Intelligent generation method for DFM analysis report of plastic part
By using an intelligent DFM analysis method for plastic parts, the problems of low efficiency and limited accuracy of traditional DFM analysis have been solved. It has achieved fully automated generation of plastic part analysis reports, improved analysis efficiency and accuracy, and realized systematic management of manufacturing knowledge.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SINO HLDG GRP
- Filing Date
- 2026-01-08
- Publication Date
- 2026-05-01
AI Technical Summary
Traditional DFM analysis of plastic parts relies on manual work, which is inefficient, has limited accuracy, and results in fragmented knowledge. It cannot adapt to the needs of rapid iteration, and existing tools cannot handle complex features and new materials and processes.
By employing a data preprocessing module, a feature intelligent recognition module, a manufacturing feasibility reasoning module, and a report generation module, combined with a knowledge graph and reasoning engine, the DFM analysis of plastic parts is fully automated, including feature recognition, defect warning, and intelligent generation of optimization suggestions.
It has achieved full automation of DFM analysis of plastic parts, improved analysis efficiency and accuracy, and enabled the systematic accumulation and reuse of manufacturing knowledge.
Smart Images
Figure CN121960397A_ABST
Abstract
Description
Intelligent Generation Method for DFM Analysis Reports of Plastic Parts Technical Field
[0001] This invention belongs to the field of plastic part processing technology and relates to an intelligent method for generating DFM analysis reports for plastic parts. Background Technology
[0002] DFM analysis for plastic parts is a crucial link between product design and manufacturing. Its core is to assess the geometric features, material properties, and process parameters of the plastic part to predict potential defects (such as shrinkage marks, warpage, weld lines, etc.) during manufacturing and propose optimization suggestions to reduce production costs and shorten cycle times. Traditional DFM analysis relies on manual labor, which has the following problems: 1. Low efficiency: Manually analyzing 3D models and matching manufacturing rules takes several days to weeks, making it difficult to adapt to rapid iteration requirements; 2. Limited accuracy: Relying on engineer experience, the evaluation of complex features (such as irregular curved surfaces and multi-cavity structures) is prone to subjective bias; 3. Fragmented knowledge: Manufacturing rules and defect cases accumulated by enterprises are scattered in documents or expert experience, making them difficult to reuse systematically.
[0003] In existing technologies, rule-based DFM tools can only handle simple features and cannot dynamically adapt to new materials and processes; some methods that introduce machine learning do not form an end-to-end report generation loop and lack structured integration of analysis results. Therefore, there is an urgent need for a method that integrates multi-source knowledge and intelligent algorithms to automatically generate DFM analysis reports throughout the entire process. Summary of the Invention
[0004] The purpose of this invention is to provide an intelligent method for generating DFM analysis reports for plastic parts, so as to solve the problems mentioned in the background art.
[0005] The objective of this invention can be achieved through the following technical solution: A method for intelligently generating DFM analysis reports for plastic parts, comprising a data preprocessing module, a feature intelligent recognition module, a manufacturing feasibility analysis module, a report structure generation module, and a user feedback and model iteration module. Specifically, it includes the following steps: S1: Converting the input plastic part design file into analyzable structured data, including geometric parameters, material properties, and process constraints; S2: Identifying key manufacturing features in the plastic part to provide a basis for subsequent feasibility analysis; S3: Based on the identified features and preset rules, reasoning about manufacturing feasibility and generating optimization suggestions, the core of which is constructing a "feature-defect-process" knowledge graph and reasoning engine; S4: Integrating the reasoning results into a standardized report, including a feature list, defect warnings, optimization suggestions, and recommended process parameters; S5: Collecting user feedback on the report for optimization of the feature recognition model and knowledge graph.
[0006] In the above-mentioned intelligent generation method for DFM analysis reports of plastic parts, the plastic part design file grid in step S1 is one of STEP, STL, or IGES format.
[0007] In the above-mentioned intelligent generation method for DFM analysis reports of plastic parts, the plastic part design file grid in step S1 is one of STEP, STL, or IGES format.
[0008] In the above-mentioned intelligent generation method for DFM analysis reports of plastic parts, the conversion of the plastic part design file into analyzable structured data in step S1 specifically includes the following steps: Step 1: 3D model analysis. To adapt to slender features such as thin-walled / undercut plastics, this method no longer limits itself to spline / parametric surface reconstruction, but instead analyzes the 3D mesh points. Set estimation of the signed distance function of two fields: interior field Outdoor Poisson implicit reconstruction with normal constraints yields the following: The topological structure and geometric parameters of the model are extracted using a non-uniform rational B-spline (NURBS) surface reconstruction algorithm.
[0009] Wall thickness calculation: Offset analysis is performed on the internal cavity and external surface of the plastic part. The wall thickness (t(p)) at any point is calculated using the following formula, where p is the point cloud normal and q is the balance coefficient. This implicit representation directly serves the stable calculation of subsequent quantitative indicators such as thickness and draft. t(p) = \min_{q \in \partial \Omega_{in}} | p - q |2 \quad \text{where p\in\partial\Omega{out ((\partial \Omega_{out}) is the external surface of the plastic part, (\partial \Omega_{in}) is the internal surface, and (|\cdot|_2) is the Euclidean distance). Second step: Data cleaning. Preprocessing uses anisotropic bilateral manifold denoising and Laplacian smoothing algorithm to remove model noise, v_i' = v_i + \lambda \sum_{j \in N(i)} (v_j - v_i)]((v_i) Let (v_i) be the original vertex coordinates, (v_i') be the smoothed coordinates, (N(i)) be the neighborhood of vertex (i), and (\lambda) be the smoothing coefficient, ranging from 0.01 to 0.1.
[0010] in Let be the suppression function for high curvature. Let be the principal curvature norm of the vertex.
[0011] In the above-mentioned intelligent generation method for DFM analysis reports of plastic parts, the identification of key manufacturing features in the plastic part in step S2 is specifically as follows: First step: preprocessed three-dimensional model voxel data; Second step: network structure, including 4 convolutional-residual blocks (Conv+Residual) and 2 upsampling blocks, and strengthening the weight of key features through attention mechanism (CBAM); Third step: loss function, using a hybrid loss function (L), which combines position error, category error and confidence error: [L = L_{loc} + \alpha L_{cls} + \beta L_{conf}].
[0012] In the above-mentioned intelligent generation method for DFM analysis reports of plastic parts, the positional error loss in the third step is: (L_{loc}=\sum_{i=1}^N\sum_{k=1}^K(x_{i,k}-\hat{x}{i,k})^2+(y{i,k}-\hat{y}{i,k})^2+(\sqrt{w{i,k}}-\sqrt{\hat{w}{i,k}})^2+(\sqrt{h{i,k}}-\sqrt{\hat{h}_{i,k}})^2)(x,y,w,h) are the feature box coordinates and dimensions, and (\hat{x},\hat{y},\hat{w},\hat{h}) are the predicted values); the category error loss is: (L_{cls} = -\sum_{i=1}^N \sum_{k=1}^K \hat{c}{i,k}\log(c{i,k})) ((c_{i,k}) is the class probability, (\hat{c}_{i,k}) is the true label); the confidence error loss is: (L_{conf}=-\sum_{i=1}^N \sum_{k=1}^K (\hat{b}{i,k} \log(b{i,k}) + (1-\hat{b}{i,k}) \log(1-b{i,k}))) ((b_{i,k}) is the feature existence confidence), with weight coefficients (\alpha=1.2), (\beta=0.8).
[0013] In the above-mentioned intelligent generation method for DFM analysis reports of plastic parts, the specific steps for constructing the "feature-defect-process" knowledge graph and reasoning engine in step S3 are as follows: First step: knowledge graph construction; Second step: hybrid reasoning algorithm, specifically including rule reasoning, using deterministic rule matching for structured features; Case reasoning: for complex features, obtaining solutions through case similarity matching.
[0014] In the above-mentioned intelligent generation method for DFM analysis reports of plastic parts, the step of integrating the reasoning results into a standardized report is as follows: Step 1: Natural language generation, using the Transformer model to convert structured data into natural language descriptions, the output probability distribution formula of the decoder layer is: [P(y_t|y_{1:t-1},x)=\text{softmax}\left(W_o\cdot\text{FFN}\left(\text{MultiHead}(Q, K, V)\right)+ b_o \right)] ((y_t) is the (t)th output word, (x) is the input feature, (W_o, b_o) are the output layer parameters, FFN is the feedforward network, and MultiHead is the multi-head attention mechanism); Step 2: Report template generation, specifically including basic information of the plastic part, feature recognition results, defect risk assessment, optimization suggestions, and recommended process parameters.
[0015] In the above-mentioned intelligent generation method for DFM analysis reports of plastic parts, the model update of the user feedback and model iteration module adopts an incremental learning algorithm to fine-tune network parameters with feedback data while retaining historical knowledge, thereby avoiding catastrophic forgetting.
[0016] Compared with existing technologies, the advantages of the intelligent generation method for DFM analysis reports of plastic parts in this invention are as follows: This intelligent generation method for DFM analysis reports of plastic parts realizes full automation from input of three-dimensional model of plastic part to output of structured analysis report, improves analysis efficiency and accuracy, and realizes systematic accumulation and reuse of manufacturing knowledge. Attached Figure Description
[0018] Figure 1 is a flowchart of the intelligent generation method for DFM analysis reports of plastic parts according to the present invention. Detailed Implementation
[0020] The following are specific embodiments of the present invention, which are described in conjunction with the accompanying drawings. However, the present invention is not limited to these embodiments.
[0021] As shown in Figure 1, the intelligent generation method for DFM analysis reports of plastic parts according to the present invention includes a data preprocessing module, a feature intelligent recognition module, a manufacturing feasibility push module, a report structure generation module, and a user feedback and model iteration module. Its key features include the following steps: S1: Converting the input plastic part design file into analyzable structured data, including geometric parameters, material properties, and process constraints; S2: Identifying key manufacturing features in the plastic part to provide a basis for subsequent feasibility analysis; S3: Based on the identified features and preset rules, reasoning about manufacturing feasibility and generating optimization suggestions, the core of which is constructing a "feature-defect-process" knowledge graph and reasoning engine; S4: Integrating the reasoning results into a standardized report, including a feature list, defect warnings, optimization suggestions, and recommended process parameters; S5: Collecting user feedback on the report for optimization of the feature recognition model and knowledge graph.
[0022] Specifically, this intelligent method for generating DFM analysis reports for plastic parts achieves full automation from inputting the 3D model of the plastic part to outputting a structured analysis report, improving analysis efficiency and accuracy, while also enabling the systematic accumulation and reuse of manufacturing knowledge.
[0023] As shown in Figure 1, the intelligent generation method for DFM analysis reports of plastic parts in this invention, in step S1, the plastic part design file grid is in one of STEP, STL, or IGES formats. The conversion of the plastic part design file into analyzable structured data in step S1 specifically includes the following steps: Step 1: 3D model analysis. To adapt to slender features such as thin-walled / undercut plastics, this method no longer limits the use of spline / parametric surface reconstruction, but instead analyzes the 3D mesh points. Set estimation of the signed distance function of two fields: interior field Outdoor Poisson implicit reconstruction with normal constraints yields the following: The topological structure and geometric parameters of the model are extracted using a non-uniform rational B-spline (NURBS) surface reconstruction algorithm.
[0024] Wall thickness calculation: Offset analysis is performed on the internal cavity and external surface of the plastic part. The wall thickness (t(p)) at any point is calculated using the following formula, where p is the point cloud normal and q is the balance coefficient. This implicit representation directly serves the stable calculation of subsequent quantitative indicators such as thickness and draft. t(p) = \min_{q \in \partial \Omega_{in}} | p - q |2 \quad \text{where p\in \partial \Omega_{out}]((\partial \Omega_{out}) is the external surface of the plastic part, (\partial \Omega_{in}) is the internal surface, and (|\cdot|_2) is the Euclidean distance); Step 2: Data cleaning. Preprocessing uses anisotropic bilateral manifold denoising and Laplacian smoothing algorithm to remove model noise, v_i' = v_i + \lambda \sum_{j \in N(i)} (v_j - v_i)]((v_i Let (v_i) be the original vertex coordinates, (v_i') be the smoothed coordinates, (N(i)) be the neighborhood of vertex (i), and (\lambda) be the smoothing coefficient, ranging from 0.01 to 0.1.
[0025] in Let be the suppression function for high curvature. Let be the principal curvature norm of the vertex.
[0026] Specifically, this method considers both spatial proximity and normal similarity weights to maintain denoising and prevent thin features from being flattened.
[0027] As shown in Figure 1, the intelligent generation method for DFM analysis reports of plastic parts of the present invention, the identification of key manufacturing features in the plastic part in step S2 is specifically as follows: First step: preprocessed three-dimensional model voxel data; Second step: network structure, including 4 convolutional-residual blocks (Conv+Residual) and 2 upsampling blocks, and strengthening the weight of key features through attention mechanism (CBAM); Third step: loss function, adopting a hybrid loss function (L), which combines position error, category error and confidence error: [L = L_{loc} + \alpha L_{cls} + \beta L_{conf}].
[0028] The positional error loss in the third step is: (L_{loc}=\sum_{i=1}^N\sum_{k=1}^K(x_{i,k}-\hat{x}{i,k})^2+(y{i,k}-\hat{y}{i,k})^2+(\sqrt{w{i,k}}-\sqrt{\hat{w}{i,k}})^2+(\sqrt{h{i,k}}-\sqrt{\hat{h}_{i,k}})^2)(x,y,w,h) are the feature box coordinates and dimensions, and (\hat{x},\hat{y},\hat{w},\hat{h}) are the predicted values); the category error loss is: (L_{cls} = -\sum_{i=1}^N \sum_{k=1}^K \hat{c}{i,k}\log(c{i,k})) ((c_{i,k}) is the class probability, (\hat{c}_{i,k}) is the true label); the confidence error loss is: (L_{conf}=-\sum_{i=1}^N \sum_{k=1}^K (\hat{b}{i,k} \log(b{i,k}) + (1-\hat{b}{i,k}) \log(1-b{i,k}))) ((b_{i,k}) is the feature existence confidence), with weight coefficients (\alpha=1.2), (\beta=0.8).
[0029] The specific steps for constructing the "feature-defect-process" knowledge graph and reasoning engine in step S3 are as follows: First step: knowledge graph construction; Second step: hybrid reasoning algorithm, specifically including rule reasoning, which uses deterministic rule matching for structured features; Case reasoning: for complex features, solutions are obtained through case similarity matching.
[0030] Specifically, the construction of the knowledge graph is as follows: knowledge is stored in the form of triples (entity, relation, entity); rule-based reasoning is: [\text{Defect}(warpage) = \begin{cases}1&\text{if} h / t>5\0&\text{else}\end{cases}]; case-based reasoning is: case similarity (S) is calculated as follows: [S(C_i, C_t) = \sum_{k=1}^M w_k\cdot \text{Sim}(f_{i,k}, f_{t,k})] ((C_i) is the historical case, (C_t) is the target case, (f_{i,k}) is the (k)th feature value, (w_k) is the feature weight, and (\text{Sim}(\cdot)) is the cosine similarity).
[0031] The steps to integrate the inference results into a standardized report are as follows: Step 1: Natural language generation. The Transformer model is used to convert structured data into natural language descriptions. The output probability distribution formula of the decoder layer is: [P(y_t |y_{1:t-1},x)=\text{softmax}\left(W_o\cdot\text{FFN}\left(\text{MultiHead}(Q, K, V)\right)+ b_o \right)] ((y_t) is the (t)th output word, (x) is the input feature, (W_o, b_o) are the output layer parameters, FFN is the feedforward network, and MultiHead is the multi-head attention mechanism); Step 2: Report template generation. Specifically, it includes basic information of the plastic part, feature recognition results, defect risk assessment, optimization suggestions, and recommended process parameters.
[0032] Model updates in the user feedback and model iteration module: An incremental learning algorithm is used to fine-tune network parameters with feedback data while retaining historical knowledge, thus avoiding catastrophic forgetting.
[0033] Contents not described in detail herein are existing technologies known to those skilled in the art. The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art can make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.
Claims
1. A method for intelligently generating DFM analysis reports for plastic parts, comprising a data preprocessing module, a feature intelligent recognition module, a manufacturing feasibility push module, a report structure generation module, and a user feedback and model iteration module, characterized in that, Specifically, the following steps are included: S1: Transform the input plastic part design file into analyzable structured data, including geometric parameters, material properties, and process constraints; S2: Identify key manufacturing features in the plastic part to provide a basis for subsequent feasibility analysis; S3: Based on the identified features and preset rules, infer manufacturing feasibility and generate optimization suggestions. The core is to build a "feature-defect-process" knowledge graph and inference engine. S4: Integrate the reasoning results into a standardized report, including a feature list, defect warnings, optimization suggestions, and process parameter recommendations; S5: Collect user feedback on the report to optimize the feature recognition model and knowledge graph.
2. The intelligent generation method for DFM analysis reports of plastic parts according to claim 1, characterized in that, The plastic part design file grid in step S1 is in one of the following formats: STEP, STL, or IGES.
3. The intelligent generation method for DFM analysis reports of plastic parts according to claim 1, characterized in that, The process of converting the plastic part design file into analyzable structured data in step S1 is specifically... The steps include: Step 1: 3D model analysis. To adapt to slender features such as thin-walled / inverted plastics, this method no longer limits itself to spline / parametric surface reconstruction, but instead analyzes the 3D mesh points. Set estimation of the signed distance function of two fields: interior field Outdoor Poisson implicit reconstruction with normal constraints yields the following: The topological structure and geometric parameters of the model are extracted using a non-uniform rational B-spline (NURBS) surface reconstruction algorithm. Wall thickness calculation: Offset analysis is performed on the internal cavity and external surface of the plastic part. The wall thickness (t(p)) at any point is calculated using the following formula, where p is the point cloud normal and q is the balance coefficient. This implicit representation directly serves the stable calculation of subsequent quantitative indicators such as thickness and draft. t(p) = \min_{q \in \partial \Omega_{in}} | p - q |2 \quad \text{where p\in \partial \Omega_{out}]((\partial \Omega_{out}) is the external surface of the plastic part, (\partial \Omega_{in}) is the internal surface, and (|\cdot|_2) is the Euclidean distance); Step 2: Data cleaning. Preprocessing uses anisotropic bilateral manifold denoising and Laplacian smoothing algorithm to remove model noise, v_i' = v_i + \lambda \sum_{j \in N(i)} (v_j - v_i)]((v_i) Let (v_i) be the original vertex coordinates, (v_i') be the smoothed coordinates, (N(i)) be the neighborhood of vertex (i), and (\lambda) be the smoothing coefficient, ranging from 0.01 to 0.
1. ;in Let be the suppression function for high curvature. Let be the principal curvature norm of the vertex.
4. The intelligent generation method for DFM analysis reports of plastic parts according to claim 1, characterized in that, The key manufacturing features identified in step S2 of the plastic part are as follows: First step: preprocessed voxelized data of the 3D model; Second step: network structure, including 4 convolutional-residual blocks (Conv+Residual) and 2 upsampling blocks, and key feature weights are enhanced through attention mechanism (CBAM); Step 3: Loss function, using a hybrid loss function (L) that combines location error, class error, and confidence error: [L = L_{loc} + \alpha L_{cls} + \beta L_{conf}].
5. The intelligent generation method for DFM analysis reports of plastic parts according to claim 4, characterized in that, The positional error loss in the third step is: (L_{loc}=\sum_{i=1}^N\sum_{k=1}^K(x_{i,k}-\hat{x}{i,k})^2+(y{i,k}-\hat{y}{i,k})^2+(\sqrt{w{i,k}}-\sqrt{\hat{w}{i,k}})^2+(\sqrt{h{i,k}}-\sqrt{\hat{h}_{i,k}})^2)(x,y,w,h) are the feature box coordinates and dimensions, and (\hat{x},\hat{y},\hat{w},\hat{h}) are the predicted values); the category error loss is: (L_{cls} = -\sum_{i=1}^N \sum_{k=1}^K \hat{c}{i,k}\log(c{i,k})) ((c_{i,k}) is the class probability, (\hat{c}_{i,k}) is the true label); the confidence error loss is: (L_{conf}=-\sum_{i=1}^N \sum_{k=1}^K (\hat{b}{i,k} \log(b{i,k}) + (1-\hat{b}{i,k}) \log(1-b{i,k}))) ((b_{i,k}) is the feature existence confidence), with weight coefficients (\alpha=1.2), (\beta=0.8).
6. The intelligent generation method for DFM analysis reports of plastic parts according to claim 1, characterized in that, The specific steps for constructing the "feature-defect-process" knowledge graph and reasoning engine in step S3 are as follows: First step: knowledge graph construction; Second step: hybrid reasoning algorithm, specifically including rule reasoning, which uses deterministic rule matching for structured features; Case reasoning: for complex features, solutions are obtained through case similarity matching.
7. The intelligent generation method for DFM analysis reports of plastic parts according to claim 1, characterized in that, The steps for integrating the inference results into a standardized report are as follows: Step 1: Natural language generation. The Transformer model is used to convert structured data into natural language descriptions. The output probability distribution formula of the decoder layer is: [P(y_t |y_{1:t-1},x)=\text{softmax}\left(W_o\cdot\text{FFN}\left(\text{MultiHead}(Q, K, V)\right)+ b_o \right)] ((y_t) is the (t)th output word, (x) is the input feature, (W_o, b_o) are the output layer parameters, FFN is the feedforward network, and MultiHead is the multi-head attention mechanism); Step 2: Report template generation. Specifically, it includes basic information of the plastic part, feature recognition results, defect risk assessment, optimization suggestions, and recommended process parameters.
8. The intelligent generation method for DFM analysis reports of plastic parts according to claim 1, characterized in that, The model update of the user feedback and model iteration module adopts an incremental learning algorithm to fine-tune network parameters with feedback data while retaining historical knowledge, thus avoiding catastrophic forgetting.