A simulation model error correction method based on discrete experimental data

Abstract

The application discloses a simulation model error correction method based on discrete experimental data, relates to the field of data fusion of multiple physical fields, and comprises the following steps: obtaining original global simulation data, limited node simulation data and discrete experimental data of a target combustion chamber; constructing a standard experimental data set in time and space by matching space nodes and aligning time interpolation; respectively calculating error correction weights for physical characteristics such as temperature and strain, and mapping the weights to global nodes based on limited node space segmentation to realize correction of global temperature and strain simulation data of the combustion chamber. The application can reduce the demand for global experimental measurement and improve the consistency of simulation data and experimental data.

Description

Technical Field

[0001] This invention relates to the field of multiphysics data fusion. More specifically, this invention relates to a simulation model error correction method based on discrete experimental data used in the design of scramjet engine combustors. Background Technology

[0002] In the field of aerospace propulsion, the safety performance and operational efficiency of scramjet engines are crucial considerations. Nonlinear thermodynamic loads (such as extremely high temperatures and strong dynamic strain), as common internal excitations during engine flight at extremely high Mach numbers, significantly impact the morphological evolution and safety of the combustion chamber structure. Directly measuring the multiphysics distribution acting on the entire combustion chamber structure is typically very difficult, as it would require installing a massive number of sensors within the structure. However, the extremely harsh operating environment inside the combustion chamber (high temperature, high pressure, and strong airflow scouring) and the extremely limited installation space make dense sensor deployment impractical or prohibitively costly in actual engineering operations. Therefore, researching simulation model error correction techniques—that is, inferring and correcting the multiphysics distribution acting on the entire structure using limited discrete experimental measurement data—has significant engineering reference value for engine morphology inversion, damage assessment, and accurate prediction of multiphysics quantities.

[0003] The acquisition and correction of global physics field data is a typical problem of multi-source data fusion and inversion, aiming to correct global theoretical predictions (numerical simulation data) using limited real responses (discrete experimental data). In the field of health monitoring and multiphysics prediction of complex equipment structures, research on global data acquisition and correction techniques is mainly based on empirical correction and global interpolation fusion methods. These methods heavily rely on massive amounts of experimental data or manual experience to trial and error for coefficients, making them difficult to implement in extreme environments with limited sensor deployment. Furthermore, existing methods often use uniform weights or global interpolation for correction, which, when dealing with multiphysics systems, can easily lead to severe distortion of small numerical features due to large differences in the magnitude of physical quantities (such as strain and temperature). At the same time, global interpolation is prone to "oversmoothing" effects, making it difficult to capture sharp spatial gradients in the structure and facing difficulties such as generalization divergence in unknown regions. Summary of the Invention

[0004] One object of the present invention is to solve at least the above-mentioned problems and / or defects, and to provide at least the advantages described below.

[0005] To achieve these objectives and other advantages of the present invention, a method for correcting simulation model errors based on discrete experimental data is provided, comprising: S1. Obtain discrete experimental datasets under excitation through data acquisition instruments and physical sensors, and combine them with the original numerical simulation datasets and finite node simulation datasets to construct a multi-source heterogeneous dataset adapted to the target combustion chamber. S2. Perform spatiotemporal consistency alignment on multi-source heterogeneous datasets to obtain a spatiotemporally unified standard experimental dataset; S3. Construct a feature-decoupled error correction weight calculation model and perform calculations based on finite node simulation data and standard experimental datasets to obtain temperature correction weights and strain correction weights. S4. Construct a full-node data correction model based on spatial topology mapping, and divide the full-domain nodes corresponding to the outer wall of the combustion chamber into multiple spatial control domains; S5. Input the original numerical simulation dataset into the full-node data correction model, and map the temperature correction weight and strain correction weight of the corresponding finite nodes to the full-node nodes according to the spatial control domain to which the full-nodes belong, and perform error correction on the temperature simulation data and strain simulation data of the full-nodes respectively. S6. Calculate the error index between the finite node simulation data before and after correction and the standard experimental data, and output the corrected global multiphysics data based on the error improvement.

[0006] Preferably, in S1, the discrete experimental dataset includes: temperature time-series data and strain time-series data acquired using temperature sensors and strain sensors; The original global simulation dataset includes: spatial coordinates of all nodes on the outer wall of the combustion chamber and corresponding temperature simulation data and strain simulation data. The data structure of the original global simulation dataset includes: timestamp, node number and node data matrix. The node data matrix includes: three-dimensional coordinates, temperature and strain of the node. The finite node simulation dataset is composed of corresponding finite nodes extracted from the original global simulation dataset based on the experimental sensor measurement point locations. The data structure of each finite node includes: simulation timestamp, finite node number, temperature simulation data and strain simulation data of the finite node. Among them, the finite nodes include: temperature measurement point corresponding nodes used to establish the correspondence between temperature experimental data and temperature simulation data; The corresponding node for strain measurement points is used to establish the correspondence between strain experimental data and strain simulation data.

[0007] Preferably, in S2, the spatiotemporal consistency alignment process includes: spatial node matching and temporal matching; The spatial node matching process involves using the node positions in the finite node simulation dataset as a spatial reference to match the measurement point positions of the experimental sensors in the discrete experimental dataset. The time series matching refers to the process of mapping experimental time series in discrete experimental datasets to simulation timestamps. The mapping method is as follows: For any simulation timestamp, a linear interpolation function is established based on its adjacent experimental sampling time points and corresponding experimental measurements. The experimental temperature time series data and experimental strain time series data are then mapped to the timestamps of the finite node simulation dataset to construct a standard experimental dataset that is consistent with the finite node simulation dataset in the time dimension.

[0008] Preferably, in S3, the error correction weight calculation model splits the standard experimental dataset and the finite node simulation dataset into temperature branches and strain branches according to the physical feature type, so as to obtain the corresponding temperature correction weights and strain correction weights through the temperature branches and strain branches. The temperature correction weight and strain correction weight are implemented using the least squares method or direct division. When using the least squares method, the corresponding weights are obtained by the following formula: in, express t Time of the first f Correction weights for each physical feature Represents the first in the simulation dataset n Each node t Time of the first f One physical characteristic value, This represents the corresponding value in the standard experimental dataset. N It represents the sum of discrete nodes.

[0009] Preferably, a preset threshold for denominator protection is set during the calculation of the error correction weight calculation model, so that when the absolute value of the denominator is less than or equal to the preset threshold, the corresponding error correction weight is set to 1.0.

[0010] Preferably, the spatial control domain is obtained in the following way: S40. By extracting the spatial coordinates of a finite number of discrete nodes along the principal scale direction, and arranging them in ascending order, a sequence set is obtained. X i ,and ; S41. Using the geometric midpoints of adjacent discrete nodes as spatial segmentation boundaries, the global nodes are divided into multiple independent spatial control domains, and the spatial segmentation boundaries... B i The calculation formula is: .

[0011] Preferably, in S4, the full-node data correction model traverses all unknown nodes in the original global numerical simulation data. k and will k Corresponding coordinates X k With all segment boundaries B i Perform a comparison to determine k The spatial control domain to which it belongs is then corrected using the following formula. k Physical field prediction values : in, Unknown node k The original simulation values, For the core discrete nodes within the spatial control domain in which it falls j Corrected weights.

[0012] The present invention has at least the following beneficial effects: Firstly, this invention introduces a feature-independent decoupling calculation mechanism for the first time in multi-physics error correction, which effectively overcomes the problem of small-value feature distortion caused by the huge difference in the magnitude of micro-strain and temperature values ​​during traditional unified weight optimization, and significantly improves the accuracy of multi-physics joint reconstruction.

[0013] Secondly, this invention innovatively proposes a spatial segmentation weight mapping algorithm based on geometric boundary division, which avoids the "over-smoothing" and unknown region divergence phenomenon caused by traditional global interpolation algorithms, and can restore the physical boundary of the nonlinear and drastic changes of the aircraft structure with high fidelity.

[0014] Third, the method of this invention can still reconstruct the spatiotemporal distribution law of multi-physics fields with high precision across the entire domain under conditions of very little sensor data, and has extremely strong engineering practical value.

[0015] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0016] Figure 1 This is an overall flowchart of a simulation model error correction method based on discrete experimental data according to the present invention; Figure 2 This is a schematic diagram illustrating the construction of the multi-source heterogeneous dataset and the correspondence between measurement points in this invention; Figure 3 This is a schematic diagram of the multi-source data spatiotemporal consistency alignment processing of the present invention; Figure 4This is a schematic diagram illustrating the calculation of the feature decoupling error correction weight in this invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0018] A simulation model error correction method based on discrete experimental data is proposed, employing a feature-independent decoupling and spatial topology mapping approach. This method utilizes a feature-independent decoupling mechanism to overcome interference from differences in physical quantities, and constructs spatial control segments by extracting the geometric boundaries of discrete measurement points. The correction weights are then scientifically mapped to unknown nodes across the entire domain, providing a new solution for global error correction in simulation models. This method demonstrates significant potential in handling problems with strong spatial nonlinearity and multi-physics coupling, significantly reducing reliance on massive amounts of data and human experience. With the development of equipment topography inversion and digital twin technologies, its application prospects are even broader. In specific implementation, this invention includes the following steps: Step 1: Obtain the multi-source heterogeneous dataset of the target aircraft. The target aircraft may be a scramjet engine combustor. The multi-source heterogeneous dataset includes: a raw global simulation dataset, a finite node simulation dataset, and a discrete experimental dataset. The data structure of the raw global simulation dataset includes: timestamps, node numbers, and a node data matrix. The node data matrix includes at least the x-coordinate, y-coordinate, z-coordinate, temperature, and strain of the nodes. The original global simulation dataset includes: the spatial coordinates of all nodes on the outer wall of the combustion chamber and the corresponding temperature and strain simulation data; The finite node simulation dataset is node data extracted from the original global simulation dataset that corresponds to the measurement points of the experimental sensors. The data structure of the finite node simulation dataset includes: timestamp, finite node number, finite node temperature simulation data, and finite node strain simulation data. The discrete experimental dataset includes temperature and strain time-series data acquired by temperature and strain sensors. Step 2: Perform consistent spatiotemporal alignment processing on multi-source heterogeneous data. Using finite-node simulation data as a spatial reference, spatial node matching is performed on discrete experimental data; spatial node matching is performed on experimental measurement points in discrete experimental dataset, and experimental time series in discrete experimental dataset are mapped to simulation timestamps in finite-node simulation dataset. That is, a spatiotemporally unified standard experimental dataset is obtained by extrapolating and aligning the time series of high-frequency data. The spatial node matching is achieved based on the correspondence between the experimental sensor measurement point positions and the spatial positions of nodes in the original global simulation dataset. The finite nodes include temperature measurement point corresponding nodes and strain measurement point corresponding nodes. The temperature measurement point corresponding nodes are used to establish the correspondence between temperature experimental data and temperature simulation data, and the strain measurement point corresponding nodes are used to establish the correspondence between strain experimental data and strain simulation data. The mapping refers to the process of linearly interpolating and aligning the time series of discrete experimental data with a high sampling rate to the time series of simulation data with a low sampling rate by setting the system sampling frequency parameter. Specifically, for any simulation timestamp, a linear interpolation function is established based on its adjacent experimental sampling time points and corresponding experimental measurement values, and the experimental temperature time series data and experimental strain time series data are mapped to the timestamps of the finite node simulation dataset, respectively, in order to construct a standard experimental dataset that is consistent with the finite node simulation dataset in the time dimension.

[0019] Step 3: Construct an error correction weight calculation model for feature decoupling. The error correction weight calculation model includes an input data organization module, a feature decoupling module, an error correction weight calculation module, and a feature-independent weight output module. The input data organization module receives a standard experimental dataset and a finite-node simulation dataset as input tensors, and organizes the input tensors according to time steps, discrete nodes, and physical features. The feature decoupling module splits the standard experimental dataset and the finite-node simulation dataset into temperature branches and strain branches according to the physical feature types (temperature and strain) to extract corresponding feature values. Specifically, the temperature branch independently extracts experimental temperature values ​​and simulated temperature values ​​and calculates temperature correction weights, while the strain branch independently extracts experimental strain values ​​and simulated strain values ​​and calculates strain correction weights, thus avoiding mutual interference in the weight calculation process caused by differences in the dimensions and numerical scales of different physical quantities.

[0020] The extracted feature values ​​are input into the error correction weight calculation model. Error correction weights are calculated for physical features of different numerical magnitudes (such as temperature and strain) to obtain temperature correction weight sets and strain correction weight sets. In practical applications, the error correction weights of the current discrete node at each time step are calculated using the feature-independent least squares method or direct division. If the temperature correction weight and strain correction weight are calculated using the least squares method, then for any time step and any physical feature, a weight optimization function is established with the objective of minimizing the sum of squared errors between the standard experimental data and the corrected finite-node simulation data, and the error correction weight of the corresponding physical feature is calculated based on the least squares analytical solution.

[0021] If the temperature correction weight and strain correction weight are calculated by direct division, then for any time step, any finite node, and any physical feature, the error correction weight is calculated based on the ratio of the standard experimental data value to the corresponding finite node simulation data value, so that each time step, each finite node, and each physical feature has an independent error correction weight.

[0022] Furthermore, a denominator protection threshold is set during the error correction weight calculation process. When the absolute value of the denominator in the weight calculation formula is less than or equal to the preset threshold, the corresponding error correction weight is set to 1.0 to avoid division by zero errors or numerical overflow caused by the simulation data being too small.

[0023] The feature-independent weight output module obtains the corresponding feature decoupling correction weight set based on the temperature correction weight set and the strain correction weight set, and outputs the feature decoupling error correction weight to provide temperature weight and strain weight as input for the full-node data correction model.

[0024] Step 4: Construct a full-node data correction model based on spatial topology mapping. This step extracts the spatial coordinates of the finite nodes and establishes spatial segmentation boundaries based on the geometric positions of the finite nodes in the principal scale direction, dividing the entire domain nodes of the combustion chamber exterior wall into multiple spatial control domains. The method for establishing the spatial segment boundary is as follows: extract the coordinates of the finite nodes in the main scale direction of the combustion chamber and arrange them in ascending order; calculate the geometric midpoint of the coordinates of adjacent finite nodes as the spatial segment boundary; for any global node in the original global simulation dataset, determine the spatial control domain to which the global node belongs based on the positional relationship between the coordinates of the global node in the main scale direction and the spatial segment boundary, and inherit the error correction weight of the corresponding finite node in the spatial control domain.

[0025] Furthermore, the spatial control domain is obtained by arranging the extracted spatial coordinates in ascending order and calculating the geometric midpoint of adjacent nodes as the spatial segmentation boundary, thus dividing the entire combustion chamber structure into multiple independent spatial control domains.

[0026] Step 5: Input the original global numerical simulation data into the full-node data correction model. This step mainly involves automatically determining the spatial control domain to which the unknown node belongs based on its spatial coordinates, mapping the temperature correction weight and strain correction weight of the corresponding finite node to the global node, and then performing error correction on the temperature simulation data and strain simulation data of the global node to obtain the corrected global multiphysics data. Specifically, when correcting errors for nodes across the entire domain, temperature simulation data is corrected using temperature correction weights, and strain simulation data is corrected using strain correction weights. For finite nodes corresponding to the measurement points of the experimental sensors, the corrected data is replaced or calibrated using the standard experimental data at the corresponding time to ensure that the correction results at the measurement point locations are consistent with the experimental measurement data.

[0027] Furthermore, the error index between the finite node simulation data before and after correction and the standard experimental data is calculated. The error index includes: mean square error and error improvement rate. When the improvement rate reaches a set threshold, the final global data is output; otherwise, the weight calculation rules are adjusted.

[0028] The mean square error is used to characterize the deviation between finite node simulation data or corrected finite node data and standard experimental data. The error improvement rate is calculated based on the mean square error before and after correction. When the error improvement rate is greater than zero or reaches a preset engineering threshold, the final corrected global multiphysics data is output. The corrected global multiphysics data includes: the spatial coordinates of global nodes on the outer wall of the combustion chamber, the corrected temperature data, and the corrected strain data. The corrected global multiphysics data is used for combustion chamber structural morphology inversion, temperature field prediction, strain field prediction, state assessment, or digital model data stream construction.

[0029] This scheme overcomes the limitations of traditional methods in handling differences in the magnitude of multi-physics characteristics and the ease with which spatial interpolation smoothing diverges. It can accurately achieve decoupling and mapping correction of multi-physics simulation data across the entire domain using a very small number of discrete measurement points. It has extremely high data fidelity and global generalization ability, providing reliable high-precision global data support for topography inversion and accurate prediction of multiple physical quantities under extreme environments.

[0030] Example: A simulation model error correction method based on discrete experimental data is applicable to multi-physics strongly coupled environments such as the combustion chamber of a scramjet engine. Figure 1 As shown, it mainly includes four parts: data acquisition stage, consistent processing stage, feature and weight calculation stage, and data mapping and reconstruction stage, specifically including the following steps: S1. Acquisition of Multi-Source Heterogeneous Datasets and System Construction This step corresponds to the data acquisition stage, which is used to acquire a multi-source heterogeneous dataset of the target aircraft (such as a scramjet engine combustor) under specific service conditions. The multi-source heterogeneous dataset includes: the original global simulation dataset, the finite node simulation dataset, and the discrete experimental dataset.

[0031] like Figure 2As shown, firstly, the global node data of the outer wall of the combustion chamber is extracted, and the global node data structure includes: timestamp, node number and node data matrix, and the node data matrix includes at least the x-coordinate, y-coordinate, z-coordinate, simulation temperature and simulation strain of the node. The original global simulation dataset is obtained by constructing the global simulation data sample. Secondly, discrete experimental datasets of the combustion chamber exterior wall under excitation are acquired using a data acquisition instrument and physical sensors. These physical sensors refer to temperature sensor points (such as thermocouples) and strain sensor points (such as strain gauges) arranged on temperature and strain measurement points on the combustion chamber exterior wall. Then, temperature and strain time-series data for a finite number of discrete measurement points are obtained from the temperature and strain time-series data obtained at each measurement point. These finite discrete measurement point temperature and strain time-series data are used as input data samples for discrete experiments to construct a discrete experimental dataset. This discrete experimental dataset contains at least two physical characteristic quantities, such as temperature and strain, with significantly different numerical orders of magnitude.

[0032] Secondly, a finite node simulation dataset corresponding to the actual physical sensor spatial locations is extracted from the original global simulation dataset through spatial node matching. For example, the simulation values ​​corresponding to the locations of four sensor measurement points are extracted as the finite node simulation dataset.

[0033] The original global simulation dataset is obtained based on the spatial coordinates (x-coordinate, y-coordinate, z-coordinate) of the global node data of the outer wall of the combustion chamber and the corresponding global simulation data samples with simulation temperature and simulation strain. The finite node simulation dataset is node data extracted from the original global simulation dataset that corresponds to the measurement points of the experimental sensors. The data structure of the finite node simulation dataset includes: timestamp, finite node number, finite node temperature simulation data, and finite node strain simulation data. S2. Spatiotemporal alignment processing for consistency of multi-source data This step corresponds to the data alignment process, where spatiotemporal alignment is achieved through linear interpolation extrapolation. Since the sampling frequency of experimental data is much higher than that of simulation data, to ensure a one-to-one correspondence between data points, a time-series-based linear interpolation function is established to map high-frequency experimental data points to low-frequency simulation timestamps, such as... Figure 3 As shown, it specifically includes: S21, Spatial Alignment Using finite node simulation data as a spatial reference, and using finite node indexes to perform spatial node matching on high-frequency discrete experimental data, the distribution of experimental sensor measurement points (i.e. temperature sensing points and strain sensing points) on the outer wall of the combustion chamber is determined for spatial node matching, thereby making the experimental measurement points correspond one-to-one with the finite nodes of the simulation.

[0034] S22, Time Alignment Since the sampling rate of actual experimental data is usually much higher than the time step of simulation data, a linear interpolation extrapolation method is used to eliminate errors caused by inconsistencies in time series. This method maps the sampled high-frequency experimental time series to low-frequency simulation data timestamps via linear interpolation. This mapping of experimental data to simulation data timestamps constructs a spatiotemporally unified standard experimental dataset. A linear interpolation function based on time series is established, and its calculation formula is as follows: in, t s It is the timestamp of the target simulation data. t 1 and t 2 are two adjacent actual timestamps in the experimental data sequence, and satisfy the following conditions: . and These are the measured values ​​of the experimental data at the corresponding time points. To align t s Standard experimental data values ​​at time points.

[0035] S3. Construct an error correction weight calculation model for feature decoupling.

[0036] This step corresponds to the feature and weight calculation stage. In practical applications, since the combustion chamber of a scramjet engine simultaneously contains physical fields such as temperature and strain, and the absolute value of strain is usually much larger than that of temperature, if weights are shared when optimizing the objective function, the large-value strain feature will dominate the weight allocation, resulting in extremely poor or even divergent correction effects for the small-value temperature feature. Therefore, a feature decoupling mechanism must be introduced to calculate the correction weights to prevent large-value features from dominating the optimization.

[0037] like Figure 4 As shown, the input data organization module in the error correction weight calculation model first organizes the input tensor according to time step, discrete node, and physical feature. Then, the feature decoupling module splits the standard experimental dataset and the finite node simulation dataset into independent temperature and strain branches according to the physical feature type, so as to extract the corresponding temperature experimental value, temperature simulation value, strain experimental value, and application simulation value, and avoid mutual interference between different dimensions and data scales. In this embodiment, the error correction weight calculation module provides two preferred calculation rules when calculating the temperature weight and strain weight respectively: Rule 1 (To avoid the optimization process being dominated by strain due to strain values ​​being much larger than temperature values, the algorithm uses the least squares method with independent features and shared nodes to calculate the error correction weights for each node): To pursue extreme local fidelity, each time point, each node, and each feature has completely independent weights. When calculating weights independently for different physical features, the objective function is to minimize the error correction weights for each specific feature. f The sum of squared errors below: The above objective function is used to target specific features. f Independent minimization of the sum of squared errors is performed to achieve decoupled calculations of physical fields of different magnitudes.

[0038] By taking the derivative, the formula for calculating the feature weights is: To achieve extreme local fidelity, the above formula is used to ensure that each time point, each node, and each feature has completely independent weights. Wherein, Indicates in t Time of the first f Correction weights for each physical characteristic (temperature or strain), Represents the first in the simulation dataset n Each node t The first moment f One physical characteristic value, This represents the corresponding value in the standard experimental dataset. N Let be the total number of discrete nodes. To prevent computational overflow due to an extremely small denominator, a condition is set such that when the denominator is ≤ 1e-6, ... = 1.0.

[0039] Rule 2 (Direct division with complete decoupling between spacetime and features): To pursue extreme local physical field fidelity, ensuring that each time point, each node, and each feature has independent weights, the calculation formula is as follows: In addition, to prevent computational overflow (division by zero error) caused by extremely small simulation feature values ​​in the calculations of the two rules mentioned above, a robust protection mechanism (also known as a minimum threshold protection mechanism) is introduced: when the absolute value of the denominator in the weight calculation formula is ≤1e-6, the error at the corresponding time step is forcibly corrected to... That is, the original simulation feature value of the node remains unchanged.

[0040] After the error correction weight calculation module independently calculates the correction weight for each feature, it outputs the temperature correction weight set and the strain correction weight set. The feature-independent weight output module obtains the corresponding feature decoupling correction weight set based on the temperature correction weight set and the strain correction weight set, and outputs the feature decoupling error correction weight to provide temperature weight and strain weight as input for the full-node data correction model.

[0041] S4. Construct a full-node data correction model based on spatial topology mapping.

[0042] This step corresponds to the spatial control domain division in the data mapping and reconstruction stage. The characteristic physical quantities of the combustion chamber exhibit a clear spatially segmented gradient distribution pattern, which traditional global interpolation or average weight allocation cannot reproduce in terms of physical reality. This embodiment employs a "spatial segmentation method" based on geometric topology for weight generalization. Its central idea is to extract the coordinates of the measuring points, calculate the adjacent geometric midpoints, and divide the spatial control domain, specifically as follows: S41. Extract the coordinate set of a finite number of discrete nodes in the principal dimension direction (such as the X-axis direction along the combustion chamber axis), and sort them in ascending order to obtain the sequence. X 1, X 2…, X m .

[0043] S42. Calculate the geometric midpoint of adjacent discrete nodes as the spatial segmentation boundary, dividing the entire physical space into multiple independent spatial control domains. Spatial segmentation boundary B i The calculation formula is: In this step, the control segment boundary is constructed by utilizing the spatial geometric midpoint of a finite number of discrete measurement points, thus overcoming the challenges of traditional complex global interpolation of physical fields.

[0044] S5, global data mapping and high-precision reconstruction output.

[0045] This step corresponds to the weight correction part in the data mapping and reconstruction stage. It mainly uses a "spatial segmentation method" to generalize the weights based on the full-node data correction model of spatial topology mapping (i.e., inheriting the correction weights of the corresponding control domain according to coordinate affiliation). In specific applications, this is achieved by traversing all unknown nodes in the original full-domain numerical simulation data. k (coordinates are) X k ), by setting its coordinates X k With all spatial segment boundaries B i By comparing the data, the spatial control domain to which the unknown node belongs is determined, and the core discrete nodes within that domain are directly inherited. j Corrected weights: in, Unknown node k Corrected physics predictions The original simulation value for the unknown node k. For the core discrete nodes within the spatial control domain in which it falls j The corrected weights are determined in this step by adjusting the weights of unknown nodes. k The control domain is automatically determined and the weight of the core measurement point j is directly inherited, ensuring the physical fidelity of the structural segmentation characteristics.

[0046] Furthermore, after completing the global mapping, the node data corresponding to the limited sensor locations in the global data are forcibly replaced with the aligned standard experimental data values ​​from step 2 to ensure 100% physical accuracy of the discrete measurement point locations.

[0047] S6. Prediction and recognition effect assessment.

[0048] This step corresponds to the high-precision global data output part in the data mapping and reconstruction stage. The error performance evaluation index is calculated by measuring the mean square error before and after correction. MSE ) and error improvement rate Improvement This is used to quantify and evaluate the quality of multiphysics reconstruction. In practical applications, when the improvement rate is greater than zero or reaches the engineering-defined threshold, it indicates that the correction is effective, and the final high-precision multiphysics data across the entire domain is output. The calculation formula is: in, MSE orig and MSE corr represents the mean square error before and after correction, respectively; M is the total number of discrete time steps; K is the total number of nodes in the entire structure; and F is the total number of physical characteristic parameters. sim corrected and exp true These represent the corrected predicted value and the true reference value for all nodes in the domain, respectively.

[0049] The above solution is merely an illustration of a preferred example and is not limited thereto. When implementing this invention, appropriate substitutions and / or modifications can be made according to the user's needs.

[0050] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. It can be applied to various fields suitable for the present invention. Other modifications can be readily made by those skilled in the art. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and examples shown and described herein.

Claims

1. A method for correcting simulation model errors based on discrete experimental data, characterized in that, include: S1. Obtain discrete experimental datasets under excitation through data acquisition instruments and physical sensors, and combine them with the original numerical simulation datasets and finite node simulation datasets to construct a multi-source heterogeneous dataset adapted to the target combustion chamber. S2. Perform spatiotemporal consistency alignment on multi-source heterogeneous datasets to obtain a spatiotemporally unified standard experimental dataset; S3. Construct a feature-decoupled error correction weight calculation model and perform calculations based on finite node simulation data and standard experimental datasets to obtain temperature correction weights and strain correction weights. S4. Construct a full-node data correction model based on spatial topology mapping, and divide the full-domain nodes corresponding to the outer wall of the combustion chamber into multiple spatial control domains; S5. Input the original numerical simulation dataset into the full-node data correction model, and map the temperature correction weight and strain correction weight of the corresponding finite nodes to the full-node nodes according to the spatial control domain to which the full-nodes belong, and perform error correction on the temperature simulation data and strain simulation data of the full-nodes respectively. S6. Calculate the error index between the finite node simulation data before and after correction and the standard experimental data, and output the corrected global multiphysics data based on the error improvement.

2. The simulation model error correction method based on discrete experimental data as described in claim 1, characterized in that, In S1, the discrete experimental dataset includes: temperature time-series data and strain time-series data acquired using temperature sensors and strain sensors; The original global simulation dataset includes: spatial coordinates of all nodes on the outer wall of the combustion chamber and corresponding temperature simulation data and strain simulation data. The data structure of the original global simulation dataset includes: timestamp, node number and node data matrix. The node data matrix includes: three-dimensional coordinates, temperature and strain of the node. The finite node simulation dataset is composed of corresponding finite nodes extracted from the original global simulation dataset based on the experimental sensor measurement point locations. The data structure of each finite node includes: simulation timestamp, finite node number, temperature simulation data and strain simulation data of the finite node. Among them, the finite nodes include: temperature measurement point corresponding nodes used to establish the correspondence between temperature experimental data and temperature simulation data; The corresponding node for strain measurement points is used to establish the correspondence between strain experimental data and strain simulation data.

3. The simulation model error correction method based on discrete experimental data as described in claim 2, characterized in that, In S2, the spatiotemporal consistency alignment process includes: spatial node matching and temporal matching; The spatial node matching process involves using the node positions in the finite node simulation dataset as a spatial reference to match the measurement point positions of the experimental sensors in the discrete experimental dataset. The time series matching refers to the process of mapping experimental time series in discrete experimental datasets to simulation timestamps. The mapping method is as follows: For any simulation timestamp, a linear interpolation function is established based on its adjacent experimental sampling time points and corresponding experimental measurements. The experimental temperature time series data and experimental strain time series data are then mapped to the timestamps of the finite node simulation dataset to construct a standard experimental dataset that is consistent with the finite node simulation dataset in the time dimension.

4. The simulation model error correction method based on discrete experimental data as described in claim 1, characterized in that, In S3, the error correction weight calculation model splits the standard experimental dataset and the finite node simulation dataset into temperature branches and strain branches according to the physical feature type, so as to obtain the corresponding temperature correction weights and strain correction weights through the temperature branches and strain branches. The temperature correction weight and strain correction weight are implemented using the least squares method or direct division. When using the least squares method, the corresponding weights are obtained using the following formula: in, express t Time of the first f Correction weights for each physical feature Represents the first in the simulation dataset n Each node t Time of the first f One physical characteristic value, This represents the corresponding value in the standard experimental dataset. N It represents the sum of discrete nodes.

5. The simulation model error correction method based on discrete experimental data as described in claim 4, characterized in that, In the calculation process of the error correction weight calculation model, a preset threshold for denominator protection is set so that when the absolute value of the denominator is less than or equal to the preset threshold, the corresponding error correction weight is set to 1.

0.

6. The simulation model error correction method based on discrete experimental data as described in claim 1, characterized in that, In S4, the spatial control domain is obtained as follows: S40. By extracting the spatial coordinates of a finite number of discrete nodes along the principal scale direction, and arranging them in ascending order, a sequence set is obtained. X i ,and ; S41. Using the geometric midpoints of adjacent discrete nodes as spatial segmentation boundaries, the global nodes are divided into multiple independent spatial control domains, and the spatial segmentation boundaries... B i The calculation formula is: .

7. The simulation model error correction method based on discrete experimental data as described in claim 1, characterized in that, In S4, the full-node data correction model traverses all unknown nodes in the original global numerical simulation data. k and will k Corresponding coordinates X k With all segment boundaries B i Perform a comparison to determine k The spatial control domain to which it belongs is then corrected using the following formula. k Physical field prediction values : in, Unknown node k The original simulation values, For the core discrete nodes within the spatial control domain in which it falls j Corrected weights.

Images