An artificial intelligence-based thermosetting mold processing method and system

Abstract

The application discloses a thermosetting mold processing method and system based on artificial intelligence, and relates to the technical field of mold processing.The application comprises the following steps: constructing a three-dimensional model of a product through drawings or 3D scanning; establishing a thermosetting material curing reaction kinetics model; predicting material flow characteristics under different temperature and pressure combinations through deep learning; adjusting mold temperature machine parameters in real time to achieve optimal control of the curing rate; performing mold surface defect detection based on computer vision; and setting an automatic compensation mechanism to automatically compensate for shrinkage deformation in subsequent processing.The application establishes a thermosetting material curing reaction kinetics model, combines material flow characteristics under different temperature / pressure combinations, adjusts mold temperature machine parameters in real time to achieve optimal control of the curing rate, detects defects on the surface of the product, automatically compensates for shrinkage deformation, improves mold processing efficiency and detection efficiency, and reduces the probability of product production rejects.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of mold processing, in particular to a thermosetting mold processing method and system based on artificial intelligence. BACKGROUND

[0002] Thermosetting mold is an industrial mold used for producing thermosetting plastic products. Thermosetting plastics undergo chemical changes during heating, forming a network structure of polymers that will not soften even if heated after molding, so thermosetting molds need to withstand high temperature and high pressure molding process and ensure the dimensional stability and surface quality of the products.

[0003] Thermosetting molds are usually processed by assembling parts according to standards and customer requirements. The quality of mold processing directly affects the service life and performance of the product. The technical bottlenecks of the prior art are as follows:

[0004] (1) Real-time defects: centralized cloud computing architecture causes a delay of more than 200ms, making it difficult to meet the 50ms level response requirement of precision machining;

[0005] (2) Data island problem: heterogeneous device protocols (Modbus / Profinet, etc.) cause more than 30% of the data to be ineffective for virtual simulation;

[0006] (3) Model distortion: the offline digital twin update cycle is as long as 24 hours, with a deviation of more than 15% from the physical production line.

[0007] Therefore, there is a problem of low detection efficiency and high false positive rate, and the subjective nature of humans leads to unsatisfactory quality inspection results. SUMMARY

[0008] The purpose of the present application is to provide a thermosetting mold processing method and system based on artificial intelligence, which establishes a thermosetting material curing reaction kinetics model, combines the material flow characteristics under different temperature / pressure combinations, and adjusts the mold temperature machine parameters in real time to achieve optimal control of the curing rate. The surface of the product is detected for defects, and the shrinkage deformation is automatically compensated. The problem of low detection efficiency and product distortion during the existing thermosetting mold processing is solved.

[0009] To solve the above technical problems, the present application is realized by the following technical scheme:

[0010] The present application is a thermosetting mold processing method based on artificial intelligence, comprising the following steps:

[0011] Step S1: Construct a three-dimensional model of the product through drawings or 3D scanning; save the three-dimensional model in stp format, open it with SOLIDWORKS to analyze the product, analyze whether the product can be smoothly demolded when the mold is opened, whether there is a reverse buckle, whether the demolding slope is sufficient, where the glue port is located, whether to make a cold runner or a hot runner, confirm that the 3D drawing will not be changed, and then convert a 2D file from the 3D drawing to make a product drawing, and the key part dimensions and tolerances are marked.

[0012] When the mold is opened and the product is made, a measurement report is made based on the 2D drawing to confirm whether the dimensions of the product made and the original model are consistent; a 2D mold structure drawing is drawn, which clearly shows the overall structure of the mold, and 3D mold splitting is completed. According to the 2D, the 3D mold splitting is completed, and each part is disassembled into a scattered piece, which is converted from the 3D drawing to a 2D drawing and distributed to the workshop for the mold master. The 3D drawing is sent to the numerical control machining department, and the CNC programmer there will program the mold core and the mold frame walking program and input it into the numerical control machining center.

[0013] Step S2: Establish a thermosetting material curing reaction kinetics model;

[0014] Step S3: Predict the material flow characteristics under different temperature and pressure combinations through deep learning;

[0015] Step S4: Real-time adjustment of mold temperature machine parameters for optimal control of curing rate;

[0016] Step S5: Mold surface defect detection based on computer vision;

[0017] Step S6: Set up an automatic compensation mechanism to automatically compensate for shrinkage and deformation in subsequent processing;

[0018] Step S7: Edge computing node real-time processing of sensor data, virtual mold testing by cloud digital twin platform, and 5G transmission to realize full-factor synchronization of the processing process.

[0019] As a preferred technical solution, in step S1, a three-dimensional model of the product is constructed by 3D scanning through a high-definition camera, and the details of the three-dimensional model are optimized as follows:

[0020] Step S01: Identify and classify the details on the surface of the three-dimensional model;

[0021] Step S02: Establish a grid index to systematically organize all partition units, and when two details are adjacent and share a network boundary, the two details are classified into the same cluster and marked;

[0022] Step S03: According to the hierarchical relationship of details, starting from the outermost details, gradually advancing to the inner layer;

[0023] Step S04: Project the target boundary point onto the grid boundary to obtain the projection point as a new point on the grid boundary; Capture the boundary triangle surface associated with each boundary point, calculate the average height of its barycenter, and assign the average height associated with the corresponding boundary point to each new point;

[0024] Step S05: Smooth the shared boundary of adjacent details in the same layer until the layer-by-layer repair is completed.

[0025] As a preferred technical solution, in step S02, each partition unit is treated as a basic unit for repairing details, the hierarchical relationship between details is calculated, and a grid index is established to systematically organize all partition units. When two details in adjacent partition units share a grid boundary, the two details will be classified into the same cluster and marked. The details in each cluster will be hierarchically divided according to the following rules: the details located at the outermost edge of the cluster are classified into the first layer, the details sharing the grid boundary with the details in the first layer are classified into the second layer, the details in the nth layer are determined by parity check reasoning, and the overall hierarchical relationship between details is established.

[0026] As a preferred technical solution, in step S03, when repairing from the outermost details, the adjustment sequence is from outside to inside, and the inner ring is adjusted based on the connection relationship of the newly added repair network. When adjusting, the reciprocal distance weighted average algorithm is used to calculate the first order neighbor and second order neighbor of the newly added outer point and the weight of each neighbor. The position of the point is adjusted from outside to layer by layer using the weighted average algorithm, and the adjustment function is as follows:

[0027] ;

[0028] In the formula, represents the position of the adjusted new point , represents the coordinates of the point being adjusted, i.e. the newly added repair point, represents the weight, represents the coordinates of the first neighbor of the newly added point ;

[0029] After completing the height distribution of the boundary points, the repair and height propagation of the details are performed layer by layer, the open repair is converted to closed repair, and the minimum area method is used to complete the detail filling.

[0030] As a preferred technical solution, in step S05, the smoothing process uses the Laplace smoothing method to adjust the height, setting the height of each point to the average height of all points within a certain buffer range. The specific formula is as follows:

[0031] ;

[0032] In the formula, For smoothing coefficients, This is the initial height. For the updated height, The specific formula for the height adjustment is as follows:

[0033] ;

[0034] In the formula, The first in the neighborhood of the point to be repaired The height value of each point. This represents the number of points in the neighborhood of the point to be repaired.

[0035] As a preferred technical solution, the specific process for establishing the curing reaction kinetics model of the thermosetting material in step S2 is as follows:

[0036] Step S21: Use DSC data to train an LSTM network to predict the evolution of curing degree, generate an adversarial network to simulate non-ideal curing behavior; use differential scanning calorimetry (DSC) to obtain the curing exothermic curve, and calculate the apparent activation energy Ea and pre-exponential factor A using the Kissinger equation;

[0037] Step S22: Obtain microscopic parameters based on molecular dynamics simulations of the cross-linking process; establish a system including autocatalytic terms. Improved kinetic equations for diffusion control terms;

[0038] Step S23: Couple the PDE module and the solid heat transfer module in COMSOL, and define the reaction rate variable as the source term;

[0039] Step S23: Real-time control based on reinforcement learning with PID minimization as the objective function;

[0040] Step S24: Combine Maxwell's equations with reaction kinetics to invert microwave curing parameters, achieving simultaneous prediction of temperature field and degree of curing; solve the thermo-chemical equations through multi-field coupling. To achieve simultaneous prediction of temperature field and degree of curing, in the formula, For material density, For specific heat capacity, Let be the partial derivative of temperature with respect to time. Thermal conductivity, For temperature gradient, is the total reaction enthalpy, is the degree of cure, is the cure rate.

[0041] In step S3, the material flow characteristics under different temperature and pressure combinations are predicted by deep learning, and the specific process is as follows:

[0042] Step S31, data preparation and preprocessing: collect the experimental data or high-precision simulation data (such as molecular dynamics simulation results) of the flow characteristics of the material under different temperature and pressure conditions, normalize / standardize the data, eliminate the dimensional influence, use wavelet denoising method to process noise data, and use regression method or mean smoothing method to process abnormal values;

[0043] Step S32, feature engineering: construct a two-dimensional input feature space of temperature-pressure combination, extract material microstructure features (such as lattice parameters, phase composition, etc.) as auxiliary input, and perform dimension reduction processing (such as PCA) or feature selection on high-dimensional features;

[0044] Step S33, LSTM model architecture: the core uses physical information neural network (PINN), embeds fluid mechanics control equations such as Navier-Stokes equation into loss function as constraint condition; combine convolutional neural network (CNN) to process material microstructure image data; use graph neural network (GNN) to model atomic / molecular interaction;

[0045] Step S34, training and optimization: adopt transfer learning strategy, pre-train on large-scale synthetic data, and fine-tune on small-scale experimental data; use Bayesian optimization to automatically adjust hyperparameters, and introduce attention mechanism to process strong nonlinear coupling of temperature / pressure parameters;

[0046] Step S35, verification and deployment: evaluate the generalization ability of the model through k-fold cross-validation, analyze the feature importance using SHAP value and other explainability methods; deploy as a digital twin system to predict material flow behavior in real time.

[0047] As a preferred technical solution, in step S4, the specific process of real-time adjustment of mold temperature machine parameters for optimal control of the cure rate is as follows:

[0048] Step S41: build thermal curing kinetics model, use Arrhenius equation to describe the relationship between temperature and cure rate;

[0049] Step S42: according to the heat transfer process of the mold-material interface, build a thermal conduction finite element model;

[0050] Step S43: initialize PID controller parameters, predict heat flow demand based on material DSC curve, roll optimize temperature estimation for next 5 seconds, start Kalman filter smoothing when temperature fluctuation is detected;

[0051] Real-time acquisition of mold temperature controller working data, including: deploying infrared thermocouple array as temperature sensing network, using piezoelectric sensor to detect pressure fluctuation in mold cavity for pressure detection, measuring dielectric loss factor through dielectric sensor for curing degree detection;

[0052] Step S44: real-time synchronization of physical sensor data and virtual model, update finite element thermal field simulation every 200ms, feedback optimization results to mold temperature controller PLC through digital thread;

[0053] Step S45: obtain current mold cavity temperature distribution , calculate local curing temperature , solve optimal temperature set point , Reaction order;

[0054] Step S46: adjust mold temperature controller heating power , verify actual curing rate and target deviation ;

[0055] Step S47: dynamically update material curing dynamic parameters;

[0056] When the temperature is monitored to exceed the material decomposition temperature, the heating is immediately cut off, when the flash risk is detected, the mold clamping force is automatically reduced for pressure compensation, and the remaining life of the heating rod is predicted based on the LSTM network.

[0057] As a preferred technical solution, in the step S5, the specific process of mold surface defect detection based on computer vision is as follows:

[0058] Step S51: adopt industrial camera with more than 20 million pixels, cooperate with multi-spectral imaging technology to capture subtle defects;

[0059] Step S52: eliminate reflection interference through ring LED light source, adopt dynamic compensation algorithm to process uneven illumination;

[0060] Step S53: use median filter to eliminate salt and pepper noise, histogram equalization to enhance defect contrast, and mean smoothing or regression method to correct abnormal pixel points;

[0061] Step S54: set in sliding window, sample 128x128 window on 512x512 image for binary classification, directly regress defect position and category, suitable for real-time production line detection;

[0062] Step S55: Use YOLO target detection to locate the defect position;

[0063] Step S56: Automatically extract macro-micro features using ResNet network, enhance feature weights of defect area, suppress background interference, and identify defects such as cracks and scratches.

[0064] As a preferred technical solution, in step S06, the digital image correlation method or optical fiber sensor is used to monitor the material shrinkage deformation in real time, and the temperature / pressure sensor is used to obtain the environmental parameters; a shrinkage constitutive equation based on physics is constructed, and is combined with a data-driven deep learning model (such as PINN) to establish a shrinkage prediction model;

[0065] The inverse solving technology is used to convert the predicted shrinkage into processing parameter correction value (such as mold size compensation, processing path offset); the BP neural network is integrated to realize nonlinear compensation under multi-parameter coupling, and the compensation coefficient is dynamically adjusted;

[0066] The compensation instruction is fed back to the processing equipment in real time through PLC or CNC system to form a "monitoring-prediction-compensation" closed loop; transfer learning is used to adapt to the characteristic differences of different batches of materials to ensure the generalization ability of compensation;

[0067] This mechanism realizes high-precision adaptive compensation by real-time monitoring, multi-modal data fusion, combining physical model and deep learning to predict deformation, and then generating compensation parameters through reverse engineering and closed-loop control of processing equipment.

[0068] As a preferred technical solution, in step S07, the edge computing node processes sensor data in real time, the virtual tryout is performed by the cloud digital twin platform, and the specific process of 5G transmission to realize synchronization of all factors in the processing process is as follows:

[0069] Step S71: Deploy an industrial-grade edge computing node and integrate temperature, pressure, and vibration multi-source sensors;

[0070] Step S72: Use the CNN model to perform data cleaning and feature extraction, and perform millisecond-level anomaly detection;

[0071] Step S73: Establish an end-to-end <10ms transmission channel through 5G URLLC; use TSN (Time Sensitive Network) protocol to ensure data timing consistency, implement AES-256 encryption, and use blockchain to ensure data security;

[0072] Step S74: Build a virtual production line three-dimensional model based on Unity3D, run multi-physics field coupling simulation, and dynamically optimize process parameters;

[0073] Step S75: the cloud issues optimization instructions to the edge controller, and the edge node adjusts the PLC parameters in real time, and the digital twin and the physical production line maintain a synchronization delay of <50ms.

[0074] The application is a thermosetting mold processing system based on artificial intelligence, which comprises an intelligent sensing layer, a decision hub layer, an execution implementation layer and a human-computer interaction layer.

[0075] The intelligent sensing layer comprises a temperature sensor, a pressure sensor, a vibration sensor and an industrial camera; the temperature sensor is used for monitoring the temperature in the mold production process; the pressure sensor is used for monitoring the internal pressure in the mold production process; the vibration sensor is used for collecting the vibration frequency in the mold production process; and the industrial camera is used for capturing the defects on the surface of the product after the mold production is completed.

[0076] The decision hub layer comprises a thermosetting reaction kinetics model, a prediction model and a mold temperature machine parameter control module; the thermosetting reaction kinetics model uses the Arrhenius equation to describe the temperature-curing rate relationship and constructs a thermosetting kinetics model; the prediction model is used for predicting the material flow characteristics under different temperature / pressure combinations through deep learning; and the mold temperature machine parameter control module is used for realizing optimal control of the curing rate by adjusting the mold temperature machine parameters in real time.

[0077] The execution implementation layer comprises a mold surface defect detection module, an LSTM network module and a shrinkage compensation module; the surface defect detection module is used for mold surface defect detection through computer vision; the LSTM network module is used for establishing a processing parameter-defect type mapping relationship through an LSTM network; and the shrinkage compensation module is used for setting an automatic compensation mechanism to automatically compensate the shrinkage deformation in subsequent processing.

[0078] The human-computer interaction layer comprises an edge node, a cloud digital twin platform and a 5G transmission module; the edge node is used for integrating temperature, pressure and vibration multi-source sensors to obtain sensor data and adjust PLC parameters in real time; the cloud digital twin platform is used for constructing a virtual production line three-dimensional model based on Unity3D, running multi-physics field coupling simulation and dynamically optimizing process parameters; and the 5G transmission module is used for realizing full-factor synchronization of the processing process.

[0079] The application has the following beneficial effects:

[0080] (1) The application establishes a thermosetting material curing reaction kinetics model, combines the material flow characteristics under different temperature / pressure combinations, adjusts the mold temperature machine parameters in real time to realize optimal control of the curing rate, detects the defects on the surface of the product, automatically compensates the shrinkage deformation, improves the mold processing efficiency and detection efficiency, and reduces the probability of product production rejects.

[0081] (2) The present application obtains the curing exothermic curve under different heating rates by differential scanning calorimetry (DSC), and uses Kissinger method to calculate apparent activation energy Ea and pre-exponential factor A1; secondly, an improved kinetic equation containing autocatalytic term and diffusion control term is established; then the generalized partial differential equation (curing kinetics) and solid heat transfer module are coupled in COMSOL, and the reaction rate variable rate is defined as a source term; finally, the heat-chemical equation is solved by multi-field coupling to realize the synchronous prediction of temperature field and curing degree.

[0082] (3) The present application realizes sub-second level synchronization of physical world and digital twin through edge-cloud collaborative architecture under the support of 5G network, so that the virtual trial result can directly guide the real-time adjustment of production line, and the traditional trial and error period is shortened from several weeks to minutes.

[0083] (4) The present application deploys industrial edge computing nodes, integrates temperature, pressure, vibration multi-source sensors, constructs a virtual production line three-dimensional model based on Unity3D, runs multi-physical field coupling simulation, dynamically optimizes process parameters, facilitates monitoring the production process of the mold, discovers production accidents in time, and improves equipment maintenance efficiency.

[0084] Of course, any product implementing the present application does not necessarily need to achieve all the advantages described above at the same time. BRIEF DESCRIPTION OF DRAWINGS

[0085] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the description of the embodiments. Obviously, the drawings in the following description only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of these drawings.

[0086] Figure 1 A flow chart of a thermosetting mold processing method based on artificial intelligence of the present application;

[0087] Figure 2 A structure schematic diagram of a thermosetting mold processing system based on artificial intelligence of the present application. DETAILED DESCRIPTION

[0088] 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 some embodiments of the present application, not all 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.

[0089] ​In addition, the technical features involved in each embodiment of the application described below can be combined with each other as long as they do not conflict with each other.

[0090] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application. Figures 1-2 and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0091] Embodiment one

[0092] Please refer to Figure 1 The present application is a thermosetting mold processing method based on artificial intelligence, which comprises the following steps:

[0093] Step S1: build a three-dimensional model of the product through drawings or 3D scanning; save the three-dimensional model as an stp format, open it with SOLIDWORKS to analyze the product, analyze whether the product can be smoothly demolded when the mold is opened, whether there is a reverse buckle, whether the demolding slope is sufficient, where the glue port is located, whether to make a cold runner or a hot runner, confirm that the 3D drawing will not be changed, and then convert the 3D drawing into a 2D file to make a product drawing, and the dimensions and tolerances of each key part shall be marked.

[0094] After the mold is opened and the product is made, a measurement report is made based on the 2D drawing to confirm whether the dimensions of the product made and the original model are consistent; a 2D mold structure drawing is drawn, which clearly shows the overall structure of the mold, and the 3D mold is divided according to the 2D drawing. After the 3D mold is divided, each part is disassembled into a scattered piece, and the 3D drawing is converted into a 2D drawing file and distributed to the workshop for the mold master. The 3D drawing file is sent to the numerical control machining department, and the CNC programmer there will program the mold core and the mold opening frame to input the numerical control machining center.

[0095] Step S2: establish a thermosetting material curing reaction kinetics model;

[0096] Step S3: predict the material flow characteristics under different temperature and pressure combinations through deep learning;

[0097] Step S4: real-time adjustment of mold temperature machine parameters for optimal control of curing rate;

[0098] Step S5: mold surface defect detection based on computer vision;

[0099] Step S6: set up an automatic compensation mechanism to automatically compensate for the shrinkage deformation in subsequent processing;

[0100] Step S7: The edge computing node processes the sensor data in real time, the virtual try-on is performed by the cloud digital twin platform, and the 5G transmission realizes the synchronization of all factors in the machining process.

[0101] In step S1, a three-dimensional model of the product is constructed by 3D scanning with a high-definition camera, and the details of the three-dimensional model are optimized as follows:

[0102] Step S01: Identify and classify the details on the surface of the three-dimensional model;

[0103] Step S02: Establish a grid index to systematically organize all partition units. When two details are adjacent and share a network boundary, the two details are classified into the same cluster and marked.

[0104] Step S03: According to the hierarchical relationship of the details, start repairing from the outermost details and gradually move inward.

[0105] Step S04: Project the target boundary point onto the grid boundary to obtain the projection point as a new point on the grid boundary. Capture the boundary triangle face associated with each boundary point and calculate the average height of its barycenter. Assign the average height associated with the corresponding boundary point to each new point.

[0106] Step S05: Smooth the shared boundary of adjacent details in the same layer until the layer-by-layer repair is completed.

[0107] In step S02, each partition unit is treated as a basic unit for repairing details, the hierarchical relationship between details is calculated, and a grid index is established to systematically organize all partition units. When two details are adjacent and share a grid boundary, the two details will be classified into the same cluster and marked. The details in each cluster will be hierarchically divided according to the following rules: the details located at the outermost edge of the cluster are classified into the first layer, the details sharing the grid boundary with the details in the first layer are classified into the second layer, the details in the nth layer are determined by parity check reasoning, and the overall hierarchical relationship between the details is established.

[0108] In step S03, when repairing from the outermost details, the adjustment sequence is from outside to inside, and the inner ring is adjusted based on the connection relationship of the newly added repair network. When adjusting, the inverse distance weighted average algorithm is used to calculate the first-order and second-order neighbor points of the newly added outer point and the weight of each neighbor point. The position of the point is adjusted from outside to layer by layer using the weighted average algorithm, and the adjustment function is as follows:

[0109] ;

[0110] In the formula, represents the position of the adjusted new point ​ coordinates of the point currently being adjusted, i.e. the newly added patch point, represents a weight, represents the newly added point coordinates of the th neighboring point of the newly added point;

[0111] After the height assignment of the boundary points is completed, the detail patching and height propagation are performed layer by layer, the open patching is converted into closed patching, and the detail filling is completed by using the minimum area method.

[0112] In step S05, the smoothing process adjusts the height by using the Laplace smoothing method, and sets the height of each point as the average height of all points within a certain buffer range of the point, and the specific formula is as follows:

[0113] ;

[0114] In the formula, is a smoothing coefficient, is an initial height, is an updated height, is a height adjustment amount, and the specific formula is as follows:

[0115] ;

[0116] In the formula, is the height value of the th point in the neighborhood of the point to be repaired, is the number of points in the neighborhood of the point to be repaired.

[0117] In step S2, the specific process of establishing the curing reaction kinetics model of the thermosetting material is as follows:

[0118] Step S21: using DSC data to train the LSTM network to predict the curing degree evolution, and generating the adversarial network to simulate the non-ideal curing behavior; using differential scanning calorimetry (DSC) to obtain the curing exothermic curve, and calculating the apparent activation energy Ea and the pre-exponential factor A by using the Kissinger equation;

[0119] Step S22: obtaining micro parameters according to the crosslinking process of molecular dynamics simulation; establishing an improved kinetics equation containing a self-catalytic term and a diffusion control term;

[0120] Step S23: coupling the PDE module and the solid heat transfer module in COMSOL, and defining the reaction rate variable as a source term;

[0121] Step S23: taking PID minimization as the objective function for real-time regulation based on reinforcement learning;

[0122] Step S24: Microwave solidification parameter inversion is realized by combining Maxwell's equations with reaction kinetics to realize simultaneous prediction of temperature field and solidification degree; the thermal-chemical equation is solved by multi-field coupling to realize simultaneous prediction of temperature field and solidification degree, wherein, is the material density, is the specific heat capacity, is the partial derivative of temperature with respect to time, is the thermal conductivity, is the temperature gradient, is the total reaction enthalpy, is the solidification degree, is the solidification rate.

[0123] In step S3, the material flow characteristics under different temperature and pressure combinations are predicted by deep learning, and the specific process is as follows:

[0124] Step S31, data preparation and preprocessing: collect experimental data or high-precision simulation data (such as molecular dynamics simulation results) of material flow characteristics under different temperature and pressure conditions, normalize / standardize the data to eliminate dimensional effects, use wavelet denoising method to process noise data, and use regression method or mean smoothing method to process abnormal values;

[0125] Step S32, feature engineering: construct a two-dimensional input feature space of temperature-pressure combination, extract material microstructure features (such as lattice parameters, phase composition, etc.) as auxiliary input, and perform dimensionality reduction processing (such as PCA) or feature selection on high-dimensional features;

[0126] Step S33, LSTM model architecture: the core uses physical information neural network (PINN), embeds fluid mechanics control equations such as Navier-Stokes equation into loss function as constraint condition; combined with convolutional neural network (CNN) to process material microstructure image data; use graph neural network (GNN) to model atomic / molecular interaction;

[0127] Step S34, training and optimization: adopt transfer learning strategy, first pre-train on large-scale synthetic data, then fine-tune on small-scale experimental data; use Bayesian optimization to automatically adjust hyperparameters, introduce attention mechanism to handle strong nonlinear coupling of temperature / pressure parameters;

[0128] Step S35, verification and deployment: evaluate the generalization ability of the model through k-fold cross-validation, use SHAP value and other explainability methods to analyze feature importance; deploy as a digital twin system to predict material flow behavior in real time.

[0129] In step S4, the specific process of real-time adjustment of mold temperature machine parameters for solidification rate optimal control is as follows:

[0130] Step S41: Construct the thermal curing kinetics model, use Arrhenius equation to describe the temperature-curing rate relationship;

[0131] Step S42: According to the heat transfer process of mold-material interface, construct the heat conduction finite element model;

[0132] Step S43: Initialize the PID controller parameters, predict the heat flow demand based on the material DSC curve, optimize the temperature estimation of the future 5 seconds, and start Kalman filter smoothing when the temperature fluctuation is detected;

[0133] Real-time acquisition of mold temperature machine working data, including: deploying infrared thermocouple array as temperature sensing network, using piezoelectric sensor to detect pressure fluctuation in mold cavity for pressure detection, measuring dielectric loss factor through dielectric sensor for curing degree detection;

[0134] Step S44: Real-time synchronization of physical sensor data and virtual model, update the finite element thermal field simulation every 200ms, and feedback the optimization results to the mold temperature machine PLC through digital thread;

[0135] Step S45: Obtain the current mold cavity temperature distribution , calculate the local curing temperature , solve the optimal temperature set point , for the reaction order;

[0136] Step S46: Adjust the mold temperature machine heating power , verify the actual curing rate and target deviation ;

[0137] Step S47: Dynamically update the material curing dynamic parameters;

[0138] When the temperature monitoring detects that the temperature exceeds the material decomposition temperature, immediately cut off the heating, when the flash risk is detected, automatically reduce the clamping force for pressure compensation, and simultaneously predict the remaining life of the heating rod based on the LSTM network.

[0139] In step S5, the specific process of mold surface defect detection based on computer vision is as follows:

[0140] Step S51: Use an industrial camera with more than 20 million pixels, and cooperate with multispectral imaging technology to capture subtle defects;

[0141] Step S52: Eliminate reflection interference through ring LED light source, and use dynamic compensation algorithm to process uneven lighting;

[0142] Step S53: Use median filtering to eliminate salt and pepper noise, histogram equalization to enhance defect contrast, and mean smoothing or regression method to correct abnormal pixel points;

[0143] Step S54: Set a sliding window, sample a 128x128 window on a 512x512 image for binary classification, and directly regress the defect position and category. It is suitable for real-time production line detection.

[0144] Step S55: Use YOLO target detection to locate the defect position;

[0145] Step S56: Use ResNet network to automatically extract macro-micro features, enhance defect area feature weights, suppress background interference, and identify cracks, scratches, and other defects.

[0146] In step S06, the digital image correlation method or optical fiber sensor is used to monitor the material shrinkage deformation in real time, and the temperature / pressure sensor is used to obtain the environmental parameters; a physical-based shrinkage constitutive equation is constructed, and a deep learning model (such as PINN) is fused to establish a shrinkage prediction model.

[0147] Using inverse solving technology, the predicted shrinkage is converted into processing parameter correction value (such as mold size compensation, processing path offset); integrating BP neural network to realize nonlinear compensation under multi-parameter coupling, dynamically adjusting compensation coefficient;

[0148] Through PLC or CNC system, the compensation instruction is fed back to the processing equipment in real time to form a "monitoring-prediction-compensation" closed loop; using transfer learning to adapt to the characteristic differences of different batches of materials, ensuring the generalization ability of compensation;

[0149] This mechanism realizes high-precision adaptive compensation by real-time monitoring and multi-modal data fusion, combining physical models and deep learning to predict deformation, and then generating compensation parameters through reverse engineering and closed-loop control of processing equipment.

[0150] In step S07, the edge computing node processes sensor data in real time, and the cloud digital twin platform performs virtual tryout, while 5G transmission realizes the synchronization of all elements in the processing process. The specific process is as follows:

[0151] Step S71: Deploy industrial-grade edge computing nodes, integrate temperature, pressure, vibration multi-source sensors;

[0152] Step S72: Use CNN model for data cleaning and feature extraction, and perform millisecond-level anomaly detection;

[0153] Step S73: Establish an end-to-end <10ms transmission channel through 5G URLLC; adopt TSN (Time Sensitive Network) protocol to guarantee data timing consistency, implement AES-256 encryption and blockchain storage to ensure data security;

[0154] Step S74: Build a virtual production line three-dimensional model based on Unity3D, run multi-physical field coupling simulation, and dynamically optimize process parameters;

[0155] Step S75: The cloud issues optimization instructions to the edge controller, and the edge node adjusts the PLC parameters in real time, and the digital twin and the physical production line maintain <50ms synchronization delay.

[0156] The application is a kind of thermosetting mold processing system based on artificial intelligence, including intelligent sensing layer, decision center layer, implementation layer and man-machine interaction layer.

[0157] The intelligent sensing layer includes temperature sensors, pressure sensors, vibration sensors and industrial cameras; the temperature sensor is used to monitor the temperature in the mold production process; the pressure sensor is used to monitor the internal pressure in the mold production process; the vibration sensor is used to collect the vibration frequency in the mold production process; the industrial camera is used to capture the defects on the surface of the product after the mold production is completed;

[0158] The decision center layer includes a thermal curing reaction kinetics model, a prediction model and a mold temperature machine parameter control module; the thermal curing reaction kinetics model uses Arrhenius equation to describe the relationship between temperature and curing rate, and constructs a thermal curing kinetics model; the prediction model is used to predict the material flow characteristics under different temperature / pressure combinations through deep learning; the mold temperature machine parameter control module is used to realize optimal control of the curing rate by adjusting the mold temperature machine parameters in real time;

[0159] The implementation layer includes a mold surface defect detection module, an LSTM network module and a shrinkage compensation module; the surface defect detection module is used for mold surface defect detection through computer vision; the LSTM network module is used to establish a mapping relationship between processing parameters and defect types through LSTM network; the shrinkage compensation module is used to set an automatic compensation mechanism to automatically compensate the shrinkage deformation in subsequent processing;

[0160] The man-machine interaction layer includes an edge node, a cloud digital twin platform and a 5G transmission module; the edge node is used to integrate temperature, pressure, vibration multi-source sensors to obtain sensor data and adjust PLC parameters in real time; the cloud digital twin platform is used to build a virtual production line three-dimensional model based on Unity3D, run multi-physical field coupling simulation, and dynamically optimize process parameters; the 5G transmission module is used to realize synchronization of all elements in the processing process.

[0161] Example two

[0162] Referring to Figure 2 The present application is a kind of thermosetting mold processing system based on artificial intelligence, which can be used to implement the method of embodiment 1 of the present application, including: intelligent perception layer, decision center layer, execution implementation layer and human-computer interaction layer.

[0163] The intelligent perception layer includes temperature sensors, pressure sensors, vibration sensors and industrial cameras; the temperature sensors are used to monitor the temperature during the mold production process; the pressure sensors are used to monitor the internal pressure during the mold production process; the vibration sensors are used to collect the vibration frequency during the mold production process; the industrial cameras are used to capture the defects on the surface of the product after the mold production is completed;

[0164] The decision center layer includes a thermal curing reaction kinetics model, a prediction model, and a mold temperature machine parameter control module; the thermal curing reaction kinetics model uses the Arrhenius equation to describe the temperature-curing rate relationship and constructs the thermal curing kinetics model; the prediction model is used to predict the material flow characteristics under different temperature / pressure combinations through deep learning; the mold temperature machine parameter control module is used to achieve optimal control of the curing rate by adjusting the mold temperature machine parameters in real time;

[0165] The execution implementation layer includes a mold surface defect detection module, an LSTM network module, and a shrinkage compensation module; the surface defect detection module is used for mold surface defect detection through computer vision; the LSTM network module is used to establish a mapping relationship between processing parameters and defect types through an LSTM network; the shrinkage compensation module is used to set an automatic compensation mechanism to automatically compensate for the shrinkage deformation in subsequent processing;

[0166] The human-computer interaction layer includes edge nodes, a cloud digital twin platform, and a 5G transmission module; the edge nodes are used to integrate temperature, pressure, and vibration multi-source sensors to obtain sensor data and adjust PLC parameters in real time; the cloud digital twin platform is used to construct a virtual production line three-dimensional model based on Unity3D, run multi-physics field coupling simulation, and dynamically optimize process parameters; the 5G transmission module is used to realize synchronization of all elements in the processing process.

[0167] It is worth noting that in the above system embodiment, each unit included is only divided according to functional logic, but is not limited to the above division, as long as the corresponding function can be realized; in addition, the specific names of each functional unit are only for easy differentiation, and do not limit the protection scope of the present application.

[0168] In addition, those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment methods can be completed by programs instructing related hardware, and the corresponding programs can be stored in a computer readable storage medium.

[0169] The preferred embodiments of the application disclosed above are only to facilitate the elucidation of the application. The preferred embodiments do not describe all the details of the application and limit the application to the specific embodiments described. Obviously, many modifications and variations can be made in light of the teachings above. The description is chosen and described in order to best explain the principles of the application and its practical application to thereby enable others skilled in the art to best utilize the application and get the best results from the application. The application is only limited by the claims and their full scope and equivalents.

Claims

1. An artificial intelligence-based thermosetting mold processing method, characterized by, Comprise the following steps: Step S1: build a three-dimensional model of the product through drawings or 3D scanning; Step S2: establish a thermosetting material curing reaction kinetics model; Step S3: predict the material flow characteristics under different temperature and pressure combinations through deep learning; Step S4: real-time adjustment of the mold temperature machine parameters for optimal control of the curing rate; In the step S4, the specific process of real-time adjustment of the mold temperature machine parameters for optimal control of the curing rate is as follows: Step S41: build a thermal curing kinetics model, and use the Arrhenius equation to describe the temperature-curing rate relationship; Step S42: according to the heat transfer process of the mold-material interface, build a heat conduction finite element model; Step S43: initialize the PID controller parameters, predict the heat flow demand based on the material DSC curve, and rollingly optimize the temperature estimation in the next 5 seconds. When the temperature fluctuation is detected, start Kalman filtering smoothing; Step S44: real-time synchronization of physical sensor data and virtual model, feedback optimization results to the mold temperature machine PLC through digital thread; Step S45: Acquire the current mold cavity temperature distribution T(x,y,z), calculate the local solidification temperature a = 1 - exp(-kt n ), solve the optimal temperature set point T * ; Step S46: Adjust the mold temperature controller heating power P = K(T * -T) + ∫(T * -T)dt, verify the actual solidification rate and target deviation; Step S47: dynamically update the material curing dynamic parameters; Step S5: mold surface defect detection based on computer vision; Step S6: set up an automatic compensation mechanism to automatically compensate for shrinkage deformation in subsequent processing; Step S7: edge computing node real-time processing of sensor data, virtual prototyping by cloud digital twin platform, and 5G transmission to realize full-factor synchronization of the processing process.

2. The artificial intelligence-based thermoset mold processing method of claim 1, wherein, In the step S1, a three-dimensional model of the product is built by 3D scanning through a high-definition camera, and the detail optimization process of the three-dimensional model is as follows: Step S01: identify and classify the details on the surface of the three-dimensional model; Step S02: establish a grid index to systematically organize all partition units. When two details are adjacent and share a network boundary, the two details are classified into the same cluster and marked; Step S03: according to the hierarchical relationship of the details, start repairing from the outermost details and gradually move inward; Step S04: project the target boundary point onto the grid boundary to obtain the projection point as a new point on the grid boundary. Capture the boundary triangle surface associated with each boundary point and calculate the average height of its barycenter. Assign the average height associated with the corresponding boundary point to each new point; Step S05: smooth the shared boundary of adjacent details in the same layer until the layer-by-layer repair is completed.

3. The artificial intelligence based thermoset mold processing method of claim 2, wherein, In the step S02, each partition unit is treated as a basic unit for repairing details, and the hierarchical relationship between details is calculated. A grid index is established to systematically organize all partition units. When two details are adjacent and share a grid boundary, the two details will be classified into the same cluster and marked. The details in each cluster will be hierarchically divided according to the following rules: the details located at the outermost edge of the cluster are classified into the first layer, the details sharing the grid boundary with the details in the first layer are classified into the second layer, the details in the nth layer are determined by parity check reasoning, and the overall hierarchical relationship between the details is established.

4. The artificial intelligence based thermoset mold processing method of claim 2, wherein, In step S03, the outermost details are repaired first, and the inner ring is adjusted based on the connection relationship of the newly added repair network in an outside-in adjustment sequence. When adjusting, the reciprocal distance weighted average algorithm is used to calculate the first-order and second-order neighbor points of the newly added outer points and the weight of each neighbor point. The position of the points is adjusted layer by layer from the outside using the weighted average algorithm, and the adjustment function is as follows: where v new represents the position of the adjusted new point vadj, v represents the coordinates of the point currently being adjusted, i.e., the newly added patch point, w i = ||v - Adj(v) || -1 represents the weight, Adj i (v) represents the coordinates of the i-th neighbor point of the newly added point v; After completing the height distribution of the boundary points, the details are repaired and the height is propagated layer by layer to convert the open repair to a closed repair, and the minimum area method is used to complete the detail filling.

5. The artificial intelligence based thermoset mold processing method of claim 2, wherein, In step S05, the height is adjusted using the Laplace smoothing method for smoothing processing, and the height of each point is set to the average height of all points within a certain buffer range. The specific formula is as follows: p new = p + βΔp; In the formula, β is a smoothing coefficient, p is an initial height, p new is an updated height, and Δp is a height adjustment amount, and the specific formula is as follows: In the formula, s i is the height value of the i-th point in the neighborhood of the point to be repaired, and m is the number of points in the neighborhood of the point to be repaired.

6. The artificial intelligence based thermoset mold processing method of claim 2, wherein, In step S2, the specific process of establishing the curing reaction kinetics model of thermosetting materials is as follows: Step S21: Use DSC data to train LSTM network to predict the evolution of curing degree and generate an adversarial network to simulate non-ideal curing behavior; Step S22: Obtain micro parameters by simulating the crosslinking process using molecular dynamics simulation; Step S23: In COMSOL, couple the PDE module and the solid heat transfer module, and define the reaction rate variable as a source term; Step S23: Real-time control based on reinforcement learning with PID minimization as the objective function; Step S24: Combine Maxwell's equations and reaction kinetics to realize microwave curing parameter inversion and synchronous prediction of temperature field and curing degree.

7. The artificial intelligence based thermoset mold processing method of claim 1, wherein, In step S5, the specific process of mold surface defect detection based on computer vision is as follows: Step S51: Use an industrial camera with more than 20 million pixels, and use multi-spectral imaging technology to capture subtle defects; Step S52: Eliminate glare interference by using a ring-shaped LED light source, and use a dynamic compensation algorithm to handle uneven lighting; Step S53: Use median filtering to eliminate salt and pepper noise, histogram equalization to enhance defect contrast, and mean smoothing or regression method to correct abnormal pixel points; Step S54: Set a sliding window and sample 128x128 windows on a 512x512 image for binary classification; Step S55: Use YOLO target detection to locate the defect position; Step S56: Use the ResNet network to automatically extract macro-micro features for defect recognition.

8. The artificial intelligence based thermoset mold processing method of claim 1, wherein, In step S7, the edge computing node processes sensor data in real time, the cloud digital twin platform performs virtual molding, and 5G transmission realizes the synchronization of all elements in the processing process. The specific process is as follows: Step S71: Deploy an industrial-grade edge computing node, integrate temperature, pressure, and vibration multi-source sensors; Step S72: Use the CNN model for data cleaning and feature extraction, and perform millisecond-level anomaly detection; Step S73: Establish an end-to-end <10ms transmission channel through 5G URLLC; Step S74: Build a virtual production line three-dimensional model based on Unity3D, run multi-physical field coupling simulation, and dynamically optimize process parameters; Step S75: The cloud issues optimization instructions to the edge controller, and the edge node adjusts the PLC parameters in real time.

9. An artificial intelligence-based thermosetting mold processing system, comprising an intelligent perception layer, a decision-making hub layer, an execution implementation layer, and a human-computer interaction layer, characterized in that: the intelligent perception layer comprises temperature sensors, pressure sensors, vibration sensors, and industrial cameras; the temperature sensors are used to monitor the temperature during mold production; the pressure sensors are used to monitor the internal pressure during mold production; the vibration sensors are used to collect the vibration frequency during mold production; the industrial cameras are used to capture defects on the surface of the product after mold production is completed; the decision-making hub layer comprises a thermal curing reaction kinetics model, a prediction model, and a mold temperature controller parameter adjustment module; the thermal curing reaction kinetics model uses the Arrhenius equation to describe the temperature-curing rate relationship and constructs a thermal curing kinetics model; the prediction model is used to predict the material flow characteristics under different temperature / pressure combinations through deep learning; the mold temperature controller parameter adjustment module is used to achieve optimal control of the curing rate by adjusting the mold temperature controller parameters in real time; the execution implementation layer comprises a mold surface defect detection module, an LSTM network module, and a shrinkage compensation module; the surface defect detection module is used for mold surface defect detection through computer vision; the LSTM network module is used to establish a mapping relationship between processing parameters and defect types through an LSTM network; the shrinkage compensation module is used to set an automatic compensation mechanism to automatically compensate for shrinkage deformation in subsequent processing; the human-computer interaction layer comprises edge nodes, a cloud digital twin platform, and a 5G transmission module; the edge nodes are used to integrate temperature, pressure, and vibration multi-source sensors to obtain sensor data and adjust PLC parameters in real time; the cloud digital twin platform is used to construct a virtual production line three-dimensional model based on Unity3D, run multi-physics field coupling simulation, and dynamically optimize process parameters; the 5G transmission module is used to realize synchronization of all factors in the processing process.

Images