Mold and mold frame design system based on virtual simulation

Abstract

The invention discloses a mold and formwork design system based on virtual simulation, and particularly relates to the field of mold design, the mold and formwork design system comprises a multi-mode perception fusion module, a digital twin modeling module, a multi-physics field coupling simulation module, a hybrid intelligent optimization module, a digital twin closed loop verification module and a data center module, and each module forms a design closed loop through a data center. The multi-modal sensing fusion module adaptively collects and fuses multi-source signals to generate high-credibility data; the digital twin modeling module constructs and iteratively corrects a model based on the data; the multi-physics field coupling simulation module realizes multi-solver co-simulation under a dynamic boundary; the hybrid intelligent optimization module accelerates to generate an optimal solution through secondary optimization and an agent model; the digital twin closed-loop verification module detects defects and generates a correction instruction; the data center module is responsible for data management, cross-module scheduling and precision control; the system improves the design precision and efficiency of the mold frame, reduces the physical mold testing cost, and is suitable for a high-precision mold development scene.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mold design, more particularly, the present application relates to a mold frame design system based on virtual simulation. BACKGROUND

[0002] As the core bearing component of injection molding, mold frame is directly related to product precision and production efficiency, and is in urgent demand in the fields of automobiles, 3C electronics, etc. The current industry digital transformation has achieved initial success, and the existing technology has significant advantages: CAD / CAE tools are widely used, and three-dimensional modeling and basic performance checking of the mold frame can be quickly completed; multi-modal sensing technology is gradually popularized, and key physical data such as temperature and stress can be collected; basic multi-physical field simulation can realize simple thermal and force coupling analysis; and the data management system can complete the storage and archiving of design files and simulation results, which greatly shortens the preliminary design cycle compared with traditional hand-drawing design and physical trial molding. However, when it is actually used, there are still some shortcomings, such as: 1. The sensing link relies on experience to plan sensing points, which is prone to blind spots in high-sensitivity areas or redundant sensors. Multi-modal data is only simply spliced, and there is a lack of dynamic weight distribution and credibility evaluation, which leads to insufficient accuracy of data sources and affects subsequent modeling basis; 2. Digital twin modeling is mostly offline static correction, and parameter updating lags behind the working condition changes, so the real-time mapping deviation between the model and the physical mold frame is large, and it is difficult to accurately reflect the actual running state of the mold frame, which restricts the reliability of simulation and optimization; 3. Optimization mostly uses single algorithm optimization, and relies on a large number of entity simulation iterations, and has not formed an efficient two-level optimization framework. The training and updating mechanism of the surrogate model is not perfect, which leads to a long optimization period and makes it difficult to balance accuracy and efficiency; 4. The data of each module is stored separately, and cross-module collaboration relies on manual connection. There is a lack of unified central scheduling and full-link precision control, and there are process breakpoints, so it is not possible to form a closed-loop iteration of design, simulation and verification. SUMMARY

[0003] In order to overcome the above-mentioned defects of the prior art, the present application provides a mold frame design system based on virtual simulation, which solves the problems in the background art by the following scheme.

[0004] In order to achieve the above-mentioned purpose, the present application provides the following technical scheme: a mold frame design system based on virtual simulation, comprising: A multi-modal sensing fusion module: adaptive planning of sensing points is combined with the characteristics of the mold frame, multi-modal signals are collected, after noise reduction, time and space synchronization processing, data fusion is realized through attention mechanism dynamic weight distribution, standardized fusion data with credibility index are generated and transmitted to the data hub; Digital twin modeling module: retrieve fusion data from data hub and construct initial model based on standard template library, adopt iterative algorithm to correct model parameters reversely, respond to external correction instructions to complete model iteration update, and return updated model to data hub; Multi-physical field coupling simulation module: based on the model and fusion data of the data hub, 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 simulation results to the data hub; Hybrid intelligent optimization module: build a multi-objective optimization model based on the simulation results of the data hub, adopt 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; Digital twin closed-loop verification module: obtain the model corresponding to the optimal solution from the data hub, 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; Data hub module: realize storage and version tracing of all types of data, dispatch cross-module collaboration based on preset event rules, and monitor the accuracy of the whole link and trigger the correction mechanism.

[0005] Technical effects and advantages of the application: 1. Perception precision: efficient fusion of multi-modal data Adopt a sensor point adaptive planning algorithm that adapts to the characteristics of the template, and realize accurate balance between monitoring coverage and resources by combining quantitative constraints; integrate multi-source preprocessed data through a dynamic weight distribution mechanism, and simultaneously build a credibility evaluation system to generate high-precision standardized fusion data, providing reliable data foundation support for subsequent links; 2. Dynamic twin modeling: real-time precision of virtual-real mapping Driven by real-time sensing data, establish a dynamic iterative correction mechanism for model parameters, breaking the limitations of offline static adjustment. Through the deviation minimization objective function optimization algorithm, continuously calibrate the mapping accuracy of the digital model and the physical template, realize real-time synchronization of their states, and improve the reliability of simulation and optimization results; 3. Intelligent optimization efficiency: precision and efficiency improvement Build a two-level algorithm architecture of global search and local optimization, taking into account the globality and fineness of the design solution. Introduce an adaptive proxy model to replace a large number of entity simulation calculations, and support efficient training and incremental update mechanism, which not only guarantees the accuracy of multi-objective optimization, but also significantly shortens the iteration period, realizing efficiency upgrade; 4. Full-chain collaborative closed loop: integrated process control The whole process data is integrated by the unified data hub, standard storage, version tracking and efficient retrieval are realized, cross-module automatic collaborative scheduling is realized based on an event-driven mechanism, combined with a whole link precision real-time monitoring and deviation correction system, manual connection breakpoints are eliminated, and a complete closed-loop process of design, simulation and verification is constructed. BRIEF DESCRIPTION OF DRAWINGS

[0006] Fig. 1 It is a schematic diagram of the overall structure of the present application.

[0007] Fig. 2 It is a schematic diagram of the mixed intelligent optimization module of the present application. DETAILED DESCRIPTION

[0008] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0009] Reference Figs. 1-2 The virtual simulation-based mold frame design system shown comprises: A multi-modal perception fusion module: through a three-stage process of perception planning, signal processing and data fusion, accurate acquisition and digital integration of physical frame state data are realized, and a high-credibility data source is provided for subsequent digital twin modeling, which specifically includes: adaptive sensor deployment, multi-source signal preprocessing, intelligent data fusion; It should be further explained that the adaptive sensor deployment sub-module includes: mold frame characteristic pre-analysis and parameter input, adaptive sensor planning, sensor adaptation and standardized deployment, post-deployment calibration and adaptation verification; the specific analysis is as follows: Mold frame characteristic pre-analysis and parameter input: the topological structure data of the target mold frame is imported through a three-dimensional modeling tool, including geometric parameters of the mold plate size, cavity distribution, runner orientation, cooling system layout, and the mold frame material properties and injection molding process parameter range are input synchronously, the mold frame material properties include: elastic modulus, thermal conductivity, yield strength, the injection molding process parameter range is: injection pressure 0-200MPa, mold temperature 20-300℃, cycle time 20-300s; Mechanics and thermal simulation: based on the mold frame geometry and material parameters, a simplified finite element algorithm is used to identify potential stress concentration areas, and a thermal conduction simulation is used to locate temperature gradient mutation areas, and a high-sensitivity area coordinate set is output.

[0010] Adaptive sensor planning: Construction of point planning objective function: with the constraints of monitoring coverage greater than 95%, the minimum number of sensors, and signal interference less than 5%, the optimal point is solved by using topology optimization algorithm; Output point planning results: generate deployment list containing sensor type, installation coordinates, monitoring parameters, and accuracy requirements, temperature sensor covers cavity surface and cooling channel key nodes, stress and strain sensor focuses on stress concentration area identified by pre-simulation, displacement sensor is deployed on the parting surface of the dynamic and static mold and the moving part, and vibration sensor is arranged at the connection part of the mold frame and the injection molding machine.

[0011] Sensor adaptation and standardized deployment: Sensor selection and adaptation: according to the deployment list, select the corresponding type of sensor, temperature sensor uses K-type thermocouple, stress and strain sensor uses resistance strain gauge, displacement sensor uses laser displacement sensor, vibration sensor uses ICP type acceleration sensor, and all sensors are configured with standardized digital interface to realize plug and play with edge nodes; The physical deployment operation is as follows: Temperature sensor is fixed on the planned point by high-temperature resistant heat-conducting glue, ensuring that the sensor end is in contact with the mold frame surface with a contact degree greater than 90%; stress and strain sensor is pasted with epoxy adhesive, and the mold frame surface is polished and degreased before pasting, and the roughness of the pasted sensor is less than 1.6 after 24 hours of curing; Displacement sensor is fixed by magnetic base, and the light path is adjusted to make the laser perpendicular to the monitoring surface with a shielding rate less than 1%; Vibration sensor is connected by screw thread, and the installation surface flatness is less than 0.02mm / m to avoid resonance with the mold frame; Calibration and adaptation verification after deployment: Static calibration: point-to-point calibration of each sensor is carried out by using standard calibration equipment, and the calibration coefficient is recorded and written into the edge node storage unit to ensure that the sensor measurement error is less than ±0.5%FS; Dynamic adaptation verification: start the mold frame empty running, collect the initial signal of the sensor, and judge the deployment effectiveness by signal integrity detection algorithm, readjust the deployment position or replace the sensor for unqualified point until all points meet the monitoring requirements; It needs to be further explained that the multi-source signal preprocessing submodule includes: signal acquisition and preliminary conditioning, multi-modal signal noise reduction processing, data space-time synchronization and outlier removal, and preprocessing data standardization output; the specific analysis is as follows: Signal acquisition and preliminary conditioning: Edge node starts signal collection: receives analog signals output by various sensors through standardized interfaces, including temperature, stress and strain, displacement, and vibration signals. The collection frequency is dynamically configured according to the sensor type: temperature 10Hz, stress and strain 100Hz, displacement 1kHz, and vibration 10kHz. A unique identifier is added to each signal for synchronization. Adaptive gain adjustment: the edge node has a signal conditioning unit that uses a peak detection algorithm to identify signal amplitude in real time. When the amplitude is below 20% of the full range, the gain is automatically increased by a maximum of 1000 times. When the amplitude is above 80% of the full range, the gain is reduced by a minimum of 100 times. This ensures that the conditioned signal amplitude is within 20%-80% of the ADC range, improving sampling accuracy. Multi-modal signal noise reduction processing: Wavelet threshold denoising: the conditioned signal is decomposed using wavelets, and the threshold of each layer of wavelet coefficients is calculated. The coefficients below the threshold are set to zero, and the denoised signal is reconstructed to reduce high-frequency noise interference. Kalman filter optimization: for low-frequency drift in the denoised signal, a second-order autoregressive state equation is constructed, with the wavelet denoising result as the observation value. Through the prediction and update iteration of Kalman filter, the signal baseline drift is eliminated, and the final output is a stable single-mode signal. Data space-time synchronization and outlier removal: Time series alignment: extract the timestamp of each signal, with the highest sampling frequency of the vibration signal 10kHz as the reference. Linear interpolation algorithm is used to complete the data of low-frequency signals, making the time step of all modal signals uniform as 0.1ms, achieving time series synchronization. Low-frequency signals include temperature, stress and strain, and displacement. Spatial alignment: based on the sensor installation coordinates output in step one, a mapping relationship between sensor position and physical coordinates of the mold is established, and each signal is bound to the three-dimensional coordinates of the mold, achieving spatial matching of multi-modal data. Outlier removal: using the sliding window method, the window size is 100 data points, and the mean and standard deviation of each signal are calculated in real time When a data point meets , it is determined as an outlier, and the data is completed through linear interpolation of the previous and subsequent data points to ensure data continuity. Preprocessing data standardization output: Convert single-mode data after space-time synchronization to a standardized digital format, organize according to signal identifier, physical coordinates, timestamp, and value structure, and temporarily store in the edge node local cache, waiting for fusion processing.

[0012] It should be further pointed out that the intelligent data fusion sub-module includes: fusion model initialization and parameter configuration, real-time multi-modal data fusion, fusion data reliability evaluation and encrypted transmission, fusion model online updating, which are analyzed as follows: Fusion model initialization and parameter configuration: Constructing attention mechanism fusion model: build a fusion network based on TensorFlow framework, input layer receives 4-way single modal standardized data, including: temperature, stress, displacement, vibration, hidden layer calculates the weight coefficient of each modal data through attention weight matrix, dimension 4xN, N is the number of fusion features, output layer generates fusion feature vector.

[0013] Model pre-training: at least 1000 groups of historical similar perception data of the model are used to pre-train the fusion model, and the multi-modal data and the physical quantity true value are used as training samples. The attention weight is optimized through back propagation, so that the fusion error of the pre-trained model is less than 5%, laying a foundation for real-time fusion.

[0014] Real-time multi-modal data fusion: Weight dynamic update: the standardized single modal data output by multi-modal signal noise reduction processing is input into the pre-trained model, and the real-time weight coefficient is calculated through attention mechanism. When the signal-to-noise ratio of a certain modal signal exceeds 45dB, the weight is increased, the highest 0.4, and when it is lower than 35dB, the weight is reduced, the lowest 0.1, to ensure that high reliability data occupies the dominant position in fusion.

[0015] Feature fusion calculation: the feature vectors of 4-way modal data are weighted and summed according to the real-time weight coefficient, generating a fusion feature vector with dimension N, which contains the comprehensive state information of the model at a certain physical coordinate.

[0016] Fusion data reliability evaluation and encrypted transmission: Reliability calculation: based on the signal-to-noise ratio, calibration error, data continuity and other indicators of each modal signal, a reliability evaluation function is constructed to output the reliability index of the fusion data, with a value of [0, 1], greater than 0.95 is qualified; the function logic is as follows: , .

[0017] Encryption and transmission: the fusion feature vector and the reliability index are encrypted by AES-128, and pushed to the data hub module through MQTT protocol. The heartbeat packet mechanism with an interval of 1s is used to monitor the connection state during transmission. If the connection is interrupted, the data is temporarily stored and automatically transmitted after recovery to ensure that the data is not lost.

[0018] Fusion model online updating: The data hub periodically feeds back the deviation of the fusion data and the subsequent simulation results, and when the average deviation of 100 consecutive groups of data is greater than 8%, the fusion model is triggered for online updating: using the latest single-mode data and simulation verification values as samples, the attention weight matrix is fine-tuned using the transfer learning strategy, without retraining, the fusion accuracy can be improved to adapt to the changes of the mold working condition.

[0019] Digital twin modeling module: driven by multi-modal perception fusion data, through the closed-loop process of basic modeling, dynamic correction, and iterative updating, a parameterized digital model that maps in real time with the physical mold is constructed, providing an accurate digital carrier for subsequent multi-physical field simulation and optimization, which specifically includes: basic model generation, adaptive parameter correction, model iterative update; It should be further explained that the basic model generation submodule includes: modeling input parameter retrieval and verification, standard mold library matching and initial model generation, model and perception point mapping association, which are analyzed as follows: Modeling input parameter retrieval and verification: Data hub interaction: send parameter retrieval requests to the data hub through the gRPC interface to obtain three types of core input data: Mold base parameters: including topological structure (template layer number, cavity number and distribution), geometric size (template length, width and thickness, guide column spacing), material properties (elastic modulus, Poisson's ratio, thermal conductivity); Product and process associated parameters: product maximum profile size, injection molding machine parameters (maximum mold capacity, clamping force), preliminary process window (mold temperature, injection pressure range); Perception point mapping data: sensor installation coordinates and mold physical coordinate mapping table output by the multi-modal perception fusion module; Parameter validity verification: build a parameter verification rule library to automatically detect the integrity and reasonableness of the input data, if there is an exception, trigger the data hub to supplement or manual correction, and after verification, enter the modeling process.

[0020] Standard mold library matching and initial model generation: Library model adaptation screening: call the standardized mold parameter library, based on matching rules, use the K-nearest neighbor algorithm to select 3-5 candidate basic models from the library, the matching rules are preset based on existing equipment conditions; Initial model parameterization construction: select the candidate model with the highest matching degree, and perform initial modeling through the parameterization modeling engine: Define core design variables: set the cavity wall thickness, support column diameter and spacing, reinforcement rib height and number, etc. as correctable parameters, and bind the parameter labels; ​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).

[0021] Model-sensor point mapping association: 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.

[0022] 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. 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: Real-time fusion data reception and coordinate matching: 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.

[0023] 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.

[0024] Model parameter adaptive correction calculation: 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: ; 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.

[0025] Least squares iterative correction: The Gauss-Newton iterative algorithm is used to solve for the optimal solution of the objective function. Initialize iteration parameters (initial model parameter value), calculate initial deviation ; construct Jacobian matrix , by update parameters, is a regularization coefficient to avoid matrix singularity, is an error vector, and k is an ordinal number; repeat iteration until deviation , output the corrected parameter vector .

[0026] Consistency check of the corrected model: Physical quantity consistency verification: Substitute the corrected model parameters into the simplified mechanics or thermal simulation (use fast finite element algorithm, calculation time ≤ 10s), output the stress distribution, temperature field and other simulation results of the model, and compare them with the corresponding physical quantities in the perception fusion data at the same time: If the deviation between the simulation value and the measured value in the key area is less than or equal to 5%, it is determined to be qualified; If the deviation exceeds 5%, backtrack to check the parameter mapping relationship and correction algorithm parameters, and execute the correction process again.

[0027] Topology structure rationality check: Call the mold structure design rule library to automatically detect whether the topology structure of the corrected model meets the process requirements, to avoid structure interference or process feasibility decline caused by parameter correction.

[0028] It should be further pointed out that the model iteration update submodule includes: external correction instruction receiving and analysis, model iteration update and version management, lightweight processing and downstream module pushing, model consistency secondary check, which are analyzed as follows: External correction instruction receiving and analysis: Multi-source instruction access: Receive two types of correction instructions through gRPC interface: Closed-loop verification module feedback: Structured instructions of problem type, location coordinates, and correction parameter requirements; Optimization module output: Parameter adjustment value in the optimal design scheme.

[0029] Instruction priority sorting: Sort according to the rule of assembly interference (priority 1) > performance defect (priority 2) > optimization parameter adjustment (priority 3), when there is instruction conflict, execute high-priority instruction first, and record conflict information to data hub at the same time.

[0030] Model iteration update and version management: Parameter batch adjustment: Convert the sorted correction instructions into model parameter adjustment values, automatically update the corresponding parameters through the batch modification interface of the parameterized modeling engine, and generate the updated three-dimensional model.

[0031] Version identification and storage: Generate a unique version number according to the model type, iteration number, and timestamp, and synchronize the updated model file and parameter change log to the version management submodule of the data hub to realize full historical version backtracking.

[0032] Lightweight processing and downstream module pushing: Lightweight optimization of iterative model: Use improved LOD technology to perform secondary lightweight processing on the updated model: Key structures are preserved at the highest precision; Non-key structures are simplified gradually to ensure that the model file size is ≤100MB, meeting the efficient loading requirements of the simulation module.

[0033] Model pushing and trigger notification: Push the lightweight model to the multi-physical field coupling simulation module through the data hub, and send a model update completion trigger signal containing the version number and correction area coordinates to trigger the simulation module to start a new round of coupling analysis, realizing the automatic linkage of modeling and simulation.

[0034] Secondary verification of model consistency: After the simulation module outputs the first round of simulation results, compare the model simulation value and the perception and measurement value from the data hub. If the deviation is less than or equal to 0.02mm, the iteration is effective; if the deviation is greater than 0.02mm, automatically trigger the adaptive parameter correction submodule to re-execute the correction process until the mapping precision of the model and the physical frame meets the requirements.

[0035] Multi-physical field coupling simulation module: The multi-physical field coupling simulation module uses the digital twin model as the carrier, combines perception fusion data to carry out thermal, force, and flow coupling analysis, and outputs high-precision simulation results to support optimization decisions; it includes: dynamic boundary condition configuration, adaptive grid optimization, collaborative simulation scheduling, simulation result processing and output; It should be further noted that the dynamic boundary condition configuration submodule includes: data retrieval and verification, dynamic modeling of boundary conditions, and real-time update configuration, which are analyzed as follows: Data retrieval and verification: Retrieve three types of data from the data hub through the gRPC interface: lightweight digital twin model, multi-modal perception fusion data (including temperature and pressure time series data), and injection molding process parameters (filling and holding time), construct data verification rules, automatically detect model topology integrity, perception data reliability, and process parameter rationality, and trigger supplementary transmission for abnormalities.

[0036] Dynamic modeling of boundary conditions: With perception data as input, build boundary conditions that change with the injection cycle: Map cooling channel inlet and outlet temperature data to thermal boundary conditions, convert injection pressure data to fluid load, and use mold clamping force data as structural load. Use linear interpolation algorithm to achieve time sequence matching between perception data and simulation time step (0.1s).

[0037] Real-time update configuration: Design a linkage mechanism for boundary conditions, subscribe to perception data updates from the data hub through MQTT protocol, and automatically trigger boundary condition reconfiguration when the rate of change of a physical quantity is greater than 10%, ensuring that the simulation input is synchronized with the physical working conditions.

[0038] Further explanation is needed. The adaptive mesh optimization submodule includes: stress concentration area identification, mixed mesh strategy implementation, and mesh independence verification, which are analyzed as follows: Stress concentration area identification: Extract stress time series data from perception fusion data, set stress value ≥ 70% of material allowable stress as threshold, identify high stress area, and output region coordinate set.

[0039] Mixed mesh strategy implementation: Call the mesh division engine, use 0.67mm tetrahedral elements to encrypt high stress areas, and use 5mm hexahedral elements to simplify non-critical areas, forming a mixed mesh model. Control the distortion rate to be less than or equal to 15% and the aspect ratio to be less than or equal to 5 through mesh quality evaluation algorithm.

[0040] Mesh independence verification: Select 3 groups of different encryption level meshes to carry out pre-simulation, compare the stress value deviation of the key area, and determine the optimal mesh parameters when the deviation is less than or equal to 5%; Further explanation is needed. The collaborative simulation scheduling submodule includes: Solver collaboration configuration: Follow the FMI2.0 standard to build a collaborative framework, integrate Calculix structural solver and OpenFOAM fluid solver, and realize data interaction interface encapsulation through functional Mock-up unit.

[0041] Simulation stage division and execution: Divide into filling, pressure maintaining, and cooling stages according to the injection cycle, and dynamically allocate solving resources: prioritize fluid solver in filling stage, step size 0.05s, and strengthen structure and thermal coupling calculation in cooling stage. Set solver interaction interval to 0.05s, and transfer temperature, pressure and other coupled physical quantities.

[0042] Simulation process monitoring: Real-time monitoring of solving convergence, residual error less than or equal to 1e-4 determines convergence, if divergence occurs, automatically adjust mesh precision or time step, restart local simulation; After simulation is completed, generate full cycle calculation log.

[0043] It should be further pointed out that the simulation result processing and output sub-module includes: result data analysis, result reliability evaluation, data pushing and triggering, the specific analysis is as follows: Result data analysis: extract the HDF5 format file of simulation output, parse to get stress cloud map (node stress value), temperature field distribution (element temperature), flow front trajectory and other core data, and generate structured data set by associating with three-dimensional coordinate of framework.

[0044] Result reliability evaluation: compare simulation value with contemporaneous sensing data, calculate deviation rate (below 8% is qualified), generate reliability report; if not qualified, backtrack optimization boundary condition or grid parameter, and re-simulate.

[0045] Data pushing and triggering: encrypt qualified simulation data set and reliability report into data hub, and send simulation completion trigger signal, automatically start optimization process of hybrid intelligent optimization module, realize simulation and optimization linkage; Hybrid intelligent optimization module: based on multi-physical field simulation results, through multi-objective algorithm and proxy model acceleration, generate optimal design scheme of framework, including: optimization target and constraint modeling, hybrid optimization algorithm scheduling, proxy model acceleration, optimization scheme generation; It should be further pointed out that the specific steps of the optimization target and constraint modeling sub-module are as follows: Objective function construction: call simulation data set from data hub, construct multi-objective optimization function: Main target: maximize weight reduction rate , maximize stiffness improvement rate , maximize life extension rate ; Target weight: support users to set weight coefficient through interactive interface , default uniform weight.

[0046] Constraint condition definition: based on material performance and process requirements, set constraints: Hard constraint: maximum stress ≤ 80% of material yield strength, maximum deformation ≤ 0.03mm, cooling time ≤ 60% of injection molding cycle; Soft constraint: design variable range, stored as XML format constraint file.

[0047] Optimization model packaging: package target function and constraint condition into standardized optimization model, push to algorithm scheduling sub-module through data hub interface, and associate with corresponding digital twin model version.

[0048] It should be further pointed out that the specific steps of the hybrid optimization algorithm scheduling sub-module are as follows: Initial parameter configuration: read optimization model, determine design variable dimension (n) and constraint number (m), initialize algorithm parameters: NSGA-II algorithm: population size 100, iteration number 50, crossover probability 0.8, mutation probability 0.01; Bayesian optimization: Matérn kernel function, exploration coefficient 2.5, maximum iteration 30 times.

[0049] Secondary optimization execution: Global search: start the NSGA-II algorithm, perform global optimization on the design variable space, generate a Pareto frontier solution set containing 100 solutions, and select the top 20% solutions with the lowest comprehensive score of the target function as the candidate set; Local optimization: take the candidate set as the initial point, run Bayesian optimization, predict the target function value through the Gaussian process model, and output 3-5 Pareto optimal solutions.

[0050] Algorithm adaptive adjustment: calculate the convergence degree of the solution every 10 iterations (optimal solution deviation ≤1% between adjacent generations to determine convergence), and automatically increase the mutation probability (up to 0.05) or expand the exploration range if convergence is slow.

[0051] It should be further explained that the specific steps of the surrogate model acceleration submodule are: 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 variable and performance indicator format, and process the data using Min-Max standardization.

[0052] Multi-scale neural network training: Model structure: input layer (n neurons), hidden layer (2 layers x 32 neurons), output layer (3 neurons corresponding to 3 target functions); Training strategy: use the Adam optimizer (learning rate 0.001), combine transfer learning to initialize network weights, fine-tune the model with new data, and make the prediction error ≤8%.

[0053] Real-time prediction and update: during optimization, 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 update (training time ≤5 minutes).

[0054] It should be further explained that the specific steps of the optimization scheme generation submodule are: Scheme evaluation and selection: perform manufacturability evaluation on the optimal solution, call the process knowledge base, calculate the manufacturing cost index and process feasibility score of each scheme, and eliminate schemes with a feasibility score less than 60.

[0055] Result encapsulation output: generate optimization scheme report, including design variable parameters, performance improvement data, parameter adjustment diagram; encapsulate in standardized JSON format, push to digital twin modeling module through data hub.

[0056] Trigger iteration mechanism: send optimization completion signal to data hub, trigger digital twin modeling module to update model parameters, start a new round of simulation verification, form an optimization, modeling, and simulation closed-loop iteration.

[0057] Digital twin closed-loop verification module: the digital twin closed-loop verification module verifies the optimization scheme through the virtual environment, generates structured correction instructions to drive design iteration, which includes: virtual verification, intelligent defect detection, closed-loop correction; It should be further pointed out that the specific steps of the virtual verification submodule are as follows: 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. Build a virtual scene at a scale of 1:1, and bind the model to the virtual space coordinates through a coordinate system alignment algorithm (error ≤0.01mm) to ensure consistency with the physical mold space position.

[0058] Physical engine parameter calibration: based on the vibration and displacement characteristics in the perception fusion data, adjust the virtual environment physical parameters: set the dynamic mold contact friction coefficient (0.15±0.02) and the component elastic collision restitution coefficient (0.2±0.05), and complete parameter calibration by comparing the virtual and measured motion trajectories (deviation ≤0.05mm).

[0059] It should be further pointed out that the specific steps of the intelligent defect detection submodule are as follows: Collision interference detection: start the GJK continuous collision detection algorithm (detection frequency 60Hz) to monitor the component contact in real time during the virtual assembly process; when interference is detected (minimum distance <0), automatically mark the interference area (red highlight), calculate the interference amount (precision 0.001mm) and record the motion stage at which the interference occurred.

[0060] Tolerance compliance analysis: call the GD&T standard database to verify the tolerance of the key mating surfaces. Calculate the probability distribution of the actual fitting gap by sampling 1000 times using the Monte Carlo simulation, and if the out-of-tolerance probability is >5%, mark it as a tolerance risk area (yellow highlight).

[0061] Process defect prediction: import temperature field and flow field data from multi-physics simulation, and visualize the melt filling process using the color gradient method. When detecting a sink mark depth ≥0.1mm or warpage ≥0.2mm, analyze the defect causes based on the cooling rate distribution, and generate defect position coordinates and severity level (1-5 levels).

[0062] It needs to be further explained that the specific steps of the closed-loop correction sub-module are: Problem priority ranking: rank the test results according to the rules of assembly interference (priority 1) > performance defects (priority 2) > process risk (priority 3); within the same priority, weight according to the impact range to generate a problem processing sequence.

[0063] Correction instruction structuring: automatically generate correction instructions for the sorted problems, with a format of problem type, location coordinates, parameter adjustment suggestion, and basis. The instructions are associated with the corresponding digital twin model parameter tags.

[0064] Closed-loop iteration triggering: push the correction instructions to the data hub through the gRPC interface to trigger the parameter correction process of the digital twin modeling module; simultaneously record the verification log (including the number of problems, correction suggestions, and verification time), if there are no high-priority problems (priority 1 or 2) for three consecutive verifications, the scheme is considered to pass and the iteration is terminated; otherwise, repeat the verification process.

[0065] 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, which includes: whole life cycle data management, cross-module collaborative scheduling, precision control; It needs to be further explained that the specific steps of the whole life cycle data management sub-module are: Multi-source data standardized access: build a unified data receiving interface, support MQTT / gRPC protocol, and receive data from each module: Sensing data: encrypted fusion data stream (including credibility index), automatically parsed into structured format of timestamp, coordinates, and numerical value; Model data: STEP / IGES format twin model and parameter change log, indexed by model frame type and version number; Simulation and optimization data: HDF5 format simulation results, JSON format optimization scheme, associated with the corresponding model version.

[0066] After access, automatically check the format compliance, and trigger the source module to retransmit abnormal data.

[0067] Hierarchical storage and version traceability: use a three-level architecture of memory, distributed database, and object storage: Hot data (near 1 hour sensing data, current model version) is stored in memory to ensure millisecond-level retrieval; Warm data (historical simulation and optimization data) is stored in MongoDB, partitioned by time dimension; Cold data (archived models and data three years ago) is transferred to object storage to reduce costs.

[0068] Generate a unique version identifier for model and scheme data, support version backtracking and difference comparison.

[0069] It should be further explained that the specific steps of the cross-module collaborative scheduling sub-module are: Event-driven rule configuration: preset trigger rules are written into the scheduling engine: Sensing data is ready, triggering parameter correction of the modeling module; Model iteration update, triggering grid division of the simulation module; Simulation results are qualified, triggering algorithm startup of the optimization module; Verification problem generation, triggering iteration of the modeling module.

[0070] Support users to add or modify rules and synchronize them to each module in real time.

[0071] Dynamic resource and process scheduling: Resource allocation: real-time monitoring of memory usage in each module, when the load of a module is greater than or equal to 80%, automatically scheduling idle node computing power support; Process monitoring: visualize the link status of sensing, modeling, simulation, optimization, and verification through a flowchart, and mark the blocked nodes; Abnormal handling: if a module does not respond for more than 5 minutes, automatically send a retry instruction, and if it fails 3 times, trigger a manual alarm.

[0072] It should be further explained that the specific steps of the precision control sub-module are: Full-link precision parameter collection: key precision indicators are retrieved from each module at 10-second intervals: Sensing layer: sensor data reliability (≥0.95); Modeling layer: twin model and measured deviation (≤0.02mm); Simulation layer: grid distortion rate (≤15%), simulation and measured deviation (≤8%); Stored as a precision time series database, generating a precision trend curve.

[0073] Deviation warning and correction trigger: set up an accuracy threshold alarm mechanism: When a certain indicator exceeds the standard for 3 consecutive times, automatically locate the abnormal source; send precision calibration instructions to the corresponding module; verify the accuracy recovery after calibration, until the indicator returns to the qualified range; Secondly, in the drawings of the disclosed embodiments, only the structures related to the disclosed embodiments are involved, other structures can be referred to the usual design, and in the case of no conflict, the same embodiment and different embodiments of the present application can be combined with each other; Finally: the above only for the preferred embodiments of the present application, and not for limiting the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application, should be included in the scope of protection of the present application.

Claims

1. A mold base design system based on virtual simulation, characterized in that, include: 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; 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; 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; 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; 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; 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.

2. The mold base design system based on virtual simulation according to claim 1, characterized in that: The adaptive planning sensing points include: By combining the topology and mechanical and thermal properties of the mold frame, highly sensitive areas are identified; and the target constraints are set as 100% coverage of highly sensitive areas, minimization of the number of sensors, and signal interference of less than 5%; then, the optimal location is solved by the topology optimization algorithm; and a deployment list containing sensor type, installation coordinates, and accuracy requirements is output.

3. The mold base design system based on virtual simulation according to claim 1, characterized in that: The attention mechanism uses dynamic weight allocation to achieve data fusion, including: The system acquires single-mode signals of temperature, stress-strain, displacement, and vibration after wavelet threshold denoising, Kalman filtering, and spatiotemporal synchronization processing; calculates the signal-to-noise ratio (SNR) of each mode based on the signal amplitude to noise ratio; dynamically allocates weights according to the SNR; and generates fused data with a confidence level greater than 0.95 by weighted summation of multimodal feature vectors and combining the calibration error and outlier ratio of each mode.

4. The mold base design system based on virtual simulation according to claim 1, characterized in that: The iterative update includes: Construct an objective function to minimize the deviation between simulated and measured values, and preset the parameters as those to be corrected; use the Gauss-Newton iterative algorithm to update the parameters; iterate until the deviation between the digital model and the physical model is less than 0.02 mm, and simultaneously verify the consistency of physical quantities and the rationality of the topology; generate a corrected model with a version identifier, and trigger the simulation module linkage.

5. The mold base design system based on virtual simulation according to claim 1, characterized in that: The optimized grid partitioning includes: Stress concentration regions were extracted from the fused data, and the stress values ​​in these regions were greater than 70% of the allowable stress of the material. High-stress areas were densified using 0.67mm tetrahedral elements, while non-critical areas were simplified using 5mm hexahedral elements. The mesh distortion rate was controlled to be less than 15%, and the aspect ratio to be less than 5. The mesh independence was verified through pre-simulation using three sets of meshes with different densification levels.

6. The mold base design system based on virtual simulation according to claim 1, characterized in that: The integrated multi-solver performs coupled simulations, including: The structural solver and fluid solver are encapsulated as functional mock-up units according to the FMI2.0 standard; the solution resources are dynamically allocated in three stages: filling, holding, and cooling, with a filling step size of 0.05s; temperature and pressure coupled physical quantities are transferred every 0.05s; the solution residual is monitored, and convergence is determined when it is less than or equal to 1e-4, and the mesh or step size is adjusted and the local simulation is restarted when divergence occurs.

7. The mold base design system based on virtual simulation according to claim 1, characterized in that: The optimization secondary algorithm includes: Initialize two-level algorithm parameters: Global search uses the NSGA-II algorithm, configured with a population size of 100, 50 iterations, a crossover probability of 0.8, and a mutation probability of 0.01; Local optimization uses the Bayesian optimization algorithm, selecting the Matérn kernel function, an exploration coefficient of 2.5, and a maximum of 30 iterations; Execute global search: The NSGA-II algorithm is used to traverse and optimize the design variable space, generating a Pareto front solution set containing 100 solutions. The top 20% of high-potential solutions are selected as candidate sets based on the comprehensive score of the objective function; Conduct local optimization: Starting from the candidate set, a Gaussian process model is constructed using Bayesian optimization to predict the objective function value, and fine optimization is performed on high-potential regions; Dynamic adjustment and output: The convergence of the solution is calculated every 10 iterations. Convergence is determined when the deviation between two adjacent optimal solutions is less than 1%. The mutation probability or exploration range is dynamically adjusted according to the convergence speed, and finally 3-5 optimal solutions are output.

8. The mold base design system based on virtual simulation according to claim 1, characterized in that: The proxy model includes: Training data preparation: Extract 1000 sets of historical simulation data and 50 sets of the latest simulation results from the data hub, construct the training set according to the format of design variables and performance indicators, and process the data using Min-Max standardization; Model Construction and Training: A multi-scale neural network surrogate model was built, with the input layer matching the design variable dimensions, the hidden layer consisting of 2 layers × 32 neurons, and the output layer corresponding to multi-objective optimization metrics. A transfer learning strategy was used to initialize the network weights, and the model was fine-tuned using the training set to reduce the prediction error to below 8%. Optimization Acceleration: During the optimization process, the proxy model is called to predict the performance indicators corresponding to the design variables in real time, replacing some direct simulation calculations and shortening the optimization iteration cycle; Incremental model update: After each round of optimization, the newly generated optimal solution is added to the training set, triggering an incremental model update. The update time is controlled within 5 minutes to adapt to dynamic changes in working conditions and data.

9. The mold base design system based on virtual simulation according to claim 1, characterized in that: The identification defects include: Assembly interference detection: Activate the continuous collision detection algorithm to monitor the contact status of components during virtual assembly at a frequency of 60Hz, mark the interference area and calculate the interference amount with an accuracy of 0.001mm, and simultaneously record the motion stage in which the interference occurs; Tolerance compliance analysis: By calling the GD&T standard database, the probability distribution of actual fit clearances is calculated using Monte Carlo simulation for key mating surfaces. Areas with an out-of-tolerance probability greater than 5% are marked as tolerance risk areas. Molding process defect identification: Import temperature field and flow field data from multiphysics simulation, present the melt filling process through visualization technology, detect defects with shrinkage depth greater than or equal to 0.1 mm or warpage greater than or equal to 0.2 mm, and trace the cause by combining heat flow parameters; Defect information integration: Output defect location coordinates, severity level and cause analysis according to detection type to form a structured defect dataset.

10. A mold base design system based on virtual simulation according to claim 1, characterized in that: The data hub module includes: A unified interface supporting MQTT / gRPC protocols is established to receive multi-source data, parse it into a structured format and verify it. A three-tier architecture of memory, distributed database and object storage is adopted for storage and version traceability is implemented. Pre-set event-driven rules schedule cross-module collaboration, monitor the load in real time and allocate computing power, and execute retry alarms for abnormal modules. Accuracy indicators are collected at fixed intervals. When the indicators exceed the standard continuously, the source is located and a calibration command is sent. The accuracy after verification and correction is ensured to guarantee closed-loop accuracy.

Images