A method for generating Moldflow parameter modification suggestions

Abstract

The application provides a Moldflow parameter modification suggestion generation method, compares a Moldflow analysis generated result with an existing engineer analysis result, generates a correction suggestion model for the Moldflow analysis result, and generates a more optimal parameter scheme suggestion, and solves the problems that in the prior art, the production guidance effect of AI on high-end mold products in Moldflow is low, and the opinions of experienced engineers cannot be converted to Moldflow technology.

Description

Technical Field

[0001] This invention relates to the field of injection molding technology, and more specifically, to a method for generating Moldflow parameter modification suggestions. Background Technology

[0002] High-end thin-walled plastic products, such as premium plastic toolboxes, are gradually evolving towards higher-end and more precise designs. Beyond essential structural elements, new toolboxes utilize stop structures, pivoting structures, or incorporate precision mechanical devices. This allows plastic toolboxes to not only hold tools but also provide functions previously unavailable, such as displaying, passing, limiting opening angles, and tool hanging. With the emergence of new industries, new tools, and new demands, the development potential of toolboxes is limitless. However, the actual production of plastic toolboxes relies heavily on existing mold-making technology. Any high-end product requiring production must first have a corresponding injection mold manufactured. The functionality of high-end products generally depends on precision structures; without injection molds meeting the required precision, even the most sophisticated product designs cannot be put into production.

[0003] When designing molds for large, thin-walled toolboxes, the surface of the toolbox is covered with textured surfaces, so imperfections in surface balance or weld lines are almost invisible. However, controlling the precision of components such as stop structures and pivot structures is a key focus in the manufacturing of high-end products. The design process must limit localized shrinkage and warping to prevent stop structures, pivots, and their mating structures from becoming completely unusable due to shrinkage and warping. Meanwhile, other parts, namely the box body, require stable quality.

[0004] Moldflow analysis technology offers significant advantages in improving productivity, ensuring product quality, reducing costs, and alleviating labor intensity. Using Moldflow, the entire injection molding process can be simulated and analyzed by computer before mold processing, accurately predicting melt filling, holding pressure, cavitation, weld lines, flow front temperature, and warpage. This allows designers to identify problems early, modify the plastic part and mold design in a timely manner, reduce or even avoid mold rework and scrap, improve part quality, and lower costs. In the manufacture of thin-walled toolboxes, this technology is mainly used to simulate and analyze non-precision parts of the box, reducing labor intensity. However, for the design of precision parts, such as stop structures, pivot structures, and the installation positions of precision machinery, the results of Moldflow simulation and optimization are basically of no reference value.

[0005] Experienced mold engineers can accurately identify design problems based on existing mold design data and predict the potential consequences of those problems with relatively high accuracy. However, their education and experience often lead to a lack of enthusiasm for learning computer simulation software. Furthermore, the accurate and referable aspects of Moldflow simulation analysis are almost meaningless for experienced engineers. The new generation of engineers can efficiently master mold flow analysis software; some have even received preliminary training in school. They can effectively use Moldflow to improve the quality of injection molded parts and reduce costs. However, their mold design experience is insufficient for handling highly complex projects.

[0006] Among various methods, using artificial intelligence-assisted mold flow analysis to improve the reference value of the results is an effective way to solve existing problems. A Chinese patent (application number CN4) discloses a method for predicting the quality of injection molded products based on an improved support vector machine. This method, based on actual production, draws a 3D model of a H-shaped fastener, sets corresponding process parameters in Moldflow, compares actual defects with simulated defects to verify the reliability of the simulation model, and predicts the quality of injection molded products by training and testing a support vector machine optimized through genetic simulated annealing. This method forms an AI model based on the comparison of actual and simulated defects, and uses this model to conduct a preliminary evaluation of the results of Moldflow simulation and optimization, avoiding engineers being misled by the simulation model. The drawback of this method is that the model only provides a reliability prediction result. In actual production, there is a significant negative correlation between design difficulty and the reliability of mold flow analysis results. This data bias leads to a final result that is an assessment of difficulty, which is of reference value for production management but has no positive effect on actual operation. Furthermore, this method requires a large amount of actual mold experimental data, making it only suitable for small molds like H-shaped fasteners. The human and material costs of a single experiment on large molds are extremely high, making it difficult to obtain the amount of data needed to establish a reliable model. Summary of the Invention

[0007] To overcome the problems of low AI guidance effectiveness for high-end mold products in existing Moldflow technologies and the inability of senior engineers' opinions to be translated into Moldflow technology, a method for generating Moldflow parameter modification suggestions is provided.

[0008] The technical solution adopted by this invention to solve its technical problem is: S1 imports the dataset into the model; The dataset is formatted as follows: process parameters, Moldflow evaluation, and engineer evaluation. The comparison yielded the following results: process parameters, engineer evaluation, and Moldflow evaluation.

[0009] The scores from engineer evaluations and Moldflow evaluations correspond to the magnitude of weld line issues, bubble issues, and shrinkage / warping issues after assessment. These scores also follow the laws of natural numbers, allowing for clear comparisons and the generation of datasets. X represents the process parameters, which are also features in the model. Rx-Rm represents the difference between the results of manual and Moldflow simulation analysis.

[0010] S2 uses this dataset to build separate models for each different problem evaluation; From a practical perspective, issues such as weld lines, bubbles, shrinkage, and warping can occur simultaneously, and their correlation is not very valuable. In practice, each evaluation model should be independent.

[0011] The evaluation results of this model can be obtained through multivariate regression or multiparameter classification. Although the mechanisms are different, from the perspective of the technical issues of model flow analysis, the output results are not significantly different when they are referenced: they all evaluate the risk of a certain problem, from no risk to high risk.

[0012] S3 uses this model to predict the differences between Moldflow simulation results and engineer annotations, provides suggestions, and displays the data.

[0013] The beneficial effects of this step are: By using a labeled dataset, the amount of actual mold test data required for the model is reduced, allowing results to be obtained even without experimental data. Furthermore, the model's accuracy is improved through a pre-training process. Data that aligns with Moldflow's conclusions is filtered out, enhancing the relevance of the results.

[0014] As a preferred option, the model in S2 is a decision tree classification prediction model. It generates multiple classifications in a quantitative way under a scoring system by using the magnitude of the bias, and establishes a decision tree classification prediction model using the dataset.

[0015] The classification probability data output by the model has extremely high reference value in model flow analysis applications. By analyzing the classification probabilities, engineers can receive a 90% probability that matches the model flow analysis conclusions, and a 10% probability that the problem outweighs the analysis results, allowing for a more thorough analysis of the design and the causes of the risks. It's important to note that neither the engineer's annotations nor the results provided by Moldflow are actual data. However, the decision tree classification prediction model itself, before any further adjustments, already possesses considerable practical value, and the suggestions it offers align with engineers' needs.

[0016] As a preferred embodiment, S2 also includes: S2.1 Randomly filter out evaluation data with excessively small differences, ensuring that they account for less than 50% of the total data. This prevents the data from becoming too large and misleading the model into classifying uncertain cases as indifferent. The filtering criteria are: Rx-Rm represents the difference between Moldflow evaluation and human evaluation, and y represents a value where the difference is too small as set by human evaluation. Because the accuracy of the results can vary greatly for different injection molded parts, for example, almost all data may be indistinguishable for simple small injection molded parts, while for large injection molded parts with intricate structures, mold flow analysis and engineer evaluation can differ significantly. Therefore, y should be defined based on the final dataset structure.

[0017] As a preferred method, the training method for the model in S2 also includes: S2.2 Obtains real data by manufacturing molds and testing production, collects parameters and results of actual mold manufacturing, introduces a new round of classification, and makes the prediction results given by the model closer to the actual mold manufacturing results.

[0018] The difference between the actual mold manufacturing results and the engineer's analysis results: Where Rt represents the actual test results, Rx represents the artificially generated evaluation, and X represents the process parameters, which are also features in the model; This method can be used to evaluate the accuracy of simulation results. After constructing the loss function using actual test data, it is also feasible to further optimize the scheme using deep learning techniques.

[0019] As a preferred method, the S1 dataset can be generated using the following methods: S1.1 Import the required product model into Moldflow S1.2 involves mold engineers manually generating several possible mold designs and using a machine to adjust 2-3 random parameters, resulting in a large number of different mold parameter schemes. Process parameters include melt temperature, mold temperature, injection temperature, injection time, injection pressure, cooling time, holding pressure, holding time, cooling time, mold opening and closing speed, demolding speed, as well as material selection and material parameters.

[0020] S1.3 generates analysis reports using Moldflow, transforming key data, including weld line length, bubble area, volume shrinkage rate, and shrinkage index, into quantitative evaluations of weld line issues, bubble issues, and shrinkage / warping issues. A questionnaire-based quantitative feedback method categorizes issues into multiple scoring values, from smallest to largest.

[0021] S1.4 generates a dataset by having engineers annotate and evaluate the same process parameters, including evaluations of issues such as weld lines, bubbles, and shrinkage / warpage. Similarly, these issues are categorized into multiple scoring values, from smallest to largest. S1.5 synthesizes it into a training sample dataset: Process parameters, Moldflow evaluation, engineer evaluation.

[0022] X represents the process parameters, Rm represents the Moldflow evaluation, and Rx represents the engineer evaluation.

[0023] The beneficial effects of this step are: By selecting 2-3 process parameters and randomly modifying them, a large amount of data can be added. This random modification method aims to expand the data volume and imbue the dataset with recognizable features after labeling. This approach generates a large amount of data quickly with minimal cost and a small workforce, thus shortening the model building time.

[0024] Furthermore, when the scores of S1.4 and S1.3 differ, a normalization method is used to adjust them to the same value range.

[0025] This step saves manpower and makes data import easier.

[0026] As a preferred approach, S3 uses this model to predict the difference between Moldflow simulation results and engineer annotations, and the recommended method is as follows: S3.1 A mold engineer designs a mold.

[0027] The S3.2 machine randomly modifies a subset of parameters to generate a complete set of mold design data. This modification adjusts 2-3 random parameters, generating numerous different mold parameter schemes. Process parameters include melt temperature, mold temperature, injection temperature, injection time, injection pressure, cooling time, holding pressure, holding time, cooling time, mold opening and closing speed, demolding speed, as well as material selection and material parameters.

[0028] S3.3 uses Moldflow analysis to derive evaluation results from the model and assesses the score of these results. The evaluation results include assessments of weld line issues, bubble issues, shrinkage and warping issues, etc. The calculation method for the score is as follows: , Among them, Rm k This represents the result of the k-th Moldflow simulation analysis, with a total of n evaluation results, Rd k The parameter represents the classification output by the decision tree classification prediction model, and 'a' represents the weight result affecting the evaluation. This score calculation model still has room for further optimization.

[0029] S3.4 outputs the predicted data for the mold, as well as the set of parameters with scores higher than the mold. It also outputs the parameter schemes from all machine-generated parameter data whose resulting scores (arranged from lowest to highest risk, from smallest to largest) are lower than the design scheme provided by the engineer.

[0030] As a preferred method, the data display method in S4 is detailed as follows: Through secondary development, AI analysis data was added to the analysis report provided by Moldflow. The data composition is as follows: Moldflow evaluates the potential flaws in its design; Evaluation of Moldflow's conclusions by a senior engineer's AI model; A set of parameter optimization suggestions for the existing design, and the potential drawbacks of these suggested parameters; This report has some technical reference value, providing evaluation and suggestions for engineers' designs, and also pointing out potential problems with the suggested solutions, making it an effective learning method.

[0031] This solution aims to promote the transformation of the mold design industry and accelerate the wider application of Moldflow. It enables a limited number of experienced mold designers to provide advice to more mold professionals based on Moldflow's secondary development plugins. It also addresses the shortcomings of experienced engineers struggling to adapt to Moldflow software and the lack of guidance from experienced engineers for new engineers proficient in using the software during high-end mold design processes. Detailed Implementation

[0032] The process of obtaining indexed datasets S1 imports the required product models into Moldflow. S2 allows mold engineers to manually generate several possible mold designs and then use the machine to adjust 2-3 random parameters, generating a large number of different mold parameter schemes. Process parameters include melt temperature, mold temperature, injection temperature, injection time, injection pressure, cooling time, holding pressure, holding time, cooling time, mold opening and closing speed, demolding speed, as well as material selection and material parameters.

[0033] S3 generates analysis reports using Moldflow, transforming key data such as weld line length, bubble area, volume shrinkage rate, and shrinkage index into evaluations of weld line issues, bubble issues, and shrinkage / warping issues. Using a standard questionnaire-based quantitative approach, issues are categorized into five scoring values ​​(1-5) for quantitative feedback.

[0034] S4 generates a dataset by having engineers annotate and evaluate the same process parameters, including evaluations of issues such as weld lines, bubbles, and shrinkage / warping. The issues are also categorized into five scoring values, from 1 to 5, from smallest to largest.

[0035] If the obtained score is not 1-5, it is standardized to form a score data structure of 1-5.

[0036] S5 synthesizes its training sample dataset: Process parameters, Moldflow evaluation, engineer evaluation.

[0037] X represents the process parameters, Rm represents the Moldflow evaluation, and Rx represents the engineer's evaluation. Model training methods S1 imports the dataset into the model; The dataset is formatted as follows: process parameters, Moldflow evaluation, and engineer evaluation. The comparison yielded the following results: process parameters, engineer evaluation, and Moldflow evaluation.

[0038] The scores from engineer evaluations and Moldflow evaluations correspond to the magnitude of weld line issues, bubble issues, and shrinkage / warping issues after assessment. This scheme is both a classification and a natural number; by comparing these scores, clear differences can be obtained and a dataset can be generated. X represents the process parameters, which are also features in the model. Rx-Rm represents the difference between the results of manual and Moldflow simulation analysis.

[0039] S2 builds the model; S21 randomly filters out data whose difference |Rx-Rm| is less than 1, so that its proportion of the total data items is less than 50%.

[0040] S2.2 Based on the magnitude of the bias, nine categories can be obtained under the 1-5 scoring system: {-4, -3, -2, -1, 0, 1, 2, 3, 4}. A decision tree classification prediction model is then established using the dataset.

[0041] S2.4 obtains real data by manufacturing molds and testing production, collects parameters and results of actual mold manufacturing, introduces a new round of classification, and makes the prediction results given by the model closer to the actual mold manufacturing results.

[0042] Differences between actual situation and engineer's analysis results: Where Rt represents the actual test results, and X represents the process parameters, which are also features in the model.

[0043] Model Implementation Methods An S1 mold engineer designs a mold.

[0044] The S2 machine randomly modifies a subset of parameters to generate a complete set of mold design data. This modification adjusts 2-3 random parameters, generating numerous different mold parameter schemes. Process parameters include melt temperature, mold temperature, injection temperature, injection time, injection pressure, cooling time, holding pressure, holding time, cooling time, mold opening and closing speed, demolding speed, as well as material selection and material parameters.

[0045] S3 uses Moldflow analysis to derive evaluation results from the model and assesses the score of these results. The evaluation results include assessments of weld line issues, bubble issues, shrinkage and warping issues, etc. The score calculation method is as follows: , Among them, Rm k This represents the result of the k-th Moldflow simulation analysis, with a total of n evaluation results, Rd k The parameter represents the classification output by the decision tree classification prediction model, and 'a' represents the weight result that affects the evaluation.

[0046] S4 outputs the predicted data for the mold, as well as the set of parameters with scores higher than the mold. It also outputs the parameter schemes with scores (arranged from low to high risk, from smallest to largest) lower than the engineer-provided design scheme from all machine-generated parameter data.

[0047] Information display Through secondary development, AI analysis data was added to the analysis report provided by Moldflow. The data composition is as follows: Moldflow evaluates the potential flaws in its design; Evaluation of Moldflow's conclusions by a senior engineer's AI model; A set of parameter optimization suggestions for existing designs, and the potential drawbacks of these suggested parameters.

Claims

1. A method for generating Moldflow parameter modification suggestions, characterized in that: S1, Import the dataset into the model; The dataset is formatted as follows: process parameters, Moldflow evaluation, and engineer evaluation. The comparison yielded the following results: process parameters, engineer evaluation, and Moldflow evaluation. The scores from engineer evaluations and Moldflow evaluations correspond to the severity of quality issues in the mold products, and also conform to the laws of natural numbers. By comparing these scores, clear differences can be obtained and a dataset can be generated. X represents the process parameter, which is also a feature in the model; Rm represents the Moldflow evaluation score; Rx represents the engineer evaluation score; and Rx-Rm represents the difference between the engineer evaluation score and the Moldflow evaluation score. S2, using this dataset, build a decision tree classification prediction model for each different problem evaluation; S3 uses this model to predict the difference between Moldflow simulation results and engineer annotations, and provides suggestions, including: generating mold design data based on randomly modified parameters; obtaining evaluation results and scoring through Moldflow analysis; and finally outputting the predicted data for the mold. S4, Present the data, which includes Moldflow's evaluation of potential defects in its design, the senior engineer's AI model's evaluation of Moldflow's conclusions, parameter optimization suggestions for the existing design, and potential defects in these suggested parameters.

2. The method for generating Moldflow parameter modification suggestions according to claim 1, characterized in that, In S2, multiple classifications are generated through quantification under a scoring system based on the magnitude of the bias. A decision tree classification prediction model is then established using the dataset.

3. The method for generating Moldflow parameter modification suggestions according to claim 2, characterized in that, S2 also includes: S2.1 Randomly remove evaluation data with excessively small differences, so that its proportion of the total data is less than 50%; The filtering criteria are: Rx-Rm represents the difference between the engineer's evaluation and the Moldflow evaluation scores, and y represents the manually set value where the difference is too small.

4. The method for generating Moldflow parameter modification suggestions according to claim 3, characterized in that, S2 also includes: S2.

2. Obtain real data by manufacturing molds and testing production, collect parameters and results of actual mold manufacturing, introduce a new round of classification, and make the prediction results given by the model closer to the actual mold manufacturing results; The difference between the actual mold manufacturing results and the engineer's analysis results: Where Rt represents the actual test result, Rx represents the engineer's evaluation score, and X represents the process parameters, which are also features in the model. This method can evaluate whether the simulation conclusions are accurate.

5. The method for generating Moldflow parameter modification suggestions according to claim 1, characterized in that, The methods for generating the S1 dataset include: S1.

1. Import the required product models into Moldflow S1.

2. Several possible mold designs are manually generated by the mold engineer, and 2-3 random parameters are adjusted by machine to generate a large number of different mold parameter schemes; S1.

3. Generate analysis reports using Moldflow to quantify and evaluate key data; The questionnaire method is used to quantify the feedback by dividing the questions into multiple scoring values ​​from small to large. S1.

4. Engineers annotate and evaluate the same process parameters, and the problems are divided into multiple scoring values ​​from small to large to obtain a quantitative evaluation; S1.

5. Synthesize them into a training sample dataset: Process parameters, Moldflow evaluation, engineer evaluation; X represents the process parameters, Rm represents the Moldflow evaluation, and Rx represents the engineer evaluation.

6. The method for generating Moldflow parameter modification suggestions according to claim 5, characterized in that, The scoring values ​​for S1.4 are different from those for S1.

3. A normalization method is used to adjust them to the same value range.

7. The method for generating Moldflow parameter modification suggestions according to claim 1, characterized in that, The method used in S3 to predict the difference between Moldflow simulation results and engineer annotations using this model is as follows: S3.

1. A mold engineer designs a mold; S3.

2. The machine randomly modifies a portion of the parameters to generate a complete set of mold design data; this modification adjusts 2-3 random parameters to generate a large number of different mold parameter schemes; Process parameters include melt temperature, mold temperature, injection temperature, injection time, injection pressure, cooling time, holding pressure, holding time, mold opening and closing speed, demolding speed, as well as material selection and material parameters; S3.

3. Through Moldflow analysis, the model derives the output evaluation results and assesses the result scores; The evaluation results include: weld line issues, bubble issues, and shrinkage / warping issues; The method for calculating the resulting score is as follows: ， Among them, Rm k This represents the result of the k-th Moldflow simulation analysis, with a total of n evaluation results, Rd k The parameter represents the classification output by the decision tree classification prediction model, and 'a' represents the weight result affecting the evaluation. S3.4 Outputs the parameter schemes whose result scores are lower than the design schemes provided by the engineer, among all the parameter data generated by the machine. The result scores are arranged from low risk to high risk, from smallest to largest.

8. The method for generating Moldflow parameter modification suggestions according to claim 7, characterized in that, The methods for displaying data in S4 are detailed as follows: Through secondary development, AI analysis data was added to the analysis report provided by Moldflow. The data composition is as follows: Moldflow evaluates the potential flaws in its design; Evaluation of Moldflow's conclusions by a senior engineer's AI model; Parameter optimization suggestions for the existing design, and the potential drawbacks of these suggested parameters.

Images