Autonomous driving method and system, and computer program product

Abstract

The application provides a autoclave process optimization design method and system based on an extended physical information neural operator, relates to the technical field of autoclave control, and comprises the following steps: collecting autoclave molding full-process multi-source data, constructing a standardized process data set through preprocessing, constructing a priori physical constraint system combined with a composite material curing multi-physical field coupling mechanism, and extracting a key feature set matching the constraint; constructing an extended physical information neural operator model embedded with the priori physical constraint, training a multi-field coupling proxy model of autoclave molding through the key feature set; based on multi-objective optimization technology, combining process parameter constraints, intelligently optimizing the optimal process parameter set based on the multi-field coupling proxy model, and converting the optimal process parameter set into an autoclave executable process curve output; and simultaneously completing the closed-loop self-correction of the multi-field coupling proxy model based on actual molding data and prediction deviation.

Description

Technical Field

[0001] This invention relates to the field of autoclave control technology, specifically to an autoclave process optimization design method and system based on extended physical information neural operators. Background Technology

[0002] The composite material autoclave molding process involves strong coupling effects of multiple physical fields such as temperature field, flow field, curing kinetics and stress-strain. The precise control of its process parameters directly determines the molding quality and efficiency.

[0003] Currently, industry and academia primarily rely on traditional numerical simulation techniques based on the finite element method (FEM) to construct coupled thermo-chemical-mechanical multi-field models for process design and verification. However, this type of method has inherent limitations in simultaneously satisfying the three objectives mentioned above: Finite element models require spatiotemporal discretization and iterative solutions for temperature field, flow field and curing dynamics. A single simulation can take several hours or even days, making it difficult to support rapid iterative optimization of process parameters and batch optimization of large-scale operating conditions. Traditional numerical simulations are highly sensitive to the strong nonlinearity, multi-scale coupling effects and dynamic changes in boundary conditions during the curing process of composite materials. The simplification assumptions of the model are prone to introducing systematic biases, and there is a lack of closed-loop correction mechanism based on actual molding data, resulting in insufficient robustness of process design to fluctuations in operating conditions. The finite element method has limited ability to perform high-precision synchronous characterization of temperature gradient, curing degree distribution and residual stress co-evolution, making it difficult to achieve high-fidelity reconstruction of multiphysics fields while ensuring computational efficiency.

[0004] Therefore, a method and system for optimizing the design of autoclave processes based on extended physical information neural operators are provided, which can achieve high efficiency, high reliability and high accuracy in the optimization design of autoclave processes. Summary of the Invention

[0005] To address the aforementioned technical problems, the present invention aims to provide a method and system for optimizing the design of autoclave processes based on extended physical information neural operators.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for optimizing the design of autoclave processes based on extended physical information neural operators, the method comprising: Multi-source data of the entire autoclave molding process were collected, and a standardized process dataset was constructed after preprocessing. A priori physical constraint system was constructed by combining the multi-physics field coupling mechanism of composite material curing, and the key feature set of matching constraints was extracted. A neural operator model with extended physical information embedded with prior physical constraints is constructed, and a multi-field coupled proxy model for autoclave molding is obtained by training the key feature set. Based on multi-objective optimization technology and combined with process parameter constraints, the optimal set of process parameters is obtained through intelligent optimization based on the multi-field coupling proxy model, and then transformed into an executable process curve output for the autoclave. Simultaneously, the closed-loop self-correction of the multi-field coupling proxy model is completed based on the actual molding data and the prediction deviation.

[0007] Preferably, the specific process for collecting multi-source data throughout the autoclave molding process is as follows: Based on the geometric characteristics of the target composite material component and the flow field distribution of the autoclave cavity, the monitoring subdomains corresponding to the multi-subdomain coupling operators of the physical information neural operators are divided and expanded, and multiple types of sensor nodes matching the curing stage characteristics are deployed in each monitoring subdomain. Based on the curing kinetics of composite materials, the process is divided into four molding stages: heating, heat preservation, pressure increase and holding, and cooling. Differentiated adaptive sampling frequencies are configured for each molding stage. Online process data of each sensor node are collected synchronously, as well as the operating conditions data of the autoclave equipment, component attribute data, and offline quality inspection data after molding for the corresponding batch. Configure a unified spatiotemporal reference stamp for all collected data to align the spatiotemporal dimensions of online process data, equipment condition data, and offline quality data, and construct the initial process dataset.

[0008] Preferably, the specific processing flow of the standardized process dataset is as follows: Based on the thermo-solid-mechanical multi-physics field coupling prior physical constraint system, a data physical consistency verification rule is constructed to identify and remove outliers from the collected multi-source data, and retain valid data that conforms to the physical laws of curing and molding. For missing values ​​in the valid data, physical compliance interpolation is performed to fill in the missing values ​​based on the constraint equations of the prior physical constraint system and the synchronous data of adjacent monitoring subdomains in the same batch, so as to obtain a complete dataset. The complete dataset is standardized in multiple dimensions to eliminate dimensional differences. At the same time, the standardized data is validated based on the prior physical constraint system to ensure that the data dimensions match the input dimensions of the extended physical information neural operator, and finally a standardized process dataset is obtained.

[0009] Preferably, the specific extraction process of the key feature set is as follows: Based on the thermo-solid-mechanical multi-physics field coupling prior physical constraint system, the physical evolution logic of the entire process of composite material curing and molding is decomposed. The independent variables, dependent variables and coupling intermediate variables of the constraint equation set of the prior physical constraint system are divided and initially selected to obtain four categories of initial physical features: process boundary feature set, multi-physics field state feature set, material property feature set and quality response feature set. Based on the coupling and correlation relationships of multiple physical fields in the prior physical constraint system, feature coupling mapping rules are constructed, cross-coupling operations are performed on the initial physical features, multi-field coupling and correlation features are extracted and added to the initial physical feature set to obtain the full physical feature set; A global sensitivity analysis model is constructed based on the constraint equations of the prior physical constraint system. The sensitivity coefficients of each feature in the full set of physical features to curing quality and multi-physics field evolution results are quantified. Key sensitive features with sensitivity coefficients higher than the preset sensitivity threshold are screened out to obtain the initial set of key features. The corresponding multi-subdomain coupled operator architecture of the extended physical information neural operator model is used to group the initially selected key feature set by subdomain dimensions and regularize the input dimensions. At the same time, the compliance verification of the physical boundary of the features is completed based on the prior physical constraint system. Finally, a key feature set that is bidirectionally matched with the prior physical constraint system and the input dimension of the extended physical information neural operator model is obtained.

[0010] Preferably, the specific extraction process of the multi-field coupling correlation features is as follows: Based on the bidirectional coupling relationship between heat conduction and curing kinetics in the prior physical constraint system, the temperature-curing degree gradient coupling characteristics are extracted. Based on the coupling relationship between degree of curing and resin rheological properties and stress evolution, the correlation features between degree of curing and residual stress are extracted; Based on the correlation between process boundary conditions and the distribution of cavity flow field and temperature field, the process parameter-field quantity distribution mapping characteristics are extracted.

[0011] Preferably, the specific conversion process of the autoclave's executable process curve is as follows: Based on the resin curing kinetics, a curing degree-viscosity linkage constraint is constructed, and the pressure timing of the optimal process parameter set is physically consistent with the calibration. The inverse dynamics model of the autoclave is used to perform feedforward pre-distortion compensation on the calibrated ideal curve; A smooth command curve for continuous velocity, acceleration, and jerk is generated using a fifth-order Bézier curve, and dynamic ripple suppression is superimposed during the heat preservation platform stage. Before the curve is released, the candidate curve is virtually run and verified. With the goal of minimizing the deviation between the predicted physical field and the expected physical field, the control points of the Bézier curve are corrected by Gaussian process regression iteration. Finally, the corrected curve is discretized according to the equipment sampling period and encoded into a multimodal process recipe file containing temperature ramp sequence, pressure valve opening sequence and safety interlock logic node.

[0012] Preferably, the closed-loop self-calibration of the multi-field coupled surrogate model based on the deviation between actual molding data and prediction includes: Collect multi-source sensor data during the actual molding process and construct a spatiotemporal deviation field with the prediction output of the multi-field coupled proxy model at spatiotemporal nodes; Based on the deviation field, the cumulative deviation trend, spatial variance, deviation time gradient, and deviation-parameter sensitivity are extracted to form a multidimensional feature index. Based on the combination of indicators, a hierarchical self-correction mode is triggered: when the spatial variance exceeds the threshold but the cumulative trend is normal, a local correction mode based on the spatial weighted loss function is triggered; when the cumulative trend exceeds the threshold and the temporal gradient is positive, a global incremental learning mode based on elastic weight consolidation and physical constraint regularization is triggered. After the correction is completed, the stability of the updated model is verified by the sliding window deviation index. Only when the corrected deviation index converges to within the threshold and no drift occurs in multiple consecutive batches, the updated model parameters are fixed and replace the currently deployed multi-field coupled proxy model; otherwise, the model is rolled back to the multi-field coupled proxy model before the update.

[0013] Preferably, the specific optimization process for the optimal set of process parameters is as follows: A multi-objective optimization function is constructed with the optimization objectives of curing degree field uniformity, residual stress peak value and process energy consumption minimization. The curing degree field uniformity is quantified by the spatiotemporal integral of the curing degree field gradient modulus output by the extended physical information neural operator model. The coupling constraints of process parameters in the prior physical constraint system are transformed into a feasible domain hyperplane in the parameter space, including the synergistic constraints of heating rate and pressurization timing, and the matching constraints of pressure amplitude and mold temperature. Based on the interpretability of the multi-field coupled proxy model, the first and second-order sensitivities of each process parameter to the output of key physical fields are automatically calculated by differentiation. A non-uniform sampling strategy for the parameter search space is constructed, and the sampling density is increased in the parameter dimension where the sensitivity is higher than the preset sensitivity threshold. During the optimization iteration process, dynamic physical field constraint indicators are calculated in real time through forward propagation of the multi-field coupled surrogate model, including maximum temperature gradient constraint, curing exothermic peak temperature rise rate constraint and pre-gel point viscosity threshold constraint, and the dynamic physical field constraint indicators are embedded as intermediate penalty terms into the multi-objective optimization function. The non-dominated sorting genetic algorithm based on reference points is combined with the non-uniform sampling strategy and dynamic physical field constraints for iterative optimization, and finally outputs the optimal set of process parameters that satisfy all hard constraints and dynamic physical field constraints.

[0014] A second aspect of the present invention also provides an extended physical information neural operator accelerated design optimization system for composite material autoclave processes, comprising: The feature extraction module collects multi-source data from the entire autoclave molding process, preprocesses it to build a standardized process dataset, and constructs a priori physical constraint system by combining the multi-physics field coupling mechanism of composite material curing, and extracts the key feature set matching the constraints. The model training module constructs an extended physical information neural operator model embedded with prior physical constraints, and obtains a multi-field coupled proxy model for autoclave molding through training with the key feature set. The control correction module, based on multi-objective optimization technology and combined with process parameter constraints, intelligently optimizes the optimal process parameter set based on the multi-field coupling proxy model and converts it into an executable process curve output for the autoclave. Simultaneously, it completes closed-loop self-correction of the multi-field coupling proxy model based on actual molding data and prediction deviation.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. By constructing an extended physical information neural operator model embedded with prior physical constraints, the coupling mechanism of thermo-solid-mechanical multi-physics fields is deeply integrated into the neural network architecture. Compared with traditional pure data-driven models, this method can still strictly follow the physical conservation laws and solidification molding rules under data sparsity or noise interference, significantly improving the prediction accuracy and generalization ability of multi-field coupling evolution processes, and providing a high-fidelity surrogate model for process optimization.

[0016] 2. Based on multi-objective optimization techniques (such as non-dominated sorting genetic algorithms) and the interpretability of surrogate models (automatic differential calculation of parameter sensitivity), efficient optimization under complex process constraints is achieved. Simultaneously, through inverse dynamics compensation, Bézier curve smoothing, and virtual operation verification, the optimal process parameter set is transformed into smooth and executable multimodal process curves that conform to the dynamic response characteristics of the equipment, thus establishing an engineering closed loop of optimization-execution and effectively shortening the process development cycle.

[0017] 3. A closed-loop self-correction mechanism based on the deviation between actual molding data and prediction was established. By constructing a spatiotemporal deviation field and designing a hierarchical triggering strategy (local correction and global incremental learning), the surrogate model can adaptively absorb uncertainties such as equipment drift and environmental fluctuations during the production process. This mechanism ensures that the model maintains high accuracy in long-term industrial deployment, avoids performance degradation, and continuously outputs robust optimal process parameters, significantly improving the quality consistency of batch production. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0019] Figure 1 This is a schematic diagram of a process optimization design method for autoclaves based on extended physical information neural operators.

[0020] Figure 2A schematic diagram of the entire process for optimizing multi-objective process parameters.

[0021] Figure 3 This is a schematic diagram illustrating the conversion of process parameters into executable curves.

[0022] Figure 4 This is a schematic diagram of an extended physical information neural operator-accelerated design optimization system for composite material autoclave processes. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other implementation methods obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0024] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0025] Example 1 like Figure 1 As shown in the figure, this embodiment discloses a method for optimizing the design of autoclave processes based on extended physical information neural operators. The method includes: Multi-source data of the entire autoclave molding process were collected, and a standardized process dataset was constructed after preprocessing. A priori physical constraint system was constructed by combining the multi-physics field coupling mechanism of composite material curing, and the key feature set of matching constraints was extracted. It should be noted that the specific data collection process for the entire autoclave molding process using multiple sources is as follows: Based on the geometric characteristics of the target composite material component and the flow field distribution of the autoclave cavity, the monitoring subdomains corresponding to the multi-subdomain coupling operators of the physical information neural operators are divided and expanded, and multiple types of sensor nodes matching the curing stage characteristics are deployed in each monitoring subdomain. Based on the curing kinetics of composite materials, the process is divided into four molding stages: heating, heat preservation, pressurization and holding, and cooling. Differentiated adaptive sampling frequencies are configured for each molding stage, and online process data from each sensor node are collected synchronously. Simultaneously, data on the operating conditions of the autoclave equipment, component properties, and offline quality inspection data after molding for the corresponding batch are also collected. In this embodiment, differentiated adaptive sampling frequencies are configured for the four molding stages: Heating stage: Resin viscosity changes rapidly, curing reaction begins, and the sampling frequency is set to... The collected data includes real-time temperature of each subdomain, cavity pressure, circulating air velocity, and heating power; during the heat preservation stage: the peak of exothermic reaction occurs and the field distribution fluctuates the most, the sampling frequency is set to... The collected data includes temperature gradients in each subdomain, estimated degrees of cure, component strain, and equipment valve openings; during the pressurization and holding stage: the resin enters the gel window and residual stress begins to accumulate, and the sampling frequency is set to... The collected data includes cavity pressure fluctuations, mold temperature, component strain changes, and pressurization timing nodes; during the cooling stage: the degree of curing tends to stabilize, and internal stress continues to be released, with the sampling frequency set to... The collected data includes cooling rate, residual strain of components, temperature uniformity of cavity, and power of cooling system.

[0026] A unified spatiotemporal reference stamp is configured for all collected data to align the spatiotemporal dimensions of online process data, equipment condition data, and offline quality data, thus constructing an initial process dataset. In this embodiment, the spatiotemporal alignment rule is: based on... To minimize the time slice, linear interpolation is performed on multi-source data within the same time slice for alignment. Offline quality data is labeled with the time interval corresponding to the forming batch, and bound one-to-one with the batch number of the process data. The final initial process dataset contains 12 core fields: batch number, spatiotemporal stamp, and monitoring subdomain. Temperature value, pressure value, strain value, equipment operating parameters, component attribute parameters, heating rate, calculated degree of curing, offline quality test value, and data validity marker.

[0027] The specific processing flow of the standardized process dataset is as follows: Based on a priori physical constraint system of thermo-solid-mechanical multiphysics coupling, a data physical consistency verification rule is constructed to identify and remove outliers from the collected multi-source data, retaining valid data that conforms to the physical laws of curing and molding. In this embodiment, the following is adopted: Outlier identification and removal are performed using a dual verification method combining criteria and physical boundaries. The specific rule is to calculate the mean of 100 consecutive sampling points within the same subdomain of the same batch. with standard deviation Exceeding Data within a given range is marked as statistical outliers; physical boundary compliance verification: range of heating rate, range of curing degree, range of temperature, and range of residual stress. Data exceeding the physical boundaries is marked as physical outliers; data that simultaneously meets both statistical and physical outlier criteria are directly removed, while data that only meets the statistical outlier criteria is retained and enters the missing value imputation stage.

[0028] For missing values ​​in the valid data, physical compliance interpolation is performed to fill in the missing values ​​based on the constraint equations of the prior physical constraint system and the synchronous data of adjacent monitoring subdomains in the same batch, so as to obtain a complete dataset. In this embodiment, the missing value imputation rule is as follows: for the duration of the missing value... For short-term missing data, linear interpolation of adjacent sampling points is used to fill in the missing data; for missing durations... arrive For missing data in the middle of the time interval, synchronous data from three adjacent monitoring subdomains in the same batch were used to fill the gaps using spatiotemporal weighted interpolation based on the heat conduction equation. The interpolation formula is as follows: In the formula: Interpolate temperature for missing points; The weights of adjacent subdomains are positively correlated with the flow field direction. To synchronize the temperatures of adjacent subdomains; This represents the spatial distance between adjacent subdomains and the missing point. The characteristic length is the width of the component. .

[0029] For missing duration Long-term missing data segments were marked as invalid batch segments and filled using numerical simulation results of prior physical constraint equations to ensure that the data conformed to solidified physical laws.

[0030] The complete dataset undergoes multi-dimensional standardization to eliminate dimensional differences. Simultaneously, based on a priori physical constraints, the standardized data's feature dimensions are validated to ensure that the data dimensions match the input dimensions of the extended physical information neural operator, ultimately yielding a standardized process dataset. The final standardized process dataset has the following dimensions: ,in The total number of sampling points is 24, and the feature dimensions are completely matched with the input dimensions of the extended physical information neural operator. It includes six core features in six monitoring subdomains: temperature, pressure, strain, heating rate, calculated degree of curing, and equipment operating conditions.

[0031] In detail, the specific extraction process of the key feature set is as follows: Based on the thermo-solid-mechanical multi-physics coupling prior physical constraint system, the physical evolution logic of the entire process of composite material curing and molding is deconstructed. Corresponding to the independent variables, dependent variables, and coupling intermediate variables of the constraint equations of the prior physical constraint system, four major categories of initial physical features are initially selected: process boundary feature set, multi-physics state feature set, material property feature set, and quality response feature set. In this embodiment, the four initially selected categories of initial physical features are specifically: Process boundary feature set: heating rate, holding temperature, holding time, initial pressurization temperature, holding pressure, holding time, and cooling rate (7 dimensions); Multi-physics state feature set: temperature field gradient of each subdomain, curing degree field distribution, cavity flow field wind speed, stress field peak value, and curing reaction rate (5 dimensions); Material property feature set: carbon fiber volume fraction, resin curing reaction parameters, thermal conductivity, elastic modulus, Poisson's ratio, and coefficient of thermal expansion (6 dimensions); Quality response feature set: final degree of curing, curing degree uniformity, residual stress peak value, component surface deviation, and delamination defect length (5 dimensions); Total 23 Initial physical characteristics.

[0032] Based on the coupling and correlation relationships of multiple physical fields in the prior physical constraint system, feature coupling mapping rules are constructed, cross-coupling operations are performed on the initial physical features, multi-field coupling and correlation features are extracted and added to the initial physical feature set to obtain the full physical feature set; In this embodiment, the specific extraction process of the multi-field coupling correlation features is as follows: Based on the bidirectional coupling relationship between heat conduction and curing kinetics in the prior physical constraint system, temperature-curing degree gradient coupling characteristics are extracted; the calculation formula for the temperature-curing degree gradient coupling characteristics is as follows: In the formula, For the temperature field gradient mode, The curing degree field gradient mode is a feature that quantifies the coupling strength between heat conduction and the curing reaction.

[0033] Based on the coupling relationship between degree of cure and resin rheological properties and stress evolution, the correlation characteristics between degree of cure and residual stress are extracted; the calculation formula for the correlation characteristics between degree of cure and residual stress is as follows: ; This represents the peak value of residual stress. This represents the average degree of curing. The gel point curing degree is 0.65 in this embodiment. This feature quantifies the degree of influence of the curing process on stress evolution.

[0034] Based on the correlation between process boundary conditions and the distribution of cavity flow field and temperature field, process parameter-field quantity distribution mapping features are extracted. The calculation formula for the process parameter-field quantity distribution mapping features is as follows: In the formula: Let the spatial standard deviation of the temperature field be . The average wind speed of the circulating air. The heating rate is a quantifiable feature that quantifies the impact of process parameters on the uniformity of the temperature field.

[0035] After extracting the above 3D multi-field coupling correlation features, the total physical feature set in this embodiment has 26 dimensions.

[0036] A global sensitivity analysis model is constructed based on the constraint equations of the prior physical constraint system. The sensitivity coefficients of each feature in the full set of physical features to curing quality and multi-physics field evolution results are quantified. Key sensitive features with sensitivity coefficients higher than a preset sensitivity threshold are selected to obtain a preliminary set of key features. In this embodiment, the following is adopted: The global sensitivity analysis method calculates the overall sensitivity coefficient for each feature using the following formula: In the formula: The total variance of the quality response objective. Features The expected conditional variance of the response target. In this embodiment, a preset sensitivity threshold is used. Features with a total sensitivity coefficient > 0.1 were included in the initial key feature set. After calculation using a standardized dataset, 16 key sensitive features were selected. Examples of core features and sensitivity coefficients include: heating rate. =0.32, insulation temperature =0.28, holding pressure =0.21, Temperature-Cure Gradient Coupling Characteristics =0.19, Curing Degree-Residual Stress Correlation Characteristics =0.17, all higher than the sensitivity threshold, and included in the initial key feature set.

[0037] Corresponding to the multi-subdomain coupled operator architecture of the extended physical information neural operator model, the initially selected key feature set is grouped by subdomain dimensions and the input dimensions are normalized. Simultaneously, based on the prior physical constraint system, the compliance of the feature physical boundaries is verified, ultimately obtaining a key feature set that bidirectionally matches the prior physical constraint system and the input dimensions of the extended physical information neural operator model. In this embodiment, there are 6 monitoring subdomains. The 16-dimensional initially selected key features are grouped by subdomain, with each subdomain corresponding to 8-dimensional input features, and finally normalized to dimensionality. The key feature set is perfectly matched with the input dimension of the multi-subdomain coupling operator of the extended physical information neural operator, and physical boundary verification is performed to ensure that all feature values ​​are within the compliance range.

[0038] A neural operator model with extended physical information embedded with prior physical constraints is constructed, and a multi-field coupled proxy model for autoclave molding is obtained by training the key feature set. In this embodiment, the extended physical information neural operator model is constructed as a three-level architecture of encoder-multi-subdomain coupled operator-decoder. The encoder employs a three-layer one-dimensional convolutional network with kernel sizes of 5, 3, and 3, and the number of neurons being 128, 64, and 32 respectively. The activation function is the Swish function. The input is the [6,8]-dimensional key feature set obtained above, and the output is a 32-dimensional deep feature encoding, achieving dimensionality reduction and abstraction of high-dimensional process features. The multi-subdomain coupling operator sets up 6 parallel operator branches for 6 monitoring subdomains. Each branch adopts a 2-layer fully connected network with 64 neurons. The activation function is the Tanh function. Cross-subdomain feature interaction is achieved between branches through an attention mechanism. The attention weight coefficient is dynamically adjusted by the flow field coupling strength between subdomains. Finally, the multi-physics prediction features of each subdomain are output. The decoder uses a 2-layer transposed convolutional network + a 1-layer fully connected output layer with convolutional kernel sizes of 3 and 5, and the number of neurons of 64 and 128, respectively. The activation function is the Swish function, and the final output dimension is [6,5], which corresponds to the prediction results of the five core physical fields of temperature field, solidification field, stress field, viscosity field, flow field and wind speed in the 6 monitoring subdomains.

[0039] Based on multi-objective optimization technology and combined with process parameter constraints, the optimal set of process parameters is obtained through intelligent optimization based on the multi-field coupling proxy model, and then transformed into an executable process curve output for the autoclave. Simultaneously, the closed-loop self-correction of the multi-field coupling proxy model is completed based on the actual molding data and the prediction deviation.

[0040] Among them, such as Figure 2 As shown, the specific optimization process for the optimal process parameter set is as follows: A multi-objective optimization function is constructed with the objectives of minimizing the uniformity of the curing field, the peak residual stress, and the process energy consumption. The uniformity of the curing field is quantified by the spatiotemporal integral of the gradient modulus of the curing field output by the extended physical information neural operator model. In this embodiment, the constructed multi-objective optimization function is as follows, where all objectives are minimized: ; To achieve uniformity of curing degree, the curing degree field gradient modulus is quantized through spatiotemporal integration. The target for residual stress peak value is the maximum residual stress value in the entire spatial domain throughout the entire forming cycle of the component. The energy consumption target for the process is quantified by integrating the total power of the heating system, circulating air system, and cooling system.

[0041] The coupling constraints of process parameters in the prior physical constraint system are transformed into a feasible domain hyperplane in the parameter space, including the synergistic constraints of heating rate and pressurization timing, and the matching constraints of pressure amplitude and mold temperature. Based on the interpretability of the multi-field coupled surrogate model, the first-order sensitivity of each process parameter to the key physical field output is calculated by automatic differentiation. A non-uniform sampling strategy for the parameter search space is constructed, increasing the sampling density in the parameter dimensions where the sensitivity is higher than a preset sensitivity threshold. The formula for calculating the first-order sensitivity is: In the formula, Let l be the l-th process parameter (e.g., heating rate, holding temperature, holding pressure, etc.). The maximum sensitivity under each target condition is taken as the overall sensitivity of that parameter. ; Let the dimension of the parameter space be D, and the total number of sampling points be... Based on sensitivity, the parameters are divided into three categories: high, medium, and low. High sensitivity parameters: The sampling density factor is 3; Medium sensitivity parameters: The sampling density factor is 2; Low sensitivity parameters: The sampling density factor is 1.

[0042] For example, setting a high sensitivity threshold Medium sensitivity threshold Calculations show that the heating rate, holding temperature, and holding pressure are highly sensitive parameters, and their sampling point allocation is as follows: , Set the sampling density multiple for the j-th parameter; =200, then the number of sampling points for the high-sensitivity parameter dimension is approximately 60, and for the low-sensitivity dimension, approximately 20. Latin hypercube sampling combined with sensitivity weights is used to generate an initial population within the parameter space: for high-sensitivity parameters, stratified sampling is performed according to probability density within their value range to increase local density; for low-sensitivity parameters, uniform stratified sampling is used; finally, the sampling points of each dimension are randomly combined to form... An initial individual.

[0043] During the optimization iteration process, dynamic physical field constraint indicators are calculated in real time through forward propagation of the multi-field coupled surrogate model, including maximum temperature gradient constraint, curing exothermic peak temperature rise rate constraint and pre-gel point viscosity threshold constraint, and the dynamic physical field constraint indicators are embedded as intermediate penalty terms into the multi-objective optimization function. The non-dominated sorting genetic algorithm based on reference points is combined with the non-uniform sampling strategy and dynamic physical field constraints for iterative optimization, and finally outputs the optimal set of process parameters that satisfy all hard constraints and dynamic physical field constraints.

[0044] Detailed, using As a multi-objective optimization framework, it addresses three conflicting optimization objectives (curing uniformity, residual stress peak, and process energy consumption) and introduces dynamic physics constraints as penalty terms. The corresponding algorithm parameters are set as follows: population size... Maximum number of iterations: Crossover probability: Simulated binary crossover (SBX) is used; mutation probability: Polynomial mutation is employed; Number of reference points: based on The method generates in three-dimensional target space Each uniformly distributed reference point is used to divide each dimension of the target into 12 equal parts.

[0045] like Figure 3 As shown, the specific conversion process of the autoclave's executable process curve is as follows: Based on the resin curing kinetics, a curing degree-viscosity linkage constraint is constructed, and the pressure timing of the optimal process parameter set is physically consistent with the calibration. The inverse dynamics model of the autoclave is used to perform feedforward pre-distortion compensation on the calibrated ideal curve; A smooth command curve for continuous velocity, acceleration, and jerk is generated using a fifth-order Bézier curve, and dynamic ripple suppression is superimposed during the heat preservation platform stage. Before the curve is issued, the candidate curve is virtually run and verified. With the goal of minimizing the deviation between the predicted physical field and the expected physical field, the control points of the Bezier curve are corrected by Gaussian process regression iteration. Finally, the corrected curve is discretized according to the equipment sampling period and encoded into a multimodal process recipe file containing temperature ramp sequence, pressure valve opening sequence and safety interlock logic node. An example of the core content of the multimodal process recipe file is shown in Table 1. Table 1. Core Contents of a Multimodal Process Formulation Document It should be noted that the closed-loop self-calibration of the multi-field coupled surrogate model based on the deviation between actual formed data and prediction includes: Collect multi-source sensor data during the actual molding process and construct a spatiotemporal deviation field with the prediction output of the multi-field coupled proxy model at spatiotemporal nodes; Based on the deviation field, the cumulative deviation trend, spatial variance, deviation time gradient, and deviation-parameter sensitivity are extracted to form a multidimensional feature index. Based on the combination of indicators, a hierarchical self-correction mode is triggered: when the spatial variance exceeds the threshold but the cumulative trend is normal, a local correction mode based on the spatial weighted loss function is triggered; when the cumulative trend exceeds the threshold and the temporal gradient is positive, a global incremental learning mode based on elastic weight consolidation and physical constraint regularization is triggered. After the correction is completed, the stability of the updated model is verified by the sliding window deviation index. Only when the corrected deviation index converges to within the threshold and no drift occurs in multiple consecutive batches, the updated model parameters are fixed and replace the currently deployed multi-field coupled proxy model; otherwise, the model is rolled back to the multi-field coupled proxy model before the update.

[0046] like Figure 4 As shown, this embodiment also discloses an extended physical information neural operator accelerated design optimization system for composite material autoclave processes, including: The feature extraction module collects multi-source data from the entire autoclave molding process, preprocesses it to build a standardized process dataset, and constructs a priori physical constraint system by combining the multi-physics field coupling mechanism of composite material curing, and extracts the key feature set matching the constraints. The model training module constructs an extended physical information neural operator model embedded with prior physical constraints, and obtains a multi-field coupled proxy model for autoclave molding through training with the key feature set. The control correction module, based on multi-objective optimization technology and combined with process parameter constraints, intelligently optimizes the optimal process parameter set based on the multi-field coupling proxy model and converts it into an executable process curve output for the autoclave. Simultaneously, it completes closed-loop self-correction of the multi-field coupling proxy model based on actual molding data and prediction deviation.

[0047] Optionally, in this embodiment, those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0048] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0049] If the integrated units in the above embodiments are implemented as software functional units and sold or used as independent products, they can be stored in the aforementioned computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause one or more electronic devices to execute all or part of the steps of the methods described in the various embodiments of this application.

[0050] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0051] In the several embodiments provided in this application, it should be understood that the disclosed application can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of units or modules may be electrical or other forms.

[0052] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0053] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0054] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for optimizing the design of autoclave processes based on extended physical information neural operators, characterized in that, The method includes: Multi-source data of the entire autoclave molding process were collected, and a standardized process dataset was constructed after preprocessing. A priori physical constraint system was constructed by combining the multi-physics field coupling mechanism of composite material curing, and the key feature set of matching constraints was extracted. A neural operator model with extended physical information embedded with prior physical constraints is constructed, and a multi-field coupled proxy model for autoclave molding is obtained by training the key feature set. Based on multi-objective optimization technology and combined with process parameter constraints, the optimal set of process parameters is obtained through intelligent optimization based on the multi-field coupling proxy model, and then transformed into an executable process curve output for the autoclave. Simultaneously, the closed-loop self-correction of the multi-field coupling proxy model is completed based on the actual molding data and the prediction deviation.

2. The autoclave process optimization design method based on extended physical information neural operators according to claim 1, characterized in that, The specific process for collecting multi-source data throughout the entire autoclave forming process is as follows: Based on the geometric characteristics of the target composite material component and the flow field distribution of the autoclave cavity, the monitoring subdomains corresponding to the multi-subdomain coupling operators of the physical information neural operators are divided and expanded, and multiple types of sensor nodes matching the curing stage characteristics are deployed in each monitoring subdomain. Based on the curing kinetics of composite materials, the process is divided into four molding stages: heating, heat preservation, pressure increase and holding, and cooling. Differentiated adaptive sampling frequencies are configured for each molding stage. Online process data of each sensor node are collected synchronously, as well as the operating conditions data of the autoclave equipment, component attribute data, and offline quality inspection data after molding for the corresponding batch. Configure a unified spatiotemporal reference stamp for all collected data to align the spatiotemporal dimensions of online process data, equipment condition data, and offline quality data, and construct the initial process dataset.

3. The autoclave process optimization design method based on extended physical information neural operators according to claim 2, characterized in that, The specific processing flow for the standardized process dataset is as follows: Based on the thermo-solid-mechanical multi-physics field coupling prior physical constraint system, a data physical consistency verification rule is constructed to identify and remove outliers from the collected multi-source data, and retain valid data that conforms to the physical laws of curing and molding. For missing values ​​in the valid data, physical compliance interpolation is performed to fill in the missing values ​​based on the constraint equations of the prior physical constraint system and the synchronous data of adjacent monitoring subdomains in the same batch, so as to obtain a complete dataset. The complete dataset is standardized in multiple dimensions to eliminate dimensional differences. At the same time, the standardized data is validated based on the prior physical constraint system to ensure that the data dimensions match the input dimensions of the extended physical information neural operator, and finally a standardized process dataset is obtained.

4. The autoclave process optimization design method based on extended physical information neural operators according to claim 3, characterized in that, The specific extraction process for the key feature set is as follows: Based on the thermo-solid-mechanical multi-physics field coupling prior physical constraint system, the physical evolution logic of the entire process of composite material curing and molding is decomposed. The independent variables, dependent variables and coupling intermediate variables of the constraint equation set of the prior physical constraint system are divided and initially selected to obtain four categories of initial physical features: process boundary feature set, multi-physics field state feature set, material property feature set and quality response feature set. Based on the coupling and correlation relationships of multiple physical fields in the prior physical constraint system, feature coupling mapping rules are constructed, cross-coupling operations are performed on the initial physical features, multi-field coupling and correlation features are extracted and added to the initial physical feature set to obtain the full physical feature set; A global sensitivity analysis model is constructed based on the constraint equations of the prior physical constraint system. The sensitivity coefficients of each feature in the full set of physical features to curing quality and multi-physics field evolution results are quantified. Key sensitive features with sensitivity coefficients higher than the preset sensitivity threshold are screened out to obtain the initial set of key features. The corresponding multi-subdomain coupled operator architecture of the extended physical information neural operator model is used to group the initially selected key feature set by subdomain dimensions and regularize the input dimensions. At the same time, the compliance verification of the physical boundary of the features is completed based on the prior physical constraint system. Finally, a key feature set that is bidirectionally matched with the prior physical constraint system and the input dimension of the extended physical information neural operator model is obtained.

5. The autoclave process optimization design method based on extended physical information neural operators according to claim 4, characterized in that, The specific extraction process of the multi-field coupled correlation features is as follows: Based on the bidirectional coupling relationship between heat conduction and curing kinetics in the prior physical constraint system, the temperature-curing degree gradient coupling characteristics are extracted. Based on the coupling relationship between degree of curing and resin rheological properties and stress evolution, the correlation features between degree of curing and residual stress are extracted; Based on the correlation between process boundary conditions and the distribution of cavity flow field and temperature field, the process parameter-field quantity distribution mapping characteristics are extracted.

6. The autoclave process optimization design method based on extended physical information neural operators according to claim 5, characterized in that, The specific conversion process of the autoclave's executable process curve is as follows: Based on the resin curing kinetics, a curing degree-viscosity linkage constraint is constructed, and the pressure timing of the optimal process parameter set is physically consistent with the calibration. The inverse dynamics model of the autoclave is used to perform feedforward pre-distortion compensation on the calibrated ideal curve; A smooth command curve for continuous velocity, acceleration, and jerk is generated using a fifth-order Bézier curve, and dynamic ripple suppression is superimposed during the heat preservation platform stage. Before the curve is released, the candidate curve is virtually run and verified. With the goal of minimizing the deviation between the predicted physical field and the expected physical field, the control points of the Bézier curve are corrected by Gaussian process regression iteration. Finally, the corrected curve is discretized according to the equipment sampling period and encoded into a multimodal process recipe file containing temperature ramp sequence, pressure valve opening sequence and safety interlock logic node.

7. The autoclave process optimization design method based on extended physical information neural operators according to claim 6, characterized in that, The closed-loop self-calibration of the multi-field coupled surrogate model based on actual molding data and prediction deviation includes: Collect multi-source sensor data during the actual molding process and construct a spatiotemporal deviation field with the prediction output of the multi-field coupled proxy model at spatiotemporal nodes; Based on the deviation field, the cumulative deviation trend, spatial variance, deviation time gradient, and deviation-parameter sensitivity are extracted to form a multidimensional feature index. Based on the combination of indicators, a hierarchical self-correction mode is triggered: when the spatial variance exceeds the threshold but the cumulative trend is normal, a local correction mode based on the spatial weighted loss function is triggered; when the cumulative trend exceeds the threshold and the temporal gradient is positive, a global incremental learning mode based on elastic weight consolidation and physical constraint regularization is triggered. After the correction is completed, the stability of the updated model is verified by the sliding window deviation index. Only when the corrected deviation index converges to within the threshold and no drift occurs in multiple consecutive batches, the updated model parameters are fixed and replace the currently deployed multi-field coupled proxy model; otherwise, the model is rolled back to the multi-field coupled proxy model before the update.

8. The autoclave process optimization design method based on extended physical information neural operators according to claim 7, characterized in that, The specific optimization process for the optimal set of process parameters is as follows: A multi-objective optimization function is constructed with the optimization objectives of curing degree field uniformity, residual stress peak value and process energy consumption minimization. The curing degree field uniformity is quantified by the spatiotemporal integral of the curing degree field gradient modulus output by the extended physical information neural operator model. The coupling constraints of process parameters in the prior physical constraint system are transformed into a feasible domain hyperplane in the parameter space, including the synergistic constraints of heating rate and pressurization timing, and the matching constraints of pressure amplitude and mold temperature. Based on the interpretability of the multi-field coupled proxy model, the first and second-order sensitivities of each process parameter to the output of key physical fields are automatically calculated by differentiation. A non-uniform sampling strategy for the parameter search space is constructed, and the sampling density is increased in the parameter dimension where the sensitivity is higher than the preset sensitivity threshold. During the optimization iteration process, dynamic physical field constraint indicators are calculated in real time through forward propagation of the multi-field coupled surrogate model, including maximum temperature gradient constraint, curing exothermic peak temperature rise rate constraint and pre-gel point viscosity threshold constraint, and the dynamic physical field constraint indicators are embedded as intermediate penalty terms into the multi-objective optimization function. The non-dominated sorting genetic algorithm based on reference points is combined with the non-uniform sampling strategy and dynamic physical field constraints for iterative optimization, and finally outputs the optimal set of process parameters that satisfy all hard constraints and dynamic physical field constraints.

9. An extended physical information neural operator-accelerated design optimization system for composite material autoclave processes, implementing the autoclave process optimization design method based on extended physical information neural operators as described in any one of claims 1 to 8, characterized in that, include: The feature extraction module collects multi-source data from the entire autoclave molding process, preprocesses it to build a standardized process dataset, and constructs a priori physical constraint system by combining the multi-physics field coupling mechanism of composite material curing, and extracts the key feature set matching the constraints. The model training module constructs an extended physical information neural operator model embedded with prior physical constraints, and obtains a multi-field coupled proxy model for autoclave molding through training with the key feature set. The control correction module, based on multi-objective optimization technology and combined with process parameter constraints, intelligently optimizes the optimal process parameter set based on the multi-field coupling proxy model and converts it into an executable process curve output for the autoclave. Simultaneously, it completes closed-loop self-correction of the multi-field coupling proxy model based on actual molding data and prediction deviation.

Images