Method and system for predicting temperature deformation of large-span aluminum alloy structure

Abstract

The application relates to the technical field of structural deformation prediction, and discloses a large-span aluminum alloy structure temperature deformation prediction method and system. The method comprises the following steps: obtaining a reduced basis matrix and a response surface interpolation weight coefficient through intrinsic orthogonal decomposition based on a finite element model, calculating a physical correlation matrix among each monitoring node, performing interpolation calculation on the basis of the temperature load vector obtained by converting measured temperature data, obtaining a prediction residual, performing spatial consistency scoring on the prediction residual according to the physical correlation matrix, determining the prediction residual as a structural damage source or a sensor drift source and processing them respectively, and finally superimposing the interpolation calculation result and the error compensation amount obtained by substituting the corrected stiffness parameter to obtain a full-field temperature deformation prediction result. The application improves the reliability of the deformation prediction result of the hybrid prediction framework under long-term operation conditions and the reliability of the structural safety evaluation.

Description

Technical Field

[0001] This application relates to the field of structural deformation prediction technology, and in particular to a method and system for predicting temperature deformation of large-span aluminum alloy structures. Background Technology

[0002] Large-span aluminum alloy structures are widely used in the roofing and curtain wall systems of large public buildings such as airport terminals, stadiums, and exhibition centers due to their advantages of being lightweight, high-strength, and corrosion-resistant. The coefficient of linear expansion of aluminum alloy materials is approximately 23 × 10⁻⁶. -6 Temperatures of ℃, approximately twice that of steel, can cause significant temperature stress and deformation in structures under the influence of complex temperature fields such as solar radiation, diurnal temperature variations, and seasonal temperature changes, directly impacting their safety and functionality. Currently, methods for predicting temperature deformation in such structures mainly fall into three categories: classical theoretical calculation methods, based on thermoelastic theory, use simplified formulas to estimate deformation caused by uniform temperature rise, assuming a uniform temperature field and idealized constraints; finite element numerical simulation methods apply temperature loads to establish a finite element model of the structure for thermo-mechanical coupling analysis, which is the mainstream approach in current engineering; and statistical regression methods utilize long-term historical monitoring data to establish a statistical mapping relationship between temperature and deformation, offering faster calculation speeds but lacking a clear physical mechanism.

[0003] However, all of the aforementioned existing methods have significant shortcomings. Classical theoretical calculations suffer from large prediction errors due to the inconsistency between the assumption of a uniform temperature field and the actual non-uniform distribution. While finite element numerical simulations offer high accuracy, their complex modeling process and the time-consuming nature of a single full-model analysis make them unsuitable for the rapid response requirements of construction control and real-time safety assessment. Furthermore, their initial models, based on design drawings, inevitably exhibit systematic deviations from the actual structure after construction in terms of material properties, connection stiffness, and boundary conditions, leading to long-term mismatches between predicted and measured values. Statistical regression methods are black-box models with unclear physical meanings, exhibiting severely insufficient generalization ability when structural conditions or environmental conditions change. All three methods fail to deeply integrate real-time temperature and deformation data acquired by structural health monitoring systems into the prediction process, resulting in the underutilization of the monitoring data's value.

[0004] When a monitoring node exhibits a continuously increasing prediction residual, this residual simultaneously matches the causes of both local structural damage and asymptotic sensor drift in its temporal statistical characteristics. Existing methods cannot distinguish between these two sources. If the residual caused by damage is mistakenly identified as sensor drift, the structural damage signal is discarded; if the spurious residual caused by sensor drift is mistakenly identified as structural damage, stiffness parameter correction is incorrectly triggered, causing the surrogate model parameters to continuously shift in the wrong direction. More critically, if the online learning of the error correction model is still driven by all residuals under the above-mentioned confused state, the systematic bias caused by damage will be mistakenly learned as normal model error and actively compensated and erased. This results in the prediction accuracy appearing normal while the true structural damage state is continuously masked, thereby posing a risk of structural safety assessment failure. Summary of the Invention

[0005] This application provides a method and system for predicting the temperature deformation of large-span aluminum alloy structures. It solves the technical problem in existing methods for predicting the temperature deformation of large-span aluminum alloy structures that cannot distinguish between structural damage residuals and sensor drift residuals, which leads to the online error correction mechanism actively masking structural damage signals. This improves the credibility of deformation prediction results and the reliability of structural safety assessment under long-term operating conditions using a hybrid prediction framework.

[0006] In a first aspect, this application provides a method for predicting the temperature deformation of a large-span aluminum alloy structure, the method comprising: Step S1: Based on the finite element model of the large-span aluminum alloy structure, the nodal displacement response under multiple temperature load modes is reduced in order through intrinsic orthogonal decomposition to obtain the reduced basis matrix and response surface interpolation weight coefficients. Step S2: Based on the reduced-order basis matrix, perform pseudo-inverse projection calculation on the unit displacement response of each monitoring node to obtain the physical correlation matrix characterizing the mechanical deformation transmission ratio between each monitoring node; Step S3: Convert the measured temperature data into a temperature load vector, then input it into the reduced-order basis matrix for interpolation calculation to obtain the predicted displacement. Subtract the predicted displacement from the measured displacement to obtain the predicted residual. Score the degree of conformity of the neighborhood cooperative response of the predicted residual according to the physical correlation matrix to obtain a spatial consistency score. Determine whether the predicted residual is a source of structural damage or a source of sensor drift based on the spatial consistency score. If it is determined to be a source of structural damage, correct the stiffness parameter. If it is determined to be a source of sensor drift, isolate the corresponding node data. Step S4: Substitute the corrected stiffness parameters into the reduced-order basis matrix and the response surface interpolation weight coefficients for interpolation calculation. Add the calculation results to the error compensation amount obtained by online regression compensation of the prediction residual to obtain the full-field temperature deformation prediction result.

[0007] Secondly, this application provides a temperature deformation prediction system for large-span aluminum alloy structures, the large-span aluminum alloy structure temperature deformation prediction system comprising: The order reduction module is used for finite element models based on large-span aluminum alloy structures. It performs order reduction processing on the nodal displacement response under multiple temperature load modes through intrinsic orthogonal decomposition to obtain the reduced basis matrix and response surface interpolation weight coefficients. The projection module is used to perform pseudo-inverse projection calculation on the unit displacement response of each monitoring node based on the reduced-order basis matrix, so as to obtain the physical correlation matrix characterizing the mechanical deformation transmission ratio between each monitoring node. The determination module is used to convert the measured temperature data into a temperature load vector, then input it into the reduced-order basis matrix for interpolation calculation to obtain the predicted displacement. The difference between the predicted displacement and the measured displacement is used to obtain the predicted residual. The degree of conformity of the neighborhood cooperative response of the predicted residual is scored according to the physical correlation matrix to obtain a spatial consistency score. Based on the spatial consistency score, the predicted residual is determined to be either a source of structural damage or a source of sensor drift. When it is determined to be a source of structural damage, the stiffness parameter is corrected. When it is determined to be a source of sensor drift, the corresponding node data is isolated. The compensation module is used to substitute the corrected stiffness parameters into the reduced-order basis matrix and the response surface interpolation weight coefficients for interpolation calculation, and then superimpose the calculation results with the error compensation amount obtained by online regression compensation of the prediction residuals to obtain the full-field temperature deformation prediction results.

[0008] Thirdly, a device for predicting the temperature deformation of a large-span aluminum alloy structure is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the device for predicting the temperature deformation of a large-span aluminum alloy structure to execute the above-described method for predicting the temperature deformation of a large-span aluminum alloy structure.

[0009] Fourthly, a computer-readable storage medium is provided, wherein instructions are stored therein, which, when executed on a computer, cause the computer to perform the above-described method for predicting temperature deformation of large-span aluminum alloy structures.

[0010] The technical solution provided in this application compresses the high-dimensional nodal displacement response to a low-dimensional reduced-order basis matrix by introducing intrinsic orthogonal decomposition into the finite element model. Combined with response surface interpolation weighting coefficients, a parameterized rapid calculation pathway is constructed. This allows the full-field nodal displacement under any combination of temperature load vector and stiffness parameters to be directly obtained through interpolation without resolving the finite element equations. This fundamentally overcomes the bottleneck of traditional finite element numerical simulation methods, which cannot meet real-time prediction requirements due to excessive computational costs. Furthermore, the introduction of the physical correlation matrix is ​​the core contribution of this application. By performing pseudo-inverse projection calculations on the reduced-order basis matrix, the deformation transfer constraints implicit in the nodal equilibrium equations of structural mechanics are explicitly transformed into a matrix form of the mechanical deformation transfer ratios between each monitored node. This converts the implicit physical constraints, which originally existed only within the finite element equations, into explicit tools that can directly participate in data judgment.

[0011] The spatial consistency score calculation quantifies the degree of agreement between the predicted residual of anomaly nodes and the expected collaborative response values ​​of each neighboring node in its mechanical response neighborhood set into a single scalar. This allows for the identification and differentiation of two types of anomalies—structural damage and sensor drift—that are indistinguishable in terms of temporal statistical characteristics, based on the spatial propagation laws of mechanical constraints. When a source of structural damage is identified, only the stiffness parameters corresponding to the damaged area are corrected in a directional manner, rather than being updated globally, thus avoiding interference from local damage to the overall model parameters. When a source of sensor drift is identified, the corresponding node data is isolated from the parameter identification dataset and reconstructed through weighted interpolation using the physical correlation matrix, ensuring that the continuity of the monitoring network data is not affected by single-point sensor failures. The above-mentioned bifurcation processing mechanism, together with the error compensation obtained from online regression compensation, works on the final superimposed output, enabling the full-field temperature deformation prediction results to maintain physical consistency even under complex conditions where sensor failures and structural damage coexist. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a schematic diagram of an embodiment of the method for predicting temperature deformation of large-span aluminum alloy structures in this application. Figure 2 This is a schematic diagram of the convergence process of gradient descent iterative correction of stiffness parameters in the damaged region in an embodiment of this application. Detailed Implementation

[0014] This application provides a method and system for predicting temperature deformation of large-span aluminum alloy structures. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings 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 described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a 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.

[0015] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the method for predicting temperature deformation of large-span aluminum alloy structures in this application includes: Step S1: Based on the finite element model of the large-span aluminum alloy structure, the nodal displacement response under multiple temperature load modes is reduced in order through intrinsic orthogonal decomposition to obtain the reduced basis matrix and response surface interpolation weight coefficients. Specifically, the finite element model is established based on the design drawings of the large-span aluminum alloy structure, and includes the equivalent radial stiffness and tangential stiffness of the main load-bearing supports, the moment release coefficient of the connection nodes, and the equivalent linear expansion coefficient of the aluminum alloy (nominal value). The parameters to be identified are defined as stiffness parameter vectors. Multiple sets of sampled parameters are generated within the feasible region of the stiffness parameter vectors using Latin hypercube sampling. Each set of sampled parameters is combined with multiple temperature load modes and then input into the finite element model for static analysis to obtain a nodal displacement response dataset. This dataset is then subjected to eigenorthogonal decomposition, and basis vectors are extracted based on a cutoff condition of at least 95% cumulative energy, resulting in a reduced-order basis matrix. The reduced-order basis matrix is ​​a set of orthogonal bases that compresses the high-dimensional nodal displacement space into a low-dimensional space; all subsequent displacement calculations are performed within this low-dimensional space. The response surface interpolation weighting coefficients are obtained through Gaussian kernel function least-squares fitting. Their function is to establish a mapping relationship between the stiffness parameter vector and the temperature load modes to the reduced-order coordinate coefficients, so that any new input can be obtained without resolving the finite element equations to obtain the corresponding nodal displacement.

[0016] Step S2: Based on the reduced-order basis matrix, perform pseudo-inverse projection calculation on the unit displacement response of each monitoring node to obtain the physical correlation matrix characterizing the mechanical deformation transmission ratio between each monitoring node. Specifically, each element of the physical correlation matrix represents the mechanical deformation transfer ratio between two monitoring nodes under temperature deformation, calculated by the pseudo-inverse projection of the reduced-order basis matrix. Specifically, for each monitoring node, a unit response vector centered at that node is constructed. This vector is multiplied by the Moore-Penrose pseudo-inverse matrix of the reduced-order basis submatrix of the monitoring node to obtain the reduced-order coordinate projection value. This projection value is then substituted into the reduced-order basis matrix to reconstruct the full-field displacement. The vertical displacement components of each neighboring node are extracted from the reconstruction result to obtain the corresponding transfer ratio. Transfer ratios with an absolute value below 0.15 are set to zero. This threshold is set based on the fact that the deformation field of the aluminum alloy large-span structure under temperature load is continuous, and there is no significant mechanical correlation between node pairs with transfer ratios below 0.15. Setting the value to zero eliminates the interference of numerical calculation noise on subsequent judgments.

[0017] Step S3: Convert the measured temperature data into a temperature load vector, then input it into a reduced-order basis matrix for interpolation calculation to obtain the predicted displacement. Subtract the predicted displacement from the measured displacement to obtain the predicted residual. Score the degree of conformity of the neighborhood cooperative response of the predicted residual based on the physical correlation matrix to obtain a spatial consistency score. Based on the spatial consistency score, determine whether the predicted residual is a source of structural damage or a source of sensor drift. If it is determined to be a source of structural damage, correct the stiffness parameter. If it is determined to be a source of sensor drift, isolate the corresponding node data. Specifically, when the predicted residual of a monitoring node exceeds a threshold, the expected collaborative response values ​​of each neighboring node in the mechanical response neighborhood set of that node are calculated based on the physical correlation matrix. The deviation rate between the measured residuals and the expected collaborative response values ​​of the neighboring nodes is averaged and subtracted from 1 to obtain a spatial consistency score. The physical meaning of this score is: the deformation of the aluminum alloy structure under temperature load satisfies the nodal equilibrium equation; the local stiffness decrease caused by actual structural damage will inevitably lead to collaborative deformation of adjacent nodes that conforms to the mechanical transmission ratio; however, sensor drift is a single-point event and will not produce collaborative responses that satisfy the physical correlation constraints in neighboring nodes. When the spatial consistency score is not lower than 0.6, it is determined to be a source of structural damage. The threshold of 0.6 means that the deviation rate between the measured residual and the expected cooperative response value of more than 60% of the nodes in the neighborhood is less than 100%, that is, the overall cooperative response mode shows physical constraint conformity. When it is lower than 0.6, it is determined to be a source of sensor drift. The corresponding node data is isolated from the parameter identification dataset, and weighted interpolation is performed according to the ratio of the measured displacement of normal nodes in the physical correlation matrix to the corresponding mechanical deformation transmission ratio to obtain the reconstructed displacement to replace the measured displacement.

[0018] Step S4: Substitute the corrected stiffness parameters into the reduced-order basis matrix and the response surface interpolation weight coefficients for interpolation calculation. Add the calculation results to the error compensation amount obtained by online regression compensation of the prediction residuals to obtain the full-field temperature deformation prediction results.

[0019] Specifically, the corrected stiffness parameters and the target temperature load vector are substituted into the reduced-order basis matrix and response surface interpolation weight coefficients for interpolation calculation to obtain the interpolation results of the nodal displacements across the entire field. Online regression compensation specifically refers to using the combination of the temperature load vectors of normal nodes and the corrected stiffness parameters as input, and the corresponding predicted residuals as output. A spatial prediction function for the system deviation is established through Gaussian process regression. The output of this function at all nodes in the entire field is used as the error compensation amount, and it is superimposed node by node with the interpolation results of the nodal displacements across the entire field to obtain the predicted results of the temperature deformation across the entire field. The role of Gaussian process regression here is to utilize the known residual information at a small number of normal nodes, establish the spatial correlation structure of the residuals between nodes through a kernel function, infer the system deviation at locations where no sensors are deployed, and thus extend the error information from a limited number of monitoring points to the entire field.

[0020] In one specific embodiment, step S1 includes: Based on the design drawings of the large-span aluminum alloy structure, the equivalent radial stiffness, tangential stiffness, moment release coefficient of the connection node, and equivalent linear expansion coefficient of the aluminum alloy of the main load-bearing support node are parametrically defined to obtain the stiffness parameter vector. Based on the feasible region of stiffness parameter vector, multiple sets of sampling parameters are generated by Latin hypercube sampling. The combination of each sampling parameter and multiple temperature load modes is substituted into the finite element model for static analysis to obtain the nodal displacement response dataset. The nodal displacement response dataset is subjected to eigenorthogonal decomposition, and the first few order basis vectors are extracted according to the truncation condition that the cumulative energy ratio is not less than 95%, so as to obtain the reduced order basis matrix. Based on the reduced-order basis matrix, the response surface interpolation weight coefficients are obtained by using the sampling parameters and the corresponding temperature load mode as inputs and the corresponding reduced-order coordinate coefficients as outputs through Gaussian kernel function least squares fitting.

[0021] Specifically, the stiffness parameter vector consists of four types of parameters: the equivalent radial stiffness of the main load-bearing support node (ranging from 50% to 150% of the design nominal value, in N / m), the tangential stiffness (ranging from the same value), the moment release coefficient of the connection node (dimensionless, ranging from 0 to 1, where 0 represents a fully hinged connection and 1 represents a fully rigid connection), and the equivalent linear expansion coefficient of the aluminum alloy (ranging from the same value). to nominal value Within the aforementioned feasible region, a number of sampling parameter sets are generated through Latin hypercube sampling. The sampling scale is 10 times the total dimension of the stiffness parameter vector. This ratio ensures uniform coverage of sampling points in the high-dimensional parameter space and avoids excessive interpolation errors in the response surface due to insufficient sampling. Each set of sampling parameters is combined with multiple temperature load modes and substituted into the finite element model for static analysis. The temperature load modes are extracted by singular value decomposition of historical measured temperature field data. The top few singular vectors with a cumulative energy ratio of not less than 95% are taken as orthogonal load bases. This truncation condition ensures that the temperature load mode set fully represents the actual temperature field changes.

[0022] When performing intrinsic orthogonal decomposition on the nodal displacement response dataset, the truncation condition is set to a cumulative energy ratio of no less than 95%. This means that the sum of the eigenvalues ​​corresponding to the retained basis vectors accounts for no less than 95% of the total sum of all eigenvalues. This threshold minimizes the computational degrees of freedom while ensuring the accuracy of displacement field reconstruction. The reduced-order basis matrix is ​​composed of the first few order basis vectors that satisfy the above truncation condition, arranged column by column. The number of columns is the number of degrees of freedom after the order reduction. The process of obtaining the response surface interpolation weight coefficients is as follows: the spliced ​​vector of each sampling parameter and the corresponding temperature load mode is used as the input feature, and the projected coordinates of the corresponding finite element analysis results on the reduced-order basis matrix (i.e., the reduced-order coordinate coefficients) are used as the output label. A similarity matrix between sampling points is constructed using a Gaussian kernel function. The least squares method is used to solve for the combination of weight coefficients that minimizes the sum of squared prediction errors of the output labels, thus obtaining the response surface interpolation weight coefficients. For the newly input stiffness parameters and temperature load vectors, the weighted summation of these parameters and the Gaussian kernel function values ​​at each sampling point is multiplied by a reduced-order basis matrix, and the prediction results of the nodal displacements across the entire field can be obtained within 0.5 seconds.

[0023] In one specific embodiment, step S2 includes: The row vectors corresponding to the vertical degrees of freedom of each monitoring node are extracted from the reduced basis matrix to obtain the reduced basis submatrix of the monitoring node; The Moore-Penrose pseudo-inverse operation is performed on the reduced-order basis submatrix of the monitoring node to obtain the reduced-order basis pseudo-inverse matrix of the monitoring node; For each monitoring node, the unit response vector centered on the node is multiplied by the reduced-order basis pseudo-inverse matrix of the monitoring node to obtain the reduced-order coordinate projection value of the node. The reduced-order coordinate projection value is substituted into the reduced-order basis matrix to reconstruct the full-field displacement. The vertical displacement components of each neighboring node are extracted from the reconstruction results to obtain the mechanical deformation transfer ratio corresponding to each monitoring node. Arrange the mechanical deformation transfer ratios corresponding to all monitoring nodes according to the node number, and set the transfer ratios with an absolute value less than 0.15 to zero to filter out weakly correlated node pairs, thereby obtaining the physical correlation matrix and the set of mechanical response neighborhoods for each monitoring node.

[0024] Specifically, the number of rows in the reduced-order basis matrix equals the total number of degrees of freedom in the finite element model, where the vertical degree of freedom of each monitoring node corresponds to a unique row index in the global degree-of-freedom numbering. The reduced-order basis submatrix for monitoring nodes is obtained by extracting the row vectors corresponding to the vertical degrees of freedom of each monitoring node from the reduced-order basis matrix according to the row index and arranging them. Its number of rows equals the total number of monitoring nodes, and its number of columns equals the number of columns in the reduced-order basis matrix, i.e., the number of degrees of freedom after reduction. Performing Moore-Penrose pseudo-inverse operation on this submatrix yields the reduced-order basis pseudo-inverse matrix for monitoring nodes. The role of the pseudo-inverse operation here is to establish an inverse mapping relationship in the sense of minimum norm between the vertical degree-of-freedom subspace of the monitoring nodes and the reduced-order coordinate space. That is, given the vertical displacement of the monitoring nodes, it solves for the compatible reduced-order coordinate vector. The operation does not require the number of monitoring nodes to be equal to the reduced-order dimension, which is suitable for the sparse distribution case where the number of monitoring nodes is less than the reduced-order dimension in this scheme.

[0025] When constructing a unit response vector for each monitoring node, the dimension of this vector is equal to the total number of monitoring nodes. Only the component corresponding to the current node has a value of 1, while all other components are 0. The physical meaning of this construction method is that it assumes a unit vertical displacement occurs at the current node while the displacements of other monitoring nodes are zero. Multiplying the unit response vector by the pseudo-inverse matrix of the reduced-order basis of the monitoring node yields the reduced-order coordinate projection value of the current node. This projection value is a coordinate vector in the reduced-order coordinate space compatible with the unit displacement assumption. Multiplying the reduced-order coordinate projection value by the reduced-order basis matrix completes the full-field displacement reconstruction. From the reconstructed full-field displacement, the displacement components at the vertical degrees of freedom of each other monitoring node are extracted, thus obtaining the mechanical deformation transfer ratio of the current node to its neighboring nodes. The physical meaning of this ratio is the amount of cooperative displacement that neighboring nodes will inevitably produce under mechanical constraints when the current node experiences a unit vertical displacement; its absolute value reflects the strength of the mechanical connection between the two nodes. The zero threshold was set to 0.15 because the temperature deformation field of large-span aluminum alloy structures is spatially continuous. The mechanical correlation between node pairs with a transfer ratio absolute value below 0.15 is negligible, and retaining such weak correlations would introduce invalid interference terms in subsequent spatial consistency scoring. The physical correlation matrix is ​​obtained by arranging the mechanical deformation transfer ratios of all monitored nodes into a matrix according to node numbers. The set of nodes corresponding to the column indices of each row with a transfer ratio absolute value not lower than 0.15 is the set of the mechanical response neighborhood of that row's node.

[0026] In one specific embodiment, step S3 involves converting the measured temperature data into a temperature load vector, then inputting it into a reduced-order basis matrix for interpolation calculation to obtain the predicted displacement. The predicted residual is obtained by subtracting the predicted displacement from the measured displacement. This includes: Based on the measured temperature data at each temperature measuring point, the temperature distribution of the entire structure is interpolated by Kriging using a spherical variogram model to obtain the temperature distribution data of the entire field. Multiply the full-field temperature distribution data with the singular vector matrix of the temperature load mode to obtain the temperature load vector; The temperature load vector and stiffness parameter vector are substituted into the reduced-order basis matrix and the response surface interpolation weight coefficients to perform Gaussian kernel function interpolation calculation, and the predicted displacement of each monitoring node is obtained. The predicted displacement is subtracted from the measured displacement at each monitoring node to obtain the predicted residual.

[0027] Specifically, the measured temperature data at each temperature measuring point are discrete point temperature values. The goal of Kriging interpolation is to extend this discrete point data to all finite element node locations of the structure to obtain the full-field temperature distribution data. The expression for the spherical variogram model is as follows: when the spatial distance does not exceed the range, the variogram value equals the nugget value plus the sill value multiplied by 1.5 times the ratio of distance to range minus 0.5 times the cube of that ratio; when the distance exceeds the range, the variogram value equals the nugget value plus the sill value. The nugget value is set to 0.1℃. 2 The variogram value reflects the measurement error variance of the temperature sensor itself; the sill value is determined by fitting historical meteorological monitoring data, reflecting the overall amplitude of spatial variation in the structural temperature field; the range is one-third of the calculated span of the structure, reflecting the effective action distance of the spatial correlation of the temperature field. This proportion is determined based on the empirical law of the spatial correlation length of the temperature field under solar radiation conditions for large-span aluminum alloy roof structures. Kriging interpolation is based on the above variogram model, and the temperature values ​​at each finite element node are weighted and estimated. The weights are obtained by solving the equation system composed of the variogram values ​​between each measuring point and the target node. The final output is a full-field temperature distribution data vector with a dimension equal to the total number of nodes in the finite element model.

[0028] The temperature load vector is obtained by multiplying the full-field temperature distribution data with the singular vector matrix of the temperature load mode. The singular vector matrix of the temperature load mode is derived from the right singular vector matrix extracted from the singular value decomposition of the historical measured temperature field data in step S1. Its number of columns is equal to the order of the selected temperature load mode, and its number of rows is equal to the total number of finite element nodes. This matrix projects the full-field temperature distribution data onto the temperature load mode space, and the resulting temperature load vector is the decomposition coefficient of the full-field temperature distribution on each orthogonal temperature load mode. Its dimension is equal to the order of the temperature load mode. During Gaussian kernel function interpolation calculation, the temperature load vector and the stiffness parameter vector are concatenated into a single input feature vector. The Gaussian kernel function value between this feature vector and the feature vector of each sampling point in step S1 is calculated. The kernel function values ​​of each sampling point are weighted and summed using the response surface interpolation weight coefficient to obtain the reduced-order coordinate coefficient corresponding to the current input. Multiplying this reduced-order coordinate coefficient with the reduced-order basis matrix yields the full-field node displacement. The displacement components at the vertical degrees of freedom of each monitoring node are extracted from this to obtain the predicted displacement. The difference between the predicted displacement and the measured displacement is calculated according to the one-to-one correspondence of the monitoring node numbers. The measured vertical displacement of the same node number is subtracted from the predicted vertical displacement to obtain the prediction residual at each monitoring node. The unit is millimeters. A positive value indicates that the measured displacement is greater than the predicted displacement, and a negative value indicates that the measured displacement is less than the predicted displacement.

[0029] In one specific embodiment, step S3 scores the degree of conformity of the predicted residual's neighborhood cooperative response based on the physical correlation matrix to obtain a spatial consistency score, including: For abnormal nodes in the prediction residual whose absolute value exceeds the residual threshold, the residual value of the abnormal node in the prediction residual is multiplied by the mechanical deformation transmission ratio corresponding to each neighboring node according to the set of mechanical response neighborhoods corresponding to the abnormal node in the physical correlation matrix, so as to obtain the expected collaborative response value of each neighboring node. The deviation rate of each neighboring node is obtained by subtracting the predicted residual of each neighboring node from the corresponding expected collaborative response value and then dividing by the residual threshold. The spatial consistency score is obtained by subtracting the mean deviation rate of each neighboring node in the set of mechanical response neighborhoods from 1.

[0030] Specifically, the residual threshold is set to three times the standard deviation of the predicted residuals of each monitoring node in historical normal operation data. The standard deviation is calculated by rolling over at least 30 days of historical monitoring data, with 2 mm as the default value in the initial stage when historical data accumulation is insufficient. This multiple setting is based on the 3σ principle of normal distribution, that is, when the structure is in a normal state and the sensor is working properly, the probability of residuals exceeding three times the standard deviation is less than 0.3%, which serves as the trigger condition for anomaly judgment. The judgment of anomaly nodes is based on the absolute value exceeding their respective residual thresholds. The residual thresholds of each node are calculated independently, reflecting the differences in the uncertainty of temperature deformation prediction at different nodes. For nodes judged as anomalies, all non-zero elements in the row where the node is located are read from the physical correlation matrix. The column index of the non-zero element is the member of the mechanical response neighborhood set of the node, and the corresponding non-zero element value is the mechanical deformation transmission ratio of each neighboring node. Multiply the predicted residual value of the abnormal node by the mechanical deformation transfer ratio of each neighboring node to obtain the expected collaborative response value of each neighboring node. The physical meaning of this product is that if the predicted residual at the abnormal node is entirely due to the change in local stiffness of the structure, then according to the mechanical deformation transfer relationship between nodes, the residual amount that should be generated synchronously at the neighboring node is the same.

[0031] The deviation rate of each neighboring node is obtained by dividing the absolute value of the difference between the measured predicted residual of that neighboring node and the corresponding expected collaborative response value by the residual threshold of the abnormal node. The reason for using the residual threshold as the normalization denominator is to eliminate the influence of the difference in residual magnitude between different nodes on the deviation rate calculation result, so that the deviation rates of all neighboring nodes are on a uniform dimensionless scale and are comparable. The arithmetic mean of the deviation rates of all neighboring nodes in the mechanical response neighborhood set is taken, and then the mean is subtracted from 1 to obtain the spatial consistency score. Theoretically, the value range of this score is from negative infinity to 1. When the measured residuals of all neighboring nodes perfectly match the expected collaborative response value, the deviation rate is 0, and the spatial consistency score takes the maximum value of 1; when the deviation between the measured residuals of each neighboring node and the expected collaborative response value reaches the residual threshold, the mean is close to 1, and the spatial consistency score is close to 0; when the deviation exceeds the residual threshold, the score is negative. In actual judgment, components with a deviation rate exceeding 1 are truncated to 1 before taking the average value to ensure that the spatial consistency score is between 0 and 1. The truncation operation is based on the fact that a deviation rate exceeding 1 indicates that the deviation between the measured residual of the neighboring node and the expected collaborative response value has exceeded the residual threshold level, and is regarded as having no collaborative response. The corresponding contribution can be included in the maximum deviation, without distinguishing the degree of difference of larger deviations.

[0032] In one specific embodiment, step S3 determines the predicted residual as either a source of structural damage or a source of sensor drift based on the spatial consistency score. When determined to be a source of structural damage, the stiffness parameter is corrected; when determined to be a source of sensor drift, the corresponding node data is isolated, including: The spatial consistency score is compared with a preset judgment threshold. When the spatial consistency score is not lower than the preset judgment threshold, the prediction residual is judged as the source of structural damage. When the spatial consistency score is lower than the preset judgment threshold, the prediction residual is judged as the source of sensor drift. When the source of structural damage is determined, the set of members in the damaged area is determined based on the set of mechanical response neighborhoods. A local loss function is constructed based on the difference between the measured displacement and the predicted displacement of each monitoring node in the damaged area. The equivalent radial stiffness corresponding to the damaged area in the stiffness parameter vector is iteratively corrected by the gradient descent method to obtain the corrected stiffness parameters. When the sensor drift is identified as the source, the measured displacement of the node is isolated from the parameter identification dataset of the stiffness parameter vector. The reconstructed displacement of the node is obtained by weighted interpolation based on the measured displacement of the normal nodes in the mechanical response neighborhood set of the node in the physical correlation matrix and the corresponding mechanical deformation transmission ratio. The reconstructed displacement is then used to replace the measured displacement in subsequent calculations.

[0033] Specifically, a preset threshold of 0.6 is set. This value physically means that at least 60% of the measured residuals of the neighboring nodes within the mechanical response neighborhood set have a deviation rate of less than 1 from the expected collaborative response value, indicating an overall collaborative response pattern consistent with physical correlation constraints. The basis for this threshold is that the deformation field of a large-span aluminum alloy structure under temperature load satisfies the linear elastic nodal equilibrium equation. The local stiffness decrease caused by actual structural damage inevitably leads to collaborative deformation at most nodes within the mechanical response neighborhood set, conforming to the physical correlation matrix constraints. Sensor drift, however, is an independent fault of a single measuring device and lacks a physical mechanism for propagation to neighboring nodes; therefore, the measured residuals of neighboring nodes will not exhibit a collaborative pattern that satisfies the mechanical deformation transmission ratio constraint. After determining the source of structural damage, the set of members within the damaged area is determined by using the spatial coordinates of each abnormal node in the mechanical response neighborhood set as the center and 1.5 times the average spacing of the structural mesh as the search radius. Members whose nodes at both ends fall within the above search range are searched in the finite element model. All members that meet the conditions constitute the set of members within the damaged area. The search radius coefficient of 1.5 is set based on the fact that the node spacing of the aluminum alloy reticulated shell structure is approximately equal to the member length, and 1.5 times the spacing can cover all members directly connected to the abnormal node.

[0034] The local loss function is defined as the sum of the squares of the differences between the measured displacements of each monitoring node within the damaged region and the predicted displacements corresponding to the current stiffness parameter vector. The learning rate for gradient descent is 0.01, and the maximum number of iterations is 50. In each iteration, the gradient is analytically calculated using the partial derivatives of the response surface interpolation weight coefficients with respect to the stiffness parameter. The updated value is obtained by subtracting the product of the learning rate and the gradient from the current equivalent radial stiffness. The iteration continues until the change in the loss function is less than... The process may stop when the maximum number of iterations is reached, outputting the corrected equivalent radial stiffness as the value of the corresponding component in the corrected stiffness parameter, while keeping the other stiffness parameter components unchanged. When a sensor drift is identified as the source, the operation of isolating the measured displacement of the node from the parameter identification dataset involves marking the displacement data of that node at that moment as unreliable and excluding it from the current stiffness parameter identification calculation, without affecting the normal data of that node at previous moments in the historical dataset. The weighted interpolation of the reconstructed displacement is based on the measured displacement of normal nodes within the mechanical response neighborhood set. The weight of each normal node is taken as the absolute value of the mechanical deformation transfer ratio between it and the drift node. The reconstructed displacement is obtained by summing the products of the measured displacement of each normal node and its corresponding weight and dividing by the sum of all weights. The basis of this weighting method is that the larger the absolute value of the mechanical deformation transfer ratio, the stronger the mechanical correlation between the two nodes, and the greater the contribution of the measured displacement of the corresponding normal node to the estimation of the true displacement of the drift node.

[0035] Figure 2 This diagram illustrates the convergence process of gradient descent iterative correction of stiffness parameters in the damaged region in an embodiment of this application. The horizontal axis represents the number of gradient descent iterations, the vertical axis represents the local loss function value, the three curves correspond to the correction loss of the equivalent radial stiffness of the three support nodes, and the dashed line represents the convergence threshold of 1.0 mm. 2 As shown in the figure, the local loss function values ​​of the three support nodes all decrease monotonically with the increase of the number of iterations. When the number of iterations reaches about 35, the loss function value drops below the convergence threshold, indicating that the gradient descent method completes the directional correction of stiffness parameters within the constraint of a maximum number of iterations of 50. The correction process of each support node is independent and the convergence behavior is consistent.

[0036] In one specific embodiment, step S4 includes: The modified stiffness parameters and the target temperature load vector are substituted into the reduced-order basis matrix and the response surface interpolation weight coefficients to perform Gaussian kernel function interpolation calculations to obtain the interpolation results of the nodal displacements in the whole field. Based on the predicted residuals of normal nodes within the mechanical response neighborhood set, the combination of temperature load vector and corrected stiffness parameter is used as input and the corresponding predicted residuals are used as output. Gaussian process regression is used to perform online regression compensation for the systematic deviation of the whole field node displacement interpolation results, and the error compensation amount is obtained. The interpolation results of the total field nodal displacements are superimposed on the error compensation amount node by node to obtain the prediction results of the total field temperature deformation.

[0037] Specifically, the target temperature load vector is obtained from three sources: when the predicted target is the measured temperature condition at the current moment, the temperature load vector calculated at the current moment is directly taken; when the predicted target is the temperature condition predicted by future weather forecasts, the weather forecast temperature data is interpolated according to the location of each temperature measuring point and then input into the Kriging interpolation and singular vector matrix projection process to obtain the corresponding target temperature load vector; when the predicted target is the assumed temperature of the design condition, the assumed temperature field is directly projected onto the singular vector matrix of the temperature load model to obtain the target temperature load vector. In all three cases, the dimension of the target temperature load vector is equal to the order of the temperature load model, which is completely consistent with the calculation process of the temperature load vector in step S3. The corrected stiffness parameters are concatenated with the target temperature load vector to form the input feature vector. The Gaussian kernel function value between this feature vector and the feature vector of the corresponding sampling point of the response surface interpolation weight coefficient is calculated. The weighted summation with the weight coefficient is used to obtain the reduced-order coordinate coefficient. The reduced-order coordinate coefficient is multiplied by the reduced-order basis matrix to output the full-field nodal displacement interpolation result with a dimension equal to the total number of degrees of freedom of all nodes in the finite element model in three directions. Unlike the predicted displacement of only the monitoring node components in step S3, the displacement components of all nodes are retained here.

[0038] The training samples for Gaussian process regression consist of data from nodes identified as normal within the mechanical response neighborhood set. Each normal node contributes one training sample. The input is the concatenated vector of the current temperature load vector and the corrected stiffness parameter at that node, and the output is the predicted residual value at that node. Gaussian process regression uses the Matérn 5 / 2 kernel function, which makes moderate assumptions about the smoothness of the input features and is suitable for modeling the spatial distribution of residuals driven by both stiffness parameters and temperature load vectors. After training, the input feature vectors corresponding to the positions of all nodes in the entire field are substituted into the trained Gaussian process regression to obtain the predicted system deviation value at each node, i.e., the error compensation amount. Its dimension is consistent with the interpolation result of the node displacement in the entire field. For nodes without sensors, the error compensation amount is inferred by Gaussian process regression from the residual information of surrounding normal nodes through the kernel function correlation structure. For nodes identified as having sensor drift, the error compensation amount at their position is zero, and only the interpolation result of the node displacement in the entire field is used as the predicted value at that position. The interpolation results of the total nodal displacements and the error compensation are added one by one according to the node number to obtain the total temperature deformation prediction result. The dimension of this result is equal to the total number of degrees of freedom in the three directions of all nodes in the finite element model, covering the predicted temperature deformation value of each finite element node in the structure in the three directions.

[0039] The above describes the method for predicting the temperature deformation of large-span aluminum alloy structures in the embodiments of this application. The following describes the system for predicting the temperature deformation of large-span aluminum alloy structures in the embodiments of this application. One embodiment of the system for predicting the temperature deformation of large-span aluminum alloy structures in the embodiments of this application includes: The order reduction module is used for finite element models based on large-span aluminum alloy structures. It performs order reduction processing on the nodal displacement response under multiple temperature load modes through intrinsic orthogonal decomposition to obtain the reduced basis matrix and response surface interpolation weight coefficients. The projection module is used to perform pseudo-inverse projection calculation on the unit displacement response of each monitoring node based on the reduced-order basis matrix, so as to obtain the physical correlation matrix characterizing the mechanical deformation transmission ratio between each monitoring node. The determination module is used to convert the measured temperature data into a temperature load vector, then input it into the reduced-order basis matrix for interpolation calculation to obtain the predicted displacement. The difference between the predicted displacement and the measured displacement is used to obtain the predicted residual. The degree of conformity of the neighborhood cooperative response of the predicted residual is scored according to the physical correlation matrix to obtain a spatial consistency score. Based on the spatial consistency score, the predicted residual is determined to be either a source of structural damage or a source of sensor drift. When it is determined to be a source of structural damage, the stiffness parameter is corrected. When it is determined to be a source of sensor drift, the corresponding node data is isolated. The compensation module is used to substitute the corrected stiffness parameters into the reduced-order basis matrix and the response surface interpolation weight coefficients for interpolation calculation, and then superimpose the calculation results with the error compensation amount obtained by online regression compensation of the prediction residuals to obtain the full-field temperature deformation prediction results.

[0040] This invention also provides a large-span aluminum alloy structure temperature deformation prediction device, which can be a server. The device includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores the data corresponding to this embodiment. The network interface communicates with external terminals via a network connection. The computer program is executed by the processor to implement the above-described method.

[0041] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the method for predicting the temperature deformation of the large-span aluminum alloy structure.

[0042] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0043] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, 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 a large-span aluminum alloy structure temperature deformation prediction device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0044] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting temperature deformation of large-span aluminum alloy structures, characterized in that, The method includes: Step S1: Based on the finite element model of the large-span aluminum alloy structure, the nodal displacement response under multiple temperature load modes is reduced in order through intrinsic orthogonal decomposition to obtain the reduced basis matrix and response surface interpolation weight coefficients. Step S2: Based on the reduced-order basis matrix, perform pseudo-inverse projection calculation on the unit displacement response of each monitoring node to obtain the physical correlation matrix characterizing the mechanical deformation transfer ratio between each monitoring node. This includes: extracting row vectors corresponding to the vertical degrees of freedom of each monitoring node from the reduced-order basis matrix to obtain the reduced-order basis submatrix of the monitoring node; performing Moore-Penrose pseudo-inverse operation on the reduced-order basis submatrix of the monitoring node to obtain the reduced-order basis pseudo-inverse matrix of the monitoring node, so as to establish an inverse mapping relationship in the sense of minimum norm between the vertical degree of freedom subspace of the monitoring node and the reduced-order coordinate space; for each monitoring node, taking that node as the center... The unit response vector of the monitoring node is multiplied by the reduced-order basis pseudo-inverse matrix to obtain the reduced-order coordinate projection value of the node. The reduced-order coordinate projection value is substituted into the reduced-order basis matrix to perform full-field displacement reconstruction. The vertical displacement components of each neighboring node are extracted from the reconstruction result to obtain the mechanical deformation transfer ratio corresponding to each monitoring node. The mechanical deformation transfer ratios corresponding to all monitoring nodes are arranged by node number. Transfer ratios with an absolute value lower than 0.15 are set to zero to filter out weakly correlated node pairs, thus obtaining the physical correlation matrix and the set of mechanical response neighborhoods of each monitoring node. The number of monitoring nodes is less than the number of columns of the reduced-order basis matrix. Step S3: Convert the measured temperature data into a temperature load vector, then input it into the reduced-order basis matrix for interpolation calculation to obtain the predicted displacement. Subtract the measured displacement from the predicted displacement to obtain the predicted residual. Score the degree of conformity of the predicted residual to the neighborhood cooperative response based on the physical correlation matrix to obtain a spatial consistency score. Determine whether the predicted residual is a source of structural damage or sensor drift based on the spatial consistency score. If it is determined to be a source of structural damage, correct the stiffness parameters; if it is determined to be a source of sensor drift, isolate the corresponding node data. Specifically, for the... For abnormal nodes whose absolute value in the prediction residual exceeds the residual threshold, based on the mechanical response neighborhood set corresponding to the abnormal node in the physical correlation matrix, the residual value of the abnormal node in the prediction residual is multiplied by the mechanical deformation transmission ratio corresponding to each neighborhood node to obtain the expected collaborative response value of each neighborhood node; the difference between the prediction residual of each neighborhood node and the corresponding expected collaborative response value is divided by the residual threshold to obtain the deviation rate of each neighborhood node; the average deviation rate of each neighborhood node in the mechanical response neighborhood set is taken and subtracted from 1 to obtain the spatial consistency score. Step S4: Substitute the corrected stiffness parameters into the reduced-order basis matrix and the response surface interpolation weight coefficients for interpolation calculation. Add the calculation results to the error compensation amount obtained by online regression compensation of the prediction residual to obtain the full-field temperature deformation prediction result.

2. The method for predicting temperature deformation of large-span aluminum alloy structures according to claim 1, characterized in that, Step S1 includes: Based on the design drawings of the large-span aluminum alloy structure, the equivalent radial stiffness, tangential stiffness, moment release coefficient of the connection node, and equivalent linear expansion coefficient of the aluminum alloy of the main load-bearing support node are parametrically defined to obtain the stiffness parameter vector. Based on the feasible region of the stiffness parameter vector, multiple sets of sampling parameters are generated by Latin hypercube sampling. The combination of each sampling parameter and multiple temperature load modes is substituted into the finite element model for static analysis to obtain the nodal displacement response dataset. The nodal displacement response dataset is subjected to eigenorthogonal decomposition, and the first few order basis vectors are extracted according to the truncation condition that the cumulative energy ratio is not less than 95%, so as to obtain the reduced order basis matrix. Based on the reduced-order basis matrix, with each sampling parameter and the corresponding temperature load mode as input and the corresponding reduced-order coordinate coefficients as output, the response surface interpolation weight coefficients are obtained by least-squares fitting using a Gaussian kernel function.

3. The method for predicting temperature deformation of large-span aluminum alloy structures according to claim 1, characterized in that, In step S3, the measured temperature data is converted into a temperature load vector and then input into the reduced-order basis matrix for interpolation calculation to obtain the predicted displacement. The difference between the predicted displacement and the measured displacement is used to obtain the predicted residual, including: Based on the measured temperature data at each temperature measuring point, the temperature distribution of the entire structure is interpolated by Kriging using a spherical variogram model to obtain the temperature distribution data of the entire field. The temperature load vector is obtained by multiplying the full-field temperature distribution data with the singular vector matrix of the temperature load mode. The temperature load vector and the stiffness parameter vector are substituted into the reduced-order basis matrix and the response surface interpolation weight coefficients to perform Gaussian kernel function interpolation calculation to obtain the predicted displacement of each monitoring node. The predicted displacement is subtracted from the measured displacement at each monitoring node to obtain the predicted residual.

4. The method for predicting temperature deformation of large-span aluminum alloy structures according to claim 3, characterized in that, In step S3, the predicted residual is determined to be either a source of structural damage or a source of sensor drift based on the spatial consistency score. When it is determined to be a source of structural damage, the stiffness parameter is corrected; when it is determined to be a source of sensor drift, the corresponding node data is isolated, including: The spatial consistency score is compared with a preset judgment threshold. When the spatial consistency score is not lower than the preset judgment threshold, the prediction residual is judged as a source of structural damage. When the spatial consistency score is lower than the preset judgment threshold, the prediction residual is judged as a source of sensor drift. When the source of structural damage is determined, the set of members in the damaged area is determined according to the set of mechanical response neighborhoods. A local loss function is constructed based on the difference between the measured displacement and the predicted displacement of each monitoring node in the damaged area. The equivalent radial stiffness corresponding to the damaged area in the stiffness parameter vector is iteratively corrected by the gradient descent method to obtain the corrected stiffness parameter. When the sensor drift source is identified, the measured displacement of the node is isolated from the parameter identification dataset of the stiffness parameter vector. The reconstructed displacement of the node is obtained by weighted interpolation based on the measured displacement of the normal nodes in the mechanical response neighborhood set of the node in the physical correlation matrix and the corresponding mechanical deformation transmission ratio. The reconstructed displacement is then used to replace the measured displacement in subsequent calculations.

5. The method for predicting temperature deformation of large-span aluminum alloy structures according to claim 4, characterized in that, Step S4 includes: The modified stiffness parameters and the target temperature load vector are substituted into the reduced-order basis matrix and the response surface interpolation weight coefficients to perform Gaussian kernel function interpolation calculation, and the full-field nodal displacement interpolation results are obtained. Based on the predicted residuals of normal nodes within the mechanical response neighborhood set, the combination of the temperature load vector and the corrected stiffness parameter is used as input, and the corresponding predicted residuals are used as output. Gaussian process regression is used to perform online regression compensation for the systematic deviation of the whole field node displacement interpolation results, and the error compensation amount is obtained. The whole-field nodal displacement interpolation results are superimposed with the error compensation amount node by node to obtain the whole-field temperature deformation prediction results.

6. A temperature deformation prediction system for large-span aluminum alloy structures, characterized in that, For implementing the method for predicting the temperature deformation of a large-span aluminum alloy structure as described in any one of claims 1-5, the large-span aluminum alloy structure temperature deformation prediction system comprises: The order reduction module is used for finite element models based on large-span aluminum alloy structures. It performs order reduction processing on the nodal displacement response under multiple temperature load modes through intrinsic orthogonal decomposition to obtain the reduced basis matrix and response surface interpolation weight coefficients. The projection module is used to perform pseudo-inverse projection calculations on the unit displacement response of each monitoring node based on the reduced-order basis matrix, to obtain a physical correlation matrix characterizing the mechanical deformation transfer ratio between each monitoring node. This includes: extracting row vectors corresponding to the vertical degrees of freedom of each monitoring node from the reduced-order basis matrix to obtain a reduced-order basis submatrix for each monitoring node; performing Moore-Penrose pseudo-inverse operations on the reduced-order basis submatrix of the monitoring nodes to obtain a reduced-order basis pseudo-inverse matrix for each monitoring node, thereby establishing an inverse mapping relationship in the sense of minimum norm between the vertical degree-of-freedom subspace and the reduced-order coordinate space of the monitoring nodes; and performing pseudo-inverse calculations on each monitoring node. For each node, the unit response vector centered at that node is multiplied by the reduced-order basis pseudo-inverse matrix of the monitoring node to obtain the reduced-order coordinate projection value of the node. This reduced-order coordinate projection value is then substituted into the reduced-order basis matrix for full-field displacement reconstruction. The vertical displacement components of each neighboring node are extracted from the reconstruction result to obtain the mechanical deformation transfer ratio corresponding to each monitoring node. All mechanical deformation transfer ratios corresponding to all monitoring nodes are arranged by node number. Transfer ratios with an absolute value less than 0.15 are set to zero to filter out weakly correlated node pairs, resulting in the physical correlation matrix and the mechanical response neighborhood set of each monitoring node. The monitoring... The number of nodes is less than the number of columns in the reduced-order basis matrix; the determination module is used to convert the measured temperature data into a temperature load vector, then input it into the reduced-order basis matrix for interpolation calculation to obtain the predicted displacement, and subtract the predicted displacement from the measured displacement to obtain the predicted residual; the degree of conformity of the neighborhood cooperative response of the predicted residual is scored according to the physical correlation matrix to obtain a spatial consistency score; based on the spatial consistency score, the predicted residual is determined to be a source of structural damage or a source of sensor drift. When it is determined to be a source of structural damage, the stiffness parameter is corrected; when it is determined to be a source of sensor drift, the corresponding number of nodes is adjusted. According to the isolation, for abnormal nodes in the predicted residuals whose absolute value exceeds the residual threshold, based on the mechanical response neighborhood set corresponding to the abnormal node in the physical correlation matrix, the residual value of the abnormal node in the predicted residuals is multiplied by the mechanical deformation transmission ratio corresponding to each neighborhood node to obtain the expected collaborative response value of each neighborhood node; the difference between the predicted residual of each neighborhood node and the corresponding expected collaborative response value is divided by the residual threshold to obtain the deviation rate of each neighborhood node; the average deviation rate of each neighborhood node in the mechanical response neighborhood set is taken and subtracted from 1 to obtain the spatial consistency score. The compensation module is used to substitute the corrected stiffness parameters into the reduced-order basis matrix and the response surface interpolation weight coefficients for interpolation calculation, and then superimpose the calculation results with the error compensation amount obtained by online regression compensation of the prediction residuals to obtain the full-field temperature deformation prediction results.

7. A device for predicting temperature deformation of large-span aluminum alloy structures, characterized in that, The method includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the method for predicting the temperature deformation of a large-span aluminum alloy structure as described in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, the processor executes the method for predicting the temperature deformation of large-span aluminum alloy structures as described in any one of claims 1 to 5.

Images