Power electronic packaging device closed-loop control method based on digital twinning

Abstract

The application relates to the technical field of device control, in particular to a power electronic packaging device closed-loop control method based on digital twinning, which comprises the following steps: a multi-physical-field digital twinning model based on a reduced-order algorithm is established in an edge computing unit, the model comprises solidification, micro-fluid dynamics and solid-phase deformation sub-models; sensor observation data such as multi-temperature-zone temperatures, servo head displacement and actual output force in a processing period are collected and input into the model; internal state variables such as solidification state parameters and local stress distribution parameters of a physical medium are output in real time by model running through proxy calculation; variable-parameter thermal control instructions and servo force control instructions are generated based on the internal state variables and are respectively sent to a heating source and a servo head mechanism. Through real-time mapping of conventional observation data into deep internal state variables, dynamic tracking of physical phase evolution is realized.

Description

Technical Field

[0001] This invention relates to the field of equipment control technology, specifically a closed-loop control method for power electronic packaging equipment based on digital twins. Background Technology

[0002] For high-end packaging equipment for high-power power electronic components, the coordinated control of actuators is crucial to ensuring the processing quality of complex controlled objects. Taking novel interconnection processes (such as nano-silver sintering) in high-end packaging equipment as an example, the physical space in which this process takes place constitutes a complex controlled object with strong multivariable coupling, high nonlinearity, and time-varying characteristics. During the processing cycle, the internal state variables of this controlled object (involving organic phase transformation pyrolysis, microfluidic dynamics, and solid-state deformation) undergo drastic dynamic evolution with external stimuli, and these deep state variables are usually difficult to observe in real time, directly, and non-destructively using conventional physical sensors.

[0003] Existing methods typically lack effective estimation of the dynamic state of the controlled object, and mostly employ open-loop control strategies based on hard-coded static time axes, or rigid threshold triggering logic relying on single variables such as time / temperature. Specifically, existing lower-level controllers simply issue fixed-intensity step or constant execution commands directly to the underlying servo driver at specific time nodes based on pre-calibrated empirical parameters.

[0004] This open-loop / rigid feedforward control mechanism, which ignores the internal state evolution of the controlled object and lacks real-time feedback mapping, suffers from serious control lag or lead problems: because the actual dynamic evolution rate of the physical system is affected by environmental disturbances and batch differences in materials, the static control timing often deviates significantly from the actual physical state of the controlled object. For example, when the controller command leads the actual drying state of the controlled object (i.e., mechanical control is applied too early), it can cause the physical medium (sintering paste) to overflow beyond its limits (leading to short-circuit failure); while when the control command lags or blindly outputs a high-amplitude control quantity when the constraints change, it can cause stress concentration phenomena that exceed physical limits in local areas, leading to irreversible rigid damage or interface failure of the processed object (chip).

[0005] To address this, a closed-loop control method for power electronic packaging equipment based on digital twins is proposed. Summary of the Invention

[0006] The purpose of this invention is to provide a closed-loop control method for power electronic packaging equipment based on digital twins, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A closed-loop control method for power electronic packaging equipment based on digital twins includes:

[0009] The power electronic packaging equipment includes a hot pressing base, a servo pressing head mechanism, a multi-temperature zone heating source, and an edge computing unit.

[0010] A multiphysics digital twin model based on a reduction-order algorithm is established in the edge computing unit. The multiphysics digital twin model includes a solidification sub-model, a microfluidic dynamics sub-model, and a solid deformation sub-model.

[0011] The sensor observation data of the power electronic packaging equipment during the processing cycle is collected. The sensor observation data includes at least the temperature of multiple temperature zones, the displacement of the servo pressure head, and the actual output force of the servo pressure head. The sensor observation data is input into the multiphysics digital twin model. The multiphysics digital twin model is run, and the internal state variables are output in real time through proxy calculation. The internal state variables include the solidification state parameters of the physical medium and the local stress distribution parameters.

[0012] Based on the curing state parameters and local stress distribution parameters, a variable parameter thermal control command for the multi-temperature zone heating source and a servo force control command for the servo pressure head mechanism are generated; the variable parameter thermal control command is sent to the multi-temperature zone heating source, and the servo force control command is sent to the servo pressure head mechanism.

[0013] Preferably, the hot-pressing base is used to support the power electronic device to be packaged and has a built-in bottom temperature sensor; the servo pressure head mechanism is disposed above the hot-pressing base and includes a displacement sensor, a force sensor and a servo driver; the multi-temperature zone heating source includes independent thermal control loops distributed in the hot-pressing base and the servo pressure head mechanism; the edge computing unit stores the multiphysics digital twin model, and the edge computing unit establishes a bidirectional data link with the temperature sensor, the force sensor, the displacement sensor, the servo driver and each of the independent thermal control loops via a communication bus.

[0014] Preferably, the proxy computation process of the multiphysics digital twin model based on the order reduction algorithm is as follows: obtain a parameter sample set including multi-temperature zone temperature, servo head displacement and actual output force of servo head, input the unreduced multiphysics digital twin model to solve and splice the physical field time series node data to construct a multivariate snapshot matrix; decouple and reduce the dimension of the multivariate snapshot matrix, extract the global spatial basis matrix, and generate multiple local time evolution operators that map different physical evolution stages according to the phase transition time nodes of the physical medium;

[0015] Based on the sensor observation data, the current state coefficient vector in the latent space is output through the proxy calculation; the target evolution operator corresponding to the current phase is matched from each of the local time evolution operators, and the current state coefficient vector is updated by the target evolution operator and multiplied by the global space basis matrix to reconstruct and output the internal state variables.

[0016] Preferably, the curing sub-model receives the multi-temperature zone temperature to perform reaction kinetic calculations and outputs curing state parameters, and defines the critical gel threshold for the physical medium to undergo phase change based on the curing state parameters.

[0017] The microfluidic dynamics sub-model operates when the solidification state parameter is below the critical gel threshold, using the multi-temperature zone temperature and the servo head displacement as boundary conditions to calculate the flow field distribution characteristics of the physical medium.

[0018] The solid-state deformation sub-model operates when the solidification state parameters reach the critical gel threshold, obtains the flow field distribution characteristics as the initial geometric boundary, and calculates and reconstructs the local stress distribution parameters based on the multi-temperature zone temperature, the servo head displacement, and the actual output force of the servo head.

[0019] Preferably, the curing state parameter output process involves interpolating the temperatures of the multi-temperature zones to extract three-dimensional spatial temperature field data; combining autocatalysis and diffusion control kinetic parameters to extract the transient crosslinking reaction rate of each spatial grid node under the three-dimensional spatial temperature field data; performing time-domain integration on the transient crosslinking reaction rate to extract the three-dimensional curing degree field matrix as the curing state parameter.

[0020] Preferably, the calculation process of the local stress distribution parameter involves extracting the chemical shrinkage strain data corresponding to the three-dimensional curing degree field matrix and the thermal mismatch strain data corresponding to the three-dimensional spatial temperature field data, and superimposing them as intrinsic distortion excitation parameters; configuring the servo head displacement and the actual output force of the servo head as normal contact stiffness constraint parameters; and under the intrinsic distortion excitation parameters and the normal contact stiffness constraint parameters, iteratively calculating the stress tensor of each micro-node within the initial geometric boundary, and extracting the corresponding numerical sequence as the local stress distribution parameter.

[0021] Preferably, the variable parameter thermal control command generation process involves extracting the curing degree spatial variance of the curing state parameter in the three-dimensional spatial domain and calculating the peak value of the thermo-coupling stress in the local stress distribution parameter.

[0022] A thermodynamic cost function is constructed with the goal of minimizing the spatial variance of the curing degree and the constraint that the peak value of the thermo-coupling stress is less than the allowable stress of the pre-set encapsulation material. The thermodynamic cost function is solved by a lightweight optimization algorithm to calculate the dynamic temperature compensation amount allocated to each of the independent thermal control loops. The dynamic temperature compensation amount is superimposed on the reference temperature control curve and converted into a pulse width modulation signal sequence corresponding to each temperature zone, which is then configured as the variable parameter thermal control command.

[0023] Preferably, the servo force control command generation process involves extracting the maximum principal stress acting on the interface of the power electronic device to be packaged from the local stress distribution parameters, and calculating the transient volume shrinkage rate of the physical medium based on the time derivative of the solidification state parameters; converting the transient volume shrinkage rate into a pressure head displacement compensation feedforward for suppressing material shrinkage pores; simultaneously, obtaining the force control feedback adjustment amount limited by the principal stress threshold based on the error between the integral resultant force of the interface traction force and the target pressure; in the control algorithm, after converting the pressure head displacement compensation feedforward amount into an equivalent force using physical equivalent stiffness, fusing it with the force control feedback adjustment amount to calculate the target output torque of the servo driver, and encoding and formatting the target output torque into the servo force control command.

[0024] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0025] 1. This invention, by deploying a multiphysics digital twin model based on a reduction-order algorithm in an edge computing unit, can map readily available conventional sensor observation data (such as temperature, displacement, and output force in multiple temperature zones) in real time into deep and difficult-to-measure internal state variables (such as solidification state parameters and local stress distribution parameters). This method can dynamically reconstruct the evolution of phase transitions, flow, and deformation of physical media without adding complex internal probes, providing a reliable internal state basis for the closed-loop control of the equipment.

[0026] 2. Unlike traditional open-loop control that sets fixed time points, this application utilizes real-time solidification state parameters output by a digital twin model to define the critical gel threshold, and dynamically switches between fluid dynamics calculations and solid-phase deformation calculations based on this. This control strategy, driven by the actual physical phase of the controlled object, can adapt to the effects of material batch differences and environmental disturbances, mitigating the problem of physical medium overflow caused by premature control commands.

[0027] 3. In the process of generating instructions for the actuator, this application incorporates the local stress distribution, thermo-coupled stress peak value, and maximum principal stress calculated by the twin model into the control loop. By generating thermal control instructions with the allowable stress of the material as a constraint and generating force control instructions with the safe crush threshold as a feedback adjustment, this method avoids the equipment blindly outputting a high-amplitude constant mechanical control quantity when the material is in a sensitive state, thereby effectively suppressing the phenomenon of local stress concentration and reducing the probability of irreversible rigid damage or interface failure of the chip. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of a closed-loop control method for power electronic packaging equipment based on digital twins.

[0029] Figure 2 This is a schematic diagram of the multiphysics sub-model switching process of the present invention;

[0030] Figure 3 This is a schematic diagram of the servo force control command generation process of the present invention. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Please see Figure 1 This invention provides a closed-loop control method for power electronic packaging equipment based on digital twins, the technical solution of which is as follows:

[0033] A closed-loop control method for power electronic packaging equipment based on digital twins includes:

[0034] The power electronic packaging equipment includes a hot pressing base, a servo pressing head mechanism, a multi-temperature zone heating source, and an edge computing unit.

[0035] A multiphysics digital twin model based on a reduction-order algorithm is established in the edge computing unit. The multiphysics digital twin model includes a solidification sub-model, a microfluidic dynamics sub-model, and a solid deformation sub-model.

[0036] The sensor observation data of the power electronic packaging equipment during the processing cycle is collected. The sensor observation data includes at least the temperature of multiple temperature zones, the displacement of the servo pressure head, and the actual output force of the servo pressure head. The sensor observation data is input into the multiphysics digital twin model. The multiphysics digital twin model is run, and the internal state variables are output in real time through proxy calculation. The internal state variables include the solidification state parameters of the physical medium and the local stress distribution parameters.

[0037] Based on the curing state parameters and local stress distribution parameters, a variable parameter thermal control command for the multi-temperature zone heating source and a servo force control command for the servo pressure head mechanism are generated; the variable parameter thermal control command is sent to the multi-temperature zone heating source, and the servo force control command is sent to the servo pressure head mechanism.

[0038] Example 1:

[0039] The power electronic packaging equipment includes a hot pressing base, a servo pressing head mechanism, a multi-temperature zone heating source, and an edge computing unit.

[0040] The hot-pressing base is used to support the power electronic device to be packaged and has a built-in bottom temperature sensor; the servo pressure head mechanism is located above the hot-pressing base and includes a displacement sensor, a force sensor, and a servo driver; the multi-temperature zone heating source includes independent thermal control loops distributed in the hot-pressing base and the servo pressure head mechanism; the edge computing unit stores the multiphysics digital twin model, and the edge computing unit establishes a bidirectional data link with the temperature sensor, the force sensor, the displacement sensor, the servo driver, and each of the independent thermal control loops via a communication bus.

[0041] Specifically, the hardware framework of the power electronic packaging equipment is jointly constructed by the hot-pressing base and the servo pressure head mechanism; the hot-pressing base is used to support the power electronic device to be packaged; a thermally conductive blind hole array with a fixed depth dimension from the upper surface of the hot-pressing base is processed inside the hot-pressing base, and the bottom temperature sensor is encapsulated in the thermally conductive blind hole array with thermally conductive silicone grease; by establishing transient heat conduction calculation logic in the edge computing unit, the transient heat flux density is calculated by combining the real-time output power of the current multi-temperature zone heating source and the heat capacity of the base material, as shown in the formula:

[0042]

[0043] in, The transient heat flux density at the current moment, This represents the real-time output power of the current multi-temperature zone heating source. The equivalent heat capacity of the base material between the temperature sensor and the heating source; The derivative of the measured temperature value collected by the temperature sensor with time, i.e., the real-time heating rate; This represents the effective heat transfer cross-sectional area corresponding to a single-zone heating source.

[0044] Based on Fourier's law of thermal conduction, the actual boundary temperature at the bottom of the power electronic device to be packaged is calculated using the following formula:

[0045]

[0046] in, This refers to the actual boundary temperature at the bottom of the power electronic device to be packaged. This is the measured temperature value collected by the temperature sensor at the current moment. This represents the temperature gradient drop. This refers to a fixed depth dimension from the upper surface of the hot-pressing base. The thermal conductivity of the base material is given. This represents the real-time temperature gradient drop caused by heat conduction from the sensor location to the surface of the hot-pressing base under transient heat flux density;

[0047] The servo pressure head mechanism is disposed above the hot pressing base. The servo pressure head mechanism includes a servo driver, a force sensor, and a displacement sensor mechanically connected in a vertically downward direction. The servo driver converts rotational motion into linear motion through a rotary motor and a ball screw mechanism, driving the pressure head body below to press down. The force sensor is fitted to the force-bearing end face of the pressure head body, and collects the pressure electrical signal applied by the pressure head body to the power electronic device in real time and converts it into the actual output force of the servo pressure head. The displacement sensor obtains the displacement of the servo pressure head by reading the absolute coordinate change of the pressure head body relative to the hot pressing base.

[0048] The multi-temperature zone heating source includes independent thermal control loops distributed within the hot pressing base and the servo pressure head mechanism. These independent thermal control loops are physically arranged in a matrix grid topology from the center region to the edge region within the hot pressing base and the servo pressure head mechanism, and each independent thermal control loop is connected to an independent power regulation actuator. By mapping and binding the spatial geometric coordinates of each independent thermal control loop in the matrix grid topology to the lower-order grid spatial coordinates of the multiphysics digital twin model in the edge computing unit, the internal state variables output by the multiphysics digital twin model can correspond to specific independent thermal control loops.

[0049] The edge computing unit stores the multiphysics digital twin model, and establishes a bidirectional data link with the temperature sensor, the force sensor, the displacement sensor, the servo driver, and each independent thermal control loop via a communication bus. The communication bus is configured as a real-time industrial local area network bus with a distributed clock synchronization mechanism. The edge computing unit acts as the master station and broadcasts a global synchronization reference clock to each sensor and driver slave station. After receiving the multi-temperature zone temperature, the servo head displacement, and the actual output force of the servo head with the same global synchronization reference clock timestamp, the edge computing unit synchronously inputs the sensor observation data at the same timestamp into the multiphysics digital twin model.

[0050] The edge computing unit sends the variable parameter thermal control command to the power adjustment execution terminal of each independent thermal control loop with a corresponding coordinate mapping relationship through the bidirectional data link to adjust the heating power; simultaneously, it sends the servo force control command to the servo driver to adjust the driving torque, thereby completing the uninterrupted synchronous acquisition of physical space data and the closed-loop control of the digital twin model within the processing cycle of the power electronic packaging equipment.

[0051] The proxy computation process of the multiphysics digital twin model based on the order reduction algorithm is as follows: obtain a parameter sample set containing multi-temperature zone temperature, servo head displacement, and actual output force of the servo head; input the unreduced multiphysics digital twin model to solve and stitch together the physical field time series node data to construct a multivariate snapshot matrix; decouple and reduce the dimension of the multivariate snapshot matrix, extract the global spatial basis matrix, and generate multiple local time evolution operators that map different physical evolution stages based on the phase transition time nodes of the physical medium.

[0052] Based on the sensor observation data, the current state coefficient vector in the latent space is output through the proxy calculation; the target evolution operator corresponding to the current phase is matched from each of the local time evolution operators, and the current state coefficient vector is updated by the target evolution operator and multiplied by the global space basis matrix to reconstruct and output the internal state variables.

[0053] Specifically, the manufacturer's specifications and specific packaging process documents of the power electronic packaging equipment are read to obtain the heating limit range of the multi-temperature zone, the maximum stroke range of the servo pressure head displacement, and the rated force range of the actual output force of the servo pressure head. Within the above range, uniform random sampling is performed using the Latin hypercube sampling algorithm to generate a parameter sample set composed of multiple sets of different operating condition parameters, so that the marginal ranges of each parameter dimension are covered in layers.

[0054] Each set of working condition parameters in the parameter sample set is input as boundary conditions into the unreduced multiphysics digital twin model constructed based on the finite element numerical solution framework; transient multiphysics coupling solution is performed with a fixed time step of the encapsulation period, and the solidification state parameters and local stress distribution parameters of all grid nodes in the computational domain of the model at each time step are extracted; the data of all grid nodes at the same time are flattened into a one-dimensional column vector according to the node coordinate number order, as the physical field time-series node data; the one-dimensional column vectors extracted under all time steps and all sample working conditions are concatenated column by column with the sample order according to the time sequence to construct the two-dimensional multivariate snapshot matrix.

[0055] Singular value decomposition (SVD) is performed on the two-dimensional multivariate snapshot matrix to obtain a left singular matrix, a diagonal matrix containing singular values, and a right singular matrix. All singular values ​​are extracted from the diagonal matrix, and the sum of squares of all singular values ​​is calculated as the total system energy. The maximum allowable temperature deviation and maximum allowable curing stress deviation of the physical medium are obtained from the packaging process specifications. The deviations of each physical field are normalized by division using dimensionless eigenvalues, and the normalized data are weighted and superimposed to serve as the accuracy tolerance standard for the digital twin model. The singular values ​​are accumulated one by one in descending order. When increasing the number of accumulated terms, an approximate multivariable snapshot matrix is ​​reconstructed using the first few columns of the left singular matrix, singular values, and right singular matrix corresponding to the current number of accumulated terms. The maximum absolute residual between the approximate multivariable snapshot matrix and the original multivariable snapshot matrix is ​​calculated. When the maximum absolute residual is less than the calculation accuracy tolerance standard for the first time, the ratio of the sum of squared singular values ​​accumulated at this time to the total energy of the system is set as the energy truncation threshold, and the current number of accumulated terms is recorded as the truncation order. All column vectors corresponding to the first truncation order in the left singular matrix are extracted and combined to form the global space basis matrix.

[0056] Extract the time-series curing degree data of the physical medium during the heating and pressurization process from the multivariate snapshot matrix; take the time when the curvature of the curvature curve reaches the gel point as the first phase transition time node, and take the time when the curing degree reaches the glass transition point as the second phase transition time node; take the first phase transition time node and the second phase transition time node as the time boundary line, and divide the entire power electronic packaging process into three physical evolution stages: viscous flow stage, rubber stage and glass stage.

[0057] Within each physical evolution stage, the corresponding physical field time-series node data is projected onto the global spatial basis matrix to obtain a dimension-reduced temporal feature coefficient matrix. The temporal feature coefficient matrix is ​​then split according to the time step to construct a previous time-step state sub-matrix containing data from time 1 to N-1, and a next time-step state sub-matrix containing data from time 2 to N. The generalized inverse matrix of the previous time-step state sub-matrix is ​​solved using the least squares method based on singular value decomposition. The next time-step state sub-matrix is ​​then multiplied with the generalized inverse matrix to calculate a linear mapping transition matrix representing the recursive relationship between adjacent time-step states. The linear mapping transition matrix corresponding to each evolution stage is used as the local time evolution operator for the corresponding stage.

[0058] A radial basis function neural network (RBFNN) is constructed as a proxy computation model. The number of centers, width, and training convergence criteria of the network are configured, and an independent test set is introduced to verify the field error and extrapolation boundary. Discrete sensor observation data are used as input features, and the time-series feature coefficient vectors obtained by solving the unreduced model at the corresponding time are used as the target output. The RBFNN is trained under supervision to establish a nonlinear mapping topology from macroscopic sensor data to low-dimensional latent space coefficients. In the online operation phase, the sensor observation data collected at the current time, including multi-temperature zone temperature, servo head displacement, and actual output force of the servo head, are input into the RBFNN after training and convergence. Through forward propagation calculation, the current time-time state coefficient vector under the latent space is mapped and output.

[0059] The equivalent degree of solidification of the physical medium at the current moment is calculated based on the real-time time integral of the multi-temperature zone temperature. The gel point and glass transition point are compared to determine the physical evolution stage of the current solidification. From the local time evolution operators corresponding to the viscous flow stage, rubbery stage, and glassy stage, the local time evolution operator consistent with the current physical evolution stage is extracted as the target evolution operator. The current state coefficient vector is multiplied by the target evolution operator to calculate the predicted state coefficient vector for the next time step. The predicted state coefficient vector is multiplied by the global spatial basis matrix to achieve a tensor inner product. The result of this inner product operation is the data back mapped to the original high-dimensional grid space nodes. The solidification state parameters and local stress distribution parameters of the physical medium in the entire twin model computational domain at the next moment are reconstructed and output as the internal state variables.

[0060] By performing singular value decomposition on the multivariate snapshot matrix to extract the global spatial basis matrix, the solution process of the high-dimensional physical field space is transformed into algebraic operations in the low-dimensional latent space. This eliminates the need to solve partial differential equations for all grid nodes during online model execution, thus reducing the computational load of the digital twin model from an algorithmic perspective. At the same time, by employing a neural network surrogate model and a local time evolution operator, the internal state variables can be reconstructed by performing only forward propagation and matrix multiplication operations of the network during the online computation phase, avoiding the complex nonlinear iterative process in traditional multiphysics numerical simulations.

[0061] The multiphysics digital twin model includes a solidification sub-model, a micro-fluid dynamics sub-model, and a solid deformation sub-model.

[0062] See Figure 2 The system collects sensor observation data of the power electronic packaging equipment during the processing cycle. The sensor observation data includes at least the temperature of multiple temperature zones, the displacement of the servo pressure head, and the actual output force of the servo pressure head. The sensor observation data is input into the multiphysics digital twin model. The multiphysics digital twin model is run, and the internal state variables are output in real time through proxy calculation. The internal state variables include the solidification state parameters of the physical medium and the local stress distribution parameters.

[0063] The curing sub-model receives the temperatures of the multi-temperature zones, performs reaction kinetic calculations, outputs curing state parameters, and defines the critical gel threshold for the physical medium to undergo a phase change based on the curing state parameters.

[0064] The microfluidic dynamics sub-model operates when the solidification state parameter is below the critical gel threshold, using the multi-temperature zone temperature and the servo head displacement as boundary conditions to calculate the flow field distribution characteristics of the physical medium.

[0065] The solid-state deformation sub-model operates when the solidification state parameters reach the critical gel threshold, obtains the flow field distribution characteristics as the initial geometric boundary, and calculates and reconstructs the local stress distribution parameters based on the multi-temperature zone temperature, the servo head displacement, and the actual output force of the servo head.

[0066] The curing state parameter output process involves interpolating the temperatures of the multiple temperature zones to extract three-dimensional spatial temperature field data; combining autocatalysis and diffusion control kinetic parameters to extract the transient crosslinking reaction rate of each spatial grid node under the three-dimensional spatial temperature field data; performing time-domain integration on the transient crosslinking reaction rate to extract the three-dimensional curing degree field matrix as the curing state parameter.

[0067] The calculation process for the local stress distribution parameters involves extracting the chemical shrinkage strain data corresponding to the three-dimensional curing degree field matrix and the thermal mismatch strain data corresponding to the three-dimensional spatial temperature field data, and superimposing them as intrinsic distortion excitation parameters; configuring the servo head displacement and the actual output force of the servo head as normal contact stiffness constraint parameters; and under the intrinsic distortion excitation parameters and the normal contact stiffness constraint parameters, iteratively calculating the stress tensor of each micro-node within the initial geometric boundary, and extracting the corresponding numerical sequence as the local stress distribution parameters.

[0068] Specifically, the solidification sub-model acquires the multi-temperature zone temperatures of the discretely distributed temperature sensors of the hot-pressing base and the multi-temperature zone heating source, as well as their spatial node coordinates in the digital twin coordinate system. For the multi-temperature zone temperatures, a spatial interpolation algorithm based on finite element shape functions is used to calculate the temperature values ​​of non-observed grid nodes in the twin computational domain according to distance attenuation weight mapping, extracting three-dimensional spatial temperature field data. In offline mode, differential scanning calorimetry is used to pre-test the encapsulated physical medium, and the parameters of the Carmel autocatalytic reaction model and the diffusion critical constant are fitted and obtained as autocatalytic and diffusion control kinetic parameters. The three-dimensional spatial temperature field data is substituted into the kinetic equation containing the autocatalytic and diffusion control kinetic parameters to calculate and extract the transient crosslinking reaction rate of each spatial grid node at the current moment. The specific formula is as follows:

[0069]

[0070] diffusion control factor The calculation formula is:

[0071]

[0072] in, At the current time step, the spatial grid coordinates are... The transient crosslinking reaction rate at the location; This refers to the three-dimensional spatial temperature field data at the spatial grid coordinates obtained by the interpolation algorithm at the current time step. This represents the current solidification degree accumulated at the previous time step for the spatial grid coordinates; It is the ideal gas constant; and These are the first and second pre-exponential factors (i.e., autocatalytic kinetic parameters) of the reaction system, respectively. and These are the activation energies of non-catalytic reactions and autocatalytic reactions (i.e., autocatalytic kinetic parameters), respectively. and These are the order of the autocatalytic reaction and the order of the non-catalytic reaction, respectively (i.e., autocatalytic kinetic parameters). is the diffusion constant (i.e., the diffusion control kinetic parameter); It is the diffusion critical solidification degree (i.e., diffusion critical constant) at which the physical medium transitions from chemical kinetic control to diffusion control.

[0073] The transient crosslinking reaction rate is subjected to Euler time-domain integration to extract the three-dimensional curing degree field matrix composed of the current cumulative curing degree of each spatial grid node as the curing state parameter; based on this, according to the intersection point of the changes in storage modulus and loss modulus with curing degree obtained by offline rheometer test, the curing degree value corresponding to the intersection point is defined as the critical gel threshold for the physical medium to undergo phase change.

[0074] When the node values ​​in the three-dimensional solidification field matrix are less than the critical gel threshold, the physical medium is determined to be in a viscous flow state. The micro-fluid dynamics sub-model is activated, and the servo head displacement is converted into the downward displacement velocity boundary of the dynamic mesh, and the multi-temperature zone temperature is converted into the viscosity-temperature phase transition boundary of the fluid. Under the above boundary conditions, the Navier-Stokes momentum conservation equation and the continuous mass conservation equation are solved jointly, and the fluid volume tracking method is used to capture the gas-liquid free surface of the physical medium in real time. The dynamic velocity vector and dynamic pressure scalar distribution of the fluid in the computational domain are calculated and output as the flow field distribution feature. When the node values ​​in the three-dimensional solidification field matrix reach the critical gel threshold, the physical medium is determined to have undergone cross-linking solidification phase transition and lost fluidity. The solution of the fluid equation is stopped and the solid deformation sub-model is activated. The free surface contour and filled area of ​​the last step of the fluid volume tracking method in the flow field distribution feature are extracted and directly solidified into the initial geometric boundary of the solid deformation sub-model.

[0075] When calculating and reconstructing the local stress distribution parameters, the difference between the three-dimensional curing degree field matrix of the current time step and the previous time step is extracted to calculate the curing degree increment. The curing degree increment is multiplied by the volume shrinkage coefficient of the physical medium to calculate the chemical shrinkage strain data. The temperature change of the three-dimensional spatial temperature field data relative to the reference ambient temperature is extracted. The temperature change is multiplied by the thermal expansion coefficient of the physical medium to calculate the thermal mismatch strain data. The chemical shrinkage strain data and the thermal mismatch strain data are tensor superimposed and configured as the intrinsic distortion excitation parameters to induce internal deformation.

[0076] The servo indenter displacement and the actual output force of the servo indenter are equivalently transformed using the penalty function method and configured as normal contact stiffness constraint parameters acting on the top surface nodes of the initial geometric boundary. Under the load vector containing the intrinsic distortion excitation parameters and the stiffness matrix containing the normal contact stiffness constraint parameters, a viscoelastic constitutive relation is assigned to each micro-node within the initial geometric boundary. The equilibrium equation is solved using the Newton-Raphson incremental iterative algorithm, and the stress tensor of each micro-node within the initial geometric boundary is iteratively calculated. The iteration stops when the difference between the stress tensors of two adjacent iterations meets the convergence criterion. The three-dimensional numerical sequence corresponding to the converged stress tensor of each micro-node is extracted and used as the output of the local stress distribution parameter.

[0077] The constructed multiphysics digital twin model provides mechanistic support for the internal physical state evolution of the power electronics packaging process. Through temporal segmentation and boundary transfer of the solidification sub-model, microfluidic dynamics sub-model, and solid deformation sub-model, the digital twin model can objectively characterize the physical behavior of the physical medium from the viscous flow state to the solid phase transition, avoiding the problem that a single physics model is difficult to characterize the cross-phase evolution process. At the same time, based on three-dimensional temperature field interpolation, reaction kinetic integration, and superposition calculation of chemical contraction and thermal mismatch strain, the digital twin model can deduce the microscopic flow velocity, dynamic pressure, and internal local stress distribution from the material physics mechanism level based on the macroscopic thermodynamic boundary.

[0078] Based on the curing state parameters and local stress distribution parameters, a variable parameter thermal control command for the multi-temperature zone heating source and a servo force control command for the servo pressure head mechanism are generated; the variable parameter thermal control command is sent to the multi-temperature zone heating source, and the servo force control command is sent to the servo pressure head mechanism.

[0079] The variable parameter thermal control command generation process involves extracting the curing degree spatial variance of the curing state parameters in the three-dimensional spatial domain and calculating the peak value of the thermo-coupling stress in the local stress distribution parameters.

[0080] A thermodynamic cost function is constructed with the goal of minimizing the spatial variance of the curing degree and the constraint that the peak value of the thermo-coupling stress is less than the allowable stress of the pre-set encapsulation material. The thermodynamic cost function is solved by a lightweight optimization algorithm to calculate the dynamic temperature compensation amount allocated to each of the independent thermal control loops. The dynamic temperature compensation amount is superimposed on the reference temperature control curve and converted into a pulse width modulation signal sequence corresponding to each temperature zone, which is then configured as the variable parameter thermal control command.

[0081] Specifically, the three-dimensional curing degree field matrix in the internal state variables is extracted as the curing state parameter, and the arithmetic mean of the curing degree values ​​of all spatial grid nodes in the three-dimensional curing degree field matrix is ​​calculated; the sum of squares of the differences between the curing degree value of each spatial grid node and the arithmetic mean is calculated and divided by the total number of spatial grid nodes to obtain the spatial variance; the spatial variance is used as the curing degree spatial variance of the curing state parameter in the three-dimensional spatial domain; at the same time, the three-dimensional numerical sequence corresponding to the stress tensor after convergence of each micro-node in the internal state variables is extracted as the local stress distribution parameter, and the largest first principal stress value is extracted as the thermo-coupling stress peak value in the local stress distribution parameter by traversing and comparing the first principal stress values ​​of each node in the three-dimensional numerical sequence;

[0082] In offline mode, standard samples of the encapsulated physical medium under different temperature gradients are acquired, and yield strength tests are performed using a material tensile testing machine. The ultimate stress test data at the point of yield failure is extracted, and the ultimate stress test data is configured as the preset allowable stress of the encapsulation material. A thermodynamic cost function is constructed, and the optimization objective of the thermodynamic cost function is set to minimize the solidification degree spatial variance. The constraint condition of the thermodynamic cost function is set to the peak value of the thermo-coupled stress being less than the preset allowable stress of the encapsulation material.

[0083] A sequential quadratic programming algorithm is used as the lightweight optimization algorithm to solve the thermodynamic cost function; the dynamic temperature compensation amount allocated to each of the independent thermal control loops is configured as the decision variable for solving; the Jacobian matrix of the curing degree space variance with respect to the decision variable and the approximate Hessian matrix of the Lagrangian function of the thermodynamic cost function are calculated; under the constraint of the preset allowable stress of the encapsulation material, the decision variable is iteratively updated along the search direction that reduces the curing degree space variance; when the difference between the decision variable in two adjacent iterations is less than the convergence tolerance and the constraint condition is met, the iterative calculation is stopped, and the converged decision variable is output as the final dynamic temperature compensation amount allocated to each of the independent thermal control loops; the standard mathematical format of the thermodynamic cost function and the sequential quadratic programming optimization problem is configured as follows:

[0084]

[0085] Constraints:

[0086]

[0087]

[0088] Specifically, It is the thermodynamic cost function; For decision variables, i.e., the dynamic temperature compensation vector allocated to each independent thermal control loop; This represents the total number of nodes in the spatial grid. Forward prediction in the time domain; To introduce temperature compensation in the forward prediction time domain, the th... Predicted solidification degree of each grid node; The global arithmetic mean prediction solidification degree in this prediction time domain; The predicted peak forward thermo-coupling stress after introducing temperature compensation; The allowable stress of the pre-set packaging material; and These are the lower and upper limits of the allowable temperature compensation, respectively.

[0089] The predicted degree of curing and the predicted peak thermo-coupling stress are obtained through a forward state equation: The decision variables are... Mapped to the model boundary conditions and combined with the state coefficients at the current moment, the corresponding phase's local time evolution operator is used in the prediction time domain within the reduced-dimensional latent space. The system performs forward evolutionary deduction and reconstructs the predicted field distribution through global spatial basis matrix reconstruction. For the calculation of the Jacobian matrix and the approximate Hessian matrix, the finite difference method is used to extract the model sensitivity online based on the aforementioned forward state equation. The control loop is also configured with execution boundaries: a maximum iteration upper limit is set for the sequential quadratic programming algorithm. If the iteration upper limit is reached during the optimization process without convergence, or if the predicted state violates the allowable stress constraint resulting in no feasible solution, a safety degradation strategy is triggered, i.e., the current dynamic temperature compensation is discarded, and the system is forced to revert to operation solely based on the baseline temperature control curve.

[0090] Obtain the time-temperature mapping spectrum provided by the physical media supplier for packaging, and discretize it to configure it as a reference temperature control curve; add the dynamic temperature compensation amount and the reference temperature value of the reference temperature control curve at the current moment as a scalar to obtain the target control temperature; calculate the temperature difference between the target control temperature and the current corresponding multi-temperature zone temperature, input the temperature difference into a proportional-integral-derivative (PID) controller, multiply it by the PID gain coefficient to calculate the duty cycle value; generate a pulse width modulation signal sequence to control the heating on / off time of each of the independent thermal control loops based on the duty cycle value, and configure it as the variable parameter thermal control command output.

[0091] By optimizing the temperature distribution within complex mold cavities by minimizing the spatial variance of curing degree in the three-dimensional domain, the synchronous cross-linking reaction of the encapsulation medium can be promoted, thereby improving the structural uniformity of the encapsulation. At the same time, by using the peak value of thermal coupling stress being less than the allowable stress of the material as a constraint condition for optimization, the local stress concentration caused by chemical volume shrinkage and thermal mismatch during the curing phase change can be actively limited, which helps to avoid potential encapsulation defects such as microcracks or interface delamination.

[0092] See Figure 3 The servo force control command generation process involves extracting the maximum principal stress acting on the interface of the power electronic device to be packaged from the local stress distribution parameters, and calculating the transient volume shrinkage rate of the physical medium based on the time derivative of the solidification state parameters; converting the transient volume shrinkage rate into a pressure head displacement compensation feedforward for suppressing material shrinkage pores; simultaneously, obtaining the force control feedback adjustment amount limited by the principal stress threshold based on the error between the integral resultant force of the interface traction force and the target pressure; in the control algorithm, after converting the pressure head displacement compensation feedforward amount into an equivalent force using physical equivalent stiffness, merging it with the force control feedback adjustment amount to calculate the target output torque of the servo driver, and encoding and formatting the target output torque into the servo force control command;

[0093] Specifically, the local stress distribution parameters output by the solid-state deformable submodel are read; a spatial geometric bounding box is established based on the physical dimensions of the power electronic device to be packaged, and the Cauchy stress tensor matrix of all micro-grid nodes falling within the coordinate system of the bottom surface of the spatial geometric bounding box is extracted; the eigenvalues ​​of the Cauchy stress tensor matrix of each extracted micro-grid node are solved to obtain the principal stress values ​​in three orthogonal directions; the principal stress values ​​of all micro-grid nodes are traversed, and the maximum eigenvalue is selected as the maximum principal stress acting on the interface of the power electronic device to be packaged.

[0094] Within the current time step of the agent calculation, the three-dimensional curing degree field matrix output by the multiphysics digital twin model is obtained as the curing state parameter; the curing degree of each spatial grid node in the three-dimensional curing degree field matrix at the previous moment and the curing degree at the current moment are calculated by first-order backward difference calculation to obtain the curing degree time derivative of each spatial grid node; the curing degree time derivative of each spatial grid node is volume integrally averaged to obtain the global average curing rate; the global average curing rate is multiplied by the intrinsic chemical shrinkage and expansion coefficient of the encapsulation material to calculate the transient volume shrinkage rate of the physical medium.

[0095] Obtain the transient thickness value of the physical medium directly below the servo indenter mechanism at the current moment; combine the lateral geometric constraints under thin-layer packaging with the material Poisson's ratio, and convert the transient volume shrinkage rate into an equivalent uniaxial shrinkage strain rate; multiply the equivalent uniaxial shrinkage strain rate, the transient thickness value, and the current time step calculated by the agent to calculate the absolute displacement compensation value used to offset the material volume shrinkage, and configure it as the indenter displacement compensation feedforward.

[0096] The system retrieves the ultimate compressive strength value of the bare die material of the power electronic device to be packaged from the locally stored material property database. It then multiplies the ultimate compressive strength value by a safety factor between 0 and 1 to calculate the preset interface safety crush threshold. Subtracting the maximum principal stress from the preset interface safety crush threshold yields the stress safety margin. The system performs area integration on the local normal traction force at the bottom contact interface of the enclosure box to obtain the current actual contact resultant force. The difference between the current actual contact resultant force and the set resultant force corresponding to the target process pressure is calculated to obtain the normal force error value. The stress safety margin is then used to limit the normal force error value. The limited normal force error value is input into the proportional-integral-derivative (PID) control algorithm model, and the force control feedback adjustment amount is calculated and output. At this point, the force control feedback adjustment amount is characterized by mechanical dimensions.

[0097] Extract the physical equivalent stiffness independently identified through offline mechanical tests; multiply the indenter displacement compensation feedforward by the physical equivalent stiffness, converting the displacement dimensions to obtain the feedforward compensation equivalent force; linearly superimpose this feedforward compensation equivalent force with the force control feedback adjustment to obtain the target output total driving force of the servo indenter; read the mechanical transmission parameters of the servo indenter mechanism, including the lead screw, reducer transmission ratio, and mechanical transmission efficiency; according to the formula:

[0098]

[0099] Perform torque calculation, where The target output torque, Output the total driving force to the target. The lead of the lead screw, The gear ratio of the reducer is given. The target output torque of the servo driver is calculated based on the mechanical transmission efficiency.

[0100] The calculated target output torque in floating-point format is multiplied by the torque resolution coefficient of the servo driver and converted into an integer torque control word. The integer torque control word is written into the register address data segment corresponding to the torque control object dictionary in the industrial field communication bus protocol. A cyclic redundancy check code is calculated and appended to the end of the data segment, and it is packaged and encapsulated into a standard low-level data frame message, which is configured as the servo force control instruction to be issued and executed.

[0101] By establishing a multiphysics digital twin model based on a reduced-order algorithm in the edge computing unit, this invention overcomes the bottleneck of long computation time in traditional multiphysics simulation and realizes real-time online solution of complex physical evolution processes. At the same time, the feedback mechanism of closed-loop control is transformed from traditional external apparent sensor data to characterize internal state variables estimated by the calibrated model (i.e., solidification state parameters and local stress distribution parameters), enabling the device to directly output variable parameter thermal control commands and servo force control commands based on the actual phase change temperature and stress state of the encapsulated physical medium.

[0102] Example 2:

[0103] This embodiment applies a digital twin-based closed-loop control method for power electronic packaging equipment to the nano-silver sintering interconnect packaging process of high-power insulated-gate bipolar transistor modules. The specific process includes:

[0104] Initialize and configure the packaging process and hardware parameters; set the object to be packaged as a high-power insulated gate bipolar transistor module and the physical medium as a cross-linkable thermosetting packaging medium; read the hardware properties of the hot-pressing base, wherein the depth dimension of the temperature sensor from the upper surface of the base is 2mm; divide the multi-temperature zone heating source into 9 independent thermal control loops: center, inner ring, and outer ring.

[0105] Within the workspace, a sample set containing different operating parameters is generated using the Latin hypercube sampling algorithm. This sample set is then input into an unreduced model with 50,000 microscopic spatial grid nodes in the computational domain for solution. A multivariate snapshot matrix is ​​constructed, and singular value decomposition is performed. The system energy cutoff threshold is set to 99.9% to extract the global spatial basis matrix. Based on discrete test curves, the critical gel threshold for phase transition is defined, the physical evolution stages are divided, and multiple corresponding local time evolution operators are generated.

[0106] During the processing cycle, the edge computing unit collects sensor data with a synchronous clock cycle of 1ms; during a certain control cycle, it collects the actual measurement value of the bottom temperature sensor, calculates the transient heat flux density by combining it with the current heating power of the heating source, and then calculates the actual boundary temperature of the bottom of the device based on Fourier's law; it also collects the actual output force of the servo pressure head synchronously, and uses the above synchronous data as the sensor observation data at the current moment.

[0107] A reduced-order digital twin model is run to perform proxy calculations and output internal state variables; the sensor observation data is input into a neural network, which maps and outputs the state coefficient vector at the current time; at this time, the average solidification degree of the physical medium in the previous time step is identified as being less than the critical gel threshold, and the medium is determined to be in a viscous flow state. The corresponding local time evolution operator is called to reconstruct the three-dimensional solidification degree field matrix and flow field distribution characteristics; as heating proceeds, when the calculation identifies that the solidification degree has reached the critical gel threshold, the current flow field distribution characteristics are frozen as the initial geometric boundary, the solid phase deformation sub-model is activated, and the local stress distribution parameters of each micro-node are iteratively calculated.

[0108] The system optimizes based on the thermodynamic cost function and generates variable parameter thermal control commands; it calculates the solidification spatial variance of the current three-dimensional spatial domain and extracts the maximum value of the first principal stress as the peak value of the thermo-coupling stress; it calls the material database to obtain the ultimate yield stress of sintered silver as the preset allowable stress of the encapsulation material; it uses a sequential quadratic programming algorithm to optimize under the constraint that the peak stress is strictly less than the allowable stress, and calculates the dynamic temperature compensation amount that the central thermal control loop needs to compensate; it then superimposes this compensation amount with the reference temperature, inputs it into the proportional-integral-derivative controller to calculate the duty cycle, and generates a pulse width modulation signal sequence to be sent to the corresponding thermal control loop.

[0109] A composite feedforward and feedback adjustment mechanism generates and issues servo force control commands; based on the curing time derivative of the digital twin output, the current transient volume shrinkage rate is calculated, and then the absolute displacement compensation value is calculated and configured as the feedforward amount for indenter displacement compensation; the ultimate compressive strength of the bare sheet is multiplied by a safety factor to obtain the interface safety crush threshold, and the current maximum principal stress is subtracted to obtain the stress safety margin, which is then converted into the corresponding force control feedback adjustment amount; the feedforward displacement is equivalent to a mechanical dimension using the normal contact stiffness, and linearly superimposed with the force control feedback adjustment amount, combined with the screw drive parameters to calculate the target output torque, which is then encoded and formatted into a low-level data frame message and sent to the servo driver for execution.

[0110] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A closed-loop control method for power electronic packaging equipment based on digital twins, characterized in that, include: The power electronic packaging equipment includes a hot pressing base, a servo pressing head mechanism, a multi-temperature zone heating source, and an edge computing unit. A multiphysics digital twin model based on a reduction-order algorithm is established in the edge computing unit. The multiphysics digital twin model includes a solidification sub-model, a microfluidic dynamics sub-model, and a solid deformation sub-model. The sensor observation data of the power electronic packaging equipment during the processing cycle is collected. The sensor observation data includes at least the temperature of multiple temperature zones, the displacement of the servo pressure head, and the actual output force of the servo pressure head. The sensor observation data is input into the multiphysics digital twin model. The multiphysics digital twin model is run, and the internal state variables are output in real time through proxy calculation. The internal state variables include the solidification state parameters of the physical medium and the local stress distribution parameters. Based on the curing state parameters and local stress distribution parameters, a variable parameter thermal control command for the multi-temperature zone heating source and a servo force control command for the servo pressure head mechanism are generated; the variable parameter thermal control command is sent to the multi-temperature zone heating source, and the servo force control command is sent to the servo pressure head mechanism.

2. The closed-loop control method for power electronic packaging equipment based on digital twins according to claim 1, characterized in that, The hot-pressing base is used to support the power electronic device to be packaged and has a built-in bottom temperature sensor; the servo pressure head mechanism is located above the hot-pressing base and includes a displacement sensor, a force sensor, and a servo driver; the multi-temperature zone heating source includes independent thermal control loops distributed in the hot-pressing base and the servo pressure head mechanism; the edge computing unit stores the multiphysics digital twin model, and the edge computing unit establishes a bidirectional data link with the temperature sensor, the force sensor, the displacement sensor, the servo driver, and each of the independent thermal control loops via a communication bus.

3. The closed-loop control method for power electronic packaging equipment based on digital twins according to claim 1, characterized in that, The proxy computation process of the multiphysics digital twin model based on the order reduction algorithm is as follows: obtain a parameter sample set containing multi-temperature zone temperature, servo head displacement, and actual output force of the servo head; input the unreduced multiphysics digital twin model to solve and stitch together the physical field time series node data to construct a multivariate snapshot matrix; decouple and reduce the dimension of the multivariate snapshot matrix, extract the global spatial basis matrix, and generate multiple local time evolution operators that map different physical evolution stages based on the phase transition time nodes of the physical medium. Based on the sensor observation data, the current state coefficient vector in the latent space is output through the proxy calculation; the target evolution operator corresponding to the current phase is matched from each of the local time evolution operators, and the current state coefficient vector is updated by the target evolution operator and multiplied by the global space basis matrix to reconstruct and output the internal state variables.

4. The closed-loop control method for power electronic packaging equipment based on digital twins according to claim 1, characterized in that, The curing sub-model receives the temperatures of the multi-temperature zones, performs reaction kinetic calculations, outputs curing state parameters, and defines the critical gel threshold for the physical medium to undergo a phase change based on the curing state parameters. The microfluidic dynamics sub-model operates when the solidification state parameter is below the critical gel threshold, using the multi-temperature zone temperature and the servo head displacement as boundary conditions to calculate the flow field distribution characteristics of the physical medium. The solid-state deformation sub-model operates when the solidification state parameters reach the critical gel threshold, obtains the flow field distribution characteristics as the initial geometric boundary, and calculates and reconstructs the local stress distribution parameters based on the multi-temperature zone temperature, the servo head displacement, and the actual output force of the servo head.

5. The closed-loop control method for power electronic packaging equipment based on digital twins according to claim 4, characterized in that, The curing state parameter output process involves interpolating the temperatures of the multiple temperature zones to extract three-dimensional spatial temperature field data; combining autocatalysis and diffusion control kinetic parameters to extract the transient crosslinking reaction rate of each spatial grid node under the three-dimensional spatial temperature field data; performing time-domain integration on the transient crosslinking reaction rate to extract the three-dimensional curing degree field matrix as the curing state parameter.

6. The closed-loop control method for power electronic packaging equipment based on digital twins according to claim 4, characterized in that, The calculation process for the local stress distribution parameters involves extracting the chemical shrinkage strain data corresponding to the three-dimensional curing degree field matrix and the thermal mismatch strain data corresponding to the three-dimensional spatial temperature field data, and superimposing them as intrinsic distortion excitation parameters; configuring the servo indenter displacement and the actual output force of the servo indenter as normal contact stiffness constraint parameters; and under the intrinsic distortion excitation parameters and the normal contact stiffness constraint parameters, iteratively calculating the stress tensor of each micro-node within the initial geometric boundary, and extracting the corresponding numerical sequence as the local stress distribution parameters.

7. The closed-loop control method for power electronic packaging equipment based on digital twins according to claim 1, characterized in that, The variable parameter thermal control command generation process involves extracting the curing degree spatial variance of the curing state parameters in the three-dimensional spatial domain and calculating the peak value of the thermo-coupling stress in the local stress distribution parameters. A thermodynamic cost function is constructed with the goal of minimizing the spatial variance of the curing degree and the constraint that the peak value of the thermo-coupling stress is less than the allowable stress of the pre-set encapsulation material. The thermodynamic cost function is solved by a lightweight optimization algorithm to calculate the dynamic temperature compensation amount allocated to each independent thermal control loop. The dynamic temperature compensation amount is superimposed with the reference temperature control curve and converted into a pulse width modulation signal sequence corresponding to each temperature zone, which is then configured as the variable parameter thermal control command.

8. The closed-loop control method for power electronic packaging equipment based on digital twins according to claim 1, characterized in that, The servo force control command generation process involves extracting the maximum principal stress acting on the interface of the power electronic device to be packaged from the local stress distribution parameters, and calculating the transient volume shrinkage rate of the physical medium based on the time derivative of the solidification state parameters. The transient volume shrinkage rate is converted into a head displacement compensation feedforward for suppressing material shrinkage pores; at the same time, the force control feedback adjustment amount limited by the principal stress threshold is obtained based on the error between the integral resultant force of the interface traction force and the target pressure. In the control algorithm, the pressure head displacement compensation feedforward is converted into an equivalent force using physical equivalent stiffness, and then fused with the force control feedback adjustment to calculate the target output torque of the servo driver. The target output torque is then encoded and formatted into the servo force control command.

Images