Automobile loudspeaker mask injection molding buckling deformation optimization method based on WOA-BP neural network
By combining orthogonal experimental design and whale optimization algorithm to improve the BP neural network, efficient optimization of the warping deformation of the automotive speaker mask is achieved, and the problems of low efficiency and poor accuracy in traditional methods are solved, and more accurate process parameters are obtained, which reduces the amount of warping deformation and improves the quality stability of injection molded parts.
Patent Information
- Application Number
- CN202510546881.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-28
- Publication Date
- 2025-08-05
AI Technical Summary
The traditional injection molding process optimization method has low efficiency and poor accuracy, making it difficult to achieve accurate optimization in the continuous parameter space of the automotive speaker mask, and traditional BP neural networks are prone to fall into local optimal solutions.
Combining orthogonal experimental design and whale optimization algorithm to improve the BP neural network (WOA-BP), traversing the process parameter combination in a continuous parameter space through grid search, optimizing the warping deformation of the car speaker mask.
It significantly improves optimization efficiency and accuracy, obtains a more accurate optimal process parameter combination, reduces the amount of warping and deformation, and improves the quality stability of injection molded parts.
Smart Images

Figure CN120430178A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of injection molding technology, and in particular to a method for optimizing injection molding warping deformation of a car speaker cover based on a WOA-BP neural network. Background Art
[0002] As a precision injection-molded part, automotive speaker covers have stringent requirements for dimensional stability in their molding quality. During injection molding process optimization, suppressing warpage is a core objective. Traditional methods rely on trial-and-error to adjust process parameters, which is inefficient and costly. To systematically analyze the nonlinear effects of process parameters on molding quality, existing technologies typically employ orthogonal experimental designs combined with CAE numerical simulations. Variance analysis is used to screen significant influencing factors and determine a preliminary process window. However, this approach only provides optimization results for discrete parameter combinations and struggles to achieve precise optimization within a continuous space.
[0003] In recent years, BP neural networks have been introduced into injection molding process modeling to establish mappings between process parameters and quality indicators. However, traditional BP networks have inherent flaws. Randomly initialized weights and thresholds lead to large fluctuations in model training results. Gradient descent methods are prone to falling into local optimal solutions, making it difficult to globally optimize complex nonlinear process relationships. Summary of the Invention
[0004] In response to the problems of low efficiency, poor precision, insufficient stability and so on in existing injection molding process optimization methods, the present invention proposes a low-warpage injection molding process optimization method for automobile speaker covers based on a whale optimization algorithm improved BP neural network (WOA-BP).
[0005] A method for optimizing the warpage deformation of an automobile speaker cover by injection molding based on a WOA-BP neural network comprises the following steps:
[0006] Step 1: Construct a mesh model for quality analysis of the car speaker cover;
[0007] Step 2: Given the process parameters that affect the warpage deformation of the injection molding of the automobile speaker cover and their corresponding numerical ranges, an orthogonal experiment is designed to obtain a series of process parameter combinations that affect the warpage deformation of the injection molding of the automobile speaker cover; based on the mesh model for quality analysis and a series of process parameter combinations, the injection molding warpage deformation of the automobile speaker cover is simulated and analyzed to obtain the corresponding warpage deformation amount;
[0008] Step 3: Normalizing the process parameter combination and the corresponding warpage deformation obtained in step 2; using the normalized data to train a BP neural network to obtain a trained BP neural network; the input of the BP neural network is the normalized process parameters affecting the injection molding warpage deformation of the car speaker cover, and the output is the warpage deformation;
[0009] Step 4: Introduce the whale optimization algorithm WOA to perform global optimization on the trained BP neural network to obtain the WOA-BP neural network;
[0010] Step 5: Based on the WOA-BP neural network, a grid search algorithm is used to traverse all possible process parameter combinations in the continuous parameter space to accurately predict the optimal process parameter combination.
[0011] Furthermore, in the step 2, the process parameters that affect the warping deformation of the car speaker cover during injection molding are injection time, melt temperature, mold temperature, holding pressure and holding time.
[0012] Furthermore, in the step 2, after obtaining a series of process parameter combinations that affect the injection molding warping deformation of the automobile speaker cover, the numerical analysis sequence is determined, the variance analysis of the process parameters is carried out, the influence intensity p value and F value are calculated, the contribution rate of the process parameters to the warping deformation is calculated, and the contribution rates are sorted from high to low; at this time, in the step 5, the three process parameters with the highest contribution rates are selected for short-step traversal optimization test.
[0013] Furthermore, the number of nodes in the hidden layer of the BP neural network is 3.
[0014] Furthermore, the implementation process of step five is as follows:
[0015] A mean analysis is performed on each process parameter that affects the injection molding warpage of the automobile speaker cover, and the numerical point of each process parameter within its numerical range corresponding to the minimum warpage deformation is obtained. A new numerical range consisting of preset positive and negative errors of the numerical point is set as the reduced numerical range of the process parameter; based on the reduced numerical range of each process parameter, the parameter combination is traversed under short steps and the parameter normalization of step three is performed, and the WOA-BP neural network obtained in step four is called to predict the normalized warpage deformation; the predicted normalized warpage deformation is denormalized to obtain the final warpage deformation; the process parameter combination corresponding to the minimum warpage deformation is selected as the optimal process parameter combination.
[0016] The beneficial effects of the present invention are:
[0017] (1) Combining orthogonal experimental design with intelligent algorithms, first narrowing the search range of process parameters through experimental design, and then using intelligent algorithms for precise optimization, significantly improving optimization efficiency;
[0018] (2) The whale optimization algorithm is used to improve the BP neural network to overcome the defects of the traditional BP network that it is easy to fall into local optimality and unstable training;
[0019] (3) The global optimization of the continuous process parameter space is achieved through the grid search method, which can obtain a more accurate optimal solution compared with the traditional discrete optimization method. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] Figure 1 Flowchart of the optimization method according to an embodiment of the present invention.
[0021] Figure 2 Schematic diagram of the car speaker cover model used in the examples.
[0022] Figure 3 This is a flow chart of the principle of the whale optimization algorithm used in the method of the present invention.
[0023] Figure 4 A comparison chart of the prediction effects of the embodiment model.
[0024] Figure 5 This is a diagram of the simulation results of the process parameters obtained from the orthogonal experiment of the embodiment.
[0025] Figure 6 This is a simulation result diagram of the process parameters after the WOA-BP model is optimized in the embodiment. DETAILED DESCRIPTION
[0026] The present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments, and the purpose and effects of the present invention will become more apparent. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0027] like Figure 1 As shown, the method for optimizing the warpage deformation of the automobile speaker cover injection molding based on the WOA-BP neural network in this embodiment includes the following steps:
[0028] Step 1: Build a mesh model for quality analysis of the car speaker cover.
[0029] Perform geometric repair on product models, identify and repair geometric defects in the model, remove some small details, simplify the model, generate high-quality mesh models for analysis, select materials, and control the maximum aspect ratio.
[0030] Step 2: Given the process parameters that affect the injection molding warpage of the car speaker cover and their corresponding numerical ranges, an orthogonal experiment is designed to obtain a series of process parameter combinations that affect the injection molding warpage of the car speaker cover; based on the mesh model used for quality analysis and a series of process parameter combinations, the injection molding warpage of the car speaker cover is simulated and analyzed to obtain the corresponding warpage deformation amount.
[0031] Among them, the process parameters that affect the warping deformation of the car speaker cover injection molding include injection time, melt temperature, mold temperature, holding pressure, and holding time.
[0032] Step 3: Normalize the process parameter combination and the corresponding warpage deformation obtained in step 2; use the normalized data to train the BP neural network to obtain a trained BP neural network; the input of the BP neural network is the normalized process parameters that affect the injection molding warpage deformation of the automobile speaker cover, and the output is the warpage deformation.
[0033] The normalized expressions of input and output data are as follows:
[0034] Input normalization:
[0035]
[0036] Output normalization:
[0037]
[0038] Where, are the normalized input and output quantities, min(X) and min(Y) are the minimum input and minimum output quantities, max(X) and max(Y) are the maximum input and maximum output quantities, respectively.
[0039] The formula for calculating the optimal number of hidden layer nodes is as follows:
[0040]
[0041] The formula for calculating the mean square error is as follows:
[0042]
[0043] Where N hidden is the number of hidden layer nodes, floor is the rounding function, N input is the number of input process parameters, N output is the number of output parameters, a is the adjustment constant, y k is the actual value, is the network prediction value, and N is the number of test samples.
[0044] In the BP neural network, the forward propagation process of the optimization network of process parameters and warpage deformation is:
[0045] Hidden layer output:
[0046]
[0047] Output layer prediction:
[0048]
[0049] Where: x i To input process parameters, is the weight matrix from the input layer to the hidden layer, is the hidden layer threshold, w j (2) is the weight from hidden layer to warping deformation, b (2) is the output layer warping threshold.
[0050] The Levenberg-Marquardt algorithm is used to back-propagate and update the warping weights:
[0051] ΔW=-(J T J+μI) -1 J T e
[0052] Where: ΔW is the weight adjustment matrix, which contains all weights (input layer to hidden layer, hidden layer to output layer) and the updated threshold value. J is the Jacobian matrix of the error with respect to the weight, μ is the damping factor (initial value 0.001), and e is the output error vector.
[0053] Step 4: Introduce the Whale Optimization Algorithm (WOA) to perform global optimization on the trained BP neural network to obtain the WOA-BP neural network. Step 4 specifically includes:
[0054] Encode the parameters to be optimized of the BP neural network into a search vector to form the search vector of the WOA, and set the upper and lower bounds of the search space;
[0055] The fitness function is defined as the mean square error (MSE) of the test set, which directly reflects the network prediction accuracy. The specific formula is the same as the previous mean square error formula;
[0056] According to the whale optimization algorithm, iterative optimization is performed according to the encirclement and predation phase mechanism, guiding other individuals to the optimal individual X. * The area shrinkage is expressed as follows:
[0057] X(t+1)=X * (t)-A·D
[0058] D=|C·X * (t)-X(t)|,A=2a·r1-a,C=2r2
[0059]
[0060] Where: X(t+1) is the current position of the individual; X *(t) is the position of the current optimal individual; A is the convergence factor; C is the disturbance factor; a is the linear decreasing coefficient, which decreases linearly from 2 to 0, controlling the search range from global to local; t is the current iteration number, T is the total iteration number, r1 and r2 are random numbers, and random disturbances are introduced to avoid premature convergence.
[0061] Simulate the bubble net feeding behavior of a humpback whale, updating its position in a spiral path:
[0062]
[0063] D'=|X * (t)-X(t)|,l∈[-1,1]
[0064] Where: D′ is the distance between the current individual and the optimal individual; b is the spiral shape definition parameter, and l controls the spiral step size.
[0065] A reflection correction strategy is used for out-of-bounds parameters to ensure that the search is always within the valid space:
[0066]
[0067] Where: X new is the corrected position, lb is the lower limit of the parameter, and ub is the upper limit of the parameter. This method avoids the loss of population diversity caused by hard truncation and maintains search vitality.
[0068] The optimization process is updated and iterated. After multiple iterations of optimization, the performance improvement of the optimization algorithm is obtained. The principle flow chart of the whale optimization algorithm is as follows: Figure 3 shown.
[0069] Step 5: Based on the WOA-BP neural network, a grid search algorithm is used to traverse all possible process parameter combinations in the continuous parameter space to accurately predict the optimal process parameter combination.
[0070] A mean analysis is performed on each process parameter that affects the warpage deformation of the injection molding of the automobile speaker cover, and the numerical point corresponding to the minimum warpage deformation of each process parameter under the influence of its single variable is obtained. A new numerical range consisting of preset positive and negative errors of the numerical point is set as the reduced numerical range of the process parameter; based on the reduced numerical range of each process parameter, the parameter combination is traversed under a short step size and the parameter normalization of step three is performed, and the WOA-BP neural network obtained in step four is called to predict the normalized warpage deformation; the predicted normalized warpage deformation is denormalized to obtain the final warpage deformation; the process parameter combination corresponding to the minimum warpage deformation is selected as the optimal process parameter combination.
[0071] The optimization process of the present invention is specifically described below through examples.
[0072] Step 1: Perform geometric repair and preprocessing on the 3D model of the car speaker cover. Figure 2 As shown, taking the speaker cover on the A-pillar in the car as an example, the material is PC+ABS, and the melt density of the material at 265°C is 0.98258g / cm 3 , the melt flow rate is 18g / 10min (220℃ / 10kg), and the heat deformation temperature reaches 98℃ (under 1.82MPa). The material size is approximately 150mm×60mm×3mm. The quality requirements for the product are as low as possible warping deformation, no obvious weld lines or pores in the appearance, and no defects such as short shots during injection. Identify and repair geometric defects in the model, including eliminating 23 short edges with a length ranging from 0.05-0.1mm, processing 39 self-intersecting surfaces, and open edges with a stitching gap of ≤0.15mm. At the same time, small features that do not affect the structural strength, such as fillets below R0.3mm, are reasonably simplified, and the continuity of key surfaces is ensured. After the repair is completed, the model is subjected to simulated injection molding simulation experiments and discretized using dual-face meshing technology. The final generated finite element model contains 142,857 units, with a maximum aspect ratio controlled within 8.0 and a matching rate of 93.5%.
[0073] Step 2: Given the process parameters and their corresponding numerical ranges that affect the warpage deformation of the automotive speaker cover during injection molding: injection time 0.8-1.6s, melt temperature 235-295°C, mold temperature 50-100°C, holding pressure 50-110%, and holding time 4-16s. A five-factor, five-level orthogonal experimental design was used to optimize the injection molding process parameters. A full "filling + holding + warpage" process analysis was performed for 25 combinations of process parameters to determine the corresponding warpage deformation.
[0074] Statistical analysis of variance was used to analyze the data. The results showed that holding pressure had the most significant effect on warpage (F=23.54, p<0.0001, contribution rate 42.1%), followed by melt temperature (F=20.15, p<0.0001, contribution rate 36.0%). Together, they explained 78.1% of the warpage. Mold temperature had a secondary significant effect (F=6.42, p=0.0032), while injection time and holding time had relatively weaker effects (p-values 0.0125 and 0.0358, respectively). Mean analysis of each factor level revealed that when holding pressure was 108%, melt temperature was 235°C, mold temperature was 50°C, injection time was 1.2s, and holding time was 7s, warpage reached a minimum value of 0.5211mm, a 31.2% reduction from the worst-case condition. The obtained process parameters were simulated and verified, and the actual simulated warpage was 0.5264mm. The accuracy of the prediction model reached 98.99%. The simulation results are as follows Figure 5 shown.
[0075] Step 3: A BP neural network prediction model was constructed based on 50 sets of process parameter data (including 25 sets of orthogonal experimental data and 25 sets of training set data). 50 sets of experimental data were read, including 5 input process parameters (injection time, melt temperature, mold temperature, holding pressure, holding time) and 1 output parameter (warpage deformation). The data set was divided into 40 training sets and 10 test sets in a 4:1 ratio. The input and output data were normalized. The test results of the mean square error of the optimal number of nodes are shown in Table 1:
[0076] Table 1. Mean square error test results of optimal number of nodes
[0077] Number of nodes 3 4 5 6 7 <![CDATA[MSE(mm 2 )]]> 0.0092 0.0239 0.0116 1.0467 0.1035 Number of nodes 8 9 10 11 12 <![CDATA[MSE(mm 2 )]]> 0.6288 4.5883 0.7692 1.4627 0.1208
[0078] Select 3 nodes hidden layer (MSE is as low as 0.0092mm 2 ), forming a 5-3-1 network structure. The initial learning rate was set to 0.01, the maximum iteration count was 1000, and the target error was 1e-5. A 5-fold cross-validation algorithm determined the number of hidden layer nodes to be 3, achieving a mean squared error of 0.0092 on the training set. The traditional BP model achieved a mean absolute error of 0.0393 mm, a mean squared error of 0.0024, and a prediction accuracy of 94.05%.
[0079] Step 4: Introduce the Whale Optimization Algorithm (WOA) to perform global optimization on the trained BP neural network. First, encode the parameters of the existing model and initialize the population. Encode the parameters to be optimized of the BP neural network into a search vector. For a 5-3-1 structure network, the following parameters need to be optimized: the weights from the input layer to the hidden layer have a total of 5 input nodes × 3 hidden nodes (15 parameters), the hidden layer threshold has 3 parameters, the weights from the hidden layer to the output layer have 3 parameters, and the output layer threshold has 1 parameter, for a total of 22 parameters, which constitute the search vector of the WOA.
[0080] The search space upper and lower bounds lb and ub are set to -3 and 3, respectively, to ensure that the activation function (tanh) works within the sensitive range. The initial population (size 30) is randomly generated, and each individual represents a set of possible network parameter combinations.
[0081] Design the fitness function and define the fitness function as the mean square error (MSE) of the test set, which directly reflects the network prediction accuracy.
[0082] In the encirclement and predation stage, in each iteration, the current optimal individual X * This method guides other individuals to converge toward the optimal region. During the bubble net attack phase, the system simulates the bubble net feeding behavior of humpback whales, updating positions in a spiral path. Boundaries are then constrained, and a reflective correction strategy is applied to out-of-bounds parameters to ensure that the search remains within the valid space. This method avoids the loss of population diversity caused by hard truncation and maintains search activity. Multiple iterations are performed, retaining the top 10% of individuals in fitness in each generation to prevent loss of high-performing genes. The maximum number of iterations is 50, and the change in optimal fitness for 10 consecutive generations is less than 0.1%.
[0083] Through optimization effect verification (compared with the BP model), after 50 iterations of optimization, the key performance improvements are shown in Table 2:
[0084] Table 2 Comparison of optimization effects of BP and WOA-BP proxy models
[0085] index BP model WOA-BP model Improvement rate MAE(mm) 0.0393 0.0181 54.0% <![CDATA[MSE(mm 2 )]]> 0.0024 0.0007 70.8% RMSE(mm) 0.0490 0.0264 46.1% Prediction accuracy 94.05% 97.19% 3.14%
[0086] The comparison shows that the performance of the optimized WOA-BP model is significantly improved, with the mean absolute error reduced by 54.0% to 0.0181mm and the mean square error reduced by 70.8% to 0.0007mm. 2 , RMSE decreased by 46.1% to 0.0264mm, and the prediction accuracy increased to 97.19%. Figure 4 shown.
[0087] Step 5: Use the trained WOA-BP network to traverse the parameter combination to find the minimum warpage deformation. Specifically: perform mean analysis on each process parameter that affects the warpage deformation of the injection molding of the car speaker cover, and set a new numerical range consisting of the preset positive and negative errors of the numerical point as the reduced numerical range of the process parameter. For example, the melt temperature is reduced from 235-295℃ to 235-265℃, and the holding pressure is reduced from 50-110% to 80-110%. The mold temperature is reduced from 50-100℃ to 60-85℃, and in this new numerical range, the step size is set to 1. In this embodiment, the ranking is carried out from large to small according to significance, and the top three are melt temperature, holding pressure, and mold temperature. These three parameters are optimized first, and at the same time, the secondary process parameters of injection time 1.2s and holding time 7s are fixed (optimal values of orthogonal experiment). Generate a parameter grid and create all combinations of three-dimensional parameter space. A total of 24986 (31×31×26) sets of parameter combinations are generated. Each set of process parameter inputs is normalized and the trained WOA-BP network is called to predict the warpage value. The results show that the optimal process parameter set with minimum warpage is: melt temperature 238°C, holding pressure 108%, mold temperature 78°C, and warpage of 0.498mm, which is 4.4% lower than the warpage of the optimal group in the original orthogonal experiment. The optimized process parameters are simulated and verified, and the actual simulated warpage is 0.502mm with an accuracy of 99.2%. The simulation results are as follows Figure 6 shown.
[0088] Those skilled in the art will understand that the foregoing descriptions are merely preferred embodiments of the invention and are not intended to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, those skilled in the art will still be able to modify the technical solutions described in the foregoing examples or substitute equivalents for some of the technical features therein. Any modifications, equivalent substitutions, etc. made within the spirit and principles of the invention shall be included within the scope of protection of the invention.
Claims
1. A method for optimizing the warpage deformation of automobile speaker cover injection molding based on WOA-BP neural network, characterized in that: The steps include: Step 1: Construct a mesh model for quality analysis of the car speaker cover; Step 2: Given the process parameters that affect the warpage deformation of the car speaker cover by injection molding and their corresponding numerical ranges, an orthogonal experiment is designed to obtain a series of process parameter combinations that affect the warpage deformation of the car speaker cover by injection molding; Based on the quality analysis mesh model and a series of process parameter combinations, a simulation analysis is performed on the injection molding warpage deformation of the car speaker cover to obtain the corresponding warpage deformation amount; Step 3: Normalize the process parameter combination and the corresponding warpage deformation obtained in step 2; The BP neural network is trained using the normalized data to obtain a trained BP neural network; The input of the BP neural network is the normalized process parameters that affect the warpage deformation of the car speaker cover during injection molding, and the output is the warpage deformation amount; Step 4: Introduce the whale optimization algorithm WOA to perform global optimization on the trained BP neural network to obtain the WOA-BP neural network; Step 5: Based on the WOA-BP neural network, a grid search algorithm is used to traverse all possible process parameter combinations in the continuous parameter space to accurately predict the optimal process parameter combination.
2. The method for optimizing the warpage deformation of automobile speaker cover injection molding based on WOA-BP neural network according to claim 1, characterized in that: In the step 2, the process parameters that affect the warping deformation of the car speaker cover during injection molding are injection time, melt temperature, mold temperature, holding pressure and holding time.
3. The method for optimizing the warpage deformation of automobile speaker cover injection molding based on WOA-BP neural network according to claim 1, characterized in that: In the second step, after obtaining a series of process parameter combinations that affect the warpage deformation of the automobile speaker cover during injection molding, a numerical analysis sequence is determined, a variance analysis of the process parameters is performed, the influence strength p value and F value are calculated, and the contribution rate of the process parameters to the warpage deformation is calculated, and the contribution rates are ranked from high to low. At this time, in step five, the three process parameters with the highest contribution rates are selected to perform a short-step traversal optimization test.
4. The method for optimizing the warpage deformation of automobile speaker cover injection molding based on WOA-BP neural network according to claim 1, characterized in that: The number of nodes in the hidden layer of the BP neural network is 3.
5. The method for optimizing the warpage deformation of automobile speaker cover injection molding based on WOA-BP neural network according to claim 1, characterized in that: The implementation process of step five is as follows: A mean analysis is performed on each process parameter that affects the injection molding warpage of the automobile speaker cover, and the numerical point of each process parameter within its numerical range corresponding to the minimum warpage deformation is obtained. A new numerical range consisting of preset positive and negative errors of the numerical point is set as the reduced numerical range of the process parameter; based on the reduced numerical range of each process parameter, the parameter combination is traversed under short steps and the parameter normalization of step three is performed, and the WOA-BP neural network obtained in step four is called to predict the normalized warpage deformation; the predicted normalized warpage deformation is denormalized to obtain the final warpage deformation; the process parameter combination corresponding to the minimum warpage deformation is selected as the optimal process parameter combination.
Citation Information
Cited By
Method for adjusting injection molding machine based on injection molding part defects and injection molding machine cleaning structure
CN121608355A
