points in the industry. Traditional parameter optimization relies on experience or single algorithm, and has problems such as weak global search ability and difficulty in balancing multiple objectives. With the growth of intelligent manufacturing demand, it is crucial to build an efficient and accurate process optimization system. SUMMARY

[0003] To solve the above problems, the present application provides an injection molding process parameter optimization method and system based on hybrid algorithm and model fusion, which combines multiple algorithms and models, optimizes the parameter combination of the support vector regression model through the simulated annealing algorithm, and realizes multi-objective optimization through the multi-objective genetic algorithm, reduces the product defect rate, reduces the production energy consumption, and realizes the multi-dimensional intelligent optimization of the injection molding process.

[0004] In order to achieve the above purpose, the present application provides the following technical scheme:

[0005] On the one hand, the present application provides an injection molding process parameter optimization method based on hybrid algorithm and model fusion, which comprises the following steps:

[0006] S100, obtaining a sample data set, training a support vector regression model through the sample data set, optimizing the parameter combination of the support vector regression model through the simulated annealing algorithm, and obtaining an optimized support vector regression model; the sample data set includes multiple process parameters and corresponding target variables, and the parameter combination includes a penalty factor and a kernel parameter of the support vector regression model;

[0007] S200, based on the optimized support vector regression model and the random forest, a weighted fusion model is constructed, the product structure complexity and the fluctuation range of the process parameters are combined, and a hybrid model of the adaptive selection weighted fusion model and the optimized support vector regression model is constructed;

[0008] S300, a three-objective optimization model containing warping deformation, volume shrinkage rate and production energy consumption is constructed, the hybrid model is used as an objective value evaluation tool, and a multi-objective genetic algorithm is used to search for a Pareto optimal solution set in the process parameter space; the Pareto optimal solution set contains multiple groups of process parameters, and each group of process parameters corresponds to a group of non-dominated predicted target values;

[0009] S400, collect the actual target value in the production process, and compare it with the corresponding predicted target value, when the deviation exceeds the preset threshold, use the gradient descent method to locally correct the key process parameters until the deviation falls within the preset threshold to obtain the optimized process parameters.

[0010] Optionally, in the S100, a sample data set is obtained, the support vector regression model is trained through the sample data set, the parameter combination of the support vector regression model is optimized through the simulated annealing algorithm, and an optimized support vector regression model is obtained, including:

[0011] S110, initialize the search range of the penalty factor and the kernel parameter, and set the initial parameter combination; the parameter combination includes the penalty factor and the kernel parameter, the penalty factor is used to control the tolerance of the support vector regression model to the mean square error, and the kernel parameter is used to map the nonlinear relationship between the process parameters and the target variable;

[0012] S120, obtain a sample data set, take the prediction mean square error of the support vector regression model on the sample data set as an optimization objective function, gradually reduce the temperature at a set cooling rate, generate a new parameter combination at each temperature through random disturbance, if the new parameter combination has a lower mean square error, accept the new parameter combination, if the error is higher, accept the new parameter combination according to the Metropolis criterion according to the probability, until the termination condition is reached, stop iteration and output the current optimal parameter combination, and substitute it into the support vector regression model to obtain an optimized support vector regression model; the termination condition is that the objective function value does not appear to be improved for a continuous preset number of iterations, reaches a preset number of iterations, or the temperature is reduced to a termination threshold.

[0013] Optionally, in the S200, a weighted fusion model is constructed based on the optimized support vector regression model and the random forest, including:

[0014] Based on the optimized support vector regression model, the interaction between the process parameters is analyzed by introducing the random forest, the optimized support vector regression model and the random forest are dynamically fused with a weight coefficient, and a weighted fusion model is constructed; when the prediction error of the optimized support vector regression model is greater than the random forest, the weight proportion of the random forest is improved, and vice versa.

[0015] Optionally, in the S200, a hybrid model of the adaptive selection of the weighted fusion model and the optimized support vector regression model is constructed by combining the product structure complexity and the fluctuation range of the process parameters, including:

[0016] When the complexity of the product structure exceeds the complexity threshold or the fluctuation range of the process parameters exceeds the fluctuation threshold, the sample dataset is clustered according to the product complexity and parameter range, and each cluster is trained separately. The support vector regression sub-model obtained by the corresponding cluster training is then called.

[0017] When the complexity of the product structure does not exceed the complexity threshold and the fluctuation range of the process parameters does not exceed the fluctuation threshold, the weighted fusion model is invoked.

[0018] Optionally, in step S300, a three-objective optimization model is constructed, including warpage deformation, volume shrinkage rate, and production energy consumption. A hybrid model is used as the objective value evaluation tool, and a multi-objective genetic algorithm is employed to search for the Pareto optimal solution set in the process parameter space, including:

[0019] S310 uses the objective function of minimizing warpage, volume shrinkage, and production energy consumption as a three-objective optimization model, introducing mold temperature constraints, holding pressure-injection time ratio constraints, and injection speed constraints. Among them, the warpage is obtained by fitting the relationship between process parameters and measured warpage values ​​through an optimized support vector regression model; the volume shrinkage rate is predicted based on the correlation between melt density change and process parameters, combined with the optimized support vector regression model; and the production energy consumption is calculated by combining equipment power and running time through an optimized support vector regression model. The mold temperature constraint is limited between the lower limit of material flowability and the upper limit of mold heat resistance; the holding pressure-injection time ratio constraint is set between the lower limit of avoiding insufficient holding pressure and the upper limit of avoiding excessive holding pressure leading to energy waste; and the injection speed constraint is set between the lower limit of preventing slow filling and cold material and the upper limit of preventing flash or bubbles.

[0020] S320, Randomly generate an initial population of process parameters, wherein the individuals in the population of process parameters are combinations of parameters;

[0021] S330 uses a multi-objective genetic algorithm to perform selection, crossover, and mutation operations on the initial process parameter population. It uses a hybrid model to calculate the target value corresponding to each individual, selects individuals with better target values ​​based on non-dominated sorting, and combines a reference point mechanism to maintain a uniform distribution of the solution set in the target space. After multiple generations of iteration, it outputs a Pareto optimal solution set containing multiple sets of non-dominated target values.

[0022] Optionally, in step S400, the gradient descent method is used to locally correct key process parameters until the deviation falls back to within a preset threshold, resulting in optimized process parameters, including:

[0023] S410 uses minimizing the deviation between the actual target value and the predicted target value as the objective function to determine the key process parameters related to the production process;

[0024] S420, small step adjustment is made to the key process parameters, and after each round of adjustment, the correction effect is quickly evaluated through the fusion prediction model to calculate the target function value corresponding to the key process parameters at present; the step range of the small step adjustment is preset according to the type of the process parameters;

[0025] S430, when the target function value is lower than the last round, continue to adjust the key process parameters in the current direction; when the target function value increases, adjust the key process parameters in the opposite direction or reduce the adjustment step; until the deviation between the actual target value and the predicted target value falls within the preset threshold, the key process parameters output at this time are taken as the optimized process parameters.

[0026] Optionally, the method further comprises:

[0027] Optionally, the method further comprises:

[0028] Optionally, the method further comprises:

[0029] at least one processor;

[0030] at least one memory for storing at least one program;

[0031] When the at least one program is executed by the at least one processor, the at least one processor implements the above method.

[0032] In another aspect, an embodiment of the present application provides a computer readable storage medium, which stores a program executable by a processor, and the program executable by the processor is used to execute the above method when executed by the processor.

[0033] The present application has the beneficial effects that: the present application discloses an injection molding process parameter optimization method and system based on hybrid algorithm and model fusion, the present application optimizes the parameter combination of the support vector regression model through the simulated annealing algorithm, effectively improves the fitting ability of the model to the complex nonlinear relationship, combines the random forest to construct the weighted fusion model, dynamically adapts to the prediction demand of different product structures, significantly enhances the generalization performance of the model, adopts the multi-objective genetic algorithm to realize the multi-objective collaborative optimization of the process parameters, and achieves a better balance between warping deformation, volume shrinkage rate and production energy consumption; the gradient descent method is used for local correction of the key parameters, and the precision of the process parameters is further improved. Experimental results show that, compared with the traditional optimization method, the method can reduce the product defect rate and reduce the production energy consumption, and provides an effective solution for the intelligent upgrading of injection molding process. BRIEF DESCRIPTION OF DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings needed in the embodiments will be briefly introduced as follows. Obviously, the accompanying drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without any creative labor.

[0035] Figure 1 is a flow diagram of a mixed algorithm and model fusion-based injection molding process parameter optimization method according to an embodiment of the present application;

[0036] Figure 2 is a structural diagram of a mixed algorithm and model fusion-based injection molding process parameter optimization system according to an embodiment of the present application. DETAILED DESCRIPTION

[0037] The disclosed concept, specific structure and generated technical effects will be described clearly and completely in the following embodiments and drawings, so as to fully understand the purpose, scheme and effect of the present application. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0038] In the injection molding process, product defects and energy consumption control are long-term pain points in the industry. Traditional parameter optimization relies on experience or single algorithm, and has problems such as weak global search ability and difficult multi-objective balance. With the growth of intelligent manufacturing demand, it is crucial to build an efficient and accurate process optimization system. The present application breaks through the limitations of existing research, integrates multiple algorithms and models, optimizes SVR parameters through simulated annealing algorithm, realizes multi-objective optimization through NSGA-III, forms a SA-SVR-NSGA-III hybrid architecture, and realizes multi-dimensional intelligent optimization of injection molding process.

[0039] In view of the defects of the prior art, the present application proposes an innovative multi-modal knowledge base construction method for physical and chemical experiment operation analysis. In view of the limitations of traditional injection molding process parameter optimization methods, an intelligent injection molding process parameter optimization method is proposed, which integrates improved algorithms and multiple models. Simulated annealing algorithm (simulated annealing algorithm) is used to optimize support vector regression (SVR) parameters, and a SA-SVR-NSGA-III hybrid framework is constructed to realize the collaborative optimization of warping deformation, volume shrinkage rate and production energy consumption. The multi-model adaptive fusion and online feedback mechanism is introduced to improve the prediction accuracy and dynamic optimization ability in complex scenarios. Experimental results show that compared with traditional methods, the present method significantly improves the optimization efficiency, multi-objective balance and industrial adaptability, providing a new path for intelligent optimization of injection molding process.

[0040] REFERENCE Figure 1 For example, Figure 1The application provides an injection molding process parameter optimization method based on a hybrid algorithm and model fusion.

[0041] In S100, a sample data set is acquired, a support vector regression model is trained through the sample data set, a parameter combination of the support vector regression model is optimized through a simulated annealing algorithm, and an optimized support vector regression model is obtained; the sample data set comprises a plurality of process parameters and corresponding target variables, and the parameter combination comprises a penalty factor and a kernel parameter of the support vector regression model;

[0042] In S200, a weighted fusion model is constructed based on the optimized support vector regression model and a random forest, a self-adaptive selection hybrid model of the weighted fusion model and the optimized support vector regression model is constructed in combination with product structure complexity and a fluctuation range of process parameters;

[0043] In S300, a three-target optimization model comprising a warping deformation amount, a volume shrinkage rate and production energy consumption is constructed, a hybrid model is used as a target value evaluation tool, and a multi-objective genetic algorithm is used to search for a Pareto optimal solution set in a process parameter space; the Pareto optimal solution set comprises a plurality of groups of process parameters, and each group of process parameters corresponds to a group of non-dominated predicted target values;

[0044] In S400, actual target values in a production process are collected and compared with corresponding predicted target values, a gradient descent method is used to locally correct key process parameters when a deviation exceeds a preset threshold value, until the deviation falls within the preset threshold value, and optimized process parameters are obtained.

[0045] In the embodiment, the optimization process of the support vector regression model parameters through the simulated annealing algorithm realizes effective exploration of an initial parameter combination. Specifically, the search ranges of the penalty factor and the kernel parameter set in the initialization stage, in combination with the set cooling rate and the random disturbance mechanism, ensure the global coverage ability of the parameter space. In the iteration process, the introduction of the Metropolis criterion makes the algorithm reasonable in probability when accepting a poor solution, and avoids falling into a local optimum.

[0046] In the construction process of the weighted fusion model, the dynamic weight adjustment mechanism realizes the complementary advantages of the optimized support vector regression model and the random forest. When the fluctuation range of the process parameters exceeds a preset threshold value, the system is automatically switched to a cluster training mode, a sub-model library for a specific working condition is generated through cluster analysis of product complexity and parameter intervals.

[0047] In the construction of the three-target optimization model, the warping deformation, the volume shrinkage and the production energy consumption are converted into quantifiable objective functions, and a parameter mapping relationship is established by combining the physical mechanism of injection molding, thereby forming a complete optimization system. The non-dominated sorting and reference point mechanism adopted by the multi-objective genetic algorithm ensures the uniformity and diversity of the Pareto front solution set.

[0048] In some embodiments, the S100, a sample data set is obtained, a support vector regression model is trained by the sample data set, a parameter combination of the support vector regression model is optimized by a simulated annealing algorithm, and an optimized support vector regression model is obtained, including:

[0049] S110, the search range of the penalty factor and the kernel parameter is initialized, and the initial parameter combination is set; the parameter combination includes the penalty factor and the kernel parameter, the penalty factor is used to control the tolerance of the support vector regression model to the mean square error, and the kernel parameter is used to map the non-linear relationship between the process parameters and the target variables;

[0050] S120, a sample data set is obtained, the prediction mean square error of the support vector regression model on the sample data set is taken as the optimization objective function, the temperature is gradually reduced at a set cooling rate, a new parameter combination is generated at each temperature by random disturbance, if the new parameter combination has a lower mean square error, the new parameter combination is accepted, if the error is higher, the new parameter combination is accepted according to the Metropolis criterion with a probability, and the iteration is stopped and the current optimal parameter combination is output when the termination condition is reached, and the optimized support vector regression model is obtained by substituting the support vector regression model into the support vector regression model; the termination condition is that the target function value does not appear to be improved for a continuous preset number of iterations, the preset number of iterations is reached, or the temperature is reduced to a termination threshold.

[0051] In this embodiment, through the fine search of the penalty factor and the kernel parameter, combined with the random disturbance characteristics of the simulated annealing algorithm, the defect that the traditional parameter optimization method is easy to fall into local optimum is effectively overcome. In the specific implementation, the parameter search range set in the initialization stage covers the commonly used numerical range in engineering practice, and the cooling rate parameter is verified by pre-experiment to ensure the balance between the convergence of the algorithm and the calculation efficiency. In the temperature decreasing process, the random disturbance mechanism generates candidate parameter combinations in Gaussian distribution, which not only ensures the exploration breadth of the parameter space, but also maintains the progressive stability of the algorithm through the probability acceptance mechanism of the Metropolis criterion. When the target function value does not appear to be improved for a continuous preset number of iterations, the system automatically triggers the termination condition to avoid the consumption of invalid calculation resources; when the temperature is reduced to the termination threshold, the current optimal parameter combination is output. Especially, for the common non-linear mapping problem in the injection molding process, the kernel parameter realizes the feature space conversion through the radial basis function, and cooperates with the adaptive adjustment of the penalty factor, so that the model shows significant advantages in the warping prediction of complex curved surface products.

[0052] In some embodiments, the S200, based on the optimization of the support vector regression model and the random forest, a weighted fusion model is constructed, comprising:

[0053] Based on the optimization of the support vector regression model, the interaction between the process parameters is analyzed by the random forest, the optimization of the support vector regression model and the random forest are dynamically weighted and fused to construct a weighted fusion model; when the prediction error of the optimization of the support vector regression model is greater than the random forest, the weight proportion of the random forest is increased, and vice versa.

[0054] In this embodiment, through the dynamic weight adjustment mechanism, the advantages of the optimization of the support vector regression model and the random forest are complementary. Specifically, the system monitors the prediction error of the two models in real time, and when the optimization of the support vector regression model has prediction deviation under certain working conditions, the weight distribution module automatically increases the contribution of the random forest; otherwise, when the random forest is insufficient in analyzing complex interactions, the weight of the optimization of the support vector regression model is increased. The performance degradation problem of a single model under local working conditions is effectively solved. In the model construction stage, the initial weight benchmark value is determined by cross-validation technology, and the real-time monitoring of the process parameter fluctuation range is combined. When the parameter standard deviation exceeds the threshold value, the system triggers the cluster training mechanism, and divides the data set into multiple subsets according to the product complexity, and trains a sub-model library for specific working conditions. In the prediction stage, the model selector automatically matches the optimal sub-model according to the product characteristics of the current production batch, ensuring the balance between prediction accuracy and calculation efficiency. The fusion model can improve the prediction accuracy compared with a single model when dealing with high-dimensional nonlinear data, especially in the warping prediction of thin-walled products. Through the dynamic tracking of the melt flow front by the random forest, the prediction delay of the optimization of the support vector regression model under time-varying working conditions is effectively compensated.

[0055] In some embodiments, the S200, combining the product structure complexity and the process parameter fluctuation range, a hybrid model of the adaptive selection of the weighted fusion model and the optimization of the support vector regression model is constructed, comprising:

[0056] When the complexity of the product structure exceeds the complexity threshold or the fluctuation range of the process parameters exceeds the fluctuation threshold, the sample data set is clustered according to the product complexity and the parameter interval, and each cluster is trained separately, and the support vector regression sub-model obtained by the corresponding cluster cluster training is called;

[0057] When the complexity of the product structure does not exceed the complexity threshold and the fluctuation range of the process parameters does not exceed the fluctuation threshold, the weighted fusion model is called.

[0058] In this embodiment, by introducing the dual judgment mechanism of product structure complexity and process parameter fluctuation range, the dynamic adaptation of the model selection strategy is realized.

[0059] In some embodiments, the S300, a three-objective optimization model is constructed to include the amount of warpage deformation, the volume shrinkage rate, and the production energy consumption, a hybrid model is used as a target value evaluation tool, and a multi-objective genetic algorithm is used to search for a Pareto optimal solution set in the process parameter space, including:

[0060] In S310, a three-objective optimization model is used to minimize the warpage deformation, the volume shrinkage rate, and the production energy consumption, mold temperature constraints, holding-injection time ratio constraints, and injection speed constraints are introduced, the amount of warpage deformation is obtained by optimizing a support vector regression model to fit the relationship between the process parameters and the measured warpage value, the volume shrinkage rate is predicted based on the correlation between the melt density change and the process parameters and by combining an optimized support vector regression model, and the production energy consumption is calculated by combining an optimized support vector regression model with the equipment power and the operating time; the mold temperature constraint is limited between the lower limit of material flowability and the upper limit of mold heat resistance, the holding-injection time ratio constraint is set between the lower limit to avoid insufficient holding pressure and the upper limit to avoid excessive holding pressure leading to energy waste, and the injection speed constraint is set between the lower limit to prevent slow filling from producing cold material and the upper limit to prevent the generation of flash or bubbles;

[0061] In S320, an initial process parameter population is randomly generated, and each individual in the process parameter population is a parameter combination.

[0062] In S330, a multi-objective genetic algorithm is used to perform selection, crossover, and mutation operations on the initial process parameter population, a hybrid model is used to calculate the target values corresponding to each individual, individuals with better target values are selected based on non-dominated sorting, a reference point mechanism is used to maintain the uniform distribution of the solution set in the target space, and after multiple iterations, a Pareto optimal solution set containing multiple groups of non-dominated target values is output.

[0063] In this embodiment, by constructing a complete three-objective optimization system, the warpage deformation, the volume shrinkage rate, and the production energy consumption are collaboratively controlled. In the optimization target function setting, not only the product quality indicators (warpage deformation, volume shrinkage rate) are considered, but also the economic indicator of production energy consumption is included, forming a multi-dimensional optimization framework. The setting of the constraint conditions takes into account the process feasibility (mold temperature range, holding-injection time ratio, injection speed interval) and the equipment protection requirements, ensuring that the optimization results can be directly applied to actual production. The multi-objective genetic algorithm effectively solves the lack of uniformity of the Pareto frontier solution set in traditional methods through non-dominated sorting and reference point mechanism, especially when dealing with complex constraint conditions, the diversity and representativeness of the solution set are still maintained.

[0064] In some embodiments, the S400, a gradient descent method is used to locally correct the key process parameters until the deviation falls within a preset threshold, and the optimized process parameters are obtained, including:

[0065] S410, determining the key process parameters related to the production process with the minimum deviation between the actual target value and the predicted target value as the objective function;

[0066] S420, adjusting the key process parameters in small steps, and after each round of adjustment, quickly evaluating the correction effect by fusing the prediction model to calculate the target function value corresponding to the key process parameters at present; the step range of the small step adjustment is preset according to the type of process parameters;

[0067] S430, when the target function value is lower than the last round, continue to adjust the key process parameters in the current direction; when the target function value increases, adjust the key process parameters in the opposite direction or reduce the adjustment step; until the deviation between the actual target value and the predicted target value falls within the preset threshold, the key process parameters output at this time are taken as the optimized process parameters.

[0068] In this embodiment, the local correction mechanism realized by the gradient descent method effectively improves the dynamic adaptability of the production process.

[0069] In some embodiments, the method further comprises:

[0070] Taking a preset production batch as a sliding window, the actual process parameters and the actual target value in the window are periodically taken as new training samples to retrain the optimized support vector regression model and the weighted fusion model.

[0071] In this embodiment, the sliding window mechanism is used to realize the real-time tracking of the dynamic changes of the production data by the model, and the incremental learning method is used to ensure the synchronous update of the model parameters and the process conditions. The prediction error of the model updated by the sliding window is lower than that of the static model in continuous production, which effectively avoids the performance degradation caused by the deviation of data distribution.

[0072] The following is a specific embodiment provided by the application:

[0073] (1) Algorithm architecture innovation: SA-SVR-NSGA-III hybrid framework

[0074] Simulated annealing algorithm (Simulated Annealing, SA) is used to optimize the parameters of the support vector regression (Support Vector Regression., SVR) model:

[0075] The simulated annealing algorithm is used to replace the traditional particle swarm optimization (Particle Swarm Optimization, PSO) to optimize the core parameters of the support vector regression model, i.e. the penalty factor C and the kernel parameter In the injection molding process parameter optimization scenario, the traditional optimization algorithm is easy to fall into local optimum and difficult to accurately optimize in the complex high-dimensional process parameter space. The simulated annealing algorithm is based on the heuristic of physical annealing process, and by introducing the probability jump mechanism, it allows to accept inferior solution with a certain probability in the search process, so as to effectively jump out of the local optimum trap.

[0076] The core principle of the simulated annealing algorithm is based on the Metropolis criterion, and the probability formula for accepting new solutions is:

[0077] ;

[0078] Wherein, is the difference between the objective function of the new solution and the current solution, and T is the current temperature. With the progress of the annealing process, the temperature T gradually decreases, and the probability of accepting inferior solutions also decreases, and finally converges to the global optimal solution.

[0079] In the SVR parameter optimization, the penalty factor C and the kernel parameter are taken as the solution space variables, and the objective function is set to minimize the mean square error (MSE) of the support vector regression model:

[0080] ;

[0081] Wherein, is the number of samples, is the actual value, is the predicted value of the support vector regression model. The simulated annealing algorithm adjusts the penalty factor C and the kernel parameter continuously, judges whether to accept the new parameter combination according to the Metropolis criterion, and gradually optimizes the performance of the support vector regression model.

[0082] In the high-dimensional process parameter space of the present application, in addition to the conventional mold temperature, melt temperature, holding pressure and other parameters, injection speed, cooling time and other key variables are also added, which further increases the complexity and difficulty of parameter optimization. The simulated annealing algorithm can realize global search in this complex space due to its unique probability jump characteristics, and can more accurately optimize the penalty factor C and the kernel parameter of the support vector regression model. The penalty factor C controls the tolerance of the support vector regression model to error, and the larger the C value, the more strict the support vector regression model is to the error; the kernel parameter determines the complexity and generalization ability of the support vector regression model, and a suitable kernel parameter value can make the support vector regression model better fit the nonlinear relationship between the process parameters and the target variables (warpage deformation, energy consumption, etc.).

[0083] Through a large number of experimental comparative analysis, it is found that the support vector regression model optimized by simulated annealing algorithm has a significant improvement in performance. Specifically, its mean squared error (MSE) convergence speed is improved by 20% compared with traditional optimization methods, and the final MSE value is lower. This means that the support vector regression model optimized by simulated annealing algorithm can reach a stable state in a shorter time, reducing the time cost of model training, and also providing a more stable and reliable foundation for the subsequent accurate prediction of injection molding process parameters; through this support vector regression model, the warpage deformation, volume shrinkage rate and production energy consumption corresponding to any process parameter can be quickly calculated, providing efficient target value evaluation support for NSGA-III multi-objective optimization, greatly improving the efficiency of the overall optimization process.

[0084] NSGA-III multi-objective collaborative optimization:

[0085] A three-objective optimization model of warpage deformation, volume shrinkage rate and production energy consumption is constructed, and the support vector regression model optimized by simulated annealing algorithm is used as the target value evaluation tool. The multi-objective genetic algorithm (NSGA-III) is used to output the Pareto optimal solution set. By introducing process feasibility constraints (such as mold temperature upper limit, holding-injection time ratio range), the optimization results are ensured to be consistent with industrial reality, realizing deep adaptation of technical solutions and production conditions.

[0086] In the injection molding process, warpage deformation, volume shrinkage rate and production energy consumption are interrelated and interdependent key indicators; for example, reducing energy consumption may lead to insufficient melt filling, thereby increasing warpage deformation; while excessively increasing the holding pressure can reduce shrinkage, but will increase energy consumption and mold wear. To achieve collaborative optimization of these three objectives, first, based on the physical mechanism of injection molding and numerical simulation results, a mapping relationship between process parameters such as mold temperature , melt temperature , injection pressure , holding pressure , injection time , holding time , injection speed , cooling time and target variables is established, which is realized by the support vector regression model optimized by simulated annealing algorithm for efficient calculation.

[0087] The mathematical expressions of each target variable are as follows:

[0088] Warpage deformation : The relationship between process parameters and measured warpage value is fitted by the support vector regression model, and the expression is , where is the random error term, with a range less than 0.05 mm.

[0089] Volume shrinkage : Based on the correlation between melt density change and process parameters, combined with support vector regression model prediction, the formula is , is the initial volume of the melt, is the SVR sub-model for predicting the volume of the product after molding.

[0090] Production energy consumption : Comprehensive equipment power and running time, calculated by support vector regression model, expression is , is the fixed energy consumption of auxiliary equipment (such as cooling system, raw material conveying system), which is a constant determined by experiment.

[0091] The optimization objective function of NSGA-III is set to minimize three target variables, namely:

[0092] ;

[0093] To avoid the optimization results deviating from the actual production conditions, multiple process feasibility constraints are introduced:

[0094] Mold temperature constraint: where (lower limit of material flowability, such as 40℃), (upper limit of mold heat resistance, such as 90℃) to prevent mold deformation or material degradation.

[0095] Pressure-injection time ratio constraint: (avoid insufficient pressure holding), (avoid long pressure holding leading to energy waste).

[0096] Injection speed constraint: (prevent slow filling from producing cold material, such as 50 mm / s), (prevent producing flash or bubbles, such as 300 mm / s).

[0097] NSGA-III searches efficiently in the process parameter space by simulating the selection, crossover, and mutation operations of biological evolution: first, randomly generate an initial population of process parameters, and quickly calculate the three objective values of each individual by the support vector regression model optimized by the simulated annealing algorithm; then, based on the "non-dominated sorting" mechanism, filter out individuals with better objective values, and combine the "reference point mechanism" to ensure the uniform distribution of the solution set in the objective space, avoiding the omission of key trade-off schemes; after multiple iterations, the final output is the Pareto optimal solution set; this Pareto optimal solution set contains multiple sets of process parameters, each corresponding to a set of non-dominated objective values (such as "warpage 3.1 mm + shrinkage 1.7% + energy consumption 122 kW·h" "warpage 3.4 mm + shrinkage 1.5% + energy consumption 108 kW·h"), which can be flexibly selected according to production priorities (such as "energy saving first" "quality first").

[0098] (II) Upgrade the prediction model based on a multi-model adaptive fusion mechanism:

[0099] Weighted fusion model of SVR-RF:

[0100] Build a weighted fusion model of support vector regression (SVR) and random forest (RF): based on the support vector regression model optimized by the simulated annealing algorithm, retain its strong fitting ability for the non-linear relationship between process parameters; introduce the interaction between random forest parameters (such as "the synergistic effect of melt temperature and cooling time on shrinkage"); fuse the outputs of the two through dynamic weight coefficients (based on real-time prediction error adjustment); when the prediction error of the support vector regression model is greater than that of the random forest, automatically increase the weight proportion of the random forest, and vice versa, to ensure that the weighted fusion model maintains high prediction accuracy in different parameter intervals.

[0101] Experimental verification shows that the prediction mean square error of the fusion model is reduced to 0.068 (the traditional single support vector regression model is 0.0827), and the accuracy is improved by 17.8%. Specifically, in the interval where the melt temperature fluctuates greatly (such as 200-260°C), the analytical ability of random forest for parameter interaction effectively makes up for the deviation of support vector regression model; while in the interval where the mold temperature is stable (such as 50-70°C), the nonlinear fitting advantage of the support vector regression model is more obvious, and the dynamic weight mechanism can adaptively balance the contributions of the two, avoiding the prediction failure of a single model in complex scenarios.

[0102] Scene adaptive selection strategy:

[0103] Based on the parameter distribution characteristics of the input (such as the melt temperature interval, product complexity clustering), design a model automatic switching mechanism to further improve the prediction robustness. The specific rules are as follows:

[0104] Complexity scenario: When the product structure is complex (e.g., the density of the automobile grid hole is >5 / cm²) or the process parameter fluctuation range is large (e.g., the injection speed change is >100 mm / s), trigger the "cluster group training" SVR sub-model. By clustering historical samples according to "product complexity + parameter interval", train an SVR sub-model for each cluster (e.g., "high-density hole + high injection speed" sub-model, "low-density hole + low injection speed" sub-model), avoiding sample cross interference in complex scenarios.

[0105] Conventional scenario: When the product structure is simple (e.g., flat parts, hole density <2 / cm²) and the process parameter fluctuation is small (each parameter change range <50%), automatically call the weighted fusion model of SVR-RF to reduce the calculation cost while ensuring accuracy.

[0106] (Three) Dynamic optimization closed loop: online feedback and drift compensation

[0107] Real-time feedback correction module:

[0108] Integrate industrial sensors (such as laser displacement sensors to detect warping values, power sensors to detect energy consumption), and collect actual target values in the production process in real time. Compare the actual value with the predicted value output by the SA-SVR-NSGA-III optimization. When the deviation exceeds 5% (e.g., predicted warping 3.2mm, actual warping 3.4mm), automatically start the gradient descent method to locally correct the key process parameters (e.g., holding pressure, cooling time): "minimize the deviation between actual value and predicted value" as the target, adjust the key process parameters in small steps (e.g., holding pressure ±2MPa, cooling time ±0.5s), and quickly evaluate the correction effect through the fusion prediction model until the deviation falls within 5%.

[0109] In actual production testing, this mechanism improves the dynamic adaptation capability of process parameters by 30%; for example, when the device heating system ages, causing the actual melt temperature to be 5℃ lower than the set value, the feedback module can adjust the holding time from 5s to 5.3s within 2 production batches, offsetting the increased warping caused by temperature deviation, effectively dealing with device fluctuations and environmental interference (e.g., workshop temperature changes, raw material batch differences).

[0110] Parameter drift compensation algorithm:

[0111] For parameter drift caused by long-term operation of the device (e.g., mold wear causing actual locking force to decrease, heating tube aging causing heating efficiency to decrease), a sliding window sample update strategy is designed: take 100 production batches as a sliding window, automatically exclude old samples outside the window (e.g., 3 months ago device status data), add "actual process parameters + actual target values" within the window as training samples, and periodically (every 100 batches) retrain the SA-SVR model and SVR-RF fusion model.

[0112] The algorithm ensures that the model always makes predictions based on the latest device state, avoiding optimization failure caused by device drift. Compared with the traditional fixed sample model (accuracy reduced to 75%), it is improved by 15 percentage points, significantly enhancing the long-term industrial adaptability of the optimization method.

[0113] Reference Figure 2 The embodiment of the present application also provides an injection molding process parameter optimization system based on a hybrid algorithm and model fusion, comprising:

[0114] At least one processor;

[0115] At least one memory for storing at least one program;

[0116] When the at least one program is executed by the at least one processor, the at least one processor implements the above method.

[0117] The contents in the above method embodiments are all applicable to the present embodiment, the function specifically implemented by the present embodiment is the same as that of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments, which will not be described here.

[0118] In addition, an embodiment of the present application also provides a computer readable storage medium, which stores computer executable instructions, and the computer executable instructions are used for executing the above method.

[0119] It is worth noting that since the computer readable storage medium of the embodiment of the present application can execute the method of any of the above embodiments, the specific implementation and technical effects of the computer readable storage medium of the embodiment of the present application can be referred to the specific implementation and technical effects of the method of any of the above embodiments.

[0120] Although the description of the present disclosure has been quite detailed and particularly described with respect to several described embodiments, it is not intended to be limited to any of these details or embodiments or any special embodiment, but should be considered to effectively cover the intended scope of the present disclosure by referring to the appended claims, taking into account the prior art to provide a broad possible interpretation of these claims. In addition, the present disclosure is described above in embodiments that the inventors can foresee, the purpose of which is to provide a useful description, and non-essential modifications to the present disclosure that have not yet been foreseen can still represent equivalent modifications of the present disclosure.

Claims

1. A method for optimizing injection molding process parameters based on hybrid algorithm and model fusion, characterized in that, The method includes the following steps: S100, Obtain a sample dataset, train the support vector regression model using the sample dataset, and optimize the parameter combination of the support vector regression model using the simulated annealing algorithm to obtain an optimized support vector regression model; the sample dataset includes multiple process parameters and corresponding target variables, and the parameter combination includes the penalty factor and kernel parameters of the support vector regression model; S200 is a hybrid model that combines an optimized support vector regression model and a random forest to construct a weighted fusion model. It also incorporates the product structure complexity and the fluctuation range of process parameters to construct an adaptive selection weighted fusion model and an optimized support vector regression model. S300 constructs a three-objective optimization model that includes warping deformation, volume shrinkage rate, and production energy consumption. A hybrid model is used as the objective value evaluation tool, and a multi-objective genetic algorithm is used to search for the Pareto optimal solution set in the process parameter space. The Pareto optimal solution set contains multiple sets of process parameters, and each set of process parameters corresponds to a set of non-dominated predicted objective values. S400 collects the actual target value during the production process and compares it with the corresponding predicted target value. When the deviation exceeds the preset threshold, the gradient descent method is used to locally correct the key process parameters until the deviation falls back to within the preset threshold, and the optimized process parameters are obtained. The S300 includes: S310 uses a three-objective optimization model that minimizes warpage, volume shrinkage, and production energy consumption. It introduces constraints on mold temperature, holding pressure-injection time ratio, and injection speed. Warpage is obtained by fitting the relationship between process parameters and measured warpage values ​​using an optimized support vector regression model. Volume shrinkage is predicted based on the correlation between melt density changes and process parameters, combined with an optimized support vector regression model. Production energy consumption is calculated using an optimized support vector regression model combined with equipment power and operating time. The mold temperature constraint is limited to the lower limit of material flowability and the upper limit of mold heat resistance. The holding pressure-injection time ratio constraint is set between the lower limit of avoiding insufficient holding pressure and the upper limit of avoiding excessive holding pressure leading to energy waste. The injection speed constraint is set between the lower limit of preventing slow filling and cold material and the upper limit of preventing flash or bubbles. S320, Randomly generate an initial population of process parameters, wherein the individuals in the population of process parameters are combinations of parameters; S330 uses a multi-objective genetic algorithm to perform selection, crossover, and mutation operations on the initial process parameter population. It uses a hybrid model to calculate the target value corresponding to each individual, selects individuals with better target values ​​based on non-dominated sorting, and combines a reference point mechanism to maintain a uniform distribution of the solution set in the target space. After multiple generations of iteration, it outputs a Pareto optimal solution set containing multiple sets of non-dominated target values.

2. The method according to claim 1, characterized in that, In step S100, a sample dataset is acquired, a support vector regression model is trained using the sample dataset, and the parameter combination of the support vector regression model is optimized using a simulated annealing algorithm to obtain an optimized support vector regression model, including: S110, Initialize the search range of the penalty factor and kernel parameter, and set the initial parameter combination; the parameter combination includes the penalty factor and kernel parameter, the penalty factor is used to control the tolerance of the support vector regression model to the mean square error, and the kernel parameter is used to map the nonlinear relationship between process parameters and target variables; S120: Obtain a sample dataset. Using the mean squared error of the support vector regression model's prediction on the sample dataset as the optimization objective function, the temperature is successively reduced at a set cooling rate. At each temperature, a new parameter combination is generated through random perturbation. If the mean squared error of the new parameter combination is lower, the new parameter combination is accepted. If the error is higher, the new parameter combination is accepted according to the Metropolis criterion with probability. The iteration stops when the termination condition is reached, and the current optimal parameter combination is output. This is then substituted into the support vector regression model to obtain the optimized support vector regression model. The termination condition is that the objective function value does not improve within a preset number of consecutive iterations, the preset number of iterations is reached, or the temperature drops to a termination threshold.

3. The method according to claim 1, characterized in that, In S200, a weighted fusion model is constructed based on an optimized support vector regression model and a random forest, including: Based on the optimized support vector regression model, the interaction between the process parameters of random forest is introduced to dynamically fuse the weight coefficients of the optimized support vector regression model and random forest, and construct a weighted fusion model. When the prediction error of the optimized support vector regression model is greater than that of the random forest, the weight ratio of the random forest is increased, and vice versa.

4. The method according to claim 1, characterized in that, In step S200, a hybrid model combining an adaptive selection weighted fusion model and an optimized support vector regression model is constructed, taking into account the product structure complexity and the fluctuation range of process parameters. This model includes: When the complexity of the product structure exceeds the complexity threshold or the fluctuation range of the process parameters exceeds the fluctuation threshold, the sample dataset is clustered according to the product complexity and parameter range, and each cluster is trained separately. The support vector regression sub-model obtained by the corresponding cluster training is then called. When the complexity of the product structure does not exceed the complexity threshold and the fluctuation range of the process parameters does not exceed the fluctuation threshold, the weighted fusion model is invoked.

5. The method according to claim 1, characterized in that, In step S400, the gradient descent method is used to locally correct key process parameters until the deviation falls back to within a preset threshold, resulting in optimized process parameters, including: S410 uses minimizing the deviation between the actual target value and the predicted target value as the objective function to determine the key process parameters related to the production process; S420, the key process parameters are adjusted in small steps. After each round of adjustment, the correction effect is quickly evaluated by the fusion prediction model, and the objective function value corresponding to the current key process parameter is calculated. The step size range of the small step adjustment is preset according to the process parameter type. S430: When the objective function value decreases compared to the previous round, continue to adjust the key process parameters in the current direction; when the objective function value increases, adjust the key process parameters in the opposite direction or reduce the adjustment step size; until the deviation between the actual target value and the predicted target value falls back to within the preset threshold, the key process parameters output at this time are used as the optimized process parameters.

6. The method according to claim 1, characterized in that, The method further includes: Using a preset production batch as a sliding window, the actual process parameters and actual target values ​​within the window are periodically used as new training samples to retrain and optimize the support vector regression model and the weighted fusion model.

7. A system for optimizing injection molding process parameters based on hybrid algorithm and model fusion, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor performs the method as described in any one of claims 1 to 6.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.

Images