Die temperature field coordinated regulation and control method based on digital twinning

Abstract

The invention discloses a die temperature field coordinated regulation and control method based on digital twinning, and relates to the technical field of die temperature control. The method comprises the following steps: constructing a mold temperature field digital twin model covering macroscopic, mesoscopic and microscopic scales; a distributed optical fiber temperature sensor is deployed to realize continuous temperature field sensing; physical-digital real-time synchronous mapping is established through ensemble Kalman filtering; performing model prediction control based on the digital twinborn body; three-dimensional visualization and multi-modal man-machine interaction of the temperature field are realized by adopting an augmented reality technology; and reinforcement learning is used to enable the digital twinborn body to have an autonomous optimization capability. Through deep fusion of a physical space and a digital space, precise sensing, real-time mapping and intelligent control of a mold temperature field are realized, and the temperature control precision and the self-adaptive capability of the system are remarkably improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of mold temperature control, and particularly relates to a mold temperature field collaborative regulation method based on digital twinning. BACKGROUND

[0002] Mold temperature control is a key link in manufacturing processes such as injection molding and die casting, and accurate control of the temperature field directly affects product quality and production efficiency.

[0003] In the prior art, mold temperature control mainly relies on a traditional PID controller to realize feedback regulation in cooperation with a point temperature sensor. However, the technical route has the following disadvantages: first, the control response is lagging, and it is difficult to adapt to rapidly changing production conditions; second, point temperature measurement cannot obtain continuous temperature field distribution information, and there is a monitoring blind area; third, there is a lack of prediction ability for the evolution trend of the temperature field, and the control precision is limited; and fourth, the physical state is not visible, and it is difficult for an operator to intuitively grasp the temperature field distribution.

[0004] Therefore, there is an urgent need for a new regulation method that can realize real-time mapping, accurate prediction and collaborative control of the temperature field. SUMMARY

[0005] In view of the deficiencies of the prior art, the application provides a mold temperature field collaborative regulation method based on digital twinning, which realizes real-time synchronous mapping and intelligent control of the physical mold and the digital model by constructing a high-fidelity digital twin of the mold.

[0006] To achieve the above object, the application adopts the following technical scheme: A mold temperature field collaborative regulation method based on digital twinning, the method comprising the following steps: S1, constructing a mold multi-scale digital twin basic model, establishing a mold temperature field digital model from macro, meso and micro three spatial scales, wherein the macro scale describes the overall temperature distribution based on a three-dimensional unsteady heat conduction equation, the meso scale introduces a convective heat transfer boundary condition for local refinement in key areas, and the micro scale considers the temperature dependence of material thermal physical property parameters; bidirectional information transmission between the three scales is realized through boundary mapping, volume averaging and homogenization methods, and an adaptive grid technology based on temperature gradient norm is used to dynamically adjust the grid density; S2, deploying a continuous temperature field perception network, realizing continuous monitoring of the mold temperature field through a distributed optical fiber temperature sensor, the sensing being based on the Raman scattering principle, obtaining temperature information by measuring the intensity ratio of Stokes light and anti-Stokes light, and determining the spatial position of the measurement point through optical time domain reflection technology; a Butterworth low-pass filter is used to filter the original signal, abnormal values are identified and corrected through local statistical inspection, and a master-slave dual-channel redundancy mechanism is configured to ensure temperature measurement continuity; S3, a physical-digital real-time synchronization mapping mechanism is established, a set Kalman filtering algorithm is used to fuse the perception data into the digital twin model, N set members are used to represent state uncertainty, model operators are used for state prediction, observation operators are used to map the model space to the observation space, and each set member is corrected according to the Kalman gain; An abnormality detection mechanism based on Mahalanobis distance is introduced, and when an abnormality is detected, the model parameter online correction is triggered; S4, implement digital twin-based predictive control, perform model predictive control based on the synchronized digital twin, obtain a control sequence by solving a constrained optimization problem, and the optimization objective includes temperature tracking accuracy, control increment and weighted sum of slack variables; A hierarchical control architecture is adopted, the global coordination layer performs overall optimization of the temperature field, the local execution layer tracks the set value through PID control, and the two layers are coupled through an adaptive coordination factor; The sequence quadratic programming algorithm is used to solve the optimization problem, and only the first control action is executed according to the rolling horizon principle; S5, realize temperature field augmented reality visualization, use ray casting algorithm to volume render the temperature field, generate color cloud image through temperature-color mapping function, and adaptively adjust the opacity according to the temperature gradient; The pose of the mold is determined through marker recognition, and the virtual temperature field is superimposed on the physical mold; Support three interaction modes of gesture recognition, voice command and touch operation, gestures are captured by a depth camera and recognized by a convolutional neural network, voice triggers control actions through keyword matching, and touch realizes accurate parameter setting through a graphical interface; Provide multi-user collaboration mechanism and time axis playback function; S6, execute digital twin autonomous learning optimization, model the temperature control as a Markov decision process, the state space includes historical temperature, historical control and environmental disturbance, the action space is the adjustment amount of each temperature control region, and the reward function integrates temperature deviation, energy consumption, temperature uniformity and production cycle; A deep neural network parameterized strategy network outputs an action distribution, and a value network estimates the expected return; Update the strategy through the proximal policy optimization algorithm, and calculate the advantage function using generalized advantage estimation; Training is carried out in the digital twin, and the new strategy must meet the performance improvement and variance reduction conditions before it can be deployed.

[0007] Further, in step S1, the macro-scale model is discretized by the finite element method, the matrix equation is obtained by the Galerkin weighted residual method, and the implicit Euler format is used for time discretization; The convection heat transfer coefficient of the mesoscale is determined by the Nusselt number, the flow state is determined according to the Reynolds number, and the Graetz solution, the Dittus-Boelter correlation or linear interpolation is used for calculation respectively; The thermal conductivity, specific heat capacity and density of the microscale are all expressed as functions of temperature.

[0008] Further, in step S2, the optical fiber is arranged in a spiral winding manner, and the spatial resolution is determined by the pulse width; the abnormal value is corrected by weighted interpolation of adjacent points, and the weight coefficient is determined according to the spatial distance; the dual-channel switching is based on statistical testing of the main and standby channel temperatures, and the switching process adopts a smooth transition function.

[0009] Further, in step S3, the model disturbance and the observation disturbance are generated according to error statistical characteristics, and the prediction error covariance is estimated by set statistics; the cross-covariance and the observation prediction covariance are calculated by set members; the model parameter correction adopts a gradient-based optimization method, and the loss function is the weighted observation error in a time window.

[0010] Further, in step S4, the prediction function is realized by recursive calculation of the digital twin; the global layer optimization target includes the weighted sum of temperature tracking, temperature uniformity and energy consumption; the coordination factor is adaptively adjusted according to the relative size of the global and local control biases; the Hessian matrix is updated by the quasi-Newton method.

[0011] Further, in step S5, the camera intrinsic matrix is obtained by calibration; the gesture control parameter mapping includes the product of gesture gain, movement amplitude and direction vector; multi-user cooperation is realized by conditional selection of master control, secondary control and current value; the time axis playback is reconstructed by linear interpolation of historical data.

[0012] Further, in step S6, the strategy network outputs the mean and standard deviation of the normal distribution; the time series difference error is used to calculate the advantage function; the importance sampling ratio is used for strategy gradient estimation; the clipping parameter prevents the policy update from being too large; the deployment criteria consider the performance improvement amplitude and stability. Advantages

[0013] 1. Enhanced full-field sensing capability: Distributed optical fiber temperature measurement realizes continuous temperature field monitoring, eliminating the monitoring blind area of traditional point-type temperature measurement.

[0014] 2. Significant improvement in control accuracy: Model predictive control based on digital twin effectively overcomes the hysteresis of traditional feedback control.

[0015] 3. Improved human-computer interaction efficiency: Augmented reality visualization and multi-modal interaction methods improve the intuitiveness and convenience of operation.

[0016] 4. Strong system self-adaptation: Reinforcement learning enables the system to have autonomous optimization capability, and the control performance continuously improves over time.

[0017] 5. High virtual-real synchronization accuracy: Data assimilation technology ensures that the digital twin accurately reflects the real-time state of the physical mold.

[0018] 6. Computational efficiency optimization: The adaptive mesh and hierarchical control architecture improves computational efficiency while ensuring accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 A flow chart showing the steps of the method described in the present invention is shown; DETAILED DESCRIPTION

[0020] The exemplary embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] In conjunction with Figure 1 , the present invention provides a mold temperature field collaborative control method based on digital twinning, which realizes real-time monitoring, accurate prediction and intelligent control of the temperature field by constructing a high-fidelity digital mapping body of the physical mold. The core of the method is to establish a digital twin that can accurately reflect the evolution law of the physical mold temperature field, and based on this, realize the deep integration and collaborative optimization of the physical space and the digital space.

[0022] Step S1: Constructing a multi-scale digital twinning basic model of the mold The multi-scale digital twinning basic model comprehensively describes the mold temperature field from three spatial scales of macro, meso and micro, and realizes effective transmission of information of different scales through scale coupling mechanism.

[0023] The macro-scale model is established based on three-dimensional unsteady heat conduction theory, and its control equation is: Wherein: is the density of the mold material; is the specific heat capacity of the material; is the temperature; is the time; is the partial derivative of temperature with respect to time; is the thermal conductivity; is the Hamiltonian operator; is the divergence operator; is the temperature gradient; is the internal heat source density.

[0024] The finite element method is used to spatially discretize the control equation, and the mold domain is divided into a finite number of elements, and in each element, the shape function is used to approximate the temperature field: Wherein: is the temperature at spatial coordinates and time ; is the shape function of the th node; is the temperature at the th node. Each node in time Temperature value; This represents the number of unit nodes. This is the summation symbol.

[0025] Applying the Galerkin weighted residual method, the discretized matrix equation is obtained:

[0026] in: This is the heat capacity matrix; The node temperature vector; The derivative of the nodal temperature vector with respect to time; This is the heat conduction matrix; This is the thermal load vector.

[0027] The elements of the heat capacity matrix are calculated using unit integration:

[0028] in: For the heat capacity matrix, the first Line 1 Column elements; Number the units; The total number of units; In the first Integral over a unit domain; For the first A shape function; For the first A shape function; It is a volumetric infinitesimal element.

[0029] The elements of the heat conduction matrix are:

[0030] in: For the heat conduction matrix, the first Line 1 Column elements; For the first The gradient of a shape function; For the first The gradient of a shape function; This is the dot product operation.

[0031] The time discretization employs an implicit Euler scheme to ensure numerical stability. in: For the first Temperature vector at any given time; For the first Temperature vector at any given time; For time step; For the first The time-dependent thermal load vector.

[0032] The meso-scale model refines the local key areas including the gate vicinity, thin-walled structure, hot spot area and the surrounding cooling channel on the basis of the macro-scale model. At the interface between the mold and the cooling medium, the convective heat transfer boundary condition is introduced: Wherein: is the heat flux density; is the derivative of temperature along the outer normal of the wall surface; denotes the value at the wall surface; is the convective heat transfer coefficient; is the wall surface temperature; is the cooling fluid temperature.

[0033] The convective heat transfer coefficient is determined by the Nusselt number :

[0034] Wherein: is the Nusselt number; is the thermal conductivity of the fluid; is the hydraulic diameter.

[0035] The calculation of the Nusselt number needs to be determined according to the flow state. The Reynolds number is defined as: Wherein: is the Reynolds number; is the fluid density; is the flow velocity; is the dynamic viscosity.

[0036] When , the flow is in a laminar state, and the Nusselt number adopts the Graetz solution: Wherein: is the pipe diameter; is the pipe length; is the Prandtl number; is the critical Reynolds number for the transition from laminar flow to turbulent flow, which is determined according to the pipe roughness and flow conditions, and is generally taken as 2300; the index denotes the cube of two-thirds.

[0037] When , the flow is in a fully developed turbulent state, and the Dittus-Boelter correlation is adopted:

[0038] Wherein: The critical Reynolds number for the transition from transitional flow to turbulent flow is typically taken as 10,000; the exponent 0.8 represents the power of 0.8. The temperature correction index is 0.4 when the fluid is heated and 0.3 when it is cooled.

[0039] Transition area Linear interpolation is used: in: For laminar flow, the Nusselt number is used. denoted as the Nusselt number for turbulent conditions.

[0040] The microscale model considers the influence of the material's microstructure on its thermal properties. The temperature dependence of the thermal conductivity is expressed as: in: For temperature Thermal conductivity at that time; Reference temperature Thermal conductivity at the specified value; The first temperature coefficient of thermal conductivity; It is the second temperature coefficient of thermal conductivity; The coefficient represents the square of the temperature difference; the coefficient is determined by material thermophysical property experiments or obtained from a material database.

[0041] The relationship between specific heat capacity and temperature is as follows: in: For temperature Specific heat capacity at time; Specific heat capacity at reference temperature; This is the primary temperature coefficient of specific heat capacity; It is the second temperature coefficient of specific heat capacity.

[0042] The temperature dependence of density is described by the coefficient of thermal expansion:

[0043] in: For temperature Density at time; Density at reference temperature; is the coefficient of volumetric thermal expansion.

[0044] Two-way information transfer between the three scales is achieved through a scale bridging algorithm. Downlink transfer from macroscopic to mesoscopic scales employs boundary condition mapping.

[0045] in: Temperature for the mesoscopic model; Indication of the meso model boundary Upper bound; is the th interpolation basis function; is the temperature of the th node of the macro model; is the number of nodes used for interpolation.

[0046] The meso-to-micro down-pass is established by volume averaging: where: is the volume averaged temperature; is the volume of the representative volume element; is the integral over the representative volume element; is the temperature distribution on the micro scale; is the volume element.

[0047] The micro-to-meso up-pass is realized by a homogenization approach: where: is the equivalent thermal conductivity; is the local thermal conductivity on the micro scale in coordinates and temperature .

[0048] The meso-to-macro up-pass is realized by a submodel contribution superposition: where: is the load vector of the macro model; is the initial load vector; is the meso submodel number; is the number of meso submodels; is the th submodel projection matrix; is the transpose of the projection matrix; is the load contribution of the th meso submodel.

[0049] To improve computational efficiency, an adaptive mesh technique is employed. The mesh adjustment is based on the temperature gradient norm: where: is the Euclidean norm of the temperature gradient; is the partial derivative of the temperature with respect to coordinate; is the partial derivative of the temperature with respect to coordinate; is the partial derivative of the temperature with respect to Partial derivative of coordinates; For square root operation.

[0050] The grid size adjustment strategy is: Where: is the adjusted grid size; is the current grid size; is the refinement / coarsening scale factor, typically 2; is the refinement threshold; is the coarsening threshold; the threshold is determined by numerical experiments.

[0051] The grid quality is monitored by the distortion index:

[0052] Where: is the grid quality index; is the actual cell volume; is the ideal cell volume.

[0053] When trigger local grid reconstruction, where is the critical distortion, typically 0.1.

[0054] Step S2: Deploy continuous temperature field sensing network Based on the digital model framework constructed in step S1, the continuous monitoring of the physical mold temperature field is realized by deploying distributed optical fiber temperature sensors. The distributed optical fiber sensing is based on the principle of Raman scattering. When laser pulses propagate in the optical fiber, the scattered light produced by the interaction with the optical fiber molecules carries temperature information.

[0055] The relationship between temperature and scattered light intensity is:

[0056] Where: is the absolute temperature; is the Planck constant; is the Raman shift; is the Boltzmann constant; is the natural logarithm; is the system calibration constant; is the Stokes light intensity; is the anti-Stokes light intensity.

[0057] The spatial positioning of the measurement points is determined by the optical time domain reflection technology:

[0058] Where: is the position of the measurement point from the starting end of the optical fiber; is the vacuum light speed; For the round-trip time of laser pulse; For the effective refractive index of optical fiber; factor 2 represents the round-trip path of light pulse.

[0059] The spatial resolution is determined by the pulse width:

[0060] Wherein: For the spatial resolution; For the laser pulse width.

[0061] The original temperature signal needs to be filtered to improve the signal-to-noise ratio, and a Butterworth low-pass filter is used: Wherein: For the filter transfer function; For the Laplace variable; For the cut-off angular frequency; For the filter order; Indicates The power of .

[0062] The filtered signal is subjected to outlier detection, and the judgment criterion is: Wherein: For the temperature of the th measurement point; For the local average temperature; For the absolute value operation; For the detection coefficient, which is set according to the noise characteristics, generally taken 3; For the local standard deviation.

[0063] When an outlier is detected, it is corrected by interpolation of adjacent points: Wherein: For the corrected temperature of the th point; For the temperature of the th point; For the temperature of the th point; For the weight coefficient of the th point; For the weight coefficient of the th point, satisfying .

[0064] The weight coefficient is determined according to the spatial distance: Wherein: For the temperature of the Point and the Spatial distance of points; For the Point and the Spatial distance of points.

[0065] In dual-redundancy configuration, the judgment of master-slave path switching is based on statistical test: Wherein: Switching criterion; Average temperature of master path; Average temperature of standby path; Standard deviation of master path temperature; Standard deviation of standby path temperature; Square root operation.

[0066] When Trigger switching, wherein The preset switching threshold value is set according to the system reliability requirement, generally in the range of 2 to 3. The switching process adopts a smooth transition function to avoid data mutation: Wherein: Temperature at Instantaneous switching time; Master path temperature at Instantaneous; Standby path temperature at Instantaneous; Transition function, from 0 to 1 smoothly within the switching time window.

[0067] The transition function adopts sigmoid form: Wherein: Exponential function; Transition rate parameter; Midpoint of switching time window.

[0068] The layout of optical fiber adopts spiral winding, and the spiral line parameter equation is: Wherein: , , Cartesian coordinates at angle On the spiral line; Spiral radius; Cosine function; Sine function; Pitch; Pi.

[0069] Step S3: Establishing physical-digital real-time synchronization mapping mechanism Based on the digital model constructed in step S1 and the perception data obtained in step S2, the real-time synchronization of the physical space and the digital space is realized through a data assimilation algorithm. The data assimilation adopts the ensemble Kalman filter method, which can effectively handle nonlinear systems and quantify uncertainties.

[0070] The state space model is represented as:

[0071] Wherein: is the state vector at the time, containing the temperature values of all grid nodes; is the state vector at the time; is the model operator, i.e. the multi-scale heat conduction model in step S1; is the model error at the time; is the observation vector at the time, from the fiber temperature measurement data in step S2; is the observation operator, mapping the state space to the observation space; is the observation error at the time.

[0072] The state uncertainty is represented by ensemble members: Wherein: is the predicted value of the ensemble member based on the information at the time at the time; is the analysis value of the ensemble member at the time; is the model disturbance of the member at the time; is the number of ensemble members, determined according to the computing resources and accuracy requirements, generally taken as 50 to 100.

[0073] The model disturbance is generated according to the statistical characteristics of the model error: Wherein: represents a multivariate normal distribution with mean 0 and covariance matrix is the time. Model error covariance matrix at time k.

[0074] The prediction error covariance is estimated by ensemble statistics: where: is the prediction error covariance matrix at time k based on information up to time k-1; is the prediction error covariance matrix at time k based on information up to time k-1; is the ensemble mean of the prediction; is the matrix transpose.

[0075] The ensemble mean is computed as: where: is the arithmetic mean of the predicted values of all ensemble members.

[0076] The Kalman gain matrix is computed as: where: is the Kalman gain matrix at time k; is the cross covariance matrix between the state and the observation; is the observation prediction covariance matrix; is the observation error covariance matrix; is the matrix inverse. The cross covariance is computed as:

[0077] where: is the predicted state of the kth ensemble member; is the ensemble mean of the state prediction; is the observation prediction of the kth member by mapping through the observation operator; is the ensemble mean of the observation prediction. The observation prediction covariance is computed as:

[0078] The update step corrects each ensemble member by the Kalman gain: where: is the updated analysis value of the kth member; is the observation disturbance of the kth member, satisfying ​​​​​​​

[0079] The anomaly detection mechanism is implemented using Mahalanobis distance: in: The Mahalanobis distance; The new information covariance matrix; It is the inverse of the new information covariance matrix.

[0080] when Online correction of model parameters is triggered at specific times, where For degrees of freedom Confidence level is The critical value of the chi-square distribution; Let be the dimension of the observation vector; It is usually taken as 0.05 or 0.01.

[0081] Online calibration of model parameters uses the gradient descent method. in: For the first The model parameter vector at time step; For the first Parameters updated in real time; The learning rate is set according to the required convergence speed. For loss function For parameters The gradient.

[0082] The loss function is defined as follows: in: The length of the time window; For the time index within the time window; For the first The observation vector at time; For parameters Next State estimation at time; It is the inverse of the observation error covariance matrix.

[0083] Step S4: Implement predictive control based on digital twins Using the real-time synchronous digital twin established in step S3, model predictive control is implemented to achieve precise regulation of the temperature field. The predictive control simulates the future evolution of the temperature field based on the digital twin, and obtains the optimal control strategy by solving an optimization problem.

[0084] The optimization problem of model predictive control is formulated as follows: in: is a target function for model predictive control; is a control sequence; is a time index within a prediction horizon; is a prediction horizon length; is a control horizon length; is a first time instant information prediction; is a first time instant temperature; is a reference temperature for the first denotes a weighted two-norm square with as weight matrix; is a control increment for the first time instant; denotes a weighted two-norm square with as weight matrix; is a relaxation factor; is a relaxation variable.

[0085] The constraints include temperature constraints: wherein: is a lower temperature limit; is an upper temperature limit; is an all-ones vector; denotes an element-wise comparison of vectors.

[0086] Control input constraints: wherein: is a lower control input limit; is an upper control input limit; is a control input for the first time instant.

[0087] Control increment constraints: wherein: is a maximum allowed value for the control increment.

[0088] The control increment is defined as: wherein: is a control input for the first time instant.

[0089] The prediction model is based on the digital twin of step S3: wherein: a multi-step prediction function for the digital twin; a temperature field estimation at the time instant.

[0090] The prediction function is implemented by a recursive computation of the digital twin: where: is a temperature prediction at the time instant based on the time instant information; is a temperature prediction at the time instant based on the time instant information; is the prediction time step; is a temperature field evolution function; is a control input at the time instant.

[0091] In the hierarchical control architecture, the optimization objective of the global coordination layer is: where: is the global layer objective function; is the temperature control zone index; is the total number of temperature control zones; is the average temperature of the zone; is the target temperature; denotes the square of the Euclidean norm; is the temperature uniformity weight coefficient; is the average temperature of all zones; is the square of the deviation of the zone temperature from the average temperature; is the energy consumption weight coefficient; is the total energy consumption.

[0092] The average temperature is computed as: The local execution layer employs a PID control to track the global layer setpoint: where: is the control quantity of the zone at the time instant; is the proportional gain of the zone; is the tracking error of the zone at the time instant; is the integral gain of the The integral gain of the region; To track error from 0 to Integral at time step; For integration variables; For the first Differential gain of the region; To track the derivative of the error with respect to time.

[0093] Tracking error is defined as: in: For the global layer, the first The area is Temperature setpoint at any time; For the first The area is The actual temperature at any given moment.

[0094] The global layer and the local layer are coupled through a coordination factor: in: For the final control output; As a coordinating factor; For global layer control variables; This is a local layer control variable.

[0095] The coordination factor is adaptively adjusted based on the control deviation: in: It is an exponential function; To adjust the parameters and control the slope of the sigmoid function; This represents the global control deviation. For local control deviation; This represents the Euclidean norm of a vector.

[0096] The optimization problem is solved using a sequential quadratic programming algorithm, which linearizes the nonlinear optimization problem at the current point: in: To control the increment vector; The Hessian matrix; To control the transpose of the increment vector; The gradient vector; This is the transpose of the gradient vector.

[0097] The Hessian matrix is ​​updated using the BFGS quasi-Newton method: in: For the first The Hessian matrix of the next iteration; For the first The Hessian matrix of the next iteration; Let be the step size vector of the decision variables; The gradient difference vector; For the first The gradient of the objective function in the next iteration; For the first The gradient of the objective function in the next iteration.

[0098] According to the rolling time domain principle, only the first control action obtained from the optimization is executed, and the optimization problem is solved again in the next time step.

[0099] Step S5: Achieve augmented reality visualization of the temperature field Based on the temperature field data of the digital twin in step S4, three-dimensional visualization and human-computer interaction are achieved through augmented reality technology.

[0100] The volume rendering of the temperature field uses a ray casting algorithm: in: screen coordinates Pixel intensity at that location; The length of the light ray; For light parameters; For position The opacity function at the location; For position Temperature The corresponding color value; It is an exponential function; From 0 to The accumulation of opacity; For integration variables; It is a light element.

[0101] The opacity function is adaptively adjusted according to the temperature gradient: in: Based on opacity; For visualization of adjustment coefficients; For position Temperature gradient norm at that location; This represents the maximum value of the temperature gradient.

[0102] The mapping from temperature to color uses a piecewise linear function: in: For temperature The corresponding RGB color value; is the blue RGB value; is the cyan RGB value; is the green RGB value; is the yellow RGB value; is the red RGB value; is the lowest temperature; is the highest temperature; , , is the temperature segmentation point, satisfying .

[0103] Augmented reality overlay requires precise spatial registration, which is determined by marker recognition to determine the pose of the mold in the camera coordinate system: where: is the 4x4 homogeneous transformation matrix representing the transformation from the mold coordinate system to the camera coordinate system; is the 3x3 rotation matrix; is the 3x1 translation vector; is the transpose of the 1x3 zero vector.

[0104] The projection of the virtual temperature field to the screen is realized by the camera intrinsic parameters: where: is the scale factor; is the screen pixel coordinate; is the three-dimensional point coordinate in the mold coordinate system; is the 3x3 camera intrinsic matrix.

[0105] The camera intrinsic matrix is defined as: where: is the focal length in the x direction; is the focal length in the y direction; is the principal point coordinate.

[0106] Gesture recognition obtains hand three-dimensional information through a depth camera, and uses a convolutional neural network for classification: where: is the predicted gesture type; represents the index that maximizes the probability ; is the conditional probability of the gesture type when given the depth image ; is the gesture type index.

[0107] The mapping relationship of gesture control parameters is: Wherein: is the parameter adjustment amount caused by the gesture; is the gesture control gain; is the gesture movement amplitude; is the gesture direction unit vector.

[0108] The voice instruction is realized through keyword recognition, and after the voice signal is converted into a mel-frequency cepstrum coefficient feature, it is matched with a predefined instruction template. The matching degree is calculated as: Wherein: is the matching degree score; is the extracted MFCC feature vector; is the template feature vector; is the transpose of the MFCC feature vector; represents the Euclidean norm of the vector.

[0109] When , the corresponding control action is triggered, wherein is the matching threshold.

[0110] Touch interaction realizes accurate parameter setting through a graphical user interface: Wherein: is the new parameter value; is the slider setting value; is the input box numerical value; is the original parameter value; is the button adjustment step.

[0111] Multi-user collaboration is realized through permission management and conflict resolution mechanisms: Wherein: is the final control value; is the primary control user setting value; is the secondary control user setting value; is the current value.

[0112] The timeline playback function reconstructs the temperature field at any time through interpolation of historical data: Wherein: is the temperature field at time playback; is the temperature field at the th historical time; is the temperature field at the The temperature field of a historical moment; For the first A timestamp of a historical moment; For the first A timestamp of a historical moment; For the moment to be played back, to satisfy .

[0113] Step S6: Perform autonomous learning optimization of the digital twin Based on the operational data accumulated in the aforementioned steps, reinforcement learning algorithms are used to continuously optimize the control strategy, giving the digital twin the ability to evolve autonomously.

[0114] The temperature control problem is modeled as a Markov decision process, and the state space is defined as follows: in: for The state vector at any given time; for Temperature field at any given moment; For the front The historical temperature field at a specific moment; For the front Historical control actions at any given moment; for Constant environmental disturbances; The length of the temperature history; To control the length of the history.

[0115] The range of motion refers to the adjustment amount of each temperature control zone: in: for Action vector at any given moment; for Time of the first Adjustment range for each temperature control zone; This represents the total number of temperature-controlled zones.

[0116] The reward function integrates multiple optimization objectives: in: for Instant rewards for each moment; for Time of the first The temperature of the area; For the first The area's reference temperature; The square of the temperature deviation; Temperature deviation weighting; is the energy consumption at time is the energy consumption weight; is the temperature variance at time is the uniformity weight; is the actual cycle time; is the target cycle time; is the cycle time over-standard penalty; is the cycle time weight. The policy network is parameterized by a deep neural network, which outputs a probability distribution over actions:

[0117] where: is the probability density of the action selected by the policy with parameters in state is a multivariate normal distribution; is the action mean vector output by the mean network; is the covariance matrix output by the covariance network. The value network estimates the expected return of a state: where:

[0118] is the value function with parameters represents the expectation under the policy is the future time index; is the discount factor, which is in the range of 0.95 to 0.99; is the reward at time is the state at time is the clipping function, which limits to the interval

[0119] The policy network is updated using the proximal policy optimization algorithm, whose objective function is: where: is the clipped objective function; represents the expectation over time is the minimum operation; is the importance sampling ratio; is the old policy; is the advantage function estimate; is the clipping function, which limits to the interval ​​​​​​​ This is the trimming parameter, with a value range of 0.1 to 0.2.

[0120] The advantage function is calculated using generalized advantage estimation: in: for Generalized advantage estimation at any given moment; Index for future steps; The moment when the trajectory terminates; This is a GAE parameter used to balance bias and variance, with a value of 0.95. for The timing difference error at each moment; express of Power of 1.

[0121] Timing difference error is defined as: in: for The timing difference error at each moment; for Value estimation of state at any given time; for Value estimation of the state at any given moment.

[0122] The loss function of the value network is: in: For the value function loss; The square of the difference between the predicted value and the target value; The target value is estimated using Monte Carlo returns.

[0123] The policy network parameters are updated as follows: in: For the first The policy network parameters for the next iteration; The learning rate of the policy network; For the trimming loss parameters The gradient.

[0124] The value network parameters are updated as follows: in: For the first Value network parameters for the next iteration; The learning rate of the value network; For the parameter of value loss The gradient.

[0125] The new strategy can only be deployed to the physical control system if both of the following conditions are met.

[0126] First, conditions for performance improvement: in: For strategy Performance metrics; For the new strategy; The original strategy; This represents the absolute value of the original strategy's performance metrics; The performance improvement threshold ranges from 0.05 to 0.1.

[0127] Second, conditions for improving stability: in: For strategy The performance variance reflects the control stability.

[0128] In summary, the mold temperature field collaborative control method based on digital twin provided by this invention forms a complete closed-loop system of "sensing-modeling-synchronization-control-interaction-optimization" through the organic coordination of six key steps.

[0129] Compared to existing technologies, this invention improves the accuracy of temperature field description through multi-scale modeling, eliminates monitoring blind spots through distributed sensing, ensures precise synchronization between virtual and real data through data assimilation, overcomes the lag of traditional feedback control through predictive control, enhances human-computer interaction efficiency through augmented reality, and endows the system with adaptive evolutionary capabilities through reinforcement learning. The method does not require large-scale modifications to existing equipment and can be widely applied to manufacturing fields requiring precise temperature control, such as injection molding, die casting, and extrusion, demonstrating good engineering practicality and application value.

[0130] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for coordinated control of mold temperature field based on digital twin, characterized in that, Includes the following steps: S1. Construct a multi-scale digital twin basic model of the mold: Establish a digital model of the temperature field from three spatial scales: macroscopic, mesoscopic and microscopic. The macroscopic scale describes the overall temperature distribution based on the three-dimensional unsteady heat conduction equation. The mesoscopic scale introduces convective heat transfer boundary conditions in key areas. The microscopic scale establishes the temperature dependence of material thermal property parameters. The information transfer between the three scales is realized through the scale bridging algorithm. S2. Deploy a continuous temperature field sensing network: Deploy distributed fiber optic temperature sensors to form a continuous sensing network, measure temperature based on the Raman scattering principle, determine the measurement point location through optical time-domain reflectometry, and filter and correct outliers in the original signal. S3. Establish a physical-digital real-time synchronization mapping mechanism: Use the ensemble Kalman filter algorithm to establish a physical-digital real-time synchronization mapping, represent uncertainty through multiple ensemble members, fuse the sensed data into the digital twin model, and introduce anomaly detection and online correction of model parameters based on Mahalanobis distance. S4. Implement predictive control based on digital twins: Implement model predictive control based on synchronized digital twins, obtain control sequences by solving multi-objective optimization problems, and adopt a hierarchical architecture to achieve coupling between global coordination and local execution. S5. Achieve augmented reality visualization of temperature field: Achieve three-dimensional visualization of temperature field through light projection algorithm, use augmented reality technology for overlay display, and support multimodal human-computer interaction with gestures, voice and touch; S6. Perform autonomous learning optimization of digital twin: Use reinforcement learning algorithms to optimize the control strategy, model temperature control as a Markov decision process, and achieve autonomous learning through policy network and value network.

2. The method for coordinated control of mold temperature field based on digital twin as described in claim 1, characterized in that, The scale bridging algorithm includes: macroscopic to mesoscopic boundary condition mapping, mapping macroscopic temperature to mesoscopic boundaries through interpolation basis functions; mesoscopic to microscopic connection through volume averaging, performing integral averaging of temperature within representative volume cells; microscopic to mesoscopic realization through homogenization, converting microscopic thermal properties into equivalent parameters; and mesoscopic to macroscopic through submodel contribution superposition, transferring mesoscopic loads to the macroscopic model through a projection matrix.

3. The method for coordinated control of mold temperature field based on digital twin as described in claim 1, characterized in that, The ensemble Kalman filtering includes: predicting the state of each ensemble member using model operators and adding model perturbations to characterize uncertainty; estimating the prediction error covariance and cross covariance using ensemble statistics; calculating the Kalman gain matrix, whose value is equal to the product of the cross covariance and the inverse of the sum of the observation covariance and the observation error covariance; and updating each ensemble member based on the Kalman gain, with the correction amount being the difference between the observed value and the predicted observed value multiplied by the Kalman gain.

4. The method for coordinated control of mold temperature field based on digital twin as described in claim 1, characterized in that, The optimization problem of the model predictive control is as follows: the objective function includes a weighted quadratic term of the deviation between the predicted temperature and the reference temperature, a weighted quadratic term of the control increment, and a penalty term for the slack variables; the constraints include upper and lower temperature limits, control input constraints, and control increment constraints; the nonlinear problem is linearized at the current point by solving the problem using a sequential quadratic programming algorithm, and the Hessian matrix is ​​updated using a quasi-Newton method.

5. The method for coordinated control of mold temperature field based on digital twin according to claim 1 or 4, characterized in that, In the hierarchical architecture, the optimization objectives of the global coordination layer include the sum of squares of the temperature deviations between each region and the target temperature, the temperature uniformity index, and the total energy consumption; the local execution layer uses PID control to track the setpoint given by the global layer; and the coordination factor is adaptively adjusted using the sigmoid function based on the relative magnitude of the global control deviation and the local control deviation.

6. The method for coordinated control of mold temperature field based on digital twin according to claim 1, characterized in that, In the 3D visualization of the temperature field, the opacity function is adaptively adjusted according to the ratio of the temperature gradient norm to the maximum gradient; the temperature-to-color mapping adopts a piecewise linear function, and linear interpolation is used to generate gradient colors in different temperature ranges; the pose transformation matrix of the mold in the camera coordinate system is obtained through marker recognition, and the virtual temperature field is projected onto the screen coordinates.

7. The method for coordinated control of mold temperature field based on digital twin as described in claim 1, characterized in that, The reinforcement learning includes: a state space containing the current and historical temperature fields, historical control actions, and environmental disturbances; an action space containing the control adjustment quantities of each temperature control zone; a reward function that is a negative weighted sum of temperature deviation, energy consumption, temperature variance, and production cycle deviation; a near-end policy optimization algorithm that limits the policy update magnitude by pruning the importance sampling ratio; and an advantage function calculated through generalized advantage estimation, combined with temporal difference error and discount factor.

8. The method for coordinated control of mold temperature field based on digital twin according to claim 1, characterized in that, It also includes adaptive mesh adjustment: the mesh density requirement is determined based on the temperature gradient norm, and when the gradient exceeds the refinement threshold, the mesh size is reduced to one-half of the original refinement ratio, and when the gradient is below the coarsening threshold, the mesh size is expanded to a multiple of the original coarsening ratio; the mesh quality is monitored through the mesh distortion index, and local mesh reconstruction is triggered when the ratio of the actual cell volume to the ideal volume is less than the set value.

9. The method for coordinated control of mold temperature field based on digital twin according to claim 1, characterized in that, The distributed fiber optic temperature sensor adopts a dual-redundant configuration: the main path performs routine temperature measurement, while the backup path remains in a preheating state; the switching criterion is the ratio of the difference between the average temperatures of the main and backup paths to the square root of the sum of the squares of their standard deviations; the switching process uses a smooth transition function to smoothly transition from the main path to the backup path within the switching time window.

10. The method for coordinated control of mold temperature field based on digital twin according to claim 1, characterized in that, The multimodal human-computer interaction includes: gesture recognition, which acquires hand depth image sequences through a depth camera, uses a convolutional neural network to identify gesture types, and uses the product of gesture amplitude, direction, and gain as the control parameter adjustment amount; voice commands, which are implemented by extracting Mel-frequency cepstral coefficient features and matching them with predefined templates; touch interaction, which provides precise parameter settings through sliders, input boxes, and buttons; and multi-user collaboration, which is implemented through permission management, selecting the final control value based on the operation status of the master user and the slave user.

Images