A mold base design system based on virtual simulation

Abstract

This invention discloses a mold base design system based on virtual simulation, specifically relating to the field of mold design. It includes: a multimodal perception fusion module, a digital twin modeling module, a multiphysics coupled simulation module, a hybrid intelligent optimization module, a digital twin closed-loop verification module, and a data hub module. Each module forms a design closed loop through the data hub. The multimodal perception fusion module adaptively acquires and fuses multi-source signals to generate high-reliability data; the digital twin modeling module constructs and iteratively corrects the model based on this data; the multiphysics coupled simulation module realizes multi-solver collaborative simulation under dynamic boundaries; the hybrid intelligent optimization module accelerates the generation of optimal solutions through secondary optimization and surrogate models; the digital twin closed-loop verification module detects defects and generates correction instructions; and the data hub module is responsible for data management, cross-module scheduling, and accuracy control. This system improves the accuracy and efficiency of mold base design, reduces the cost of physical mold trials, and is suitable for high-precision mold development scenarios.

Description

Technical Field

[0001] This invention relates to the field of mold design technology, and more specifically, to a mold base design system based on virtual simulation. Background Technology

[0002] As the core load-bearing component in injection molding, the mold base directly affects the precision of the product and production efficiency, and there is an urgent need for it in the automotive, 3C electronics and other fields. The industry's digital transformation has begun to show results, and existing technologies have significant advantages: CAD / CAE tools are widely used, which can quickly complete the 3D modeling of the mold base and the basic performance verification; multimodal sensing technology is gradually becoming popular, which can collect key physical data such as temperature and stress; basic multiphysics simulation can realize simple thermal and mechanical coupling analysis; data management systems can complete the storage and archiving of design files and simulation results, which significantly shortens the preliminary design cycle compared with traditional hand-drawn design and physical mold trials.

[0003] However, it still has some drawbacks in practical use, such as:

[0004] 1. The sensing process relies on experience to plan sensor locations, which can easily lead to blind spots in highly sensitive areas or sensor redundancy. Multimodal data is simply stitched together without dynamic weight allocation and reliability assessment, resulting in insufficient data source accuracy and affecting the foundation for subsequent modeling.

[0005] 2. Digital twin modeling is mostly offline static correction, and parameter updates lag behind changes in working conditions. The real-time mapping deviation between the model and the physical mold is large, making it difficult to accurately reflect the actual operating status of the mold and restricting the reliability of simulation and optimization.

[0006] 3. Optimization often uses a single algorithm for optimization and relies on a large number of entity simulation iterations. It has not formed an efficient two-level optimization framework, and the training and update mechanism of the proxy model is not perfect, resulting in a long optimization cycle and difficulty in balancing accuracy and efficiency.

[0007] 4. Data from each module is stored in a scattered manner, and cross-module collaboration relies on manual coordination. There is a lack of unified central scheduling and end-to-end accuracy control, resulting in process breakpoints and the inability to form a closed-loop iteration of design, simulation and verification. Summary of the Invention

[0008] In order to overcome the above-mentioned defects of the prior art, the present invention provides a mold base design system based on virtual simulation, which solves the problems mentioned in the background art through the following solutions.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a mold base design system based on virtual simulation, comprising:

[0010] Multimodal perception fusion module: Adaptively plans sensor locations based on the characteristics of the model frame, collects multimodal signals, and after noise reduction and spatiotemporal synchronization processing, achieves data fusion through dynamic weight allocation via an attention mechanism, generates standardized fused data with a credibility index, and transmits it to the data center;

[0011] Digital twin modeling module: retrieves fused data and standard template library from the data center to build an initial model, uses iterative algorithm to correct model parameters in reverse, responds to external correction commands to complete model iterative updates, and the updated model is sent back to the data center;

[0012] Multiphysics Coupled Simulation Module: Based on the model and fused data of the data hub, the fused data is transformed into dynamic boundary conditions, hybrid mesh optimization is implemented, multiple solvers are integrated to carry out coupled simulation in accordance with the FMI standard, and the simulation results are output to the data hub;

[0013] Hybrid intelligent optimization module: Based on the simulation results of the data center, a multi-objective optimization model is constructed. It adopts a two-level algorithm framework of global search and local optimization, combined with a surrogate model to accelerate the optimization process, and outputs the optimal solution to the data center;

[0014] Digital twin closed-loop verification module: Obtain the model corresponding to the optimal solution from the data center, conduct immersive verification in a virtual environment, identify defects through multi-dimensional detection, and generate structured correction instructions to be fed back to the data center;

[0015] Data Hub Module: Enables storage and version tracking of all types of data, schedules cross-module collaboration based on preset event rules, synchronously monitors the accuracy of the entire link, and triggers correction mechanisms.

[0016] The technical effects and advantages of this invention are as follows:

[0017] 1. Enhanced Perception Precision: Efficient Fusion of Multimodal Data

[0018] An adaptive planning algorithm for sensor locations, adapted to the characteristics of the template, is adopted, and quantitative constraints are combined to achieve a precise balance between monitoring coverage and resources. Multi-source preprocessed data is integrated through a dynamic weight allocation mechanism, and a credibility assessment system is constructed simultaneously to generate high-precision standardized fusion data, providing a reliable data foundation for subsequent steps.

[0019] 2. Dynamic twin modeling: Real-time and accurate mapping between virtual and real worlds.

[0020] Driven by real-time sensing data, a dynamic iterative correction mechanism for model parameters is established, overcoming the limitations of offline static adjustment. Through a deviation minimization objective function optimization algorithm, the mapping accuracy between the digital model and the physical model is continuously calibrated, achieving real-time synchronous matching of their states and improving the reliability of simulation and optimization results.

[0021] 3. Intelligent optimization for high efficiency: improving both accuracy and efficiency.

[0022] A two-tier algorithm architecture combining global search and local optimization is constructed to balance the globality and precision of the design solution. An adaptive surrogate model is introduced to replace a large amount of entity simulation calculations, along with an efficient training and incremental update mechanism. This significantly shortens the iteration cycle and achieves performance upgrades while ensuring the accuracy of multi-objective optimization.

[0023] 4. End-to-end collaborative closed-loop system: integrated process control

[0024] A unified data hub integrates data from the entire process, enabling standardized storage, version traceability, and efficient retrieval. An event-driven mechanism enables automated collaborative scheduling across modules. Combined with a real-time end-to-end accuracy monitoring and deviation correction system, manual connection points are eliminated, and a complete closed-loop process for design, simulation, and verification is constructed. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the overall structure of the present invention.

[0026] Figure 2 This is a schematic diagram of the hybrid intelligent optimization module of the present invention. Detailed Implementation

[0027] 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.

[0028] refer to Figures 1-2 The mold base design system based on virtual simulation shown includes:

[0029] Multimodal perception fusion module: Through a three-level process of perception planning, signal processing, and data fusion, it realizes the accurate acquisition and digital integration of physical model status data, providing a highly reliable data source for subsequent digital twin modeling. Specifically, it includes: adaptive sensor deployment, multi-source signal preprocessing, and intelligent data fusion.

[0030] It should be further explained that the adaptive sensor deployment submodule includes: pre-analysis of module characteristics and parameter input, adaptive planning of sensor locations, sensor adaptation and standardized deployment, and post-deployment calibration and adaptation verification; its specific analysis is as follows:

[0031] Mold base characteristic pre-analysis and parameter input: Import the topological structure data of the target mold base through 3D modeling tools, including the geometric parameters of mold plate size, cavity distribution, runner direction, and cooling system layout. Simultaneously input the mold base material properties and injection molding process parameter range. Mold base material properties include: elastic modulus, thermal conductivity, and yield strength. The injection molding process parameter range is: injection pressure 0-200MPa, mold temperature 20-300℃, and cycle time 20-300s.

[0032] Operational mechanics and thermal pre-simulation: Based on the geometry and material parameters of the mold frame, a simplified finite element algorithm is used to identify potential stress concentration areas, and thermal conduction simulation is used to locate areas of abrupt temperature gradient changes, outputting a coordinate set of highly sensitive areas.

[0033] Sensor location adaptive planning:

[0034] Construct the objective function for site planning: with constraints of monitoring coverage greater than 95%, minimizing the number of sensors, and signal interference less than 5%, use a topology optimization algorithm to solve for the optimal location;

[0035] Output point planning results: Generate a deployment list including sensor type, installation coordinates, monitoring parameters, and accuracy requirements. Temperature sensors cover the cavity surface and key nodes of the cooling channel. Stress and strain sensors focus on the stress concentration areas identified by pre-simulation. Displacement sensors are deployed on the mold parting surface and moving parts. Vibration sensors are arranged at the connection between the mold frame and the injection molding machine.

[0036] Sensor adaptation and standardized deployment:

[0037] Sensor selection and adaptation: Select the corresponding type of sensor according to the deployment list. Temperature sensing uses K-type thermocouples, stress and strain sensing uses resistance strain gauges, displacement sensing uses laser displacement sensors, and vibration sensing uses ICP-type accelerometers. All sensors are equipped with standardized digital interfaces to achieve plug-and-play with edge nodes.

[0038] The physical deployment process is as follows:

[0039] Temperature sensors are fixed to designated locations using high-temperature resistant thermally conductive adhesive, ensuring a contact ratio of over 90% between the sensing end and the mold base surface. Stress-strain sensors are bonded using epoxy adhesive; the mold base surface is sanded and degreased before bonding, and then allowed to cure for 24 hours after bonding to achieve the desired surface roughness. Required to be less than 1.6 ;

[0040] The displacement sensor is fixed by a magnetic base, and the optical path is adjusted so that the laser is perpendicular to the monitoring surface, with an obstruction rate of less than 1%.

[0041] The vibration sensor is rigidly connected by a thread, and the flatness of the mounting surface is less than 0.02 mm / m to avoid resonance with the mold frame;

[0042] Post-deployment calibration and adaptation verification:

[0043] Static calibration: Each sensor is calibrated point-to-point using standard calibration equipment, the calibration coefficients are recorded and written to the edge node storage unit to ensure that the sensor measurement error is less than ±0.5%FS;

[0044] Dynamic adaptation verification: Start the formwork to run under no-load, collect the initial signals of the sensors, and use the signal integrity detection algorithm to determine the effectiveness of the deployment. For unqualified points, readjust the deployment position or replace the sensors until all points meet the monitoring requirements.

[0045] It should be further explained that the multi-source signal preprocessing submodule includes: signal acquisition and preliminary conditioning, multimodal signal noise reduction, data spatiotemporal synchronization and outlier removal, and preprocessed data standardization output; its specific analysis is as follows:

[0046] Signal acquisition and preliminary conditioning:

[0047] Edge node signal acquisition: Receives analog signals from various sensors, including temperature, stress-strain, displacement, and vibration signals, through a standardized interface. The acquisition frequency is dynamically configured according to the sensor type: 10Hz for temperature, 100Hz for stress-strain, 1kHz for displacement, and 10kHz for vibration. A unique identifier is added to each signal simultaneously.

[0048] Adaptive gain adjustment: The edge node has a built-in signal conditioning unit that identifies the signal amplitude in real time through a peak detection algorithm. When the amplitude is less than 20% of the full scale, the gain is automatically increased, with a maximum amplification of 1000 times. When the amplitude is more than 80% of the full scale, the gain is reduced, with a minimum amplification of 100 times, ensuring that the signal amplitude after conditioning is within 20%-80% of the ADC range, thus improving sampling accuracy.

[0049] Multimodal signal noise reduction processing:

[0050] Wavelet threshold denoising: The conditioned signal is decomposed into wavelets, the threshold of wavelet coefficients at each level is calculated, coefficients below the threshold are set to zero, and the denoised signal is reconstructed to reduce high-frequency noise interference.

[0051] Kalman filter optimization: To address the low-frequency drift in the denoised signal, a second-order autoregressive state equation is constructed. Using the wavelet denoising result as the observation, the signal baseline drift is eliminated through Kalman filter prediction and update iteration, and finally a stable single-mode signal is output.

[0052] Data spatiotemporal synchronization and outlier removal:

[0053] Timing alignment: Extract the timestamp of each signal, and use the highest sampling frequency of the vibration signal, 10kHz, as a reference. Use a linear interpolation algorithm to complete the data of the low-frequency signals, so that the time step of all modal signals is unified to 0.1ms, and timing synchronization is achieved. The low-frequency signals are: temperature, stress-strain, and displacement.

[0054] Spatial alignment: Based on the sensor installation coordinates output in step one, establish a mapping relationship between the sensor position and the physical coordinates of the mold frame, bind each signal to the three-dimensional coordinates of the mold frame, and realize spatial matching of multimodal data;

[0055] Outlier removal: A sliding window method is used, with a window size of 100 data points, to calculate the mean of each signal in real time. with standard deviation When a data point satisfies When an outlier occurs, it is identified as an outlier and is filled in by linear interpolation of the preceding and following data points to ensure data continuity.

[0056] Standardized output of preprocessed data:

[0057] The spatiotemporally synchronized single-modal data is converted into a standardized digital format, organized according to the structure of signal identifier, physical coordinates, timestamp, and numerical value, and temporarily stored in the local cache of the edge node, awaiting fusion processing.

[0058] It should be further explained that the intelligent data fusion submodule includes: fusion model initialization and parameter configuration, real-time multimodal data fusion, fusion data credibility assessment and encrypted transmission, and online fusion model update, which are analyzed in detail below:

[0059] Fusion model initialization and parameter configuration:

[0060] Construct an attention mechanism fusion model: Based on the TensorFlow framework, a fusion network is built. The input layer receives four single-modal standardized data, including temperature, stress, displacement, and vibration. The hidden layer calculates the weight coefficients of each modality data through an attention weight matrix with dimensions 4×N, where N is the number of fusion features. The output layer generates a fusion feature vector.

[0061] Model pre-training: The fusion model is pre-trained using at least 1,000 sets of historical perception data of the same model. Multimodal data and measured real values ​​of physical quantities are used as training samples. Attention weights are optimized through backpropagation to make the fusion error of the pre-trained model less than 5%, laying the foundation for real-time fusion.

[0062] Real-time multimodal data fusion:

[0063] Dynamic weight update: Standardized single-modal data output from multimodal signal denoising is input into the pre-trained model, and real-time weight coefficients are calculated through an attention mechanism. When the signal-to-noise ratio of a certain modality signal exceeds 45dB, the weight is increased, up to a maximum of 0.4. When it is below 35dB, the weight is decreased, down to a minimum of 0.1, ensuring that high-reliability data dominates the fusion process.

[0064] Feature fusion calculation: The feature vectors of the four modal data are weighted and summed according to the real-time weight coefficients to generate a fusion feature vector of dimension N, which contains the comprehensive state information of the model at a certain physical coordinate.

[0065] Integrating data trustworthiness assessment with encrypted transmission:

[0066] Credibility Calculation: Based on indicators such as signal-to-noise ratio, calibration error, and data continuity of each modal signal, a credibility evaluation function is constructed to output the credibility index of the fused data, with a value of [0, 1], and a value greater than 0.95 is considered acceptable; the function logic is as follows:

[0067] ),

[0068] .

[0069] Encryption and Transmission: The fused feature vector and credibility index are encrypted using AES-128 and pushed to the data hub module via the MQTT protocol. During transmission, a heartbeat mechanism with a 1-second interval is used to monitor the connection status. If the connection is interrupted, the data is temporarily stored and automatically retransmitted after the connection is restored to ensure that no data is lost.

[0070] Online updates for the fusion model:

[0071] The data center periodically reports the deviation between the fused data and the subsequent simulation results. When the average deviation of 100 consecutive sets of data is greater than 8%, the fusion model is triggered to update online: using the latest single-modal data and simulation verification values ​​as samples, the attention weight matrix is ​​fine-tuned using a transfer learning strategy, which can improve the fusion accuracy and adapt to changes in the working conditions of the mold without retraining.

[0072] Digital twin modeling module: Driven by multimodal perception fusion data, it constructs a parameterized digital model that is mapped in real time to the physical model through a closed-loop process of basic modeling, dynamic correction and iterative update, providing an accurate digital carrier for subsequent multiphysics simulation and optimization. Specifically, it includes: basic model generation, adaptive parameter correction and model iterative update.

[0073] It should be further explained that the basic model generation submodule includes: modeling input parameter retrieval and verification, standard template library matching and initial model generation, and model-sensor point mapping association, which are analyzed in detail below:

[0074] Modeling input parameter retrieval and verification:

[0075] Data Hub Interaction: Parameter retrieval requests are sent to the data hub via the gRPC interface to obtain three types of core input data:

[0076] Basic parameters of the mold frame: including topology (number of template layers, number and distribution of cavities), geometric dimensions (length, width and thickness of template, spacing between guide pillars), and material properties (elastic modulus, Poisson's ratio, thermal conductivity);

[0077] Product and process related parameters: maximum product outline dimensions, injection molding machine parameters (maximum mold capacity, clamping force), preliminary process window (mold temperature, injection pressure range);

[0078] Sensing point mapping data: A mapping table of sensor installation coordinates and physical coordinates of the mold frame output by the multimodal sensing fusion module;

[0079] Parameter validity verification: Construct a parameter verification rule base to automatically detect the completeness and rationality of input data. If an anomaly is found, trigger the data center to retransmit or manually correct it. After the verification is successful, proceed to the modeling process.

[0080] Standard template library matching and initial model generation:

[0081] Model adaptation and screening within the library: The standardized template parameter library is called, and based on the matching rules, the K-nearest neighbor algorithm is used to screen 3 to 5 candidate basic models from the library. The matching rules are preset based on the existing equipment conditions.

[0082] Initial model parameterization: Select the candidate model with the highest matching degree and perform initial modeling through the parameterized modeling engine: ( )

[0083] Define core design variables: set cavity wall thickness, support column diameter and spacing, stiffener height and number, etc. as modifiable parameters and bind parameter labels;

[0084] Generate initial 3D model: Automatically adjust candidate model parameters according to input geometry to build a complete 3D model including mold cavity, flow channel, cooling system and ejection mechanism, and output initial STEP format file (accuracy level 0.01mm).

[0085] Model-sensor point mapping association:

[0086] Parameter and sensor binding: Based on the sensing point mapping data obtained from the modeling input parameters, the physical location of each sensor is marked in the initial model, a two-way mapping relationship between the key parameters of the model and the sensor monitoring points is established, and stored as an XML format mapping table.

[0087] Initial model lightweight preprocessing: The initial model is lightweighted using simplified LOD technology: the geometric details of key structures such as cavities, guide pillars, and support pillars are preserved, and features of non-critical areas are simplified, reducing the number of triangular facets in the model by at least 40%, which facilitates subsequent real-time correction and transmission.

[0088] It should be further explained that the adaptive parameter correction submodule includes: real-time fused data reception and coordinate matching, adaptive correction calculation of model parameters, and consistency verification of the corrected model. A detailed analysis follows:

[0089] Real-time fusion data reception and coordinate matching:

[0090] Dynamic data subscription: Subscribe to real-time data output by the multimodal perception fusion module through the streaming data interface of the data hub. The data format includes signal identifier, physical coordinates, timestamp, numerical value, and confidence index. The subscription frequency is synchronized with the perception data update frequency.

[0091] Model coordinate matching: The mapping table generated by the mapping association between the model and the sensing points is called to convert the physical coordinates in the real-time fused data into parameter-related coordinates in the model, locate the key parameters of the model corresponding to the data, and only select high-confidence data with a confidence index greater than 0.95 for correction.

[0092] Model parameter adaptive correction calculation:

[0093] Correction objective function construction: With the goal of minimizing the deviation between the model simulation values ​​and the perceived measured values, a correction objective function is constructed as follows: ;

[0094] in, This is the vector of model parameters to be corrected. For parameters The model simulation values ​​are as follows. To perceive the measured value, The weighting coefficients for each parameter are set based on the sensitivity of the parameter to the mold performance.

[0095] Least squares iterative correction: The Gauss-Newton iterative algorithm is used to solve for the optimal solution of the objective function.

[0096] Initialize iteration parameters (Initial model parameter values), calculate initial deviations ;

[0097] Constructing the Jacobian matrix ,pass Update parameters, These are regularization coefficients to avoid matrix singularities. Let k be the error vector, and k be the ordinal number.

[0098] Iterate repeatedly until the deviation , and output the corrected parameter vector .

[0099] Consistency verification of the corrected model:

[0100] Consistency verification of physical quantities: Substitute the corrected model parameters into the simplified mechanics or thermodynamics simulation (using the fast finite element algorithm, calculation time ≤ 10s), output the simulation results such as the stress distribution and temperature field of the model, and compare them with the corresponding physical quantities in the同期感知融合数据 (synchronous perception fusion data):

[0101] If the deviation between the simulation value and the measured value in the key area is less than or equal to 5%, the verification is determined to be qualified;

[0102] If the deviation exceeds 5%, then trace back and check the parameter mapping relationship and the correction algorithm parameters, and re - execute the correction process.

[0103] Rationality verification of the topological structure: Call the die structure design rule library to automatically detect whether the topological structure of the corrected model meets the process requirements, and avoid structural interference or decreased process feasibility caused by parameter correction.

[0104] It should be further noted that the model iteration update sub - module includes: external correction instruction reception and parsing, model iteration update and version management, lightweight processing and downstream module push, and secondary model consistency verification. The specific analysis is as follows:

[0105] External correction instruction reception and parsing:

[0106] Multi - source instruction access: Receive two types of correction instructions through the gRPC interface:

[0107] Feedback from the closed - loop verification module: Structured instructions of problem type, position coordinates, and correction parameter requirements;

[0108] Output from the optimization module: Parameter adjustment values in the optimal design scheme.

[0109] Instruction priority sorting: Sort according to the rule of assembly interference (priority 1) > performance defect (priority 2) > optimization parameter adjustment (priority 3). When there are conflicting instructions, execute the high - priority instructions first, and synchronously record the conflict information in the data center.

[0110] Model iteration update and version management:

[0111] Batch parameter adjustment: Convert the sorted correction instructions into model parameter adjustment values, and automatically update the corresponding parameters through the batch modification interface of the parametric modeling engine to generate an updated three - dimensional model.

[0112] Version identification and storage: Generate a unique version number according to the model model, iteration number, and timestamp rules, and push the updated model file and parameter change log to the version management submodule of the data center to realize full historical version backtracking.

[0113] Lightweight processing and downstream module push:

[0114] Iterative model lightweight optimization: A second lightweighting process is performed on the updated model using an improved LOD technique.

[0115] The critical structure retains the highest precision;

[0116] Non-critical structures are simplified progressively to ensure that the model file size is ≤100MB, meeting the high-efficiency loading requirements of the simulation module.

[0117] Model push and trigger notification: The lightweight model is pushed to the multiphysics coupling simulation module through the data hub. At the same time, a model update completion trigger signal is sent, which includes the version number and the coordinates of the correction area. This triggers the simulation module to start a new round of coupling analysis, realizing automatic linkage between modeling and simulation.

[0118] Model consistency secondary verification:

[0119] After the simulation module outputs the first round of simulation results, it retrieves the model simulation values ​​and the perceived measured values ​​from the data center for comparison. If the deviation is less than or equal to 0.02mm, the iteration is deemed valid; if the deviation is greater than 0.02mm, the adaptive parameter correction submodule is automatically triggered to re-execute the correction process until the mapping accuracy between the model and the physical template meets the requirements.

[0120] Multiphysics Coupled Simulation Module: The multiphysics coupled simulation module uses a digital twin model as a carrier and combines sensor fusion data to carry out thermal, mechanical, and fluid coupling analysis, outputting high-precision simulation results to support optimization decisions; it includes: dynamic boundary condition configuration, adaptive mesh optimization, collaborative simulation scheduling, simulation result processing and output;

[0121] It should be further explained that the dynamic boundary condition configuration submodule includes: data retrieval and verification, dynamic modeling of boundary conditions, and real-time update of condition configuration, which are analyzed in detail below:

[0122] Data retrieval and verification: Three types of data are retrieved from the data hub via the gRPC interface: lightweight digital twin model, multimodal sensing fusion data (including temperature and pressure time series data), and injection molding process parameters (filling and holding time). Data verification rules are constructed to automatically detect the integrity of the model topology, the credibility of the sensing data, and the rationality of the process parameters. If an anomaly occurs, retransmission is triggered.

[0123] Dynamic modeling of boundary conditions: Using sensing data as input, boundary conditions that change with the injection cycle are constructed: the inlet and outlet temperature data of the cooling channel are mapped as thermal boundary conditions, the injection pressure data is converted into fluid loads, and the clamping force data is used as structural loads. A linear interpolation algorithm is used to achieve time-series matching between the sensing data and the simulation time step (0.1s).

[0124] Real-time configuration update of conditions: Design a boundary condition linkage mechanism, subscribe to the sensing data update signal of the data center through the MQTT protocol, and automatically trigger boundary condition reconfiguration when the rate of change of a certain physical quantity is greater than 10%, so as to ensure that the simulation input is synchronized with the physical conditions.

[0125] It should be further explained that the adaptive mesh optimization submodule includes: stress concentration region identification, hybrid mesh strategy execution, and mesh independence verification, the specific analysis of which is as follows:

[0126] Stress concentration area identification: Extract stress time series data from the sensing fusion data, set the stress value ≥ 70% of the material's allowable stress as the threshold, identify high stress areas, and output the area coordinate set.

[0127] Hybrid mesh strategy execution: The mesh generation engine is invoked, using 0.67mm tetrahedral elements for high-stress areas and 5mm hexahedral elements for non-critical areas to create a hybrid mesh model. A mesh quality assessment algorithm controls the distortion rate to be less than or equal to 15% and the aspect ratio to be less than or equal to 5.

[0128] Mesh independence verification: Three sets of meshes with different refinement levels were selected for pre-simulation. The stress value deviations in key areas were compared. When the deviation was less than or equal to 5%, the optimal mesh parameters were determined.

[0129] It should be further noted that the collaborative simulation scheduling submodule includes:

[0130] Solver Collaborative Configuration: A collaborative framework is built in accordance with the FMI2.0 standard, integrating the Calculix structural solver and the OpenFOAM fluid solver, and data interaction interface encapsulation is implemented through functional mock-up units.

[0131] Simulation phase division and execution: Based on the injection molding cycle, the simulation is divided into three phases: filling, holding pressure, and cooling. Solver resources are dynamically allocated: during the filling phase, the fluid solver is prioritized with a step size of 0.05s; during the cooling phase, structural and thermal coupling calculations are strengthened. A solver interaction interval of 0.05s is set to transfer coupled physical quantities such as temperature and pressure.

[0132] Simulation process monitoring: Real-time monitoring of solution convergence; convergence is determined when the residual is less than or equal to 1e-4; if divergence occurs, the mesh accuracy or time step is automatically adjusted and the local simulation is restarted; a full-cycle calculation log is generated after the simulation is completed.

[0133] It should be further explained that the simulation result processing and output submodule includes: result data parsing, result credibility assessment, and data push and triggering, which are analyzed in detail below:

[0134] Results data analysis: Extract the HDF5 format file of the simulation output, and analyze it to obtain core data such as stress cloud map (nodal stress value), temperature field distribution (unit temperature), and flow front trajectory. Then, associate the three-dimensional coordinates of the mold frame to generate a structured dataset.

[0135] Result credibility assessment: Compare the simulation values ​​with the synchronous sensing data, calculate the deviation rate (below 8% is acceptable), and generate a credibility report; if unacceptable, backtrack to optimize the boundary conditions or mesh parameters and re-simulate.

[0136] Data push and trigger: The qualified simulation dataset and credibility report are encrypted and written into the data center. At the same time, a simulation completion trigger signal is sent to automatically start the optimization process of the hybrid intelligent optimization module, realizing the linkage between simulation and optimization.

[0137] Hybrid Intelligent Optimization Module: Based on multiphysics simulation results, it generates the optimal design scheme of the mold frame through multi-objective algorithms and surrogate models. It includes: optimization objective and constraint modeling, hybrid optimization algorithm scheduling, surrogate model acceleration, and optimization scheme generation.

[0138] It should be further explained that the specific steps of the optimization objective and constraint modeling submodule are as follows:

[0139] Objective function construction: Retrieve the simulation dataset from the data hub and construct a multi-objective optimization function:

[0140] Main objective: Maximize weight loss rate ( Maximizing stiffness improvement rate Maximizing lifespan extension ;

[0141] Target weights: Allows users to set weight coefficients through an interactive interface. The default setting is uniform weighting.

[0142] Constraint Definition: Based on material properties and process requirements, constraints are set as follows:

[0143] Hard constraints: maximum stress ≤ 80% of material yield strength, maximum deformation ≤ 0.03 mm, cooling time ≤ 60% of injection molding cycle;

[0144] Soft constraints: Design the range of variables, stored as an XML constraint file.

[0145] Model optimization encapsulation: The objective function and constraints are encapsulated into a standardized optimization model, which is then pushed to the algorithm scheduling submodule through the data hub interface and associated with the corresponding digital twin model version.

[0146] It should be further explained that the specific steps of the hybrid optimization algorithm scheduling submodule are as follows:

[0147] Initial parameter configuration: Read the optimization model, determine the dimension of design variables (n) and the number of constraints (m), and initialize the algorithm parameters:

[0148] NSGA-II algorithm: population size 100, number of iterations 50, crossover probability 0.8, mutation probability 0.01;

[0149] Bayesian optimization: using the Matérn kernel function, exploration coefficient 2.5, and a maximum of 30 iterations.

[0150] Second-level optimization execution:

[0151] Global Search: The NSGA-II algorithm is launched to perform global optimization on the design variable space, generating a Pareto front solution set containing 100 solutions, and selecting the top 20% of solutions with the comprehensive score of the objective function as the candidate set;

[0152] Local optimization: Starting with the candidate set, run Bayesian optimization, predict the objective function value through the Gaussian process model, and output 3-5 Pareto optimal solutions.

[0153] Algorithm adaptive adjustment: The convergence of the solution is calculated every 10 iterations (convergence is determined when the deviation between the optimal solutions of two adjacent generations is ≤1%). If the convergence is slow, the mutation probability is automatically increased (up to 0.05) or the exploration range is expanded.

[0154] It should be further explained that the specific steps of the proxy model acceleration submodule are as follows:

[0155] Training data preparation: Extract 1000 sets of historical simulation data and the latest 50 sets of simulation results from the data hub, construct a training set according to the design variables and performance index formats, and process the data using Min-Max standardization.

[0156] Multi-scale neural network training:

[0157] Model structure: Input layer (n neurons), hidden layer (2 layers × 32 neurons), output layer (3 neurons, corresponding to 3 objective functions);

[0158] Training strategy: The Adam optimizer (learning rate 0.001) is used, and the network weights are initialized by transfer learning. The model is then fine-tuned with new data to make the prediction error ≤8%.

[0159] Real-time prediction and update: During the optimization process, the surrogate model predicts the performance indicators corresponding to the design variables in real time, replacing part of the simulation calculation; after each round of optimization, the newly generated optimal solution is added to the training set, triggering incremental model updates (training time ≤ 5 minutes).

[0160] It should be further explained that the specific steps for generating the optimization scheme submodule are as follows:

[0161] Solution evaluation and screening: The manufacturability of the optimal solution is evaluated by calling the process knowledge base, calculating the manufacturing cost index and process feasibility score of each solution, and eliminating solutions with a feasibility score of less than 60.

[0162] Results encapsulation and output: Generate an optimization plan report, including design variable parameters, performance improvement data, and parameter adjustment diagrams; encapsulate in a standardized JSON format and push it to the digital twin modeling module through the data hub.

[0163] Triggering Iteration Mechanism: Sending an optimization completion signal to the data center triggers the digital twin modeling module to update model parameters, starts a new round of simulation verification, and forms a closed-loop iteration of optimization, modeling, and simulation.

[0164] Digital twin closed-loop verification module: The digital twin closed-loop verification module verifies the optimization scheme throughout the entire process in a virtual environment, and generates structured correction instructions to drive design iteration. It includes: virtual verification, intelligent defect detection, and closed-loop correction.

[0165] It should be further explained that the specific steps of the virtual verification submodule are as follows:

[0166] Virtual environment initialization: Retrieve the latest lightweight digital twin model (STEP format) and physical property parameters (density, friction coefficient, etc.) from the data hub and import them into the Unity 3D engine. Construct a virtual scene at a 1:1 scale, and bind the model to the virtual space coordinates using a coordinate system alignment algorithm (error ≤ 0.01mm) to ensure consistency with the physical model's spatial position.

[0167] Physics engine parameter calibration: Based on the vibration and displacement characteristics in the sensor fusion data, adjust the physical parameters of the virtual environment: set the contact friction coefficient of the dynamic and fixed molds (0.15±0.02) and the elastic collision recovery coefficient of the components (0.2±0.05), and complete the parameter calibration by comparing the virtual and measured motion trajectories (deviation ≤0.05mm).

[0168] It should be further explained that the specific steps of the intelligent defect detection submodule are as follows:

[0169] Collision interference detection: The GJK continuous collision detection algorithm (detection frequency 60Hz) is activated to monitor the contact of components in the virtual assembly process in real time; when interference is detected (minimum distance < 0), the interference area is automatically marked (highlighted in red), the interference amount is calculated (accuracy 0.001mm), and the motion stage of the interference is recorded.

[0170] Tolerance compliance analysis: The GD&T standard database is used to verify the tolerances of key mating surfaces. The probability distribution of actual mating clearances is calculated through 1000 Monte Carlo simulations. If the probability of exceeding the tolerance is >5%, it is marked as a tolerance risk area (highlighted in yellow).

[0171] Process defect prediction: Import temperature and flow field data from multiphysics simulation and visualize the melt filling process using the color gradient method. When a shrinkage mark depth ≥0.1mm or a warpage ≥0.2mm is detected, analyze the cause of the defect based on the cooling rate distribution, and generate the defect location coordinates and severity level (1-5).

[0172] It should be further explained that the specific steps of the closed-loop correction submodule are as follows:

[0173] Problem priority sorting: The test results are sorted according to the rule of assembly interference (priority 1) > performance defects (priority 2) > process risks (priority 3); within the same priority, the impact range is weighted to generate a problem handling sequence.

[0174] Structured Correction Instructions: Correction instructions are automatically generated for sorted issues, formatted as issue type, location coordinates, parameter adjustment suggestions, and basis. Instructions are associated with corresponding digital twin model parameter labels.

[0175] Closed-loop iteration trigger: The correction command is pushed to the data center through the gRPC interface to trigger the parameter correction process of the digital twin modeling module; the verification log (including the number of issues, correction suggestions, and verification time) is recorded synchronously. If there are no high-priority issues (priority 1 or 2) in 3 consecutive verifications, the solution is deemed to be passed and the iteration is terminated; otherwise, the verification process is repeated.

[0176] Data Hub Module: As the core hub of the system, it realizes the management of all types of data and cross-module collaborative scheduling, ensuring the efficient operation of the design closed loop. It includes: full life cycle data management, cross-module collaborative scheduling, and precision control.

[0177] It should be further explained that the specific steps of the full lifecycle data management submodule are as follows:

[0178] Standardized access to multi-source data: Establish a unified data receiving interface that supports MQTT / gRPC protocols to receive data from various modules.

[0179] Sensing data: Encrypted fused data stream (including credibility index) is automatically parsed into a structured format of timestamps, coordinates, and numerical values;

[0180] Model data: STEP / IGES format twin model and parameter change log, indexed by mold model and version number;

[0181] Simulation and optimization data: simulation results in HDF5 format, optimization schemes in JSON format, and associated with the corresponding model version.

[0182] After access, the system automatically verifies the format compliance, and triggers retransmission of abnormal data from the source module.

[0183] Tiered storage and version tracking: Employing a three-tier architecture of in-memory, distributed database, and object storage.

[0184] Hot data (sensing data from the past hour and the current model version) is stored in memory, ensuring millisecond-level retrieval;

[0185] Warm data (historical simulation and optimization data) is stored in MongoDB and partitioned by time dimension;

[0186] Cold data (archived models and data from three years ago) is transferred to object storage to reduce costs.

[0187] Generate unique version identifiers for model and solution data, and support version backtracking and difference comparison.

[0188] It should be further explained that the specific steps of the cross-module collaborative scheduling submodule are as follows:

[0189] Event-driven rule configuration: Preset trigger rules are written to the scheduling engine.

[0190] Once the sensor data is ready, it triggers parameter correction in the modeling module.

[0191] Model iteration updates trigger mesh generation for simulation modules;

[0192] The simulation results are satisfactory, triggering the optimization module algorithm to start.

[0193] The verification question is generated, triggering the modeling module iteration.

[0194] Users can add or modify rules, which are then synchronized to all modules in real time.

[0195] Dynamic resource and process scheduling:

[0196] Resource allocation: Real-time monitoring of memory usage of each module; when the load of a module is ≥80%, automatic scheduling of idle node computing power to support it.

[0197] Process monitoring: Visualizes the status of the perception, modeling, simulation, optimization, and verification links through flowcharts, and marks blocked nodes;

[0198] Error handling: If a module does not respond within 5 minutes, a retry command will be automatically sent. If it fails 3 times, a manual alarm will be triggered.

[0199] It should be further explained that the specific steps of the precision control submodule are as follows:

[0200] End-to-end accuracy parameter acquisition: Key accuracy indicators are retrieved from each module at 10-second intervals.

[0201] Perception layer: Sensor data reliability (≥0.95);

[0202] Modeling layer: Deviation between twin model and actual measurement (≤0.02mm);

[0203] Simulation layer: Mesh distortion rate (≤15%), simulation-to-measurement deviation (≤8%);

[0204] Stored as a precision time-series database, generating precision trend curves.

[0205] Deviation warning and correction trigger: Set accuracy threshold alarm mechanism:

[0206] If a certain indicator exceeds the standard three times in a row, the system will automatically locate the source of the anomaly; send an accuracy calibration command to the corresponding module; and verify the accuracy recovery after calibration until the indicator returns to the acceptable range.

[0207] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0208] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A virtual simulation-based mold-die design system, characterized by comprising: Comprise: Multi-modal perception fusion module: combine the characteristics of the frame to adaptively plan the sensing points, collect multi-modal signals, after noise reduction, space-time synchronization processing, realize data fusion through attention mechanism dynamic weight distribution, generate standardized fusion data with credibility index and transmit to data hub module; The adaptive planning of the sensing point, comprising: Combined with the topological structure and mechanical and thermal characteristics of the frame, identify the high sensitivity area; and set the target constraints of 100% coverage of the high sensitivity area, minimization of the number of sensors, and signal interference less than 5%; then use a topology optimization algorithm to solve the optimal point; output the deployment list containing sensor type, installation coordinates and precision requirements; Digital twin modeling module: retrieve the fusion data and standard frame library from the data hub module to construct the initial model, use iterative algorithm to correct the model parameters, respond to external correction instructions to complete model iteration update, and return the updated model to the data hub module; Multi-physical field coupling simulation module: based on the model and fusion data of the data hub module, convert the fusion data into dynamic boundary conditions, implement hybrid mesh optimization division, integrate multiple solvers in accordance with the FMI standard to carry out coupling simulation, and output the simulation results to the data hub module; Hybrid intelligent optimization module: build a multi-objective optimization model based on the simulation results of the data hub module, use a two-level algorithm framework of global search and local optimization, combine with the proxy model to accelerate the optimization process, and output the optimal solution to the data hub module; Digital twin closed-loop verification module: obtain the model corresponding to the optimal solution from the data hub module, carry out immersive verification in a virtual environment, identify defects through multi-dimensional detection, and generate structured correction instructions to feed back to the data hub module; Data hub module: realize the storage and version tracing of all types of data, schedule cross-module collaboration based on preset event rules, monitor the accuracy of the whole link synchronously and trigger the correction mechanism.

2. A virtual simulation-based mold-die set design system according to claim 1, wherein: The attention mechanism dynamic weight distribution realizes data fusion, comprising: Obtain temperature, stress and strain, displacement and vibration single modal signals after wavelet threshold denoising, Kalman filtering and space-time synchronization processing; calculate the signal-to-noise ratio of each mode based on the signal amplitude and noise ratio; dynamically allocate weights according to the signal-to-noise ratio; weight the sum of multi-modal feature vectors, combine the calibration error and abnormal value proportion of each mode, and generate fusion data with a credibility greater than 0.

95.

3. The mold-tool set design system based on virtual simulation according to claim 1, wherein: The iterative update comprises: Construct a target function that minimizes the deviation between simulation values and measured values, and preset the parameters as the parameters to be corrected; update the parameters using the Gauss-Newton iterative algorithm; iterate until the deviation between the digital model and the physical frame is less than 0.02mm, and verify the physical quantity consistency and topological rationality at the same time; generate a corrected model with version identification and trigger the simulation module linkage.

4. The mold-tool set design system based on virtual simulation according to claim 1, wherein: The grid optimization division comprises: Extract the stress concentration area from the fusion data, and the stress value in the stress concentration area is greater than 70% of the allowable stress of the material; use 0.67mm tetrahedral elements to encrypt the high stress area, and use 5mm hexahedral elements to simplify the non-critical area; control the grid distortion rate to be less than 15% and the aspect ratio to be less than 5; verify the grid independence through 3 groups of different encryption level grids pre-simulation.

5. The mold-tool set design system based on virtual simulation according to claim 1, wherein: The integrated multi-solver coupling simulation comprises: The structure solver and the fluid solver are packaged as functional Mock-up units according to the FMI2.0 standard; the solving resources are dynamically allocated in three stages of filling, pressure maintaining and cooling, and the filling step is 0.05s; the temperature and pressure coupling physical quantities are transmitted every 0.05s; the solving residual is monitored, and when the residual is less than or equal to 1e-4, it is determined that the convergence is achieved, and when divergence occurs, the grid or step is adjusted to restart the local simulation.

6. The virtual simulation-based mold-die set design system according to claim 1, wherein: The optimization secondary algorithm comprises: The two-level algorithm parameters are initialized: the global search adopts the NSGA-II algorithm, the population size is configured as 100, the iteration number is 50, the crossover probability is 0.8, and the mutation probability is 0.01; the local optimization adopts the Bayesian optimization algorithm, the Matérn kernel function is selected, the exploration coefficient is 2.5, and the maximum iteration is 30 times; the global search is performed: the NSGA-II algorithm is used to traverse and optimize the design variable space, a Pareto frontier solution set containing 100 solutions is generated, and the top 20% of high-potential solutions are selected as the candidate set according to the target function comprehensive score; the local optimization is carried out: the candidate set is taken as the initial point, the Gaussian process model is constructed by the Bayesian optimization to predict the target function value, and the high-potential area is finely optimized; the convergence degree of the solution is calculated every 10 iterations, the deviation between the optimal solutions of adjacent generations is less than 1% to determine the convergence, and the mutation probability or exploration range is dynamically adjusted according to the convergence speed, and finally 3-5 optimal solutions are output.

7. The virtual simulation-based mold-die set design system according to claim 1, wherein: The proxy model comprises: The training data preparation: 1000 groups of historical simulation data and 50 groups of latest simulation results are extracted from the data hub, the training set is constructed according to the format of the design variables and the performance indicators, and the data is processed by using the Min-Max standardization; The model construction and training: a multi-scale neural network proxy model is built, the input layer matches the design variable dimension, the hidden layer is set as 2 layers × 32 neurons, and the output layer corresponds to the multi-objective optimization indicators; the network weight is initialized by using the transfer learning strategy, and the model is fine-tuned by using the training set to make the prediction error less than 8%; The optimization acceleration execution: the proxy model is called in the optimization process to predict the performance indicators corresponding to the design variables in real time, to replace part of the direct simulation calculation, and to shorten the optimization iteration period; The model incremental update: after each optimization round, the newly generated optimal solution is supplemented to the training set to trigger the model incremental update, and the update time is controlled within 5 minutes to adapt to the dynamic changes of the working conditions and data.

8. The virtual simulation-based mold-die set design system according to claim 1, wherein: The defect identification comprises: Assembly interference detection: a continuous collision detection algorithm is started to monitor the component contact state in the virtual assembly process at a frequency of 60Hz, to mark the interference area and calculate the interference amount with an accuracy of 0.001mm, and to record the motion stage at which the interference occurs; Tolerance compliance analysis: a GD&T standard database is called to calculate the probability distribution of the actual fitting gap by using Monte Carlo simulation for the key fitting surface, and the tolerance risk area is marked when the out-of-tolerance probability is greater than 5%; Forming process defect identification: the temperature field and flow field data of the multi-physical field simulation are imported, the melt filling process is presented by using the visualization technology, the defects with a shrinkage depth greater than or equal to 0.1mm or a warpage amount greater than or equal to 0.2mm are detected, and the causes are traced in combination with the heat flow parameters; Defect information integration: output defect position coordinates, severity level and cause analysis according to detection type, form structured defect data set.

9. The virtual simulation-based mold-die set design system according to claim 1, characterized by: The data hub module comprises: A unified interface supporting MQTT / gRPC protocol is built to receive multi-source data, which is parsed into a structured format and verified, stored in a three-level architecture of memory, distributed database and object storage, and version traceability is realized; preset event-driven rules are used to schedule cross-module collaboration, real-time monitoring of load and scheduling of computing power, and retry alarm for abnormal modules; precision indicators are collected at fixed intervals, and when the indicators continuously exceed the standard, the source is located and calibration instructions are sent, and the corrected precision is verified to ensure closed-loop precision.

Images