Semi-solid forming parameter optimization method and system for complex inner cavity of automobile universal joint
By constructing a semi-solid forming parameter optimization method and system for the complex internal cavity of automotive universal joints, and combining CAE simulation and knowledge graph, intelligent optimization and closed-loop control of process parameters are realized. This solves the problems of fragmented process knowledge and disconnected control modes in existing technologies, and improves product quality stability and forming quality.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-09
- Publication Date
- 2026-04-07
AI Technical Summary
In the existing technology, during the semi-solid forming process of the complex internal cavity of automotive universal joints, CAE simulation data and expert experience are in a "fragmented" state, which makes it impossible to systematically inherit process knowledge. Moreover, the control mode is biased towards "offline optimization and online execution", which cannot adapt to the fluctuation of working conditions in the production process, resulting in unstable product quality and high scrap rate.
A cavity model was constructed using 3D CAD software, and the slurry filling and solidification process was simulated using CAE simulation software. An initial parameter database was built, and principal component analysis was used to extract the dominant process parameters. A semi-solid forming process knowledge graph was constructed, and a graph neural network model was trained. The injection speed and holding pressure were adjusted in real time by monitoring data through sensors to achieve closed-loop control.
It enables intelligent optimization of process parameters, solves the problems of blind selection of process parameters and difficulty in fully releasing the value of CAE data, breaks down the "fragmented" barrier of process knowledge, realizes proactive intervention in quality risks, and improves product quality stability and forming quality.
Smart Images
Figure CN121808968A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of semi-solid forming parameter optimization for universal joints, specifically to a method and system for optimizing semi-solid forming parameters of complex internal cavities in automotive universal joints. Background Technology
[0002] Semi-solid forming technology, a near-net-shape forming process between liquid casting and solid forging, has shown great potential in the manufacturing of complex components in the automotive and aerospace industries due to its advantages such as stable filling, dense structure, and fewer forming defects. Automotive universal joints, as a key component of the transmission system, have complex internal structures and uneven wall thicknesses, requiring extremely high forming quality. Traditional semi-solid forming process development heavily relies on a "trial and error" approach, involving repeated modifications to mold design and process parameters through physical experiments. This method is time-consuming, costly, and struggles to deeply understand the complex intrinsic relationship between process parameters and forming quality. With the development of numerical simulation technology, Computer-Aided Engineering (CAE) technology has been widely applied to semi-solid forming processes. In process simulation, the industry typically uses 3D CAD software (such as SolidWorks and UG) to model the cavity and uses dedicated CAE software (such as AnyCasting and ProCAST) to simulate the filling, solidification, and defect formation process of semi-solid slurry. By setting different combinations of process parameters (such as injection speed, mold temperature, and holding pressure), CAE simulation can predict defects such as filling rate, stress distribution, and shrinkage cavities to a certain extent, thereby reducing the number of physical mold trials. To further optimize process parameters, researchers often use Design of Experiments (DOE) methods, such as the Taguchi method, to systematically arrange simulation experiments to analyze the significance of the impact of each parameter on the quality target and select key process parameters accordingly.
[0003] In recent years, data-driven approaches have provided new ideas for process optimization. Some studies have attempted to use historical data to build statistical or machine learning models (such as neural networks and support vector machines) between process parameters and quality indicators in order to find the optimal parameter window. In addition, knowledge graphs, as an emerging knowledge representation and reasoning technology, have begun to be explored in the manufacturing industry to structure and associate diverse information such as process knowledge and defect patterns, providing support for process decision-making.
[0004] Existing technologies for constructing complex internal cavities in automotive universal joints primarily focus on using 3D CAD software to build 3D models of these cavities and then using analysis software to assess their stability and quality. The analysis results are then combined with semi-solid forming technology. However, these technologies suffer from a lack of effective integration of complex relationships between process parameters, quality indicators, and defect types. CAE simulation data and expert experience are fragmented, hindering the systematic transmission of process knowledge. Furthermore, the technology's preference for an open-loop control mode of "offline optimization, online execution" is severely disconnected from the real-time dynamic forming process. This results in a fixed parameter set that cannot adapt to fluctuations in production conditions, preventing proactive intervention in quality risks and leading to unstable product quality and high scrap rates. Summary of the Invention
[0005] To address the aforementioned technical problems, a semi-solid forming parameter optimization method and system for the complex internal cavity of automotive universal joints is provided. This technical solution solves the problem mentioned in the background technology that CAE simulation data and expert experience are "fragmented," resulting in the inability to systematically inherit process knowledge. At the same time, the technology's preference for an open-loop control mode of "offline optimization and online execution" is seriously disconnected from the real-time dynamic forming process, causing the fixed parameter set to be unable to adapt to the fluctuations in the working conditions during the production process and unable to proactively intervene in quality risks, resulting in unstable product quality and a high scrap rate.
[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:
[0007] A semi-solid forming parameter optimization method for complex internal cavities of automotive universal joints includes:
[0008] Based on the structural design drawings of the automotive universal joint, a cavity model including the gating system, venting groove, overflow groove and core was constructed using 3D CAD software.
[0009] Based on the cavity model, the filling and solidification process of semi-solid slurry is simulated using CAE simulation software to obtain multiple sets of process parameter combinations and their corresponding forming quality results, forming an initial parameter database.
[0010] Based on multiple combinations of process parameters in the initial parameter database, a set of dominant process parameters is constructed using principal component analysis.
[0011] A knowledge graph of semi-solid forming process is constructed, with dominant process parameters, forming quality results and defect types as nodes and physical and statistical correlations between parameters as edges.
[0012] Based on the knowledge graph of semi-solid forming process and the initial parameter database, a graph neural network model is trained to output the predicted values of cavity filling rate, maximum stress and shrinkage defect probability in the future.
[0013] A sensor array is set up to collect monitoring data of the semi-solid forming of the cavity and input it into a trained graph neural network model. Based on the prediction results, the injection speed and / or holding pressure are dynamically adjusted through a PID control algorithm.
[0014] Preferably, the step of simulating the filling and solidification process of semi-solid slurry using CAE simulation software based on the cavity model to obtain multiple sets of process parameter combinations and their corresponding forming quality results, forming an initial parameter database, specifically includes:
[0015] Import the cavity model into the CAE simulation software;
[0016] The constitutive model of the semi-solid slurry is set as the Kaufmann non-Newtonian fluid model, and the relationship expression between its apparent viscosity and solid fraction and shear rate is constructed.
[0017] Set the initial value range for slurry parameters, process parameters, and mold / material parameters;
[0018] Based on the Taguchi method in experimental design, an L25 orthogonal experimental table was designed, with solid fraction, injection speed and mold temperature as control factors. Each factor was set with 5 levels to simulate the filling and solidification process.
[0019] Record the filling rate, maximum internal stress, shrinkage cavity defect volume fraction, and deformation in the simulation results of each set of parameters as the forming quality results;
[0020] Integrate all process parameter combinations and their corresponding forming quality results to form an initial parameter database containing both qualified and defective cases.
[0021] Preferably, the construction of a semi-solid forming process knowledge graph, using dominant process parameters, forming quality results, and defect types as nodes and physical and statistical correlations between parameters as edges, specifically includes:
[0022] The dominant process parameters, forming quality results, and typical defect types are defined as nodes in the knowledge graph;
[0023] A multi-index hybrid strategy is used to construct statistical correlation edges between nodes, and the correlation coefficient between any two dominant process parameter nodes is calculated.
[0024] Based on CAE simulation results and expert experience rules, semantic relationship edges are established between each process parameter node pair. The semantic relationship edge types include at least: causing, inhibiting, influencing, and belonging.
[0025] Based on the Apriori algorithm, we mine the association rules between process parameter combinations and defect types, set the minimum support to s4% and the minimum confidence to s5%, and transform the association rules that meet the conditions into association relationship edges.
[0026] Based on all defined nodes and relation edges, a knowledge graph of semi-solid forming process is constructed and stored in the Neo4j graph database;
[0027] Among them, the node attributes include: process parameter name, value range, and unit; the relationship edge attributes are determined according to the relationship type: statistically related edges record their correlation coefficients, semantically related edges record their semantically related edge types, and associated edges record their confidence and support.
[0028] The multi-indicator hybrid strategy specifically includes:
[0029] Calculate the Pearson correlation coefficient between any two numerical process parameter nodes. When the absolute value of the coefficient is greater than s3, establish a strongly linear correlation edge.
[0030] Simultaneously, calculate the Spearman rank correlation coefficient between any two numerical nodes, and establish a monotonically strongly correlated edge when the absolute value of the coefficient is greater than s3.
[0031] Furthermore, this solution proposes a semi-solid forming parameter optimization system for complex internal cavities of automotive universal joints, used to implement the semi-solid forming parameter optimization method for complex internal cavities of automotive universal joints as described above, including:
[0032] Cavity model module, which is used to construct a cavity model including runner, venting groove, overflow groove and core based on the structural design drawings of automotive universal joint and 3D CAD software;
[0033] The data processing module is used to simulate the filling and solidification process of semi-solid slurry based on the cavity model using CAE simulation software, obtain multiple sets of process parameter combinations and their corresponding forming quality results, and form an initial parameter database; based on the multiple sets of process parameter combinations in the initial parameter database, a set of dominant process parameters is constructed using principal component analysis; and a semi-solid forming process knowledge graph is constructed with dominant process parameters, forming quality results, and defect types as nodes and physical correlations and statistical correlations between parameters as edges.
[0034] The parameter prediction and optimization module is used to train a graph neural network model based on the semi-solid forming process knowledge graph and initial parameter database, and output the predicted values of cavity filling rate, maximum stress and shrinkage defect probability in the future period; set up a sensor group to collect cavity semi-solid forming monitoring data and input it into the trained graph neural network model; and dynamically adjust the injection speed and / or holding pressure based on the prediction results through a PID control algorithm.
[0035] Preferably, the data processing module includes:
[0036] An initial parameter acquisition unit is used to simulate the filling and solidification process of semi-solid slurry based on the cavity model using CAE simulation software, acquire multiple sets of process parameter combinations and their corresponding forming quality results, and form an initial parameter database.
[0037] The principal component extraction unit is used to construct a set of dominant process parameters based on multiple combinations of process parameters in the initial parameter database and the principal component analysis method.
[0038] The knowledge graph unit is used to construct a semi-solid forming process knowledge graph with dominant process parameters, forming quality results and defect types as nodes and physical correlations and statistical correlations between parameters as edges.
[0039] Preferably, the parameter prediction and optimization module includes:
[0040] The parameter prediction unit is used to train a graph neural network model based on a semi-solid forming process knowledge graph and an initial parameter database, and output predicted values of cavity filling rate, maximum stress and shrinkage defect probability for a future period of time.
[0041] The parameter optimization unit is used to set the sensor group to collect semi-solid forming monitoring data of the cavity and input it into the trained graph neural network model. Based on the prediction results, the injection speed and / or holding pressure are dynamically adjusted through the PID control algorithm.
[0042] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0043] This invention provides a method and system for optimizing semi-solid forming parameters for complex internal cavities in automotive universal joints. The proposed solution employs an intelligent optimization strategy integrating a structured simulation database and a process knowledge graph. First, an L25 orthogonal experiment is designed using the Taguchi method, combined with a Kaufmann non-Newtonian fluid model to accurately simulate the filling and solidification process of semi-solid slurry. This systematically constructs an initial parameter database containing multiple sets of process parameters and their forming quality results. Based on this database, a semi-solid forming process knowledge graph is built: using dominant process parameters, forming quality results, and defect types as nodes, a multi-index hybrid strategy (Pearson and Spearman coefficients), CAE simulation, expert experience rules, and April... The i-association rule algorithm constructs a multi-relation network containing statistically relevant edges, semantically related edges, and associative edges, and stores it in the Neo4j graph database. Through deep integration of the database and knowledge graph, this solution not only solves the problems of blind selection of process parameters and difficulty in fully releasing the value of CAE data in existing technologies, but also breaks down the barrier of "fragmented" process knowledge, realizing structured storage and computable reasoning of knowledge. Even in the face of multi-variable strongly coupled scenarios of complex internal cavity forming of universal joints, this solution can provide accurate and reliable data and knowledge support for subsequent intelligent prediction and closed-loop adaptive control of graph neural networks, fundamentally overcoming the shortcomings of existing technologies that rely on static experience and cannot dynamically respond to fluctuations in operating conditions. Attached Figure Description
[0044] Figure 1 This is a flowchart of a semi-solid forming parameter optimization method for a complex internal cavity of an automotive universal joint according to the present invention.
[0045] Figure 2 To obtain multiple combinations of process parameters and their corresponding forming quality results for this invention, an initial parameter database flowchart is formed;
[0046] Figure 3 This is a flowchart illustrating the knowledge graph of the semi-solid forming process of the present invention. Detailed Implementation
[0047] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.
[0048] Reference Figure 1 As shown, a semi-solid forming parameter optimization method for complex internal cavities of automotive universal joints includes:
[0049] Based on the structural design drawings of the automotive universal joint, a cavity model including the gating system, venting groove, overflow groove and core was constructed using 3D CAD software.
[0050] Based on the cavity model, the filling and solidification process of semi-solid slurry is simulated using CAE simulation software to obtain multiple sets of process parameter combinations and their corresponding forming quality results, forming an initial parameter database.
[0051] Based on multiple combinations of process parameters in the initial parameter database, a set of dominant process parameters is constructed using principal component analysis.
[0052] A knowledge graph of semi-solid forming process is constructed, with dominant process parameters, forming quality results and defect types as nodes and physical and statistical correlations between parameters as edges.
[0053] Based on the knowledge graph of semi-solid forming process and the initial parameter database, a graph neural network model is trained to output the predicted values of cavity filling rate, maximum stress and shrinkage defect probability in the future.
[0054] A sensor array is set up to collect monitoring data of the semi-solid forming of the cavity and input it into a trained graph neural network model. Based on the prediction results, the injection speed and / or holding pressure are dynamically adjusted through a PID control algorithm.
[0055] This solution utilizes a CAE simulation module to perform multi-parameter combination simulation analysis of the semi-solid forming process of the complex internal cavity of an automotive universal joint. It then extracts a set of key process parameters affecting forming quality using principal component analysis, and constructs a process knowledge graph that integrates parameter correlations and defect patterns. Based on this knowledge graph, a graph neural network model is trained to intelligently predict cavity filling rate, maximum stress, and shrinkage cavity probability. Finally, a sensor array collects forming process data in real time, and the injection speed and holding pressure are dynamically adjusted based on the prediction results, forming a closed-loop optimization system from virtual simulation to real-time control. This transforms quality control from passive remediation to proactive intervention, effectively solving the quality instability problem caused by fluctuations in operating conditions during the semi-solid forming process, and achieving precise control and stable improvement of the forming quality of complex internal cavity components in automotive universal joints.
[0056] The construction of a cavity model, including a gating system, venting groove, overflow groove, and core, based on the structural design drawings of the automotive universal joint and using 3D CAD software, specifically includes:
[0057] Using SolidWorks 3D CAD software, a 1:1 parametric model was created based on the structural design drawings of the automotive universal joint to generate a cavity geometry model of the complex internal cavity of the automotive universal joint.
[0058] In the cavity geometry model, a gating system is constructed, which includes a sprue with a circular cross-section, a horizontal sprue with a trapezoidal cross-section, and an venting groove and an overflow groove located at the end of the cavity.
[0059] In the cavity geometry model, a detachable core model corresponding to the shape of the universal joint cavity is placed, and the fitting clearance range between the core model and the cavity is set to [a,b].
[0060] Perform virtual assembly interference checks on the completed cavity geometry model to ensure that there is no geometric interference between the core, the mold cavity, and the ejector mechanism, and export the model as an STP or IGES format.
[0061] This can be explained by using 3D CAD software such as SolidWorks, strictly adhering to the final design drawings of the universal joint product, and associating the designed parts with the mold base through cavity tools. After setting the scaling factor, 1:1 parametric modeling is performed to generate an accurate mold cavity geometric model. The core feature of this cavity geometric model is that its inner surface reproduces the final geometric shape of the universal joint forging, especially the complex internal structure of the ball cage track, providing a foundation for all subsequent simulation analyses and ensuring the geometric authenticity of the simulation. The sprue refers to a circular cross-section, serving as the initial inlet for the molten metal, with its axis typically perpendicular to the mold parting surface. The runner refers to a trapezoidal cross-section, connected to the end of the sprue and distributed on the parting surface, used to smoothly distribute the molten metal to each ingate. The inclined structure of the trapezoidal cross-section facilitates demolding and utilizes gravity to achieve slag blocking, reducing the entry of oxide inclusions into the cavity. The venting and overflow system is located at the end of the cavity or in the area where the molten metal is last filled, specifically including narrow venting channels for discharging gas from the cavity (quickly exporting air through tiny gaps). The filling process is facilitated by two components: a gas-filling system (to prevent porosity defects) and an overflow groove (to collect low-temperature metals and impurities at the front end of the slurry, improving the quality of the forging body). These components work together to ensure a smooth filling process and improve forging quality. The fit clearance between the detachable core model and the cavity geometry model is set based on the difference in thermal expansion between the alloy and core materials at semi-solid forming temperatures, such as 580℃-650℃, combined with the processing requirements for core surface roughness. Its value range [a,b] can be set to [0.05mm,0...]. [0.1mm], this range is optimized with reference to the universal joint cross shaft axial clearance design standard. This clearance is designed to ensure that: at the high temperature of forging, the mating surfaces of the core and the cavity fit tightly, effectively sealing and preventing semi-solid slurry from entering the gap and forming burrs; at the same time, after cooling, the gap is used to compensate for the shrinkage, so that the core can be smoothly pulled out from the inner cavity of the forging, avoiding jamming or damage to the surface of the forging. Among them, the detachable core model refers to the moving part in the mold used to form the complex cavity inside the product (such as the universal joint) and can be removed separately after the product is formed.
[0062] Reference Figure 2 As shown, the process of obtaining multiple sets of process parameter combinations and their corresponding forming quality results to form an initial parameter database specifically includes:
[0063] Import the cavity model into the CAE simulation software;
[0064] The constitutive model of the semi-solid slurry is set as the Kaufmann non-Newtonian fluid model, and the relationship expression between its apparent viscosity and solid fraction and shear rate is constructed.
[0065] Set the initial value range for slurry parameters, process parameters, and mold / material parameters;
[0066] Based on the Taguchi method in experimental design, an L25 orthogonal experimental table was designed, with solid fraction, injection speed and mold temperature as control factors. Each factor was set with 5 levels to simulate the filling and solidification process.
[0067] Record the filling rate, maximum internal stress, shrinkage cavity defect volume fraction, and deformation in the simulation results of each set of parameters as the forming quality results;
[0068] Integrate all process parameter combinations and their corresponding forming quality results to form an initial parameter database containing both qualified and defective cases.
[0069] This solution utilizes systematic computer simulation to replace the traditional "trial and error" method, establishing a "knowledge base"—an initial parameter database—that efficiently and cost-effectively covers the causal relationships between various process conditions and forming results. This database forms the data foundation and prerequisite for optimizing the semi-solid forming parameters of the complex internal cavity of automotive universal joints. Specifically, the initial value ranges for setting slurry parameters, process parameters, and mold / material parameters are as follows: Before conducting formal orthogonal experiments, a series of single-factor sensitivity analyses can be performed. This involves fixing other parameters and varying one parameter within a set range. Through a small number of CAE simulations, the impact on one or more key quality indicators (such as filling rate) is observed. The parameter value corresponding to the inflection point where the quality indicator begins to deteriorate sharply or tends to stabilize is determined as the reasonable lower and upper limits of that parameter. The relationship between apparent viscosity, solid fraction, and shear rate is expressed as follows:
[0070]
[0071] In the formula, The apparent viscosity of the slurry. As the reference viscosity, Solid fraction, i.e., the volume fraction of solid particles. For the maximum flowable solid fraction, The rheological index characterizes the intensity of the effect of solid fraction on viscosity. Shear rate, For reference shear rate, The shear thinning index, It is the activation energy for viscous flow. The gas constant is This refers to the actual slurry temperature. For reference temperature;
[0072] This expression considers the effects of apparent viscosity, solid fraction, shear rate, and temperature on slurry viscosity. Compared to a simple model that only considers a single factor, it can more accurately describe the complex rheological behavior of semi-solid slurry during the filling process, thereby ensuring the fidelity of the digital twin simulation and providing a reliable data foundation for subsequent optimization. The parameters in the expression ( , , , , , , The data can be obtained by fitting semi-solid rheological experimental data or directly from the validated material database built into the CAE software.
[0073] The construction of the dominant process parameter set based on multiple combinations of process parameters in the initial parameter database and using principal component analysis specifically includes:
[0074] Extract the original data matrix consisting of all numerical process parameters from the initial parameter database;
[0075] Z-score standardization is performed on the original data matrix to eliminate the influence of dimensions;
[0076] Calculate the covariance matrix of the standardized data, and solve for the eigenvalues and eigenvectors of the covariance matrix;
[0077] Sort the eigenvalues from largest to smallest, and calculate the variance contribution rate and cumulative variance contribution rate of each principal component;
[0078] Select the eigenvectors corresponding to the first k principal components whose cumulative variance contribution rate is greater than or equal to s1%;
[0079] Analyze the original process parameters whose absolute load value is greater than s2 among the first k eigenvectors, and determine them as the dominant process parameters;
[0080] Based on the selected dominant process parameters, construct a set of dominant process parameters;
[0081] The set of dominant process parameters includes at least: injection speed, holding pressure, pressure switching point, and mold temperature field uniformity.
[0082] This solution addresses the "curse of dimensionality" problem caused by the high dimensionality and strong multicollinearity of process parameters in the initial parameter database. While CAE simulation data is accurate and reliable, strong correlations between parameters (such as injection rates at different times) can lead to overfitting, computational complexity, and poor stability in subsequent prediction models. This solution uses principal component analysis (PCA) to reduce the dimensionality of several original parameters into eigenvectors corresponding to k principal components, ensuring the cumulative variance contribution rate is greater than or equal to s1% (e.g., s1=95). By analyzing the first k eigenvectors, the original process parameters with absolute load values greater than s2 (e.g., s2=0.7) are identified as the dominant process parameters (e.g., injection rate curves, holding pressure curves, etc.). These dominant parameters... The core process information carried by the previous k independent principal components is represented, thereby effectively filtering out redundant information and noise (parts with a contribution rate lower than 1-s1%) in the original parameters, extracting the core features that characterize the process rules, and significantly improving the efficiency and generalization ability of subsequent modeling while preserving the accuracy of simulation data to the maximum extent. Among them, the variance contribution rate refers to the proportion of the variance of a single principal component to the total variance of all principal components, which is used to measure the amount of information it carries. The cumulative variance contribution rate is the sum of the variance contribution rates of the first k principal components, which is used to measure the total amount of information they carry together. The loading in the eigenvector reflects the degree of correlation between the original process parameters and the principal components. The larger the absolute value of the loading, the greater the contribution of the parameter to the composition of this principal component. These parameters are the standard process quantities in principal component analysis.
[0083] Reference Figure 3 As shown, the construction of the semi-solid forming process knowledge graph specifically includes:
[0084] The dominant process parameters, forming quality results, and typical defect types are defined as nodes in the knowledge graph;
[0085] A multi-index hybrid strategy is used to construct statistical correlation edges between nodes, and the correlation coefficient between any two dominant process parameter nodes is calculated.
[0086] Based on CAE simulation results and expert experience rules, semantic relationship edges are established between each process parameter node pair. The semantic relationship edge types include at least: causing, inhibiting, influencing, and belonging.
[0087] Based on the Apriori algorithm, we mine the association rules between process parameter combinations and defect types, set the minimum support to s4% and the minimum confidence to s5%, and transform the association rules that meet the conditions into association relationship edges.
[0088] Based on all defined nodes and relation edges, a knowledge graph of semi-solid forming process is constructed and stored in the Neo4j graph database;
[0089] Among them, the node attributes include: process parameter name, value range, and unit; the relationship edge attributes are determined according to the relationship type: statistically related edges record their correlation coefficients, semantically related edges record their semantically related edge types, and associated edges record their confidence and support.
[0090] The multi-indicator hybrid strategy specifically includes:
[0091] Calculate the Pearson correlation coefficient between any two numerical process parameter nodes. When the absolute value of the coefficient is greater than s3, establish a strongly linear correlation edge.
[0092] Simultaneously, calculate the Spearman rank correlation coefficient between any two numerical nodes, and establish a monotonically strongly correlated edge when the absolute value of the coefficient is greater than s3.
[0093] This solution, by constructing a semi-solid forming process knowledge graph based on CAE simulation data and utilizing statistical laws and expert experience rules, effectively breaks down the barriers between CAE data, statistical laws, and expert experience, integrating them into a unified, interconnected, and computable knowledge network. This effectively solves the problems of "data silos" and "fragmented experience," while providing a crucial structured data foundation for subsequent intelligent process diagnosis, parameter optimization recommendations, and expert knowledge transfer. This drives a paradigm shift in process development from experience-based trial and error to intelligent diagnosis. Here, s3, s4, and s5 are preset threshold parameters used to control the strictness and significance level of the generation of statistically relevant edges and association rule edges, respectively. Furthermore, the multi-index hybrid strategy allows these relationship edges to instantly locate and present the cluster of key process parameters most relevant to a specific quality defect or optimization target when querying the graph. This not only rapidly narrows the investigation scope from dozens of parameters to at least a few core factors, greatly improving diagnostic efficiency, but also... The study distinguishes between linear and monotonic correlations, providing deeper insights into the specific patterns of interaction between parameters (such as linear proportionality or nonlinear saturation). This guides engineers in precise, data-driven process debugging and optimization. Specifically, when constructing the initial knowledge graph, statistical correlation calculations are required for several dominant process parameters. While the Pearson correlation coefficient, as a computationally efficient and meaningful indicator, can quickly identify the most significant monotonic trends between parameters, it may miss complex nonlinear relationships. Therefore, this approach introduces the Spearman rank correlation coefficient to overcome the limitations of the Pearson correlation coefficient as a single linear correlation indicator. This allows for a comprehensive capture of both linear and nonlinear monotonic relationships between process parameters, effectively avoiding the risk of missing important nonlinear laws (such as saturation and threshold effects) due to using only Pearson correlation. This ensures that the constructed knowledge graph more realistically and completely reflects the complex interactions within the process system, thus providing a more solid and reliable data relationship foundation for subsequent quality prediction, defect tracing, and process optimization.
[0094] The specific steps for training a graph neural network model based on a semi-solid forming process knowledge graph and initial parameter database to output predicted values for cavity filling rate, maximum stress, and shrinkage cavity probability over a future period include:
[0095] Adjacency matrix and node feature matrix are extracted from the knowledge graph of semi-solid forming process and used as the basic input of graph neural network;
[0096] Construct a graph neural network model containing a K-layer graph convolutional network;
[0097] The node features obtained after K layers of convolution are subjected to global average pooling to obtain the global embedding vector of the graph.
[0098] The global embedding vector is input into a fully connected layer, and the predicted values of cavity filling rate, maximum stress and shrinkage defect probability are output over a period of time.
[0099] The graph neural network model is trained using the actual simulation results in the initial parameter database as labels, the mean square error between the predicted and actual values as the loss function, and the Adam optimizer.
[0100] Explained by this, cavity filling rate, maximum stress, and shrinkage cavity probability are key parameters for assessing the core quality risks of complex internal cavities in automotive universal joints during semi-solid forming. The filling rate is used to evaluate whether the universal joint cavity can be completely formed, avoiding direct scrap due to insufficient filling. Maximum stress is used to assess the structural integrity of the formed universal joint cavity, controlling residual stress to prevent early fatigue failure. The shrinkage cavity probability assesses the internal tightness of the universal joint cavity, preventing the influence of hidden internal defects. These three parameters together constitute the three core indicators for evaluating the usability, durability, and reliability of formed universal joint cavity parts. Accurate prediction of these parameters is crucial for achieving industrial... The key objective of process parameter optimization is that the adjacency matrix and the node feature matrix together constitute a numerical representation of the semi-solid forming process knowledge graph. The adjacency matrix quantitatively describes the structural information of the graph, namely the correlation strength and topological relationship between various process parameters, quality indicators and defect types, i.e., relation edge attributes. The node feature matrix quantitatively describes the self-attribute information of each node, such as the parameter type, typical values, etc., i.e. node attributes. These two matrices serve as a pair of structured inputs, enabling the subsequent graph neural network to simultaneously perceive the structure of the network and the attribute features of the nodes, thereby deeply integrating process knowledge for accurate prediction and optimization.
[0101] In this graph neural network model containing K layers of graph convolutional networks, each layer of graph convolutional operation updates the node features using the following formula:
[0102]
[0103] In the formula, For the first The node feature matrix of the layer For the first The node feature matrix of the layer It is a ReLU nonlinear activation function. It is an adjacency matrix. It is the identity matrix. for The degree matrix, For the first The trainable weight matrix of the layer;
[0104] further, For the first The node feature matrix of the layer, where each row corresponds to the feature vector of a node. Representative after the first The first layer obtained after convolution calculation of the layer graph Layer node feature matrix, The adjacency matrix extracted from the semi-solid forming process knowledge graph is an N×N matrix (N is the number of nodes) used to represent the connection relationships between nodes. If a node... and nodes If there is an edge connecting them, then =1, otherwise 0 The identity matrix and the adjacency matrix Same dimensions It's an adjacency matrix with added self-connections, which includes the node's own information when aggregating neighbor information. for The degree matrix is a diagonal matrix, and the elements on the diagonal are... Represents a node The degree (i.e., the number of edges connected to the node, including self-joins). For self-connected adjacency matrices Symmetric normalization is performed to prevent nodes with high degree from occupying too much weight during feature propagation, thus making the training process more stable.
[0105] The process of setting up a sensor group to collect semi-solid forming monitoring data of the cavity and inputting it into a trained graph neural network model, and dynamically adjusting the injection speed and / or holding pressure based on the prediction results using a PID control algorithm, specifically includes:
[0106] Displacement sensors are installed on the injection cylinder, and pressure and temperature sensors are installed inside the cavity to synchronously collect injection speed, cavity pressure, and temperature data at a fixed sampling frequency.
[0107] Outlier removal is performed on the collected raw data based on the Raida criterion, and Kalman filtering algorithm is used for smoothing and noise reduction.
[0108] The processed data is Z-score standardized to eliminate the influence of units;
[0109] The preprocessed real-time data is aligned with the adjacency matrix and node feature matrix of the process knowledge graph to construct the process state vector at the current moment, which is then input into the trained graph neural network model.
[0110] Obtain the predicted values of cavity filling rate, maximum stress, and shrinkage cavity probability for a future period of time from the model output;
[0111] Based on the predicted values of cavity filling rate, maximum stress, and shrinkage defect probability, the injection speed and / or holding pressure are dynamically adjusted through a PID control algorithm.
[0112] The adjusted injection speed and holding pressure commands are sent to the die-casting machine's PLC control system for closed-loop adaptive control of the cavity forming process.
[0113] This solution works by using sensors to monitor dynamic changes in the production process (such as speed, pressure, and temperature) in real time. A trained graph neural network model proactively predicts future trends of key quality indicators (fill rate, stress, and defect probability). Based on these predictions, a PID controller dynamically adjusts key process parameters such as injection speed and holding pressure. Specifically, target values are set for fill rate, maximum stress, and shrinkage cavity probability. The deviation between the predicted and target values is used as the controller input and mapped according to the following rules: the predicted values of key quality indicators output by the model are compared with preset target thresholds, and mapped to adjustments of process parameters according to clear quantitative rules. For example, if the predicted fill rate is lower than the set threshold s6, the injection speed setting is increased by a fixed step D1 (or proportionally increased by P1) to ensure the cavity is filled; if the predicted maximum stress exceeds the safety threshold, the target value is adjusted accordingly. If the threshold s7 is reached, the holding pressure will be reduced by a fixed step D2 to reduce internal stress. If the predicted shrinkage probability exceeds the allowable threshold s8, the holding pressure will be increased by a fixed step D3 to promote feeding. When multiple indicators are abnormal at the same time, the system will perform adjustments according to the priority order of 'filling rate > stress > defect probability', thereby achieving stable, closed-loop adaptive control of key quality targets. This step effectively solves the problem of unstable product quality caused by uncertain factors such as material state fluctuations and mold temperature changes during semi-solid forming. This transforms the traditional "open-loop" experience-based production that relies on fixed parameters into "closed-loop" intelligent production based on real-time status and predictive feedback. Ultimately, it achieves precise, stable, and adaptive control of the forming quality of complex internal cavity components, thus combining the offline trained graph neural network model with online real-time control to form a closed-loop system that can adaptively optimize.
[0114] Furthermore, based on the same inventive concept as the aforementioned method for optimizing semi-solid forming parameters of complex internal cavities in automotive universal joints, this solution proposes a semi-solid forming parameter optimization system for complex internal cavities in automotive universal joints, comprising:
[0115] Cavity model module, which is used to construct a cavity model including runner, venting groove, overflow groove and core based on the structural design drawings of automotive universal joint and 3D CAD software;
[0116] The data processing module is used to simulate the filling and solidification process of semi-solid slurry based on the cavity model using CAE simulation software, obtain multiple sets of process parameter combinations and their corresponding forming quality results, and form an initial parameter database; based on the multiple sets of process parameter combinations in the initial parameter database, a set of dominant process parameters is constructed using principal component analysis; and a semi-solid forming process knowledge graph is constructed with dominant process parameters, forming quality results, and defect types as nodes and physical correlations and statistical correlations between parameters as edges.
[0117] The parameter prediction and optimization module is used to train a graph neural network model based on the semi-solid forming process knowledge graph and initial parameter database, and output the predicted values of cavity filling rate, maximum stress and shrinkage defect probability in the future period; set up a sensor group to collect cavity semi-solid forming monitoring data and input it into the trained graph neural network model; and dynamically adjust the injection speed and / or holding pressure based on the prediction results through PID control algorithm.
[0118] The data processing module includes:
[0119] An initial parameter acquisition unit is used to simulate the filling and solidification process of semi-solid slurry based on the cavity model using CAE simulation software, acquire multiple sets of process parameter combinations and their corresponding forming quality results, and form an initial parameter database.
[0120] The principal component extraction unit is used to construct a set of dominant process parameters based on multiple combinations of process parameters in the initial parameter database and the principal component analysis method.
[0121] The knowledge graph unit is used to construct a semi-solid forming process knowledge graph with dominant process parameters, forming quality results and defect types as nodes and physical and statistical correlations between parameters as edges.
[0122] The parameter prediction and optimization module includes:
[0123] The parameter prediction unit is used to train a graph neural network model based on a semi-solid forming process knowledge graph and an initial parameter database, and output predicted values of cavity filling rate, maximum stress and shrinkage defect probability for a future period of time.
[0124] The parameter optimization unit is used to set the sensor group to collect semi-solid forming monitoring data of the cavity and input it into the trained graph neural network model. Based on the prediction results, the injection speed and / or holding pressure are dynamically adjusted through the PID control algorithm.
[0125] In summary, the advantages of this invention are: it effectively realizes a paradigm shift from "experience-based trial and error" to "intelligent closed-loop" in the semi-solid forming of complex internal cavities in automotive universal joints, thereby improving the forming quality and stability of complex internal cavities.
[0126] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.
Claims
1. A semi-solid forming parameter optimization method for complex internal cavities of automotive universal joints, characterized in that, include: Based on the structural design drawings of the automotive universal joint, a cavity model including the gating system, venting groove, overflow groove and core was constructed using 3D CAD software. Based on the cavity model, the filling and solidification process of semi-solid slurry is simulated using CAE simulation software to obtain multiple sets of process parameter combinations and their corresponding forming quality results, forming an initial parameter database. Based on multiple combinations of process parameters in the initial parameter database, a set of dominant process parameters is constructed using principal component analysis. A knowledge graph of semi-solid forming process is constructed, with dominant process parameters, forming quality results and defect types as nodes and physical and statistical correlations between parameters as edges. Based on the knowledge graph of semi-solid forming process and the initial parameter database, a graph neural network model is trained to output the predicted values of cavity filling rate, maximum stress and shrinkage defect probability in the future. A sensor array is set up to collect monitoring data of the semi-solid forming of the cavity and input it into a trained graph neural network model. Based on the prediction results, the injection speed and / or holding pressure are dynamically adjusted through a PID control algorithm.
2. The semi-solid forming parameter optimization method for complex internal cavities of automotive universal joints according to claim 1, characterized in that, The construction of a cavity model, including a gating system, venting groove, overflow groove, and core, based on the structural design drawings of the automotive universal joint and using 3D CAD software, specifically includes: Using SolidWorks 3D CAD software, a 1:1 parametric model was created based on the structural design drawings of the automotive universal joint to generate a cavity geometry model of the complex internal cavity of the automotive universal joint. In the cavity geometry model, a gating system is constructed, which includes a sprue with a circular cross-section, a horizontal sprue with a trapezoidal cross-section, and an venting groove and an overflow groove located at the end of the cavity. In the cavity geometry model, a detachable core model corresponding to the shape of the universal joint cavity is placed, and the fitting clearance range between the core model and the cavity is set to [a,b]. Perform virtual assembly interference checks on the completed cavity geometry model to ensure that there is no geometric interference between the core, the mold cavity, and the ejector mechanism, and export the model as an STP or IGES format.
3. The semi-solid forming parameter optimization method for complex internal cavities of automotive universal joints according to claim 2, characterized in that, The process of filling and solidifying semi-solid slurry based on the cavity model is simulated using CAE simulation software to obtain multiple sets of process parameter combinations and their corresponding forming quality results, forming an initial parameter database, specifically including: Import the cavity model into the CAE simulation software; The constitutive model of the semi-solid slurry is set as the Kaufmann non-Newtonian fluid model, and the relationship expression between its apparent viscosity and solid fraction and shear rate is constructed. Set the initial value range for slurry parameters, process parameters, and mold / material parameters; Based on the Taguchi method in experimental design, an L25 orthogonal experimental table was designed, with solid fraction, injection speed and mold temperature as control factors. Each factor was set with 5 levels to simulate the filling and solidification process. Record the filling rate, maximum internal stress, shrinkage cavity defect volume fraction, and deformation in the simulation results of each set of parameters as the forming quality results; Integrate all process parameter combinations and their corresponding forming quality results to form an initial parameter database containing both qualified and defective cases.
4. The semi-solid forming parameter optimization method for complex internal cavities of automotive universal joints according to claim 3, characterized in that, The construction of the dominant process parameter set based on multiple combinations of process parameters in the initial parameter database and using principal component analysis specifically includes: Extract the original data matrix consisting of all numerical process parameters from the initial parameter database; Z-score standardization is performed on the original data matrix to eliminate the influence of dimensions; Calculate the covariance matrix of the standardized data, and solve for the eigenvalues and eigenvectors of the covariance matrix; Sort the eigenvalues from largest to smallest, and calculate the variance contribution rate and cumulative variance contribution rate of each principal component; Select the eigenvectors corresponding to the first k principal components whose cumulative variance contribution rate is greater than or equal to s1%; Analyze the original process parameters whose absolute load value is greater than s2 among the first k eigenvectors, and determine them as the dominant process parameters; Based on the selected dominant process parameters, construct a set of dominant process parameters; The set of dominant process parameters includes at least: injection speed, holding pressure, pressure switching point, and mold temperature field uniformity.
5. The semi-solid forming parameter optimization method for complex internal cavities of automotive universal joints according to claim 4, characterized in that, The construction of a semi-solid forming process knowledge graph, using dominant process parameters, forming quality results, and defect types as nodes and physical and statistical correlations between parameters as edges, specifically includes: The dominant process parameters, forming quality results, and typical defect types are defined as nodes in the knowledge graph; A multi-index hybrid strategy is used to construct statistical correlation edges between nodes, and the correlation coefficient between any two dominant process parameter nodes is calculated. Based on CAE simulation results and expert experience rules, semantic relationship edges are established between each process parameter node pair. The semantic relationship edge types include at least: causing, inhibiting, influencing, and belonging. Based on the Apriori algorithm, we mine the association rules between process parameter combinations and defect types, set the minimum support to s4% and the minimum confidence to s5%, and transform the association rules that meet the conditions into association relationship edges. Based on all defined nodes and relation edges, a knowledge graph of semi-solid forming process is constructed and stored in the Neo4j graph database; Among them, the node attributes include: process parameter name, value range, and unit; the relationship edge attributes are determined according to the relationship type: statistically related edges record their correlation coefficients, semantically related edges record their semantically related edge types, and associated edges record their confidence and support. The multi-indicator hybrid strategy specifically includes: Calculate the Pearson correlation coefficient between any two numerical process parameter nodes. When the absolute value of the coefficient is greater than s3, establish a strongly linear correlation edge. Simultaneously, calculate the Spearman rank correlation coefficient between any two numerical nodes, and establish a monotonically strongly correlated edge when the absolute value of the coefficient is greater than s3.
6. The semi-solid forming parameter optimization method for complex internal cavities of automotive universal joints according to claim 5, characterized in that, The specific steps for training a graph neural network model based on a semi-solid forming process knowledge graph and initial parameter database to output predicted values for cavity filling rate, maximum stress, and shrinkage cavity probability over a future period include: Adjacency matrix and node feature matrix are extracted from the knowledge graph of semi-solid forming process and used as the basic input of graph neural network; Construct a graph neural network model containing a K-layer graph convolutional network; The node features obtained after K layers of convolution are subjected to global average pooling to obtain the global embedding vector of the graph. The global embedding vector is input into a fully connected layer, and the predicted values of cavity filling rate, maximum stress and shrinkage defect probability are output over a period of time. The graph neural network model is trained using the actual simulation results in the initial parameter database as labels, the mean square error between the predicted and actual values as the loss function, and the Adam optimizer.
7. The semi-solid forming parameter optimization method for complex internal cavities of automotive universal joints according to claim 6, characterized in that, The process of setting up a sensor group to collect semi-solid forming monitoring data of the cavity and inputting it into a trained graph neural network model, and dynamically adjusting the injection speed and / or holding pressure based on the prediction results using a PID control algorithm, specifically includes: Displacement sensors are installed on the injection cylinder, and pressure and temperature sensors are installed inside the cavity to synchronously collect injection speed, cavity pressure, and temperature data at a fixed sampling frequency. Outlier removal is performed on the collected raw data based on the Raida criterion, and Kalman filtering algorithm is used for smoothing and noise reduction. The processed data is Z-score standardized to eliminate the influence of units; The preprocessed real-time data is aligned with the adjacency matrix and node feature matrix of the process knowledge graph to construct the process state vector at the current moment, which is then input into the trained graph neural network model. Obtain the predicted values of cavity filling rate, maximum stress, and shrinkage cavity probability for a future period of time from the model output; Based on the predicted values of cavity filling rate, maximum stress, and shrinkage defect probability, the injection speed and / or holding pressure are dynamically adjusted through a PID control algorithm. The adjusted injection speed and holding pressure commands are sent to the die-casting machine's PLC control system for closed-loop adaptive control of the cavity forming process.
8. A semi-solid forming parameter optimization system for complex internal cavities of automotive universal joints, characterized in that, A semi-solid forming parameter optimization method for realizing the complex internal cavity of an automotive universal joint as described in any one of claims 1-7, comprising: Cavity model module, which is used to construct a cavity model including runner, venting groove, overflow groove and core based on the structural design drawings of automotive universal joint and 3D CAD software; The data processing module is used to simulate the filling and solidification process of semi-solid slurry based on the cavity model using CAE simulation software, obtain multiple sets of process parameter combinations and their corresponding forming quality results, and form an initial parameter database; based on the multiple sets of process parameter combinations in the initial parameter database, a set of dominant process parameters is constructed using principal component analysis; and a semi-solid forming process knowledge graph is constructed with dominant process parameters, forming quality results, and defect types as nodes and physical correlations and statistical correlations between parameters as edges. The parameter prediction and optimization module is used to train a graph neural network model based on the semi-solid forming process knowledge graph and initial parameter database, and output the predicted values of cavity filling rate, maximum stress and shrinkage defect probability in the future period; set up a sensor group to collect cavity semi-solid forming monitoring data and input it into the trained graph neural network model; and dynamically adjust the injection speed and / or holding pressure based on the prediction results through a PID control algorithm.
9. The semi-solid forming parameter optimization system for complex internal cavities of automotive universal joints according to claim 8, characterized in that, The data processing module includes: An initial parameter acquisition unit is used to simulate the filling and solidification process of semi-solid slurry based on the cavity model using CAE simulation software, acquire multiple sets of process parameter combinations and their corresponding forming quality results, and form an initial parameter database. The principal component extraction unit is used to construct a set of dominant process parameters based on multiple combinations of process parameters in the initial parameter database and the principal component analysis method. The knowledge graph unit is used to construct a semi-solid forming process knowledge graph with dominant process parameters, forming quality results and defect types as nodes and physical correlations and statistical correlations between parameters as edges.
10. The semi-solid forming parameter optimization system for complex internal cavities of automotive universal joints according to claim 9, characterized in that, The parameter prediction and optimization module includes: The parameter prediction unit is used to train a graph neural network model based on a semi-solid forming process knowledge graph and an initial parameter database, and output predicted values of cavity filling rate, maximum stress and shrinkage defect probability for a future period of time. The parameter optimization unit is used to set the sensor group to collect semi-solid forming monitoring data of the cavity and input it into the trained graph neural network model. Based on the prediction results, the injection speed and / or holding pressure are dynamically adjusted through the PID control algorithm.