Relay Housing Molding Parameter Retrieval and Process Optimization Based on Pressure Distribution at
Time:2026-11-09 08:21:23 / Popularity: / Source:
0 Introduction
A relay is an automatic switching element with isolation function. Its housing is a plastic molded product, which is an important protective component of relay. In injection molding process of relay housing parts, defects such as concave openings, burrs, scorching, and weld lines often occur. To eliminate these problems, multiple mold repairs and trial moldings are usually required, repeatedly adjusting molding process parameters, increasing trial-and-error costs. Therefore, quickly obtaining optimal process parameters for relay housings is crucial for improving production efficiency of these parts.
Data-driven intelligent reasoning technology for process parameters is key to overcoming challenges of process parameter optimization. Simultaneously, quality control of injection molded products is also trending towards intelligence. In recent years, intelligent methods such as case-based reasoning, expert systems, machine learning, deep learning have provided broad insights into modeling and analyzing nonlinear, strongly coupled, time-varying relationships between injection molding process parameters and plastic product quality. Case-based reasoning, in particular, solves new problems by drawing on experience and methods of previous examples. Steps include case retrieval, reuse, modification, and storage, similar to process by which trial molders adjust part process parameters based on experience. Since case retrieval depends on feature similarity, feature selection is crucial. Yu et al. proposed a mold forming feature representation method based on similarity of injection pressure profiles, reducing dimensionality of molding features and improving retrieval accuracy. Guo et al. improved retrieval efficiency by characterizing geometric and process information through pressure distribution. However, these studies did not perform automatic preprocessing of pressure profile data, nor did they consider impact of different processing methods on retrieval accuracy, and they did not further utilize pressure profile data to correct process parameters of similar parts.
To address these issues, this paper, based on injection molding process parameters and model data of a large number of relay housing parts in enterprises, proposes a preprocessing method suitable for pressure distribution curves. Subsequently, particle swarm optimization is used to analyze pressure data, obtain combined injection positions, a neural network model is used to fit relationship between pressure data and actual injection speed to predict injection speed. This method has faster retrieval efficiency and higher accuracy, can improve production efficiency of parts.
Data-driven intelligent reasoning technology for process parameters is key to overcoming challenges of process parameter optimization. Simultaneously, quality control of injection molded products is also trending towards intelligence. In recent years, intelligent methods such as case-based reasoning, expert systems, machine learning, deep learning have provided broad insights into modeling and analyzing nonlinear, strongly coupled, time-varying relationships between injection molding process parameters and plastic product quality. Case-based reasoning, in particular, solves new problems by drawing on experience and methods of previous examples. Steps include case retrieval, reuse, modification, and storage, similar to process by which trial molders adjust part process parameters based on experience. Since case retrieval depends on feature similarity, feature selection is crucial. Yu et al. proposed a mold forming feature representation method based on similarity of injection pressure profiles, reducing dimensionality of molding features and improving retrieval accuracy. Guo et al. improved retrieval efficiency by characterizing geometric and process information through pressure distribution. However, these studies did not perform automatic preprocessing of pressure profile data, nor did they consider impact of different processing methods on retrieval accuracy, and they did not further utilize pressure profile data to correct process parameters of similar parts.
To address these issues, this paper, based on injection molding process parameters and model data of a large number of relay housing parts in enterprises, proposes a preprocessing method suitable for pressure distribution curves. Subsequently, particle swarm optimization is used to analyze pressure data, obtain combined injection positions, a neural network model is used to fit relationship between pressure data and actual injection speed to predict injection speed. This method has faster retrieval efficiency and higher accuracy, can improve production efficiency of parts.
1 Experimental Methods
To achieve retrieval of similar relay housing parts, a database of injection gate pressure data needs to be constructed. Therefore, filling flow analysis is performed on each existing housing part under same process conditions to obtain their injection gate pressure distribution data. After preprocessing data, it is stored in database. Using inlet pressure distribution data as an evaluation index for part similarity, case retrieval was conducted, and different subsequent processing was performed based on similarity level. When similarity was high enough, process parameters were slightly modified and reused; if similarity was low, process parameters needed to be significantly modified. The overall method flow is shown in Figure 1.
Figure 1: Method Flowchart
Using Moldex3D software, mold flow analysis was performed on relay housing parts that had been successfully trial-molded by company. Material was set as PBT-kingfa-RG301, machine was ARBURG 1200T 1600-350-35, and process parameter settings are shown in Table 1.
Using Moldex3D software, mold flow analysis was performed on relay housing parts that had been successfully trial-molded by company. Material was set as PBT-kingfa-RG301, machine was ARBURG 1200T 1600-350-35, and process parameter settings are shown in Table 1.
| Process Parameters | Value |
| Injection Speed/mm.s-1 | 80 |
| Injection Pressure/MPa | 250 |
| V/P Switching Point | Fill Volume Percentage (excluding runner) -99% |
| Holding Time/s | 0.5 |
| Holding Pressure/MPa | 40% of End-of-Fill Pressure |
| Mold Temperature/℃ | 60 |
| Melt Temperature/℃ | 275 |
| Cooling Time/s | 0.1 |
Table 1: Process Parameter Settings
To ensure data comparability, following rules were executed during mold flow analysis: cold runner simulation of full cavity filling analysis, semi-hot runner simulation of filling analysis excluding hot runner portion, and hot runner simulation of single cavity filling analysis. Gate pressure data file can be exported through mold flow analysis. Given complexity of exported data, preprocessing is necessary.
Preprocessing method for gate pressure curve is as follows: First, pressure data is classified into five categories based on part volume range: 0~14 mL, 10~24 mL, 20~34 mL, 30~44 mL, and 40~54 mL. Second, initial data points with a pressure value of 0 are removed, and pressure data curve is sampled at 0.01 s intervals. Finally, end pressure values within same volume range need to be extended until pressure data within that volume range has same number of data points. Three methods for extending end pressure values are considered: 1) setting extension point pressure value to zero; 2) setting extension point pressure value to end pressure value; and 3) extending end pressure values of extension points with same slope.
To compare performance of three methods in matching part similarity, two similar models were selected for mold flow filling analysis. Model appearance is shown in Figure 2, pressure curve in Figure 3, and Figure 4 shows processing methods for three different end-pressure data.
To ensure data comparability, following rules were executed during mold flow analysis: cold runner simulation of full cavity filling analysis, semi-hot runner simulation of filling analysis excluding hot runner portion, and hot runner simulation of single cavity filling analysis. Gate pressure data file can be exported through mold flow analysis. Given complexity of exported data, preprocessing is necessary.
Preprocessing method for gate pressure curve is as follows: First, pressure data is classified into five categories based on part volume range: 0~14 mL, 10~24 mL, 20~34 mL, 30~44 mL, and 40~54 mL. Second, initial data points with a pressure value of 0 are removed, and pressure data curve is sampled at 0.01 s intervals. Finally, end pressure values within same volume range need to be extended until pressure data within that volume range has same number of data points. Three methods for extending end pressure values are considered: 1) setting extension point pressure value to zero; 2) setting extension point pressure value to end pressure value; and 3) extending end pressure values of extension points with same slope.
To compare performance of three methods in matching part similarity, two similar models were selected for mold flow filling analysis. Model appearance is shown in Figure 2, pressure curve in Figure 3, and Figure 4 shows processing methods for three different end-pressure data.
Figure 2 Similar part model
Figure 3 Pressure curve of similar model exported by software
Figure 4 Processing method for end-pressure values
After processing pressure data using above preprocessing methods, similarity between two models is shown in Table 2. Method 1 is prone to errors in judging part similarity. This is because end-pressure data of parts is generally large, setting extension point pressure value to 0 easily reduces similarity. Methods 2 and 3 can achieve better matching, but considering amount of data collected and difficulty of processing, this paper chooses Method 2 for data preprocessing.
After processing pressure data using above preprocessing methods, similarity between two models is shown in Table 2. Method 1 is prone to errors in judging part similarity. This is because end-pressure data of parts is generally large, setting extension point pressure value to 0 easily reduces similarity. Methods 2 and 3 can achieve better matching, but considering amount of data collected and difficulty of processing, this paper chooses Method 2 for data preprocessing.
| Methods | Method 1 | Method 2 | Method 3 |
| Similarity | 0.715 | 0.989 | 0.988 |
Table 2 Model similarity of different data preprocessing methods
To improve efficiency of data processing and retrieval, this paper automates above methods. Algorithms are used to clean, sample, and reduce dimensions of pressure data, improving matching accuracy of data similarity. Process is visualized to avoid discrepancies between matching and actual data due to improper data processing, thus improving accuracy and efficiency of retrieval. Figure 5 shows sample data obtained after automatic preprocessing of pressure curve of inlet of a shell-type part.
To improve efficiency of data processing and retrieval, this paper automates above methods. Algorithms are used to clean, sample, and reduce dimensions of pressure data, improving matching accuracy of data similarity. Process is visualized to avoid discrepancies between matching and actual data due to improper data processing, thus improving accuracy and efficiency of retrieval. Figure 5 shows sample data obtained after automatic preprocessing of pressure curve of inlet of a shell-type part.
Figure 5 Comparison of pressure curve preprocessing
The nearest neighbor retrieval method achieves the highest similarity among various instance retrieval methods. Its steps include data preparation, distance measurement, querying instances, calculating distance, sorting, and selection. Among them, distance measurement method uses Euclidean distance to evaluate similarity of pressure data based on distance between data points. Assume there are two pressure curves. Sample points of curve 1 after automatic data preprocessing are [(0, y10), (1, y11), ..., (n, y1n)], and sample points of curve 2 after automatic data preprocessing are [(0, y20), (1, y21), ..., (n, y2n)], as shown in Figure 6. Distance formula between two curves is shown in Equation (1).
The nearest neighbor retrieval method achieves the highest similarity among various instance retrieval methods. Its steps include data preparation, distance measurement, querying instances, calculating distance, sorting, and selection. Among them, distance measurement method uses Euclidean distance to evaluate similarity of pressure data based on distance between data points. Assume there are two pressure curves. Sample points of curve 1 after automatic data preprocessing are [(0, y10), (1, y11), ..., (n, y1n)], and sample points of curve 2 after automatic data preprocessing are [(0, y20), (1, y21), ..., (n, y2n)], as shown in Figure 6. Distance formula between two curves is shown in Equation (1).
Figure 6 Explanation of Curve Distance Calculation
Similarity calculation formula between target part and source part is shown in Equation (2).
Similarity calculation formula between target part and source part is shown in Equation (2).
In addition, to improve retrieval efficiency, pressure distribution data of inlet of parts were observed. It was found that pressure curves of different parts all showed an upward trend over time, and end pressures of similar parts were also mostly similar. Therefore, only similarity of parts with similar end pressures can be calculated. Specific steps are as follows: All parts in database are classified according to volume size range; Select parts with similar end pressures within range and calculate similarity; System will sort results according to similarity value and select top 10 parts with the highest similarity value for display.
2 Process Parameter Optimization, Experimental Results and Analysis
Case retrieval usually cannot find completely similar parts, so it is necessary to optimize process parameters of the most similar parts to adapt to target parts. Relay housing parts are mostly thin-walled components. Common defects during initial mold trials include scratches, burning, whitening of gate, and burrs. Addressing these defects typically involves adjusting injection speed. Furthermore, data analysis of existing process parameters for relay housing parts reveals that selectable ranges for material temperature, holding time, and holding pressure are relatively narrow, while selectable ranges for injection speed, mold temperature, and cooling time are relatively wide, especially injection speed. Therefore, injection speed has a significant impact on relay housing parts, necessitating adjustments to process parameters related to injection position and speed.
Geometric characteristics of part directly determine melt flow path, and differences in flow paths lead to significant changes in melt pressure during filling stage. In closed-loop controlled injection molding, derivative of melt pressure with respect to filling stroke is a function of injection speed. Therefore, points of slope change in pressure curve can be used to optimize injection position and speed.
Injection position, also known as screw position, is related to filling time of mold cavity by injection molding machine; it represents screw position at which injection speed transitions. Gate pressure curve is related to filling time. By setting a constant injection speed through mold flow analysis, gate pressure curve of part is obtained. Under constant injection speed conditions, location where slope of gate pressure curve changes significantly is also location where melt pressure changes significantly during melt flow process during mold filling. This location indicates a large geometric change in part and is a key point for injection speed conversion, which can be used as injection speed conversion location in subsequent actual injection molding process. Gate pressure curve data under constant injection speed, after time interval sampling, will generate a scatter plot. These scatter plots can be linearly fitted into a straight line. Maximum value of sum of fitting degrees when each data segment is fitted into a straight line is defined as slope inflection point. Injection position is represented by finding slope inflection point.
(1) Symbol definition: pi(i, yi): data points after pressure data sampling; k: number of inflection points.
(2) Objective function: In problem of determining optimal value of injection position parameters of part, finding slope inflection point of gate pressure data of part under constant injection speed conditions is crucial.
① Linear fit of (x+1)th segment Rx2: Assuming the first inflection point is at n, linear fit formula for the first segment is shown in equation (3).
Geometric characteristics of part directly determine melt flow path, and differences in flow paths lead to significant changes in melt pressure during filling stage. In closed-loop controlled injection molding, derivative of melt pressure with respect to filling stroke is a function of injection speed. Therefore, points of slope change in pressure curve can be used to optimize injection position and speed.
Injection position, also known as screw position, is related to filling time of mold cavity by injection molding machine; it represents screw position at which injection speed transitions. Gate pressure curve is related to filling time. By setting a constant injection speed through mold flow analysis, gate pressure curve of part is obtained. Under constant injection speed conditions, location where slope of gate pressure curve changes significantly is also location where melt pressure changes significantly during melt flow process during mold filling. This location indicates a large geometric change in part and is a key point for injection speed conversion, which can be used as injection speed conversion location in subsequent actual injection molding process. Gate pressure curve data under constant injection speed, after time interval sampling, will generate a scatter plot. These scatter plots can be linearly fitted into a straight line. Maximum value of sum of fitting degrees when each data segment is fitted into a straight line is defined as slope inflection point. Injection position is represented by finding slope inflection point.
(1) Symbol definition: pi(i, yi): data points after pressure data sampling; k: number of inflection points.
(2) Objective function: In problem of determining optimal value of injection position parameters of part, finding slope inflection point of gate pressure data of part under constant injection speed conditions is crucial.
① Linear fit of (x+1)th segment Rx2: Assuming the first inflection point is at n, linear fit formula for the first segment is shown in equation (3).
In formula, yi represents actual pressure value at point i;
represents fitted pressure value at point i; and y represents average of actual pressure values.
Data for other segments follow same pattern.
② Objective function r: Subtracting average of sums of fits of each segment from 1 gives objective function, expressed as equation (4).
represents fitted pressure value at point i; and y represents average of actual pressure values.Data for other segments follow same pattern.
② Objective function r: Subtracting average of sums of fits of each segment from 1 gives objective function, expressed as equation (4).
Considering response speed of injection molding machine and fact that only two speed change points are usually set in actual production (i.e., three injection speeds, occasionally two or four injection speeds), number of inflection points in each pressure curve should be controlled between 1 and 3. There should be a sequence between inflection points, and distance between each inflection point should be at least 2.
As shown in the previous section, injection position problem is equivalent to obtaining combined point position; that is, when objective function is minimized, it is required injection position. Optimizing combination of injection position points using Particle Swarm Optimization (PSO) algorithm is a good decision-making scheme.
In this paper, inertia weight w of PSO algorithm adopts a linear variation strategy, and other main parameter settings are shown in Table 3.
As shown in the previous section, injection position problem is equivalent to obtaining combined point position; that is, when objective function is minimized, it is required injection position. Optimizing combination of injection position points using Particle Swarm Optimization (PSO) algorithm is a good decision-making scheme.
In this paper, inertia weight w of PSO algorithm adopts a linear variation strategy, and other main parameter settings are shown in Table 3.
| Parameters | Value | Parameters | Value |
| Wmax | 1.5 | tmax | 30 |
| Wmin | 0.8 | Particle dimension D | 4 |
| Particle swarm size N | 40 | C1, C2 | 2 |
Table 3. Main Parameters of Algorithm
Slope inflection point is searched based on particle swarm optimization algorithm. Specific steps are as follows: Preprocess gate pressure data of target part, and take points at intervals of t=0.01 s to obtain pressure-time data points; Assume there are k=1~3 slope inflection points, and limit distance between inflection points; Use PSO to randomly generate slope inflection point positions, linearly fit each data point, and obtain determination coefficient Rx2 of each fitted line; Use PSO to obtain combined slope change point when objective function value r is minimized, as shown in equation (4).
Slope inflection point obtained by above method only represents time of slope change position point of preprocessed pressure distribution map. To obtain recommended injection position, a conversion is required. Conversion formula is given in equation (5):
Slope inflection point is searched based on particle swarm optimization algorithm. Specific steps are as follows: Preprocess gate pressure data of target part, and take points at intervals of t=0.01 s to obtain pressure-time data points; Assume there are k=1~3 slope inflection points, and limit distance between inflection points; Use PSO to randomly generate slope inflection point positions, linearly fit each data point, and obtain determination coefficient Rx2 of each fitted line; Use PSO to obtain combined slope change point when objective function value r is minimized, as shown in equation (4).
Slope inflection point obtained by above method only represents time of slope change position point of preprocessed pressure distribution map. To obtain recommended injection position, a conversion is required. Conversion formula is given in equation (5):
In formula, Saim - injection position of target part, mm; Ssim - injection position of similar part, mm; Dsim - screw diameter of similar part, mm; Daim - screw diameter of target part, mm; baim - multiple of target part; bsim - multiple of similar part; taim - filling time of target part during mold flow analysis, s; tsim - filling time of similar part during mold flow analysis, s.
Derivative of gate pressure curve with respect to filling stroke has a non-linear relationship with melt volume flow rate. Therefore, a neural network model can be used to construct relationship between two. BP neural network model uses error between actual output of network and target vector to modify its weights, making output value as close to target value as possible. This paper uses BP neural network model to predict and optimize injection speed of parts.
Based on collected gate pressure data and inflection point of pressure curve slope, three situations are summarized: first, actual speed switching point is completely different from pressure curve slope change point; second, actual speed switching point is close to pressure curve slope change point; and third, two coincide. This paper only analyzes second and third situations. Based on above analysis, pressure curve is related to injection speed. To simplify data and optimize input value of network model, taking HFE97 shell as an example, six sample values in Figure 7 are selected as input of BP neural network, and three segments of actual injection speed are used as output to establish a prediction model.
Derivative of gate pressure curve with respect to filling stroke has a non-linear relationship with melt volume flow rate. Therefore, a neural network model can be used to construct relationship between two. BP neural network model uses error between actual output of network and target vector to modify its weights, making output value as close to target value as possible. This paper uses BP neural network model to predict and optimize injection speed of parts.
Based on collected gate pressure data and inflection point of pressure curve slope, three situations are summarized: first, actual speed switching point is completely different from pressure curve slope change point; second, actual speed switching point is close to pressure curve slope change point; and third, two coincide. This paper only analyzes second and third situations. Based on above analysis, pressure curve is related to injection speed. To simplify data and optimize input value of network model, taking HFE97 shell as an example, six sample values in Figure 7 are selected as input of BP neural network, and three segments of actual injection speed are used as output to establish a prediction model.
Figure 7 Schematic diagram of BP model input data
Actual injection speed of each part needs to be converted to injection speed under same screw diameter. Conversion formula is shown in equation (6).
Actual injection speed of each part needs to be converted to injection speed under same screw diameter. Conversion formula is shown in equation (6).
In formula, v -- converted speed, mm/s; vo -- original speed, mm/s; D -- converted screw diameter, mm; Do -- original screw diameter, mm.
After filtering data in situations two and three, a total of 84 sets of data were used for training and testing. Training and test sets are divided in an 8:2 ratio, with 68 sets of data used for training and remaining 16 sets used for testing. To eliminate impact of magnitude difference between data sets on network prediction error, data needs to be normalized.
Nodes for each layer are determined: 6 nodes for input layer, 3 nodes for output layer, 1 node for hidden layer. Empirical formula for designing hidden layer nodes is shown in equation (7).
After filtering data in situations two and three, a total of 84 sets of data were used for training and testing. Training and test sets are divided in an 8:2 ratio, with 68 sets of data used for training and remaining 16 sets used for testing. To eliminate impact of magnitude difference between data sets on network prediction error, data needs to be normalized.
Nodes for each layer are determined: 6 nodes for input layer, 3 nodes for output layer, 1 node for hidden layer. Empirical formula for designing hidden layer nodes is shown in equation (7).
In formula, m represents number of neurons in input layer, n represents number of neurons in output layer, and a is a constant between [0, 10].
Number of hidden layer neurons is selected using an empirical formula, ranging from [3, 13]. After trying each option and observing network performance, final number of hidden layer neurons is determined to be 10. Sigmoid activation function is used for hidden layer transfer function, as shown in equation (8), and linear transfer function is used for output layer. Network structure of velocity prediction model is 6-10-3, error target is set to 1×10-3, and model learning rate is 0.01. After iteration, neural network training stopped once it met set error target.
Number of hidden layer neurons is selected using an empirical formula, ranging from [3, 13]. After trying each option and observing network performance, final number of hidden layer neurons is determined to be 10. Sigmoid activation function is used for hidden layer transfer function, as shown in equation (8), and linear transfer function is used for output layer. Network structure of velocity prediction model is 6-10-3, error target is set to 1×10-3, and model learning rate is 0.01. After iteration, neural network training stopped once it met set error target.
Figure 8 shows that most of predicted values match expected values. Prediction error is shown in Figure 9. As can be seen from figure, prediction errors for each injection speed segment are basically within ±10 mm/s, which is consistent with feasible range of injection speed in actual production.
Figure 8: BP neural network predicted values and actual values
Figure 9: Predicted injection speed error variation
RSME (root mean squared error, RMSE) can be used to measure deviation between predicted value and actual value. Its formula is shown in equation (9).
RSME (root mean squared error, RMSE) can be used to measure deviation between predicted value and actual value. Its formula is shown in equation (9).
In formula, Yi represents predicted value, y represents actual value, and n represents number of predicted values.
This paper uses root mean square error (RSME) and coefficient of determination to evaluate predictive performance of model. According to above formula, measured root mean square error of BP neural network prediction value is 5.535, and coefficient of determination is 95.16%, indicating accuracy of model in predicting injection molding speed of product, which can provide a basis for setting injection speed. Following relay housing information is provided:
Target part: HFE82-500M housing; Plastic material: PBT-4308G29-1135-9; Injection molding machine: ARBURG 470H 1000-400-35. Part model is shown in Figure 10.
This paper uses root mean square error (RSME) and coefficient of determination to evaluate predictive performance of model. According to above formula, measured root mean square error of BP neural network prediction value is 5.535, and coefficient of determination is 95.16%, indicating accuracy of model in predicting injection molding speed of product, which can provide a basis for setting injection speed. Following relay housing information is provided:
Target part: HFE82-500M housing; Plastic material: PBT-4308G29-1135-9; Injection molding machine: ARBURG 470H 1000-400-35. Part model is shown in Figure 10.
Figure 10 HFE82-500M part model and runner.
Process parameters generated using method described in this paper are used as initial mold trial parameters for part. First, gate pressure distribution data from mold flow analysis of target part is exported and automatically preprocessed. Then, a similar part search is performed. Results show that HFE9 housing has the highest similarity to target part, with a similarity of 88.0%. This indicates that molding characteristics of two are quite similar. Therefore, process parameters of HFE9 housing are used as initial reference parameters (screw diameter 40 mm). Screw diameter of injection molding machine used for target part was 35 mm. After conversion using formulas (5) and (6), it was used for trial molding. Although part could be formed, there were obvious surface defects, as shown in Figure 11, and surface quality did not meet production requirements. Therefore, injection position and injection speed were optimized, and process parameters for each stage are shown in Table 4.
Process parameters generated using method described in this paper are used as initial mold trial parameters for part. First, gate pressure distribution data from mold flow analysis of target part is exported and automatically preprocessed. Then, a similar part search is performed. Results show that HFE9 housing has the highest similarity to target part, with a similarity of 88.0%. This indicates that molding characteristics of two are quite similar. Therefore, process parameters of HFE9 housing are used as initial reference parameters (screw diameter 40 mm). Screw diameter of injection molding machine used for target part was 35 mm. After conversion using formulas (5) and (6), it was used for trial molding. Although part could be formed, there were obvious surface defects, as shown in Figure 11, and surface quality did not meet production requirements. Therefore, injection position and injection speed were optimized, and process parameters for each stage are shown in Table 4.
Figure 11 Side defect diagram of part
| Parameters | Initial Reference Parameters | Initial Trial Mold Parameters | Final Actual Parameters | Compliance Rate/% |
| Temperature/℃ | 265~275 | 265~275 | 260-280 | 98 |
| Mold Temperature/℃ | 45 | 45 | 45 | 100 |
| Feed Rate/mm | 33 | 54 | 51 | 94 |
| Injection Position/mm | 20/10/4.5 | 29/9/6 | 29/9/6 | 100 |
| Injection Speed/mm*s-1 | 80/90/55 | 116/111/73 | 116/111/73 | 100 |
| Holding Pressure/bar | 450/550/200 | 450/550/200 | 450/550/200 | 100 |
| Holding Time/s | 0.4/0.8/0.5 | 0.4/0.8/0.5 | 0.4/0.8/0.5 | 100 |
| Cooling Time/s | 12 | 12 | 12 | 100 |
| Overall Compliance Rate/% | 99 | |||
Table 4 Process parameters and compliance rate
Corrected process parameters were used as initial trial molding parameters for target part. After parameter fine-tuning, part was successfully formed without obvious appearance defects, as shown in Figure 12. Figure 13 shows that side defects have been eliminated, indicating that corrected parameters are more suitable for injection molding of this part. Finally, compliance rate between actual trial molding parameters and initial parameters reached 99%, dimensions and appearance of produced relay housing met requirements. This result shows that even if similarity of parts is insufficient, intelligent parameter optimization method can still effectively recommend suitable processes and manufacture high-quality products.
Corrected process parameters were used as initial trial molding parameters for target part. After parameter fine-tuning, part was successfully formed without obvious appearance defects, as shown in Figure 12. Figure 13 shows that side defects have been eliminated, indicating that corrected parameters are more suitable for injection molding of this part. Finally, compliance rate between actual trial molding parameters and initial parameters reached 99%, dimensions and appearance of produced relay housing met requirements. This result shows that even if similarity of parts is insufficient, intelligent parameter optimization method can still effectively recommend suitable processes and manufacture high-quality products.
Figure 12 Actual model of HFE82-500M shell
Figure 13 Side view of HFE82-500M shell
3 Conclusion
(1) Proposed preprocessing method for injection position pressure distribution data not only improves efficiency of retrieval but also improves accuracy of retrieval;
(2) Retrieval method for example is the nearest neighbor retrieval method. To address shortcomings of this method, algorithm is improved by utilizing characteristic that each group of pressure data shows an upward trend over time, which can improve retrieval efficiency;
(3) For parts with high similarity that cannot be retrieved, PSO algorithm is used to search for slope inflection point of inlet pressure data and convert it into injection position of each segment. A BP neural network is constructed based on shell-type pressure curve and actual injection speed to predict injection speed of target part. Predicted process parameters are used in actual mold trials to verify feasibility.
(2) Retrieval method for example is the nearest neighbor retrieval method. To address shortcomings of this method, algorithm is improved by utilizing characteristic that each group of pressure data shows an upward trend over time, which can improve retrieval efficiency;
(3) For parts with high similarity that cannot be retrieved, PSO algorithm is used to search for slope inflection point of inlet pressure data and convert it into injection position of each segment. A BP neural network is constructed based on shell-type pressure curve and actual injection speed to predict injection speed of target part. Predicted process parameters are used in actual mold trials to verify feasibility.
Last article:Injection Mold Gate Design Q&A Notes
Next article:Branch Runner Design












