Dynamic mold flow intelligent analysis method and system for die-casting mold

Abstract

The embodiment of the application relates to the technical field of intelligent casting, and discloses a dynamic mold flow intelligent analysis method for a die-casting die, comprising the following steps: feature recognition is performed on a die-casting engineering drawing to generate a three-dimensional model with pre-deformation compensation; a warping deformation intelligent prediction model is constructed according to obtained die material data, historical mold flow analysis data and real-time sensing data; die state data in a die-casting process is collected in real time based on an Internet of Things sensor, and the die state data is input into the warping deformation intelligent prediction model to perform dynamic warping prediction to obtain corresponding warping deformation amount prediction values and warping form distribution states; and the warping deformation amount prediction values and the warping form distribution states are matched with set conditions. In the embodiment of the application, the method can not only offset possible shrinkage and warping after forming in advance in a design stage, but also can realize continuous monitoring of the die-casting process by using real-time collection of die state data, so that batch defects are avoided.

Description

Technical Field

[0001] This invention relates to the field of intelligent casting technology, specifically to a dynamic mold flow intelligent analysis method and system for die casting molds. Background Technology

[0002] Currently, die casting is a key process in the production of automotive parts, especially for the application of large integrated die-cast parts in new energy vehicles, which places higher demands on mold design and process control. Warpage is one of the most common quality defects in die-cast parts, directly affecting the assembly sealing and service life of the product. Therefore, how to achieve intelligent analysis of die-casting molds to improve yield has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0003] To address the aforementioned shortcomings, this invention discloses a dynamic mold flow intelligent analysis method for die casting molds, which enables dynamic mold flow intelligent analysis of die casting molds.

[0004] The first aspect of this invention discloses a dynamic mold flow intelligent analysis method for die-casting molds, including: Feature recognition is performed on the engineering drawings of die-cast parts to generate a 3D model with pre-deformation compensation; Based on the acquired mold material data, historical mold flow analysis data, and real-time sensor data, an intelligent prediction model for warpage deformation is constructed. The mold status data during the die casting process is collected in real time by IoT sensors, and the mold status data is input into the intelligent warping deformation prediction model to perform dynamic warping prediction in order to obtain the corresponding predicted value of warping deformation and the distribution state of warping morphology. The predicted warpage deformation and warpage morphology distribution are matched with set conditions. If the set conditions are not met, an early warning is issued and the mold structure or process parameters are automatically adjusted.

[0005] The second aspect of this invention discloses a dynamic mold flow intelligent analysis system for die casting molds, comprising: Generation module: Used to perform feature recognition on die-casting engineering drawings to generate a 3D model with pre-deformation compensation; The building module is used to construct an intelligent prediction model for warpage deformation based on the acquired mold material data, historical mold flow analysis data, and real-time sensor data. Prediction module: used to collect mold status data in real time during the die casting process based on IoT sensors, and input the mold status data into the warping deformation intelligent prediction model to perform dynamic warping prediction in order to obtain the corresponding warping deformation prediction value and warping morphology distribution status. Adjustment module: used to match the predicted warpage deformation and warpage morphology distribution with set conditions. If the set conditions are not met, an early warning is issued and the mold structure or process parameters are automatically adjusted.

[0006] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: The dynamic mold flow intelligent analysis method for die casting molds in this embodiment of the invention identifies features in the die casting engineering drawings and generates a three-dimensional model with pre-deformation compensation. This not only enables the pre-compensation of shrinkage and warping that may occur after molding during the design stage, reducing defects from the source and improving product dimensional accuracy and first-pass yield, but also utilizes IoT sensors to collect mold status data in real time and inputs the data into a prediction model for dynamic warping analysis. This achieves continuous monitoring of the die casting process, enabling timely detection of abnormal trends and preventing the generation of batch defects. Attached Figure Description

[0007] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is a flowchart illustrating the dynamic mold flow intelligent analysis method for die-casting molds disclosed in an embodiment of the present invention; Figure 2 This is a schematic diagram of the process for constructing a three-dimensional model as disclosed in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a dynamic mold flow intelligent analysis system for die casting molds provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0009] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0010] It should be noted that the terms "first," "second," "third," "fourth," etc., in the specification and claims of this invention are used to distinguish different objects, not to describe a specific order. The terms "comprising" and "having," and any variations thereof, in the embodiments of this invention are intended to cover non-exclusive inclusion. Exemplarily, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or devices.

[0011] Example 1 Please see Figure 1 , Figure 1 This is a flowchart illustrating the dynamic mold flow intelligent analysis method for die-casting molds disclosed in this invention. The execution entity of the method described in this embodiment is an execution entity composed of software and / or hardware. This execution entity can receive relevant information via wired or / or wireless means and can send certain instructions. It can also have certain processing and storage functions. This execution entity can control multiple devices, such as remote physical servers or cloud servers and related software, or local hosts or servers and related software that perform related operations on devices located in a certain location. In some scenarios, it can also control multiple storage devices, which can be placed in the same location as the devices or in different locations. Figure 1 As shown, this dynamic mold flow intelligent analysis method based on die-casting molds includes the following steps: S101: Perform feature recognition on die-casting engineering drawings to generate a 3D model with pre-deformation compensation; S102: Construct an intelligent prediction model for warpage deformation based on the acquired mold material data, historical mold flow analysis data, and real-time sensor data; S103: Real-time acquisition of mold status data during the die casting process based on IoT sensors, and input of the mold status data into the intelligent warping deformation prediction model to perform dynamic warping prediction in order to obtain the corresponding predicted value of warping deformation and the distribution state of warping morphology. S104: Match the predicted warpage deformation and warpage morphology distribution with the set conditions. If the set conditions are not met, issue an early warning and automatically adjust the mold structure or process parameters.

[0012] The solution of this invention generates a three-dimensional model with pre-deformation compensation by performing feature recognition on the engineering drawings of die castings. This allows for the consideration of possible deformation in advance, making the model closer to the shape of die castings in actual production. This provides a reliable basis for subsequent accurate analysis and helps improve the accuracy and reliability of the entire analysis process.

[0013] In practical implementation, a warpage deformation intelligent prediction model is constructed based on mold material data, historical mold flow analysis data, and real-time sensor data. By integrating multiple data sources and making full use of historical experience and real-time information, the model can more accurately predict the warpage deformation of the mold during the die casting process, thereby providing comprehensive data.

[0014] By using IoT sensors to collect mold status data in real time during the die casting process and inputting it into the intelligent warping deformation prediction model for dynamic warping prediction, the system can obtain the predicted value of warping deformation and the distribution of warping morphology at different times, realize dynamic monitoring and prediction of the die casting process, promptly identify potential problems and make corrections, and improve the yield rate.

[0015] In this embodiment of the invention, the predicted results are matched with set conditions. If the set conditions are not met, an early warning is issued, and the mold structure or process parameters are automatically adjusted. This mechanism can quickly respond to abnormal situations in the production process, take timely measures to avoid or reduce quality problems caused by warping deformation, improve production efficiency and product quality, and reduce production costs.

[0016] Specifically, the set conditions include warpage deformation threshold, solids distribution threshold, and multi-dimensional defect criterion threshold; among them, the warpage deformation threshold is preset based on product design requirements; the solids distribution threshold includes a safety threshold, a warning threshold, and a danger threshold, with corresponding temperature ranges of safe temperature range, warning temperature range, and danger temperature range, respectively; the multi-dimensional defect criterion threshold includes the solids difference criterion threshold in the junction area, the solids at the front end of the junction area, the filling time deviation rate criterion threshold, the shrinkage cavity ratio criterion threshold, and the clamping force load rate criterion threshold.

[0017] The mold material data includes: material density, specific heat capacity, thermal conductivity, viscosity as a function of shear rate, surface tension as a function of temperature, solid fraction as a function of temperature, liquidus temperature, solidus temperature, and critical solid fraction. More preferably, such as Figure 2 As shown, the step of performing feature recognition on the die-casting engineering drawing to generate a 3D model with pre-deformation compensation includes: S1011: Perform feature recognition on the acquired two-dimensional engineering drawings or three-dimensional point cloud data to determine the geometric topology and feature parameters of the corresponding die casting; S1012: Based on the geometric topology, characteristic parameters, and pre-built feature database, determine the corresponding solidity sensitive region; the solidity sensitive region includes the first wall thickness region, the second wall thickness region, the corner region, and the gate distal region; S1013: In the 3D modeling environment, a model sketch with pre-deformation compensation is generated based on the spatial coordinates of the solidity sensitive region and the solidity evolution sensitivity coefficient. The calculation formula for the pre-deformation compensation is: ;in, Let i be the compensation amount for the i-th sensitive region. For compensation coefficient, This is the sensitivity coefficient for the evolution of the solid fraction in the corresponding region. For mold temperature, The melting temperature is... It is the critical solid fraction.

[0018] This invention performs feature recognition on two-dimensional engineering drawings or three-dimensional point cloud data, enabling accurate determination of the geometric topology and characteristic parameters of die-cast parts. This provides precise basic data for subsequent analysis, helping to comprehensively understand the key characteristics of die-cast parts such as shape and size, ensuring that the entire modeling and analysis process is based on reliable information, and improving the accuracy and relevance of subsequent steps.

[0019] In the specific implementation process, based on the geometric topology, characteristic parameters, and a pre-built feature database, the solidity-sensitive areas were identified, clearly indicating key locations such as the first wall thickness region, the second wall thickness region, corner regions, and the region far from the gate. Here, the first wall thickness region refers to the thin-walled area, and the second wall thickness region refers to the thick-walled area. These regions are more prone to deformation and other problems during die casting due to factors such as changes in solidity. Accurately locating these regions provides direction for subsequent targeted processing, helps to concentrate resources to solve key problems, and improves analysis efficiency.

[0020] In a 3D modeling environment, a model sketch with pre-deformation compensation is generated based on the spatial coordinates of the solidity-sensitive region and the solidity evolution sensitivity coefficient. A calculation formula for the pre-deformation compensation is also provided. This formula comprehensively considers multiple factors, including the compensation coefficient, the solidity evolution sensitivity coefficient of the corresponding region, mold temperature, melt temperature, and critical solidity, making the calculation of the pre-deformation compensation more scientific and reasonable. The 3D model with pre-deformation compensation generated in this way can more accurately simulate the deformation that may occur in die castings during actual production, compensating for deformation in advance. This effectively reduces deformation errors in actual production, improves the quality and dimensional accuracy of die castings, reduces scrap rates, and increases production efficiency.

[0021] More preferably, determining the corresponding solidity-sensitive region based on the geometric topology, feature parameters, and a pre-constructed feature database includes: Based on the solid fraction and temperature correlation curves in the database, the solid fraction sensitive areas of the casting to be analyzed are identified and determined. The solid fraction sensitive areas include the abrupt change in solid fraction evolution rate caused by wall thickness difference, the filling end area at the far end of the gate, and the heat concentration area caused by geometric abrupt change. Based on the identification results of the solidity sensitive region, local mesh refinement is performed on the solidity sensitive region during the three-dimensional solid mesh generation process, wherein the mesh refinement density is positively correlated with the solidity gradient. Based on the solid fraction evolution prediction results, differentiated pre-deformation compensation amounts are generated in adjacent regions where the difference in solid fraction evolution rate exceeds a preset threshold.

[0022] In practical implementation, based on the solids-temperature correlation curves in the database, regions of abrupt changes in solids evolution rate caused by wall thickness differences, the filling end region at the far end of the gate, and heat concentration regions caused by geometrical abrupt changes were identified as solids-sensitive areas. These areas cover key parts of the die-casting process where solids changes are complex and prone to problems due to various factors, providing an accurate target range for subsequent precise analysis and processing, and helping to focus on solving the core issues affecting die-casting quality. Using solids-temperature correlation curves for identification, this method based on scientific data and patterns makes the determination of sensitive areas more accurate and reliable, avoiding errors that may be caused by subjective judgment, and providing a solid foundation for subsequent mesh refinement and pre-deformation compensation.

[0023] In the process of generating the 3D solid mesh, local mesh refinement is performed in solidity-sensitive regions, and the mesh refinement density is positively correlated with the solidity gradient. Regions with large solidity gradients indicate drastic changes in solidity. Refining the mesh in these regions allows for a more detailed description of the physical changes, such as temperature and stress fields, thereby improving the accuracy of numerical simulations and more accurately reflecting the actual conditions in these critical areas during die casting. Refining the mesh locally only in sensitive regions, rather than using a uniform high-density mesh for the entire model, ensures computational accuracy while rationally allocating computational resources, reducing unnecessary computational load, improving computational efficiency, shortening the analysis cycle, and lowering computational costs.

[0024] Based on the solid fraction evolution prediction results, differentiated pre-deformation compensation amounts are generated in adjacent regions where the difference in solid fraction evolution rates exceeds a preset threshold. Since the solid fraction evolution rates differ across regions, the resulting deformation varies. Differentiated pre-deformation compensation can accurately compensate for the specific deformation trends in each region, better offsetting deformations that may occur in actual production, and further improving the dimensional and shape accuracy of die-cast parts. Deformation during die casting is often complex, with different regions influencing each other. By generating differentiated pre-deformation compensation amounts, this complex deformation situation can be better adapted, making the pre-deformation compensation more consistent with the deformation patterns in actual production, effectively reducing quality problems caused by deformation, and improving product yield and quality stability.

[0025] More preferably, the intelligent prediction model for warping deformation is constructed in the following manner: A training dataset is constructed by collecting warping data sources, including material solid fraction versus temperature curve data, solid fraction evolution cloud map sequence data extracted from historical model flow analysis, and warping deformation measured data under the same working conditions as the solid fraction evolution cloud map sequence. A three-dimensional convolutional neural network is constructed. Taking the solid fraction evolution cloud map sequence as input, the evolution features of solid fraction in time and space are extracted simultaneously through the three-dimensional convolution kernel, and the solid fraction evolution feature vector is output. A feature parsing layer is connected after the three-dimensional convolutional neural network. The feature parsing layer converts the solid fraction evolution feature vector into solid fraction evolution feature parameters. The solid fraction evolution feature parameters include the time for each region to reach the critical solid fraction, the solid fraction spatial distribution non-uniformity index, and the solid fraction evolution time difference between the thick-walled region and the thin-walled region. A multilayer perceptron network is constructed as a warpage deformation prediction subnetwork. The input layer of the multilayer perceptron network receives the solid fraction evolution characteristic parameters, and the output layer outputs the predicted warpage deformation. The three-dimensional convolutional neural network, feature parsing layer and multilayer perceptron network are connected in series to form a complete intelligent prediction model for warping deformation. The model is then jointly trained end-to-end using the solid fraction evolution cloud map sequence as input and the corresponding measured warping deformation amount as supervision label.

[0026] Specifically, a training dataset was constructed by collecting data on the relationship between material solid fraction and temperature, solid fraction evolution cloud map sequences extracted from historical model flow analysis, and measured warping deformation data under the same working conditions. This fusion of multi-source data covers multiple dimensions, including material properties, historical simulation results, and actual production results, providing the model with comprehensive and rich information. This enables the model to learn the complex relationships between different factors and warping deformation, improving the model's generalization ability and accuracy.

[0027] The collected data are closely correlated; the solid fraction evolution contour map sequence and the measured warpage deformation data come from the same working condition, accurately reflecting the correspondence between solid fraction changes and warpage deformation under that specific condition. This correlation helps the model better understand the physical process, thereby enabling more accurate prediction of warpage deformation.

[0028] By constructing a three-dimensional convolutional neural network and taking the solids fraction evolution cloud map sequence as input, the evolution features of the solids fraction in both time and space are extracted simultaneously through a three-dimensional convolutional kernel. During die casting, the change in solids fraction not only varies across different spatial regions but is also a dynamic process over time. The three-dimensional convolutional neural network can simultaneously capture features in both dimensions, providing a more comprehensive description of the solids fraction evolution and offering richer feature information for subsequent warpage prediction.

[0029] Compared to traditional manual feature extraction methods, 3D convolutional neural networks can automatically learn features from solid fraction evolution contour sequences without the need for manually designing complex feature extraction rules. This not only reduces human intervention and improves the efficiency of feature extraction, but also enables the discovery of feature patterns that are difficult for humans to detect, thus helping to improve the model's prediction accuracy.

[0030] In the feature parsing layer of this invention, the solid fraction evolution feature vector output by the three-dimensional convolutional neural network is converted into solid fraction evolution feature parameters, such as the time for each region to reach the critical solid fraction, the solid fraction spatial distribution non-uniformity index, and the solid fraction evolution time difference between thick-walled and thin-walled regions. These feature parameters are a further refinement and quantification of the solid fraction evolution characteristics, which can more intuitively reflect the key factors that have an important impact on warping deformation during the solid fraction evolution process, and provide more targeted input for warping deformation prediction.

[0031] Converting feature vectors into feature parameters reduces the dimensionality of the data to some extent, thereby reducing the input complexity of subsequent multilayer perceptron networks. This helps improve the training efficiency and prediction speed of the network, while avoiding overfitting caused by excessively high data dimensionality.

[0032] As a sub-network for predicting warpage deformation, the multilayer perceptron (MLP) network possesses powerful nonlinear mapping capabilities. It can learn the complex nonlinear relationship between solid fraction evolution characteristic parameters and warpage deformation. Through the combination of multiple layers of neurons and the action of activation functions, it performs in-depth processing and transformation on the input feature parameters, thereby outputting accurate predicted values ​​for warpage deformation. The structure of the MLP network can be adjusted according to actual needs, adjusting the number of layers and neurons to adapt to prediction tasks of varying complexity. This flexibility allows the model to better adapt to different die-casting processes and material properties, improving its applicability and versatility.

[0033] A complete intelligent prediction model for warping deformation is constructed by concatenating a 3D convolutional neural network, a feature parsing layer, and a multilayer perceptron network, and then jointly trained end-to-end. This training method allows the entire model to be optimized as a whole, with each part cooperating to adjust parameters and minimize prediction error. Compared to training each part separately and then combining them, end-to-end joint training can better coordinate the relationships between the parts and improve the overall performance of the model.

[0034] Through end-to-end joint training, the model can make full use of the information in the input data and learn the prediction of warping deformation directly from the original solid fraction evolution contour sequence, reducing the information loss and error accumulation that may be caused by intermediate links, thereby further improving the accuracy of warping deformation prediction.

[0035] More preferably, the process by which the feature parsing layer converts the solid fraction evolution feature vector into solid fraction evolution feature parameters includes: By performing time-series analysis on the solid fraction evolution cloud map sequence, the time point at which the solid fraction of each pixel reaches the preset critical value is identified, and a critical time distribution map is formed. Calculate the root mean square error between the solid fraction at each location at the same time and the global average solid fraction, and use the value at the time of completion of filling as the non-uniformity index. Based on the pre-annotated thick-walled and thin-walled regions using geometric topology information, the average time for the two regions to reach the critical solid fraction is calculated, and the difference between them is determined. The end-to-end joint training uses mean squared error as the loss function, and performs joint optimization of the parameters of the 3D convolutional neural network, feature parsing layer and multilayer perceptron network through backpropagation algorithm, and uses cross-validation to select the optimal model parameters.

[0036] The solution of this invention performs time-series analysis on the solids fraction evolution cloud map sequence to identify the time point at which the solids fraction at each pixel location reaches a preset critical value and forms a critical time distribution map. This process can accurately determine the time when the solids fraction at different locations of the die casting changes to the critical state, providing detailed information in the time dimension for subsequent analysis of the relationship between solids fraction evolution and warpage deformation, and helping to deeply understand the physical change process of each region during the die casting process. The critical time distribution map can intuitively show the differences in the time when different regions of the die casting reach the critical solids fraction. This difference is closely related to factors such as the geometry, material properties, and process parameters of the die casting, and is one of the important potential factors leading to warpage deformation, providing an important basis for subsequent feature parameter extraction and warpage deformation prediction.

[0037] Specifically, the root mean square error of the solid fraction at each location at the same time compared to the global average solid fraction is calculated, and the value at the time of filling completion is used as the non-uniformity index. This index can quantify the degree of spatial non-uniformity of the solid fraction distribution in the die casting when filling is complete. The greater the non-uniformity, the more significant the difference in solid fraction between different regions inside the die casting, and the greater the possibility of internal stress and warping deformation during cooling. This provides an important quantitative indicator for assessing the risk of warping deformation. Simplifying the complex spatial distribution of solid fraction into a single non-uniformity index reduces the dimensionality and complexity of the data, facilitating subsequent processing and analysis by a multilayer perceptron network, while retaining the characteristic information of solid fraction non-uniformity distribution that has a significant impact on warping deformation.

[0038] Determining the time difference between thick-walled and thin-walled regions: Based on geometric topological information, thick-walled and thin-walled regions are pre-labeled, and the average time and difference between the two regions reaching the critical solid fraction are calculated. Due to different heat dissipation conditions, the solid fraction evolution processes of thick-walled and thin-walled regions differ significantly. This difference leads to inconsistent contraction during cooling, resulting in warping deformation. By calculating the time difference, this difference between key regions can be accurately captured, providing targeted characteristic parameters for warping deformation prediction.

[0039] By combining geometric topology information for region labeling and time difference calculation, the influence of the geometry of the die casting on the evolution of solid fraction and warpage deformation is fully considered, making the characteristic parameters more reflective of the physical phenomena in the actual die casting process and improving the representativeness and effectiveness of the characteristic parameters.

[0040] Specifically, mean squared error (MSE) can intuitively measure the difference between the model's predicted warpage and the actual measured value. By minimizing the MSE, the model's prediction can be made as close as possible to the true value, thereby improving the accuracy and precision of warpage prediction. The MSE is continuously differentiable, allowing for convenient gradient calculation in the backpropagation algorithm, thus enabling optimized adjustment of model parameters and making the training process more stable and efficient.

[0041] This invention uses the backpropagation algorithm to jointly optimize the parameters of the three-dimensional convolutional neural network, the feature parsing layer, and the multilayer perceptron network. This enables the entire model to learn and adjust as a whole, with each part cooperating with the others to adapt to the warping deformation prediction task. This avoids the local optima problem that may result from optimizing each part individually, thereby improving the overall performance and generalization ability of the model.

[0042] During joint optimization, the parameters of different network layers can be adjusted collaboratively based on the feedback from the loss function, making the various stages such as feature extraction, feature parsing, and warping prediction more closely integrated. This ensures that the model can learn the most effective information from the input data and make accurate predictions.

[0043] Specifically, the model for predicting warpage deformation takes three-dimensional data (such as three-dimensional volume data containing multi-channel information, covering the measured values ​​of different physical quantities of the workpiece, presented in the form of a C×D×H×W tensor) as input. First, it enters a three-dimensional convolutional neural network, where the convolutional layers use multiple three-dimensional convolutional kernels (e.g., the first convolutional layer uses 32 3×3×3 convolutional kernels, with a stride of 1 or 2) to perform convolution operations on the input data. Local features are extracted by sliding the convolutional kernels in three dimensions and calculating the dot product. Each convolutional layer is followed by a ReLU activation function to introduce nonlinearity. This is followed by pooling layers, commonly using 2×2×2 max pooling layers (with a stride of 2), which maximize the value in local regions to reduce the data space size, decrease computational load, and enhance translation invariance. If necessary, fully connected layers are used to flatten the extracted three-dimensional features for preliminary integration.

[0044] Next, the data enters the feature parsing layer, which can consist of multiple fully connected layers (such as two fully connected layers with 64 and 32 neurons respectively). Linear transformations and non-linear activations are used to further analyze and refine the features. To enhance the model's focus on key features, a self-attention mechanism can be introduced, allowing the model to automatically learn the importance relationships between different features and assign higher weights to key features.

[0045] The data is then input into a multilayer perceptron (MLP) network. First, the input layer receives the feature vector output from the feature parsing layer, and then it enters multiple hidden layers (e.g., two hidden layers with 64 and 32 neurons respectively, each followed by a ReLU activation function) for deep processing. Finally, the output layer outputs the result based on the task type. For regression tasks (predicting continuous warping deformation), the output layer may have only one neuron and use a linear activation function; for classification tasks (e.g., determining whether warping deformation exceeds a threshold), the number of neurons in the output layer is increased accordingly, and a Softmax activation function is used.

[0046] Throughout the model training process, an appropriate loss function is used to measure the difference between the predicted and true values. Mean squared error is commonly used for regression tasks, while cross-entropy loss is used for classification tasks. The gradient of the loss function with respect to the parameters of each network layer is calculated using the backpropagation algorithm. Then, optimization algorithms (such as the Adam algorithm, which combines the advantages of momentum and adaptive learning rate) are used to update the parameters based on the gradient, thereby achieving joint optimization of the parameters of each network layer, avoiding local optima problems, and improving the overall performance and generalization ability of the model.

[0047] Specifically, cross-validation selects the optimal model parameters. By dividing the dataset into multiple subsets and performing multiple training and validation cycles, cross-validation can more comprehensively evaluate the model's performance under different data distributions, avoiding overfitting caused by inconsistent training and test data distributions, and improving the model's reliability and stability. Cross-validation compares the performance of the model under different parameter combinations, thereby selecting the optimal model parameters. This allows the model to have better predictive performance on unknown data, further improving the practicality and accuracy of the intelligent warp deformation prediction model.

[0048] More preferably, the intelligent analysis method further includes: A digital twin of the die-cast part to be formed is constructed, the digital twin including a three-dimensional model with pre-deformation compensation, a warping deformation intelligent prediction model and a real-time error correction model based on a recurrent neural network; The real-time error correction model calibrates the simulation prediction results based on the real-time data stream from sensors deployed at key locations in the mold, and dynamically adjusts the optimal process parameters to compensate for warping deformation in actual production. The real-time error correction model module is a time-series prediction model built on a long short-term memory network. Its input includes: real-time data streams from sensors deployed at key locations in the mold. The sensors include pressure sensors inside the mold cavity, temperature sensors, mold strain sensors, and temperature sensors at the inlet and outlet of the cooling pipes. The simulated warping prediction value is output by the intelligent warping deformation prediction model module; the corrected warping prediction value is output by the real-time error correction model module. The corrected warping prediction value is the result of dynamically compensating for the systematic deviation between the simulated warping prediction value and the actual measured warping value in production. The dynamic adjustment of optimal process parameters includes: Based on the output deviation value of the real-time error correction model, the process parameter settings for several future production cycles are optimized using a model predictive control algorithm. The optimization objective of the model predictive control algorithm is to minimize the weighted sum of prediction error and control cost, wherein the constraints include injection speed, holding pressure, and mold temperature. The optimized process parameters are automatically sent to the die-casting machine control system via industrial automation protocols to achieve pressure adjustment.

[0049] Specifically, a digital twin of the die-casting part to be formed was constructed, including a 3D model with pre-deformation compensation, an intelligent prediction model for warpage deformation, and a real-time error correction model based on a recurrent neural network. This achieved comprehensive simulation from geometry and warpage deformation prediction to real-time error correction. This comprehensive simulation can more accurately reflect the various states and changes of the die-casting part during actual production, providing a solid foundation for the optimization and control of the production process.

[0050] Digital twins tightly integrate virtual models with actual production. Through real-time data interaction, the virtual model can reflect the actual production situation in real time, and the predictions from the virtual model can guide and adjust actual production. This fusion and interaction of virtual and real elements breaks down the separation between virtual and reality in traditional production, improving the flexibility and intelligence of production.

[0051] Specifically, the real-time error correction model calibrates the simulation prediction results based on real-time data streams from multiple sensors deployed at key locations in the mold (instrument pressure sensors, temperature sensors, mold strain sensors, and temperature sensors at the inlet and outlet of cooling pipes). The fusion of multi-source data provides more comprehensive and accurate information, reflecting the actual state of the mold and die-cast parts from different perspectives, thereby improving the accuracy and reliability of the calibration.

[0052] Due to various uncertainties in actual production processes, such as fluctuations in material properties and changes in ambient temperature, deviations can occur between simulation predictions and actual production conditions. Real-time error correction models can dynamically adjust based on real-time data streams, promptly capturing these changes and correcting deviations, ensuring that predictions remain consistent with actual production conditions and improving the model's adaptability and real-time performance. The real-time error correction model uses the systematic deviation between the simulated warpage prediction value output by the intelligent warpage deformation prediction model and the actual measured warpage value from production as the correction target, outputting a corrected warpage prediction value. Through dynamic compensation for systematic deviations, the inherent differences between the simulation model and actual production can be eliminated, further improving the accuracy of warpage deformation prediction and providing a more reliable basis for optimizing process parameters.

[0053] In this embodiment of the invention, a long short-term memory (LSTM) network is used to construct a time-series prediction model, which can better handle sensor data with time-series characteristics. LSTM networks have the ability to remember long-term information and can capture long-term dependencies in the data. This has significant advantages in analyzing the state changes of molds and die-cast parts at different points in time and predicting future deviation trends, thereby improving the accuracy and stability of real-time error correction.

[0054] The effect of dynamically adjusting optimal process parameters is achieved by using a model predictive control algorithm to continuously optimize the process parameter settings for several future production cycles, based on the output deviation value of the real-time error correction model. This rolling optimization approach is forward-looking, capable of predicting future deviation trends based on current and past information and adjusting process parameters in advance, thereby more effectively controlling warpage and improving product quality. Rolling optimization is an iterative process; parameters are adjusted and optimized each production cycle based on new real-time data. This continuous improvement mechanism ensures that process parameters remain at their optimal state, adapting to various changes in actual production and continuously improving production stability and consistency.

[0055] Based on the overall optimization objectives and constraints, the optimization objective of the model predictive control algorithm is to minimize the weighted sum of prediction bias and control cost, taking into account both the accuracy of warpage prediction and the cost of adjusting process parameters. By reasonably setting the weights, it is possible to minimize control costs and maximize production efficiency while ensuring product quality.

[0056] The optimization process considered constraints such as injection speed, holding pressure, and mold temperature to ensure that the optimized process parameters met the requirements of actual production and the performance limitations of the equipment. This constraint improved the feasibility and practicality of the optimization results, avoiding production problems caused by unreasonable parameters. The optimized process parameters were automatically transmitted to the die-casting machine control system via an industrial automation protocol for adjustment, achieving automated control of the production process. Automated control not only improved production efficiency and reduced manual intervention and operational errors, but also enabled real-time response to changes in the production process, ensuring timely adjustment of process parameters and continuous, stable production.

[0057] More preferably, the intelligent analysis method further includes: Infrared temperature image sequences during the melt filling process are acquired in real time by using an infrared thermal imager installed at the die-casting mold. The infrared temperature image sequence is subjected to time-series analysis, and the instantaneous advancing velocity and temperature distribution of the melt front are extracted based on optical flow method or feature point tracking algorithm; Based on the real-time temperature data collected by thermocouple sensors deployed in the mold cavity and the phase transition temperature range of the die-cast material, the local solid fraction at each thermocouple location is calculated using the first calculation model or the second calculation model. Based on the temperature distribution at the melt front and the pre-defined temperature solid fraction mapping curve of the material, the instantaneous solid fraction at each position when the melt front reaches is calculated in real time. The point solid fraction calculated by thermocouples is fused with the surface solid fraction obtained by infrared inversion to obtain a high-precision solid fraction distribution. The step of inputting mold state data into the intelligent warping deformation prediction model for dynamic warping prediction specifically includes: The fused high-precision solid fraction distribution is input into the intelligent prediction model for warping deformation. Based on the correlation between the melt front advance velocity and the actual flow velocity in the gate region, the velocity distribution of the entire flow field is calculated and input into the intelligent prediction model for warping deformation. Based on the calculated flow velocity distribution and solid fraction distribution, the data are input into the intelligent prediction model for warping deformation. The intelligent prediction model for warping deformation dynamically outputs the predicted value of warping deformation and the distribution state of warping morphology at the current moment based on the above multi-source input.

[0058] Specifically, by using an infrared thermal imager installed at the die-casting mold to collect real-time infrared temperature image sequences during the melt filling process, it is possible to intuitively and comprehensively obtain the temperature distribution of the melt within the cavity and the dynamic process of filling. Compared to traditional single-point temperature measurement methods, the infrared thermal imager can cover the entire melt area, providing richer information and helping to gain a deeper understanding of the melt filling behavior.

[0059] Because infrared temperature image sequences have high temporal resolution, they can capture instantaneous temperature changes during melt filling. This is of great significance for analyzing the melt's flow characteristics, heat transfer process, and potential defects (such as cold shuts and flow marks), providing accurate basic data for subsequent solid fraction calculations and warpage prediction.

[0060] In practical implementation, time-series analysis of infrared temperature image sequences based on optical flow methods or feature point tracking algorithms can accurately extract the instantaneous propulsion velocity of the melt front. Optical flow methods calculate the velocity of an object by analyzing the movement of pixels in the image, while feature point tracking algorithms track specific feature points on the melt front, thus accurately determining its position and speed. This is crucial for understanding the filling progress and flow state of the melt. Besides propulsion velocity, temperature distribution information at the melt front can also be extracted. The temperature distribution at the melt front directly affects the change in solids fraction and the solidification process of the melt, significantly influencing the occurrence of warpage. By obtaining the temperature distribution at the front, a more comprehensive understanding of the melt's thermal state can be achieved, providing more detailed data support for subsequent solids fraction calculations and warpage prediction.

[0061] Based on real-time temperature data collected by thermocouple sensors deployed within the die cavity and the phase transition temperature range of the die-cast material, the local solid fraction at each thermocouple location is calculated using either a first or second calculation model. Different calculation models can be selected and adjusted according to specific material properties and process conditions, enabling flexible handling of various complex die-casting conditions and improving the accuracy and adaptability of solid fraction calculation.

[0062] Thermocouple sensors are commonly used temperature measurement devices in die-casting production. They are used to calculate local solid fraction, eliminating the need for numerous additional sensors and reducing costs and equipment complexity. Furthermore, thermocouple sensors offer high measurement accuracy and stability, providing a reliable data foundation for solid fraction calculations.

[0063] Real-time calculation of the frontal solids fraction is achieved by using a temperature-solids fraction mapping curve between the melt front temperature distribution and the material's pre-defined temperature-solids fraction mapping curve. This curve accurately reflects the relationship between temperature and solids fraction. By comparing the actual temperature of the melt front with the mapping curve, the frontal solids fraction can be calculated in real time, providing crucial dynamic parameters for predicting warpage deformation.

[0064] Specifically, the extraction of the melt front propulsion velocity includes: Temperature threshold segmentation is performed on the infrared temperature image sequence to identify the melt front position; Calculate the displacement of the melt front edge position between adjacent frames, and combine it with the inter-frame time interval to obtain the instantaneous propulsion velocity at each position point; Generate a full-view velocity distribution cloud map of the melt front and identify velocity anomaly regions, including velocity drop regions, velocity change regions, and flow stagnation regions.

[0065] In addition to the speed extraction mentioned above, the following early warning methods are also included in the specific implementation: The first early warning processing method is as follows: Based on the temperature distribution of the melt front edge identified by infrared temperature image recognition, the cold material convergence area is automatically identified: When two or more melt front edges are detected to meet, the temperature gradient and temperature difference at the convergence point are analyzed; if the temperature difference at the convergence point exceeds the preset threshold, or the temperature of any melt front edge is lower than the preset critical filling temperature, it is determined to be a high-risk area of ​​cold material convergence; the location coordinates and temperature difference of the cold material convergence area are recorded for subsequent defect early warning and processing path generation.

[0066] The second early warning processing method is as follows: Based on the melt front advancement speed identified by infrared temperature image, the filling time deviation is calculated in real time: The actual filling time of the melt from the gate to each position is calculated by integrating the melt front advancement speed; The actual filling time is compared with the preset theoretical filling time to calculate the filling time deviation rate; When the filling time deviation rate at any position exceeds the preset threshold, an abnormal filling time early warning is triggered.

[0067] The third early warning processing method is as follows: Based on the melt front temperature distribution identified by infrared temperature image recognition, the end solid fraction is calculated in real time; based on the melt front temperature and the preset temperature-solid fraction mapping curve of the material, the instantaneous solid fraction when the melt front reaches each position is calculated in real time; when the filling of the end region exceeds the preset warning threshold, the end overcooling warning is triggered; based on historical data, a correlation model between the end solid fraction and defects such as missing material and cold shut is established to predict the probability of end defects in real time.

[0068] In practical implementation, the automatic adjustment of mold structure or process parameters includes: When the actual flow rate at the ingate deviates from the preset velocity tolerance window, multi-variable collaborative optimization of velocity, pressure, and temperature is initiated. The velocity tolerance window is dynamically adjusted according to the filling process, and its width during the main filling stage is greater than that during the initial filling stage and the final filling stage. An optimization algorithm constrained by solids ratio is invoked, with the joint optimization objective of minimizing the warpage deformation and the non-uniformity of solids ratio distribution in each region at the solidification completion time. The optimal combination of process parameters is searched in the continuous parameter space. The objective function of the optimization algorithm includes the predicted warpage deformation value, the standard deviation of solids ratio in each region at the solidification completion time, the filling time deviation rate, and the shrinkage cavity ratio as a weighted sum.

[0069] Specifically, when a high-risk area of ​​cold material convergence is identified, the active control mechanism for the convergence area is activated, including at least one of the following actions: The local induction heating coil in the junction area is activated to heat the mold surface momentarily before the melt arrives, thereby increasing the melt temperature at the junction. Adjust the opening sequence of the needle valve hot nozzle of the multi-gate system to change the melt confluence position and move the confluence point to the preset overflow groove area. Increase the melt injection temperature by 5-15℃ to reduce melt viscosity and delay solidification; Reduce the local cooling intensity near the intersection area to slow down the solidification rate of the melt in that area.

[0070] When a region of sudden velocity drop or flow stagnation at the melt front is identified, the flow optimization actuator is activated, including at least one of the following actions: Increase hammer speed to increase filling flow rate; The flow distribution is changed by adjusting the cross-sectional area of ​​the gate through intelligent flow-throttling pins; Activate the vacuum-assisted filling system to reduce flow resistance; Optimize the gate location or runner design to eliminate flow dead zones.

[0071] When abnormal melt front advance speed or abnormal melt front temperature is detected based on infrared temperature image, multi-parameter collaborative optimization is initiated. The optimized parameters include at least: hammer speed, melt injection temperature, mold temperature distribution, and holding pressure curve.

[0072] In this embodiment of the invention, the dynamic warpage prediction further includes: The current filling mode is automatically identified based on the real-time velocity value. The filling modes include laminar flow filling mode, transition flow filling mode, and turbulent atomization mode. Based on the fused high-precision solid fraction distribution, isolated liquid phase regions and shrinkage cavity risk areas are identified. When a local area is found to have a higher temperature than the surrounding area and a lower solid fraction than the surrounding area, it is determined that an isolated liquid phase region exists. Based on pressure monitoring data at the filling end, a gas extrusion red zone is identified. When a pressure peak is detected that exceeds a preset threshold and is located in a fixed position, it is determined that a gas extrusion red zone exists.

[0073] Specifically, fusing the point solids content calculated by thermocouples with the surface solids content retrieved by infrared spectroscopy leverages the advantages of both methods. The point solids content calculated by thermocouples offers high local accuracy, while the surface solids content retrieved by infrared spectroscopy provides a general distribution across the entire melt region. Data fusion compensates for their respective shortcomings, yielding a high-precision solids content distribution that more accurately reflects the solidification state of the melt within the mold cavity. Individual measurement methods may contain inherent errors, such as those from thermocouple sensors or infrared thermal imagers. Data fusion, by integrating multiple data sources, reduces the impact of measurement errors on the solids content distribution, improves data reliability and accuracy, and provides a more reliable basis for predicting warpage deformation.

[0074] The integrated high-precision solids distribution and velocity distribution across the entire flow field are input into the intelligent warpage prediction model, comprehensively considering various factors influencing warpage. The solids distribution reflects the solidification state of the melt, while the velocity distribution reflects its flow characteristics. These factors are interrelated and mutually influential, jointly determining the generation and development of warpage. By comprehensively considering these factors, the accuracy and reliability of warpage prediction can be improved. Because the input data is acquired and calculated in real time, the intelligent warpage prediction model can dynamically predict based on actual changes during the production process. In die-casting production, process parameters, material properties, and other factors may change, leading to alterations in warpage. Dynamic prediction can promptly capture these changes, providing timely feedback for production process adjustments and control, thus contributing to improved product quality and production efficiency.

[0075] Specifically, the intelligent warpage prediction model is based on multi-source input and dynamically outputs the predicted warpage amount and warpage morphology distribution at the current moment. The predicted warpage amount quantitatively presents the degree of warpage, allowing production personnel to intuitively understand the product's deformation. The warpage morphology distribution shows the specific distribution of warpage on the product, aiding in analyzing the causes of warpage and determining improvement measures. This detailed output provides strong support for production process optimization and quality control.

[0076] More preferably, the step of calculating the local solid fraction at each thermocouple location using the first or second calculation model includes: The solid fraction calculation model is dynamically selected based on real-time requirements. During the first response phase, the first calculation model is used to calculate the local solid fraction at each thermocouple location. The calculation formula for the first calculation model is as follows: ,in, The calculated φ value, based on the real-time temperature data acquired by the thermocouple, is limited to the range of 0 to 1. φ represents the solid fraction. Liquidus temperature This refers to the solidus temperature. When in the second response stage, the local solid fraction at each thermocouple location is calculated using the first calculation model. The second calculation model is as follows: The initial alloy composition, liquidus temperature, solidus temperature, and equilibrium distribution coefficient of the die-cast material are obtained, wherein the equilibrium distribution coefficient is the ratio of solid phase composition to liquid phase composition. Starting with the liquidus temperature, the calculation is iterated downwards step by step according to the preset temperature step size; Within each temperature step, the solid phase increment is calculated based on the equilibrium distribution coefficient at the current temperature. The solid phase increment is determined based on the current liquid phase fraction, the current liquid phase composition, the initial alloy composition, and the equilibrium distribution coefficient. The cumulative solid fraction at the current temperature is obtained by summing the solid fraction increments at each temperature step. The composition of the remaining liquid phase is updated. The updated liquid phase composition is calculated based on the liquid phase composition and liquid phase fraction of the previous temperature step, the solid phase increment of the current temperature step, and the solid phase composition. The formula for calculating the solid phase increment is as follows: ,in, This represents the solid phase increment at the current temperature step. The liquid phase fraction at the previous temperature step. The liquid phase composition of the previous temperature step. Here, k represents the initial composition of the alloy, and k is the equilibrium distribution coefficient at the current temperature. Repeat the above iterative calculation until the preset termination condition is met. The termination condition includes the current temperature reaching the solidus temperature or the cumulative solid fraction reaching the preset complete solidification threshold. The final cumulative solid fraction is output as the local solid fraction at the location of the thermocouple sensor.

[0077] The first calculation model is adopted in the first response stage. The formula of this model is simple, which enables the calculation to be completed quickly in scenarios that require rapid acquisition of solid fraction information, and provides timely data support for subsequent production control, meeting the needs of production links with high real-time requirements.

[0078] A more complex second calculation model is employed in the second response stage. This stage may require higher accuracy in the solids fraction calculation, or the production process may enter a more complex state, necessitating a more precise reflection of the material's solidification process. By considering factors such as the initial alloy composition and equilibrium distribution coefficient, and through iterative calculations, the second calculation model can provide more accurate solids fraction information.

[0079] Dynamically selecting the calculation model based on different response stages demonstrates the flexibility of the method. In the die-casting production process, the solidification characteristics and process requirements of the material may differ at different stages. By dynamically selecting the model, these changes can be better adapted to, ensuring that appropriate solids ratio data can be obtained at each stage, thereby improving the stability and controllability of the entire production process.

[0080] Specifically, the first calculation model has a simple formula structure and clear physical meaning. The solid fraction can be obtained by performing simple mathematical operations on the measured thermocouple temperature values ​​and the liquidus and solidus temperatures. This intuitive calculation method is easy for engineers to understand and apply, reducing the complexity and probability of errors in the calculation process.

[0081] Because the calculation process is simple and does not require complex iterations or large amounts of data computation, the solid fraction result can be obtained in a short time. This is very important for die casting production, which requires real-time monitoring and rapid adjustment of production parameters, as it can provide timely feedback on the solidification state of the material and provide timely basis for production decisions.

[0082] The calculated φ value is limited to between 0 and 1, which conforms to the physical definition of solid fraction. 0 indicates that the material is completely in a liquid state, and 1 indicates that the material is completely in a solid state. This reasonable range limitation gives the calculation results clear physical meaning, making it easy to correlate and analyze with other production parameters, and providing a reliable data foundation for subsequent warpage prediction, etc.

[0083] The second calculation model acquires multiple parameters of the die-cast material, including the initial alloy composition, liquidus temperature, solidus temperature, and equilibrium partition coefficient, and performs calculations based on these parameters. The alloy composition significantly influences the solidification process, while the equilibrium partition coefficient reflects the relationship between the solid and liquid phase compositions. By comprehensively considering these factors, the behavior of the material during solidification can be described more accurately, improving the precision of solid fraction calculations.

[0084] This model starts at the liquidus temperature and iteratively calculates stepwise according to a preset temperature step. Within each temperature step, it calculates the solid phase increment based on the equilibrium distribution coefficient and updates the composition of the remaining liquid phase. This calculation method simulates the process of a material gradually transforming from a liquid to a solid state during actual solidification, considering the impact of compositional changes on solidification. This is more consistent with the physical nature of material solidification and thus provides more accurate solids fraction information. By iteratively accumulating the solids fraction within each temperature step, the model can more accurately track the material's solidification process. As the number of iterations increases, the calculation results gradually approach the true solids fraction value. Compared to some simpler calculation methods, it provides higher precision solids fraction data, making it particularly suitable for die-casting production scenarios with high precision requirements.

[0085] In some complex die-casting processes, the solidification process of materials can be affected by various factors, leading to complex changes in the solid fraction. The iterative calculation method of the second computational model can better adapt to this complexity. By continuously adjusting and updating the calculation parameters, it accurately reflects the solidification state of the material at different temperatures, providing reliable data support for the optimization and control of the production process. Termination conditions include the current temperature reaching the solidus temperature or the cumulative solid fraction reaching a preset threshold for complete solidification. When the temperature reaches the solidus temperature, the material has essentially completed the solidification process; while reaching the preset threshold for cumulative solid fraction allows for flexible setting of the standard for complete solidification based on actual production needs. Reasonable termination conditions ensure timely stopping of iterative calculations, avoiding unnecessary calculations, and guaranteeing the accuracy and completeness of the calculation results.

[0086] More preferably, the intelligent analysis method further includes: The melt velocity at the gate is obtained, and the corresponding air entrapment characteristic parameters are calculated based on the melt velocity and a pre-set air entrapment calculation formula. The air entrapment calculation formula is as follows: Where G is the characteristic parameter of the gas vortex. The fluid density of the melt. For melt velocity, The surface tension coefficient of the molten fluid. The characteristic length; The calculated air entrapment characteristic parameters are compared with the set conditions. If the calculated parameters are greater than the set air entrapment parameters, air entrapment is determined to have occurred, and an early warning is issued.

[0087] Specifically, the gas entrapment calculation formula comprehensively considers several key physical quantities, including melt density, melt velocity, surface tension coefficient, and characteristic length. These physical quantities affect the generation and extent of gas entrapment from different perspectives. For example, melt density reflects the mass characteristics of the melt, velocity determines the kinetic energy of the melt, surface tension coefficient affects the stability of the melt surface, and characteristic length is related to the geometric dimensions of structures such as gates. By comprehensively calculating these factors, the actual situation of gas entrapment can be more accurately reflected, providing a reliable basis for subsequent judgments.

[0088] Adaptable to various process conditions: Because this formula covers multiple factors affecting gas entrapment, it can adapt to different die-casting process conditions. Whether it is different material types (different materials have different densities and surface tension coefficients), different gate designs (different feature lengths), or different injection speeds (different melt speeds), the corresponding gas entrapment characteristic parameters can be calculated using this formula, demonstrating strong versatility and adaptability.

[0089] The calculated gas entrainment characteristic parameter is a specific numerical value that can quantitatively characterize the degree of gas entrainment. Compared to judging gas entrainment conditions solely through observation or experience, numerical indicators are more objective and accurate, facilitating quantitative analysis and comparison. For example, by comparing the gas entrainment characteristic parameters under different process parameters, we can analyze which factors have a greater impact on gas entrainment, thereby optimizing the process in a targeted manner.

[0090] Facilitates the establishment of evaluation standards: Based on quantified entrapment characteristic parameters, a unified entrapment evaluation standard can be established. A suitable set entrapment parameter is established as a threshold for determining whether entrapment has occurred. When the calculated entrapment characteristic parameter exceeds this set value, entrapment is confirmed to have occurred. This standardized evaluation method helps improve the consistency and stability of the production process and reduces interference from human factors.

[0091] In practice, by acquiring the melt velocity at the gate and calculating the gas entrapment characteristic parameters in real time, and then comparing them with set conditions, the gas entrapment situation can be monitored in real time. Once the gas entrapment characteristic parameters exceed the set value, gas entrapment can be immediately identified, allowing for the timely detection of potential problems in the production process. This is crucial for ensuring product quality, as gas entrapment can lead to defects such as porosity and looseness inside die-cast parts, affecting the mechanical properties and appearance quality of the product. Timely gas entrapment warnings allow for adjustments to be made before defects escalate significantly. For example, process parameters such as injection speed and pressure can be adjusted in a timely manner to change the flow state of the melt and reduce the occurrence of gas entrapment; or the mold can be optimized, and the gate design improved to reduce the risk of gas entrapment. Through early intervention, the further development of defects can be effectively prevented, improving the product yield.

[0092] Gas entrapment warning functionality helps production personnel promptly detect and address gas entrapment issues, preventing the production of a large number of defective die-cast parts. This reduces scrap, lowers production costs, and improves production efficiency. Because it eliminates the need to spend significant time and effort reworking or scrapping products with serious defects, more resources can be allocated to normal production processes. Continuous gas entrapment monitoring and warnings ensure that gas entrapment conditions remain within a controllable range during production, thus guaranteeing product quality stability. Whether in large-scale or small-batch production, it enables the production of consistent, compliant die-cast parts, enhancing the company's market competitiveness.

[0093] The dynamic mold flow intelligent analysis method for die casting molds in this embodiment of the invention identifies features in the die casting engineering drawings and generates a three-dimensional model with pre-deformation compensation. This not only enables the pre-compensation of shrinkage and warping that may occur after molding during the design stage, reducing defects from the source and improving product dimensional accuracy and first-pass yield, but also utilizes IoT sensors to collect mold status data in real time and inputs the data into a prediction model for dynamic warping analysis. This achieves continuous monitoring of the die casting process, enabling timely detection of abnormal trends and preventing the generation of batch defects.

[0094] Example 2 Please see Figure 3 , Figure 3 This is a schematic diagram of the dynamic mold flow intelligent analysis system for die-casting molds disclosed in an embodiment of the present invention. Figure 3 As shown, the dynamic mold flow intelligent analysis system for this die-casting mold may include: Generation module 21: Used to perform feature recognition on the die-casting engineering drawings to generate a 3D model with pre-deformation compensation; Module 22: Used to build an intelligent prediction model for warpage deformation based on the acquired mold material data, historical mold flow analysis data and real-time sensor data; Prediction module 23: is used to collect mold status data in real time during the die casting process based on IoT sensors, and input the mold status data into the warping deformation intelligent prediction model to perform dynamic warping prediction in order to obtain the corresponding warping deformation prediction value and warping morphology distribution status. Adjustment module 24: is used to match the predicted warpage deformation and warpage morphology distribution with the set conditions. If the set conditions are not met, an early warning is issued and the mold structure or process parameters are automatically adjusted.

[0095] The dynamic mold flow intelligent analysis method for die casting molds in this embodiment of the invention identifies features in the die casting engineering drawings and generates a three-dimensional model with pre-deformation compensation. This not only enables the pre-compensation of shrinkage and warping that may occur after molding during the design stage, reducing defects from the source and improving product dimensional accuracy and first-pass yield, but also utilizes IoT sensors to collect mold status data in real time and inputs the data into a prediction model for dynamic warping analysis. This achieves continuous monitoring of the die casting process, enabling timely detection of abnormal trends and preventing the generation of batch defects.

[0096] Example 3 Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of the present invention. The electronic device can be a computer, a server, etc. Of course, in certain cases, it can also be a mobile phone, tablet computer, monitoring terminal, or other smart device, as well as an image acquisition device with processing capabilities. Figure 4 As shown, the electronic device may include: Memory 510 storing executable program code; Processor 520 coupled to memory 510; The processor 520 calls the executable program code stored in the memory 510 to execute some or all of the steps in the dynamic mold flow intelligent analysis method for die casting molds in Embodiment 1.

[0097] This invention discloses a computer-readable storage medium storing a computer program that enables a computer to perform some or all of the steps in the dynamic mold flow intelligent analysis method for die-casting molds in Embodiment 1.

[0098] This invention also discloses a computer program product, wherein when the computer program product is run on a computer, the computer executes some or all of the steps in the dynamic mold flow intelligent analysis method for die casting molds in Embodiment 1.

[0099] This invention also discloses an application publishing platform, which is used to publish computer program products. When the computer program products are run on a computer, the computer executes some or all of the steps in the dynamic mold flow intelligent analysis method for die casting molds in Embodiment 1.

[0100] In various embodiments of the present invention, it should be understood that the sequence number of each process does not necessarily imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0101] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they can be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0102] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0103] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-accessible memory. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several requests to cause a computer device (which can be a personal computer, server, or network device, specifically a processor in the computer device) to execute some or all of the steps of the methods described in the various embodiments of the present invention.

[0104] In the embodiments provided by this invention, it should be understood that "B corresponding to A" means that B is associated with A, and B can be determined based on A. However, it should also be understood that determining B based on A does not mean determining B solely based on A; B can also be determined based on A and / or other information.

[0105] Those skilled in the art will understand that some or all of the steps in the various methods of the embodiments described can be implemented by a program instructing related hardware. This program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0106] The above provides a detailed description of the dynamic mold flow intelligent analysis method, system, electronic device, and storage medium for die casting molds disclosed in the embodiments of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A dynamic mold flow intelligent analysis method for die-casting molds, characterized in that, include: Feature recognition is performed on the engineering drawings of die-cast parts to generate a 3D model with pre-deformation compensation; Based on the acquired mold material data, historical mold flow analysis data, and real-time sensor data, an intelligent prediction model for warpage deformation is constructed. The mold status data during the die casting process is collected in real time by IoT sensors, and the mold status data is input into the intelligent warping deformation prediction model to perform dynamic warping prediction in order to obtain the corresponding predicted value of warping deformation and the distribution state of warping morphology. The predicted warpage deformation and warpage morphology distribution are matched with set conditions. If the set conditions are not met, an early warning is issued and the mold structure or process parameters are automatically adjusted.

2. The dynamic mold flow intelligent analysis method for die-casting molds as described in claim 1, characterized in that, The process of performing feature recognition on the die-casting engineering drawing to generate a 3D model with pre-deformation compensation includes: Feature recognition is performed on the acquired two-dimensional engineering drawings or three-dimensional point cloud data to determine the geometric topology and feature parameters of the corresponding die castings; Based on the geometric topology, characteristic parameters, and pre-built feature database, the corresponding solidity sensitive region is determined; the solidity sensitive region includes the first wall thickness region, the second wall thickness region, the corner region, and the gate distal region; In a 3D modeling environment, a model sketch with pre-deformation compensation is generated based on the spatial coordinates of the solidity-sensitive region and the solidity evolution sensitivity coefficient. The formula for calculating the pre-deformation compensation is as follows: ;in, Let i be the compensation amount for the i-th sensitive region. For compensation coefficient, This is the sensitivity coefficient for the evolution of the solid fraction in the corresponding region. For mold temperature, The melting temperature is... It is the critical solid fraction.

3. The dynamic mold flow intelligent analysis method for die-casting molds as described in claim 2, characterized in that, The step of determining the corresponding solidity-sensitive region based on the geometric topology, characteristic parameters, and pre-constructed feature database includes: Based on the solid fraction and temperature correlation curves in the database, the solid fraction sensitive areas of the casting to be analyzed are identified and determined. The solid fraction sensitive areas include the abrupt change in solid fraction evolution rate caused by wall thickness difference, the filling end area at the far end of the gate, and the heat concentration area caused by geometric abrupt change. Based on the identification results of the solidity sensitive region, local mesh refinement is performed on the solidity sensitive region during the three-dimensional solid mesh generation process, wherein the mesh refinement density is positively correlated with the solidity gradient. Based on the solid fraction evolution prediction results, differentiated pre-deformation compensation amounts are generated in adjacent regions where the difference in solid fraction evolution rate exceeds a preset threshold.

4. The dynamic mold flow intelligent analysis method for die-casting molds as described in claim 2, characterized in that, The intelligent prediction model for warping deformation is constructed in the following manner: A training dataset is constructed by collecting warping data sources, including material solid fraction versus temperature curve data, solid fraction evolution cloud map sequence data extracted from historical model flow analysis, and warping deformation measured data under the same working conditions as the solid fraction evolution cloud map sequence. A three-dimensional convolutional neural network is constructed. Taking the solid fraction evolution cloud map sequence as input, the evolution features of solid fraction in time and space are extracted simultaneously through the three-dimensional convolution kernel, and the solid fraction evolution feature vector is output. A feature parsing layer is connected after the three-dimensional convolutional neural network. The feature parsing layer converts the solid fraction evolution feature vector into solid fraction evolution feature parameters. The solid fraction evolution feature parameters include the time for each region to reach the critical solid fraction, the solid fraction spatial distribution non-uniformity index, and the solid fraction evolution time difference between the thick-walled region and the thin-walled region. A multilayer perceptron network is constructed as a warpage deformation prediction subnetwork. The input layer of the multilayer perceptron network receives the solid fraction evolution characteristic parameters, and the output layer outputs the predicted warpage deformation. The three-dimensional convolutional neural network, feature parsing layer and multilayer perceptron network are connected in series to form a complete intelligent prediction model for warping deformation. The model is then jointly trained end-to-end using the solid fraction evolution cloud map sequence as input and the corresponding measured warping deformation amount as supervision label.

5. The dynamic mold flow intelligent analysis method for die-casting molds as described in claim 4, characterized in that, The process by which the feature parsing layer converts the solid fraction evolution feature vector into solid fraction evolution feature parameters includes: By performing time-series analysis on the solid fraction evolution cloud map sequence, the time point at which the solid fraction of each pixel reaches the preset critical value is identified, and a critical time distribution map is formed. Calculate the root mean square error between the solid fraction at each location at the same time and the global average solid fraction, and use the value at the time of completion of filling as the non-uniformity index. Based on the pre-annotated thick-walled and thin-walled regions using geometric topology information, the average time for the two regions to reach the critical solid fraction is calculated, and the difference between them is determined. The end-to-end joint training uses mean squared error as the loss function, and performs joint optimization of the parameters of the 3D convolutional neural network, feature parsing layer and multilayer perceptron network through backpropagation algorithm, and uses cross-validation to select the optimal model parameters.

6. The dynamic mold flow intelligent analysis method for die-casting molds as described in claim 2, characterized in that, The intelligent analysis method also includes: A digital twin of the die-cast part to be formed is constructed, the digital twin including a three-dimensional model with pre-deformation compensation, a warping deformation intelligent prediction model and a real-time error correction model based on a recurrent neural network; The real-time error correction model calibrates the simulation prediction results based on the real-time data stream from sensors deployed at key locations in the mold, and dynamically adjusts the optimal process parameters to compensate for warping deformation in actual production. The real-time error correction model module is a time-series prediction model built on a long short-term memory network. Its input includes: real-time data streams from sensors deployed at key locations in the mold. The sensors include pressure sensors inside the mold cavity, temperature sensors, mold strain sensors, and temperature sensors at the inlet and outlet of the cooling pipes. The simulated warping prediction value is output by the intelligent warping deformation prediction model module; the corrected warping prediction value is output by the real-time error correction model module. The corrected warping prediction value is the result of dynamically compensating for the systematic deviation between the simulated warping prediction value and the actual measured warping value in production. The dynamic adjustment of optimal process parameters includes: Based on the output deviation value of the real-time error correction model, the process parameter settings for several future production cycles are optimized using a model predictive control algorithm. The optimization objective of the model predictive control algorithm is to minimize the weighted sum of prediction error and control cost, wherein the constraints include injection speed, holding pressure, and mold temperature. The optimized process parameters are automatically sent to the die-casting machine control system via industrial automation protocols to achieve pressure adjustment.

7. The dynamic mold flow intelligent analysis method for die-casting molds as described in claim 2, characterized in that, The intelligent analysis method also includes: Infrared temperature image sequences during the melt filling process are acquired in real time by using an infrared thermal imager installed at the die-casting mold. The infrared temperature image sequence is subjected to time-series analysis, and the instantaneous advancing velocity and temperature distribution of the melt front are extracted based on optical flow method or feature point tracking algorithm; Based on the real-time temperature data collected by thermocouple sensors deployed in the mold cavity and the phase transition temperature range of the die-cast material, the local solid fraction at each thermocouple location is calculated using the first calculation model or the second calculation model. Based on the temperature distribution at the melt front and the pre-defined temperature solid fraction mapping curve of the material, the instantaneous solid fraction at each position when the melt front reaches is calculated in real time. The point solid fraction calculated by thermocouples is fused with the surface solid fraction obtained by infrared inversion to obtain a high-precision solid fraction distribution. The step of inputting mold state data into the intelligent warping deformation prediction model for dynamic warping prediction specifically includes: The fused high-precision solid fraction distribution is input into the intelligent prediction model for warping deformation. Based on the correlation between the melt front advance velocity and the actual flow velocity in the gate region, the velocity distribution of the entire flow field is calculated and input into the intelligent prediction model for warping deformation. Based on the calculated flow velocity distribution and solid fraction distribution, the data are input into the intelligent prediction model for warping deformation. The intelligent prediction model for warping deformation dynamically outputs the predicted value of warping deformation and the distribution state of warping morphology at the current moment based on the above multi-source input.

8. The dynamic mold flow intelligent analysis method for die-casting molds as described in claim 7, characterized in that, The calculation of the local solid fraction at each thermocouple location using the first or second calculation model includes: The solid fraction calculation model is dynamically selected based on real-time requirements. During the first response phase, the first calculation model is used to calculate the local solid fraction at each thermocouple location. The calculation formula for the first calculation model is as follows: ,in, The calculated φ value, based on the real-time temperature data acquired by the thermocouple, is limited to the range of 0 to 1, where φ represents the solid fraction. Liquidus temperature This refers to the solidus temperature. When in the second response stage, the local solid fraction at each thermocouple location is calculated using the first calculation model. The second calculation model is as follows: The initial alloy composition, liquidus temperature, solidus temperature, and equilibrium distribution coefficient of the die-cast material are obtained, wherein the equilibrium distribution coefficient is the ratio of solid phase composition to liquid phase composition. Starting with the liquidus temperature, the calculation is iterated downwards step by step according to the preset temperature step size; Within each temperature step, the solid phase increment is calculated based on the equilibrium distribution coefficient at the current temperature. The solid phase increment is determined based on the current liquid phase fraction, the current liquid phase composition, the initial alloy composition, and the equilibrium distribution coefficient. The cumulative solid fraction at the current temperature is obtained by summing the solid fraction increments at each temperature step. The composition of the remaining liquid phase is updated. The updated liquid phase composition is calculated based on the liquid phase composition and liquid phase fraction of the previous temperature step, the solid phase increment of the current temperature step, and the solid phase composition. The formula for calculating the solid phase increment is: ,in, This represents the solid phase increment at the current temperature step. The liquid phase fraction at the previous temperature step. The liquid phase composition of the previous temperature step. Here, k represents the initial composition of the alloy, and k is the equilibrium distribution coefficient at the current temperature. Repeat the above iterative calculation until the preset termination condition is met. The termination condition includes the current temperature reaching the solidus temperature or the cumulative solid fraction reaching the preset complete solidification threshold. The final cumulative solid fraction is output as the local solid fraction at the location of the thermocouple sensor.

9. The dynamic mold flow intelligent analysis method for die-casting molds as described in claim 1, characterized in that, The intelligent analysis method also includes: The melt velocity at the gate is obtained, and the corresponding air entrapment characteristic parameters are calculated based on the melt velocity and a pre-set air entrapment calculation formula. The air entrapment calculation formula is as follows: Where G is the characteristic parameter of the gas vortex. The fluid density of the melt. For melt velocity, The surface tension coefficient of the molten fluid. The characteristic length; The calculated air entrapment characteristic parameters are compared with the set conditions. If the calculated parameters are greater than the set air entrapment parameters, air entrapment is determined to have occurred, and an early warning is issued.

10. A dynamic mold flow intelligent analysis system for die-casting molds, characterized in that, include: Generation module: Used to perform feature recognition on die-casting engineering drawings to generate a 3D model with pre-deformation compensation; The building module is used to construct an intelligent prediction model for warpage deformation based on the acquired mold material data, historical mold flow analysis data, and real-time sensor data. Prediction module: used to collect mold status data in real time during the die casting process based on IoT sensors, and input the mold status data into the warping deformation intelligent prediction model to perform dynamic warping prediction in order to obtain the corresponding warping deformation prediction value and warping morphology distribution status. Adjustment module: used to match the predicted warpage deformation and warpage morphology distribution with set conditions. If the set conditions are not met, an early warning is issued and the mold structure or process parameters are automatically adjusted.

Images