Intelligent prediction method and system for deformation performance index of axially confined concrete beam

Abstract

The invention relates to the technical field of civil engineering structure anti-seismic performance evaluation and artificial intelligence technology crossing, in particular to an intelligent prediction method and system for deformation performance indexes of an axially confined concrete beam. The prediction method comprises the following steps: acquiring an initial sparse real data set formed by key design parameters and key deformation performance index parameters of a plurality of concrete target beams; the key design parameters comprise an axial constraint characterization coefficient, a shear span ratio and a bending shear ratio; training a physical information neural network model by using the initial sparse real data set, and generating a physically consistent expansion data set based on the trained physical information neural network model; training a prediction model based on the extended data set, the real training set and the real verification set; and performing performance evaluation on the prediction model by using the real test set. According to the technical scheme, a complete technical closed loop of'constraint quantization-physical enhancement-small sample learning 'is constructed, so that the deformation performance prediction problem of the axially confined concrete beam under the condition of sample data scarcity is solved.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of seismic performance evaluation of civil engineering structures and artificial intelligence technology, and in particular to an intelligent prediction method and system for deformation performance indicators of axially constrained concrete beams. Background Technology

[0002] In seismic design and safety assessment of building structures, the deformation performance of concrete beams, especially the displacement angle limits under cyclic loading at states such as equivalent yield, initial capacity degradation, and significant capacity degradation, is a crucial basis for evaluating their seismic performance and damage state. Traditional theoretical analysis typically simplifies beams as pure bending members, neglecting the fact that in actual structures, due to the integral casting and connection between beams and floor slabs, shear walls, or frame columns, their axial deformation is constrained, placing them in a complex stress state of compression-bending-shear coupling. This axial constraint effect significantly alters the beam's stiffness, capacity, and ductility; ignoring this effect will lead to serious deviations in the prediction of the beam's seismic deformation performance.

[0003] Obtaining data on the deformation performance of beams considering axial constraints requires complex quasi-static tests to simulate real boundary conditions. This process is costly and time-consuming, resulting in extremely sparse reliable experimental data samples (typically only a few dozen sets), creating a "data shortage" bottleneck. Under these extremely sparse sample conditions, existing technologies face severe challenges: simplified models based on classical mechanics theory cannot accurately invert the compression-bending coupling nonlinearity of axially constrained concrete beams; refined finite element simulations are costly and dependent on empirical parameters; purely data-driven machine learning models suffer from severe overfitting due to insufficient samples, exhibiting poor generalization ability and potentially violating physical laws in their predictions, leading to low reliability; and traditional data interpolation and other enhancement methods cannot guarantee the physical authenticity of the generated data. Summary of the Invention

[0004] To address the aforementioned technical problems, or at least partially address them, this invention provides an intelligent prediction method and system for the deformation performance of axially constrained concrete beams. By constructing a complete technical closed loop of "constraint quantification → physical enhancement → small sample learning," the method solves the problem of predicting the deformation performance of axially constrained concrete beams under conditions of scarce sample data.

[0005] In a first aspect, the present invention provides an intelligent prediction method for the deformation performance index of axially constrained concrete beams, comprising:

[0006] S1. Obtain the key design parameters and key deformation performance index parameters of the concrete target beam. The key design parameters and key deformation performance index parameters of the multiple concrete target beams form an initial sparse real dataset. The key design parameters include the axial constraint characterization coefficient η, shear span ratio λ, and bending-shear ratio m. The key deformation performance index parameters are the component displacement angle limits corresponding to the performance points.

[0007] S2. Preprocess the initial sparse real dataset, and then preprocess the real dataset D. real Divided into real training set D real_train Real verification set D real_val and the real test set D real_test ;

[0008] S3. Construct a physical information neural network model, train the physical information neural network model using the initial sparse real dataset, and generate physically consistent synthetic data based on the trained physical information neural network model to form an expanded dataset D. syn ;

[0009] S4. Based on the expanded dataset D syn The real training set D real_train and the real verification set D real_val A three-stage transfer learning strategy was used to train the prediction model.

[0010] S5. Using the aforementioned real test set D real_test The trained prediction model is evaluated to obtain a qualified prediction model.

[0011] S6. Input the key design parameters of the concrete beam to be predicted into the standard prediction model, and output the key deformation performance index parameters of the concrete beam to be predicted.

[0012] In some embodiments, obtaining the axial constraint characterization coefficient η of the concrete target beam includes:

[0013] For the actual structural environment of the target concrete beam, establish an equivalent substructure finite element model or an overall structural finite element model that includes itself, adjacent vertical members and connecting floor slabs.

[0014] The equivalent axial constraint stiffness K at the beam end of the target concrete beam was obtained by performing mechanical analysis on the finite element model. a The formula η=K is used. a / bf c Calculate the dimensionless axial constraint characterization coefficient η;

[0015] Where b is the beam cross-section width, f c This refers to the compressive strength of concrete.

[0016] In some embodiments, the equivalent axial restraint stiffness K at the beam end of the target concrete beam is obtained. a ,include:

[0017] A load is applied to the target concrete beam in the finite element model;

[0018] Obtain the equivalent axial displacement δ at the beam end caused by the applied load on the target concrete beam. A and the end axial reaction force N c ;

[0019] Using formula K a =N c / δ A Obtain the equivalent axial restraint stiffness K at the beam end of the target concrete beam. a .

[0020] In some implementations, key deformation performance parameters of the target concrete beam are obtained, including:

[0021] Identify three key performance points;

[0022] The component displacement angle limits corresponding to each key performance point are obtained as θ1, θ5 and θ6, and the component displacement angle limits θ1, θ5 and θ6 are used as key deformation performance index parameters;

[0023] Wherein, θ1 is the equivalent yield displacement angle limit corresponding to the component in an undamaged state, θ5 is the displacement angle limit corresponding to the bearing capacity decreasing to 80% corresponding to the component in a relatively severely damaged state, and θ6 is the displacement angle limit corresponding to the bearing capacity decreasing to 50% corresponding to the component in a severely damaged state.

[0024] In some implementations, the key deformation performance parameters of the concrete beam to be predicted include: the component displacement angle limits θ1, θ5, and θ6 corresponding to the three key performance points of the concrete beam to be predicted; and the ordinary deformation performance parameters of the concrete beam to be predicted are determined based on the key deformation performance parameters, including:

[0025] Based on the component displacement angle limits θ1, θ5, and θ6 corresponding to the component displacement angles of the concrete beam to be predicted, determine the component displacement angle limits θ2, θ3, and θ4 corresponding to the ordinary performance points of the concrete beam to be predicted, and obtain them using the following formula:

[0026] θ2 = 0.25 × (θ5 - θ1)

[0027] θ3 = 0.50 × (θ5 - θ1)

[0028] θ4 = 0.75 × (θ5 - θ1)

[0029] Wherein, θ2 is the displacement angle limit corresponding to the component under slight damage, θ3 is the displacement angle limit corresponding to the component under mild damage, and θ4 is the displacement angle limit corresponding to the component under moderate damage.

[0030] In some implementations, physically consistent synthetic data is generated based on the trained physical information neural network model to form an augmented dataset D. syn ,include:

[0031] Based on the Latin hypercube sampling method, a large-scale virtual critical design parameter P is generated within a reasonable range of critical design parameters. syn Virtual key design parameter P syn Input the physical information into the neural network model and output the corresponding deformation performance prediction parameter R. syn To obtain the augmented dataset D syn ={P syn R syn}

[0032] In some implementations, the three-stage transfer learning strategy includes a pre-training stage, a fine-tuning stage, and a validation and early stopping stage; based on the augmented dataset, the three-stage transfer learning strategy is used to train the prediction model, including:

[0033] During the pre-training phase, the real training set D real_train and augmented dataset D syn Merge and construct a hybrid training set D mix_train Using a mixed training set D mix_train Train the prediction model;

[0034] During the fine-tuning phase, the real training set D is used. real_train Perform fine calibration on the prediction model;

[0035] During the verification and early stopping phases, the real verification set D is used. real_val Validate the performance of the prediction model.

[0036] In some implementations, constructing a physical information neural network model involves embedding the deformation control equation of a concrete beam under compression-bending-shear coupling into the loss function of the physical information neural network in the form of residual terms, so that the network output satisfies the physical laws represented by the deformation control equation.

[0037] In some implementations, the real test set D is used. real_test The trained prediction model is then evaluated to obtain a qualified prediction model, including:

[0038] During the testing process, multi-dimensional evaluation metrics are obtained, including prediction accuracy metrics, physical consistency metrics, and uncertainty quantification metrics.

[0039] Determine whether the multi-dimensional evaluation indicators meet the standards;

[0040] If the multi-dimensional evaluation metrics meet the standards, the trained prediction model is deemed to have met the standards; if the multi-dimensional evaluation metrics do not meet the standards, return to S3 to adjust the physical control equations and / or return to S4 to adjust the model architecture or training strategy to obtain a qualified prediction model.

[0041] Secondly, the present invention also provides an intelligent prediction system for the deformation performance index of axially constrained concrete beams, comprising:

[0042] The acquisition module is used to acquire the key design parameters and key deformation performance index parameters of the concrete target beam. The key design parameters and key deformation performance index parameters of the multiple concrete target beams form an initial sparse real dataset. The key design parameters include the axial constraint characterization coefficient η, shear span ratio λ, and bending shear ratio m. The key deformation performance index parameters are the component displacement angle limits corresponding to the key performance points.

[0043] The preprocessing module is used to preprocess the initial sparse real dataset, and to preprocess the real dataset D. real Divided into real training set D real_train Real verification set D real_val and the real test set D real_test ;

[0044] An expanded dataset generation module is used to construct a physical information neural network model, train the physical information neural network model using the initial sparse real dataset, and generate physically consistent synthetic data based on the trained physical information neural network model to form an expanded dataset D. syn ;

[0045] Training module, used to train based on the augmented dataset D syn The real training set D real_train and the real verification set D real_val A three-stage transfer learning strategy was used to train the prediction model.

[0046] Evaluation module, used to evaluate the real test set D real_test The trained prediction model is evaluated to obtain a qualified prediction model.

[0047] The prediction module is used to input the key design parameters of the concrete beam to be predicted into the standard prediction model, output the key deformation performance index parameters of the concrete beam to be predicted, and determine the ordinary deformation performance index parameters of the concrete beam to be predicted based on the key deformation performance index parameters of the concrete beam to be predicted. The ordinary deformation performance index parameters are the component displacement angle limits corresponding to the ordinary performance points.

[0048] Thirdly, the present invention provides an electronic device, including a processor and a memory, wherein the processor executes the steps of the intelligent prediction method for axially constrained concrete beam deformation performance indicators as described in the first aspect by calling programs or instructions stored in the memory.

[0049] The technical solution provided by the embodiments of the present invention has the following advantages compared with the prior art:

[0050] 1. Effectively overcome data bottlenecks: Through the innovative method of physical information augmentation, sparse experimental data can be expanded into large-scale training samples while ensuring the physical authenticity of the data, laying the foundation for the application of deep learning models.

[0051] 2. Ensure the physical credibility of the predictions: By deeply embedding the fundamental laws of mechanics into the model learning process, the physical rationality of the generated data and prediction results is fundamentally constrained, avoiding the "physical absurdity" output of pure black box models.

[0052] 3. Precise quantification of complex constraints: The proposed axial constraint quantification method based on refined finite element analysis transforms engineering boundary conditions that are difficult to consider and handle in existing theories and techniques into calculable and comparable characteristic parameters, thereby improving the engineering applicability and reliability of the model. Attached Figure Description

[0053] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 A flowchart illustrating an intelligent prediction method for deformation performance indicators of axially constrained concrete beams provided in an embodiment of the present invention;

[0056] Figure 2 A schematic diagram illustrating the calculation of axial constraint stiffness provided in an embodiment of the present invention;

[0057] Figure 3 A schematic diagram of the structure of an intelligent prediction system for the deformation performance index of an axially constrained concrete beam provided in an embodiment of the present invention;

[0058] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0059] To better understand the above-mentioned objectives, features, and advantages of the present invention, the solutions of the present invention will be further described below. It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other.

[0060] Many specific details are set forth in the following description in order to provide a full understanding of the invention, but the invention may also be practiced in other ways different from those described herein; obviously, the embodiments in the specification are only some embodiments of the invention, and not all embodiments.

[0061] The intelligent prediction method for deformation performance indicators of axially constrained concrete beams provided in this invention can accurately quantify the complex axial constraint effect of beams in spatial structures under sparse test data conditions. It effectively integrates physical mechanisms and data patterns to achieve reliable and efficient prediction of deformation performance indicators of axially constrained concrete beams, providing a technical foundation for improving the accuracy and reliability of structural seismic analysis and performance evaluation.

[0062] Figure 1 This is a flowchart illustrating an intelligent prediction method for the deformation performance indicators of axially constrained concrete beams provided in an embodiment of the present invention. This method can be executed by the intelligent prediction system for the deformation performance indicators of axially constrained concrete beams provided in this embodiment of the invention. This intelligent prediction system for the deformation performance indicators of axially constrained concrete beams can be implemented using software and / or hardware. Figure 1 As shown, the prediction method includes the following steps:

[0063] S1. Obtain the key design parameters and key deformation performance index parameters of the concrete target beam. The key design parameters and key deformation performance index parameters of multiple concrete target beams form an initial sparse real dataset. The key design parameters include the axial constraint characterization coefficient η, shear span ratio λ, and bending-shear ratio m. The key deformation performance index parameters are the component displacement angle limits corresponding to the key performance points.

[0064] Specifically, in this step, key design parameters and key deformation performance parameters of concrete target beams can be obtained. The key design parameters and key deformation performance parameters of multiple concrete target beams form an initial sparse real dataset, thereby constructing a preliminary quasi-static test database for axially constrained concrete beams.

[0065] Based on physical concepts, the shear span ratio λ, bending-shear ratio m, and axial constraint characterization coefficient η, which best represent the axially constrained stress state of the concrete target beam, are selected as key design parameters P=[η,λ,m]. Multiple key design parameters constitute the key design parameter vector P. i = [η i , λ i , mi ] T .

[0066] Among them, the shear span ratio λ reflects the relative magnitude of bending moment and shear force on the beam section and is directly related to the stress mode of the beam; the bending-shear ratio m is an index for comparing the relative relationship between the bending capacity and shear capacity of the beam and is used to determine the dominant type of failure mode of the beam.

[0067] In some embodiments, obtaining the axial constraint characterization coefficient η of the concrete target beam includes:

[0068] For the actual structural environment of the target concrete beam, establish an equivalent substructure finite element model or an overall structural finite element model that includes itself, adjacent vertical members and connecting floor slabs.

[0069] The equivalent axial constraint stiffness K at the beam end of the target concrete beam was obtained by performing mechanical analysis on the finite element model. a The formula η=K is used. a / bf c Calculate the dimensionless axial constraint characterization coefficient η.

[0070] Specifically, in this embodiment, a finite element model is established for the actual structural environment of the target concrete beam. Then, a mechanical analysis is performed on the finite element model using the formula η=K. a / bf c Calculate the dimensionless axial constraint characterization coefficient η.

[0071] Where b is the beam cross-section width, f c η represents the compressive strength of concrete; η is a dimensionless parameter whose value directly reflects the axial constraint strength of the actual structure on the beam. η=0 corresponds to an unconstrained pure bending state, and the larger the value of η, the stronger the constraint.

[0072] In some embodiments, the equivalent axial restraint stiffness K at the beam end of the target concrete beam is obtained. a ,include:

[0073] Apply a load to the target concrete beam in the finite element model;

[0074] Obtain the equivalent axial displacement δ at the beam end caused by the applied load on the target concrete beam. A and the end axial reaction force N c ;

[0075] Using formula K a =N c / δ A Obtain the equivalent axial restraint stiffness K at the beam end of the target concrete beam. a .

[0076] Specifically, one of the following equivalent loading methods can be selected to apply a load to the target concrete beam and perform calculation and analysis:

[0077] (1) Elastic heating loading method (applicable to elastic stage estimation): Apply a uniform heating load to the target concrete beam, perform linear analysis, and calculate the equivalent axial displacement δ at the beam end caused by thermal expansion of the target concrete beam. A And the axial reaction force N at the end caused by the constraint of the surrounding structure. c .

[0078] (2) Elastic "force-displacement" loading method (applicable to elastic stage estimation): A small unit axial displacement is applied to the end of the target concrete beam as the equivalent axial displacement δ at the beam end. A Perform linear analysis to calculate the end axial reaction force N of the target concrete beam due to the constraint of the surrounding structure caused by slight axial deformation. c .

[0079] (3) Elastoplastic force-displacement loading method (applicable to considering material nonlinearity): Apply a vertical downward displacement to the middle of the target concrete beam, perform nonlinear analysis, and directly extract the equivalent axial displacement δ at the beam end. A and the axial reaction force N at the end c .

[0080] For example, such as Figure 2 As shown, Figure 2 The equivalent axial displacement δ at the beam end A and the resulting end axial reaction force N c Due to the relationship between the beam and surrounding structural members (cast-in-place floor slabs and vertical members), once the beam undergoes axial elongation, it will generate a constraint axial reaction force. Therefore, in obtaining the equivalent axial displacement δ at the beam end... A and the end axial reaction force N c Then, formula K is used. a =N c / δ A The equivalent axial constraint stiffness K at the beam end of the target concrete beam can then be obtained. a .

[0081] In some implementations, key deformation performance parameters of the target concrete beam are obtained, including:

[0082] Identify three key performance points;

[0083] The component displacement angle limits corresponding to each key performance point are obtained as θ1, θ5 and θ6, and the component displacement angle limits θ1, θ5 and θ6 are used as key deformation performance index parameters;

[0084] Wherein, θ1 is the equivalent yield displacement angle limit corresponding to the component in an undamaged state, θ5 is the displacement angle limit corresponding to the bearing capacity decreasing to 80% corresponding to the component in a relatively severely damaged state, and θ6 is the displacement angle limit corresponding to the bearing capacity decreasing to 50% corresponding to the component in a severely damaged state.

[0085] Specifically, the deformation performance index parameter is the component displacement angle limit corresponding to the performance point. Based on the component performance status, it is divided into 7 performance states: no damage, minor damage, slight damage, moderate damage, relatively severe damage, severe damage, and failure.

[0086] In this embodiment, three key performance points are determined based on mainstream performance design methods both domestically and internationally, corresponding to the undamaged state, the moderately damaged state, and the severely damaged state, respectively. The corresponding displacement angle limits for the three components are θ1, θ5, and θ6, respectively. The undamaged state corresponds to the equivalent yield displacement angle limit θ1 predefined according to the skeleton curve corresponding to the experimental hysteresis curve, using the Park method or energy method. The moderately damaged state corresponds to the displacement angle limit θ5 when the bearing capacity decreases to 80%, and the severely damaged state corresponds to the displacement angle limit θ6 when the bearing capacity decreases to 50%. The component displacement angle limits of the above three key performance points are selected as key deformation performance index parameters R=[θ1, θ5, θ6]. Multiple key deformation performance index parameters constitute the key deformation performance index parameter vector R. i =[θ 1,i , θ 5,i , θ 6,i ] T .

[0087] In some implementations, the component displacement angle limits for the three common performance points (minor damage, slight damage, and moderate damage) can be interpolated based on the component displacement angle limits for three key performance points θ1, θ5, and θ6, respectively, and are calculated as follows:

[0088] θ2 = 0.25 × (θ5 - θ1)

[0089] θ3 = 0.50 × (θ5 - θ1)

[0090] θ4 = 0.75 × (θ5 - θ1)

[0091] S2. Preprocess the initial sparse real dataset, and then process the preprocessed real dataset D. real Divided into real training set D real_train Real verification set D real_val and the real test set D real_test .

[0092] Specifically, preprocessing the initial sparse real dataset may include: standardizing the key design parameters [η,λ,m] and key deformation performance index parameters [θ1, θ5, θ6] respectively, and then introducing an outlier handling mechanism to remove outlier data points based on physical common sense, such as obviously unreasonable data like θ6 / θ1 > 50 or θ1 < 0.001.

[0093] Next, the dataset is partitioned: the initial sparse dataset D is divided into two parts. real The training set and the real training set are divided according to a preset ratio. real_train Real verification set D real_val and the real test set D real_test Preferably, the partitioning ratio can be 70% for the training set, 15% for the validation set, and 15% for the test set.

[0094] S3. Construct a physical information neural network model. Train the physical information neural network model using an initial sparse real dataset. Based on the trained physical information neural network model, generate physically consistent synthetic data to form an expanded dataset D. syn .

[0095] In some implementations, constructing a physical information neural network model involves embedding the deformation control equation of a concrete beam under compression-bending-shear coupling into the loss function of the physical information neural network in the form of residual terms, so that the network output satisfies the physical laws represented by the deformation control equation.

[0096] Specifically, constructing a physical information neural network model NN PINN The network consists of (P, R), with input key design parameters P = [η, λ, m] and output key deformation performance parameters R = [θ1, θ5, θ6]. The network structure contains 3-5 hidden layers, each with 20-50 neurons, using the tanh activation function. The output layer uses the Softplus activation function to ensure a positive output.

[0097] Based on mechanical principles and engineering experience, the following types of physical constraint equations are established:

[0098] (1) Relationship constraint (strong constraint): The limit value of the component displacement angle should satisfy θ1<θ5<θ6. In the loss function, this is achieved by introducing an inequality penalty term, which imposes a significant penalty when the predicted value violates this relationship.

[0099] (2) Trend constraints (weak constraints): An increase in the axial constraint characterization coefficient η usually leads to an increase in the displacement angle limit θ1 (strengthened constraint, delayed yield time), and a decrease in the displacement angle limits θ5 and θ6 (strengthened constraint, deteriorated deformation capacity); an increase in the shear span ratio λ usually leads to a decrease in the displacement angle limit θ1 (tends towards bending, earlier yield time), and an increase in the displacement angle limits θ5 and θ6 (tends towards bending, improved deformation capacity). A decrease in the bending-shear ratio m usually leads to a decrease in the displacement angle limit θ1 (tends towards bending, earlier yield time), and an increase in the displacement angle limits θ5 and θ6 (tends towards bending, improved deformation capacity).

[0100] The partial derivatives of the output parameters with respect to each input parameter are calculated using automatic differentiation techniques, and the signs of these partial derivatives are constrained in the loss function to conform to the aforementioned trend.

[0101] (3) Empirical formula constraints (weak constraints): Based on dimensional analysis or regression analysis, an approximate linear relationship such as θ1=Aη+Bλ+Cm can be established, where A, B, and C are parameter ranges estimated based on limited data or theory. The mean square error of this linear relationship and the predicted value is added to the loss function as a constraint term.

[0102] The above constraint equations can be encoded as a set of equations X relating to the inputs and outputs of a physical information neural network model.

[0103] Among them, the loss function design is as follows: the loss function Ф includes the data fitting loss term Ф d and physical residual loss term Ф p That is, the loss function Ф = κ1 Ф d + κ2 Ф p Where κ1 and κ2 are weighting coefficients. Specifically, the data fitting loss term Ф d Calculate the mean squared error on the actual data points (Pi, Ri); physical residual loss term Ф p , is used to calculate the extent to which the network prediction violates the above physical constraint equation X for all inputs (including real data points and virtual input points), and is measured in the form of mean square error.

[0104] Specifically, in the physical information neural network model NN PINN During the training process (P, R), an adaptive weight adjustment strategy is adopted: in the early stage of training, κ1 = 1.0 and κ2 = 0.1 are set to prioritize fitting the real data. In the later stage of training, the physical constraint weight κ2 is gradually increased until κ1 = 1.0 and κ2 = 1.0 are finally achieved to enhance the consistency of physical laws.

[0105] After training, the physical information neural network model NN PINN(P, R) becomes a surrogate model that predicts performance under given parameters and whose prediction results strictly follow the preset physical relationships.

[0106] In some implementations, physically consistent synthetic data is generated based on a trained physical information neural network model to form an augmented dataset D. syn ,include:

[0107] Based on the Latin hypercube sampling method, a large-scale virtual critical design parameter P is generated within a reasonable range of critical design parameters. syn Virtual key design parameter P syn Input the physical information into the neural network model and output the corresponding deformation performance prediction parameters R. syn To obtain the augmented dataset D syn ={P syn R syn}

[0108] The reasonable range of key design parameters is set based on actual engineering experience. Specifically, the shear span ratio λ ∈ [0.5, 6], the bending-shear ratio m ∈ [0.1, 3], and the axial constraint characterization coefficient η ∈ [0, 100]. A large-scale virtual key design parameter P is generated. syn At that time, the number of sampling points was much greater than the actual amount of data.

[0109] Specifically, the virtual key design parameter P syn Input the physical information into the neural network model and output the corresponding deformation performance prediction parameters R. syn Thus, the augmented dataset D can be obtained. syn ={P syn R syn}

[0110] For example, the initial sparse real dataset has 100 elements. Through embodiments of this disclosure, an expanded dataset D can be formed. syn The number is 1000.

[0111] In some implementations, to reflect prediction uncertainty, the standard deviation σ is estimated based on the model's prediction error on real data, and then applied to R. syn By adding appropriate noise, the R-value after adding appropriate noise can be obtained. syn =NN PINN (P syn ) + ε, where ε ~ N(0, σ2).

[0112] S4, Based on augmented dataset D syn Real training set D real_train and the true verification set D real_val A three-stage transfer learning strategy was used to train the prediction model.

[0113] Specifically, in this embodiment, based on the expanded dataset D syn Real training set D real_train and the true verification set D real_val The model is trained using a three-stage transfer learning strategy, with key design parameters as input and key deformation performance index parameters as output, forming an end-to-end prediction model.

[0114] In some implementations, the three-stage transfer learning strategy includes a pre-training stage, a fine-tuning stage, and a validation and early stopping stage; based on an augmented dataset, the three-stage transfer learning strategy is used to train the prediction model, including:

[0115] During the pre-training phase, the real training set D real_train and augmented dataset D syn Merge and construct a hybrid training set D mix_train Using a mixed training set D mix_train Train the prediction model;

[0116] During the fine-tuning phase, the real training set D is used. real_train Perform fine calibration on the prediction model;

[0117] During the verification and early stopping phases, the real verification set D is used. real_val Validate the performance of the prediction model.

[0118] Specifically, in this embodiment, during the pre-training phase, a mixed data training set D is first used. mix_train The initial prediction model is trained to learn the basic mapping relationship from design parameters to performance metrics; then, in the fine-tuning phase, the real training set D is used. real_train The trained model is finely calibrated to ensure its predictions accurately align with the real experimental distribution; finally, in the validation and early stopping phase: the real validation set D is used. real_val Monitor model performance, employ early stopping strategies to prevent overfitting, and optimize hyperparameters.

[0119] S5, Using the real test set D real_test The trained prediction model is evaluated to obtain a qualified prediction model.

[0120] In some implementations, the real test set D is used. real_test The trained prediction model is then evaluated to obtain a qualified prediction model, including:

[0121] During the testing process, multi-dimensional evaluation metrics are obtained, including prediction accuracy metrics, physical consistency metrics, and uncertainty quantification metrics.

[0122] Determine whether the multi-dimensional evaluation indicators meet the standards;

[0123] If the multi-dimensional evaluation metrics meet the standards, the trained prediction model is deemed to have met the standards; if the multi-dimensional evaluation metrics do not meet the standards, return to S3 to adjust the physical control equations and / or return to S4 to adjust the model architecture or training strategy to obtain a qualified prediction model.

[0124] Specifically, prediction accuracy indicators may include: mean absolute error (MAE), mean absolute percentage error (MAPE), and root mean square error (RMSE); physical consistency indicators may include: relationship constraint violation rate and trend constraint compliance; uncertainty quantification indicators may include: 95% prediction interval coverage and prediction interval width.

[0125] For example, during the testing process, the training of the prediction model is considered successful when the following conditions are met simultaneously: (1) the MAPE of θ1, θ5, and θ6 are all less than 25%; (2) the relationship constraint violation rate is less than 5%; and (3) the 95% prediction interval coverage is greater than 90%. If the above conditions are not met, the reasons need to be analyzed and iterative optimization is required: if it is due to insufficient data quality or physical constraints, return to S3 to adjust the physical constraint equations or regenerate synthetic data; if it is due to model architecture or training strategy problems, return to S4 to adjust the model hyperparameters or training strategy.

[0126] S6. Input the key design parameters of the concrete beam to be predicted into the standard prediction model, output the key deformation performance index parameters of the concrete beam to be predicted, and determine the ordinary deformation performance index parameters of the concrete beam to be predicted based on the key deformation performance index parameters of the concrete beam to be predicted. The ordinary deformation performance index parameters are the component displacement angle limits corresponding to the ordinary performance points.

[0127] Specifically, for the concrete beam to be predicted, its key design parameters are calculated and input into the standard prediction model. The key deformation performance index parameters are obtained through the standard prediction model, that is, the predicted value of the component displacement angle limit of the key performance point of the concrete beam to be predicted is obtained, and the 95% confidence interval is output to provide a reliable reference for engineering decision-making.

[0128] In some embodiments, the key deformation performance parameters of the concrete beam to be predicted include: the component displacement angle limits θ1, θ5, and θ6 at three key performance points corresponding to the concrete beam to be predicted; determining the ordinary deformation performance parameters of the concrete beam to be predicted based on the key deformation performance parameters includes:

[0129] Based on the component displacement angle limits θ1, θ5, and θ6 corresponding to the component displacement angles of the concrete beam to be predicted, determine the component displacement angle limits θ2, θ3, and θ4 corresponding to the ordinary performance points of the concrete beam to be predicted, and obtain them using the following formula:

[0130] θ2 = 0.25 × (θ5 - θ1)

[0131] θ3 = 0.50 × (θ5 - θ1)

[0132] θ4 = 0.75 × (θ5 - θ1)

[0133] Wherein, θ2 is the displacement angle limit corresponding to the component under slight damage, θ3 is the displacement angle limit corresponding to the component under mild damage, and θ4 is the displacement angle limit corresponding to the component under moderate damage.

[0134] Specifically, based on the component displacement angle limits θ1, θ5, and θ6 corresponding to the three key performance points of the concrete beam to be predicted, the component displacement angle limits θ2, θ3, and θ4 corresponding to the three ordinary performance points of slight damage, mild damage, and moderate damage can be calculated by interpolation using the above formula.

[0135] Therefore, the component displacement angle limits θ1, θ5 and θ6 of the three key performance points corresponding to the concrete beam to be predicted are obtained, and the component displacement angle limits θ2, θ3 and θ4 of the three ordinary performance points corresponding to the concrete beam to be predicted are added to form a complete prediction system.

[0136] By comparing the component displacement angle limit response θ of the target beam under seismic loading with the component displacement angle limit values ​​at different performance points, the component damage state of the target beam can be determined. For example, if θ2 < θ < θ3, the target beam can be judged to be in a state of slight damage; if θ5 < θ < θ6, the target beam can be judged to be in a state of severe damage.

[0137] Therefore, the intelligent prediction method for axially constrained concrete beam deformation performance provided in this disclosure deeply embeds mechanical laws into the model learning process to generate large-scale, physically consistent synthetic data. This fundamentally solves the problem of insufficient model training caused by sparse experimental data, while ensuring the physical rationality of the data augmentation process and the final prediction results. Furthermore, the three-stage transfer learning strategy adopted in this application can efficiently integrate the universal laws contained in large-scale synthetic data with the precise distribution represented by small-scale real data. Through pre-training and fine-tuning, it significantly improves the model's prediction accuracy and generalization ability on unseen data. In addition, the proposed axial constraint quantification method transforms complex engineering boundary conditions into a clear, computable dimensionless feature parameter, simplifying model input and enabling the model to be easily applied to the performance prediction of concrete beams in different structural environments, thus improving the engineering adaptability of the technology.

[0138] Based on the same inventive concept, the present invention also provides an intelligent prediction system for the deformation performance index of axially constrained concrete beams. Figure 3 This is a structural schematic diagram of an intelligent prediction system for the deformation performance index of an axially constrained concrete beam, provided as an embodiment of the present invention. Figure 3 As shown, the prediction system includes: an acquisition module 31, used to acquire key design parameters and key deformation performance index parameters of concrete target beams, forming an initial sparse real dataset from multiple concrete target beams; the key design parameters include the axial constraint characterization coefficient η, shear span ratio λ, and bending-shear ratio m, and the key deformation performance index parameters are the component displacement angle limits corresponding to the key performance points; and a preprocessing module 32, used to preprocess the initial sparse real dataset, and to preprocess the real dataset D. real Divided into real training set D real_train Real verification set D real_val and the real test set D real_test The extended dataset generation module 33 is used to construct a physical information neural network model. It trains the physical information neural network model using an initial sparse real dataset and generates physically consistent synthetic data based on the trained model to form the extended dataset D. syn Training module 34 is used for training based on the augmented dataset D. syn Real training set D real_train and the true verification set D real_val A three-stage transfer learning strategy is used to train the prediction model; evaluation module 35 is used to evaluate the model using the real test set D. real_test The trained prediction model is evaluated to obtain a compliant prediction model. Prediction module 36 is used to input the key design parameters of the concrete beam to be predicted into the compliant prediction model, output the key deformation performance index parameters of the concrete beam to be predicted, and determine the ordinary deformation performance index parameters of the concrete beam to be predicted based on the key deformation performance index parameters. The ordinary deformation performance index parameters are the component displacement angle limits corresponding to the ordinary performance points.

[0139] The intelligent prediction system for the deformation performance index of axially constrained concrete beams provided in the above embodiments can execute the intelligent prediction method for the deformation performance index of axially constrained concrete beams provided in the above embodiments, and has the same or corresponding beneficial effects, which will not be elaborated here.

[0140] This invention also provides a storage medium that stores a program or instructions that cause a computer to execute the steps of the intelligent prediction method for axially constrained concrete beam deformation performance indicators provided in the above embodiments.

[0141] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0142] Based on the above embodiments, this invention also provides an electronic device. Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Figure 4 As shown, the electronic device includes a processor 401 and a memory 402. The processor 401 executes programs or instructions stored in the memory, such as... Figure 1 The steps of the intelligent prediction method for the deformation performance index of axially constrained concrete beams described above have the beneficial effects of the above embodiments, and will not be repeated here.

[0143] like Figure 4 As shown, an electronic device may include at least one processor 401, at least one memory 402, and at least one communication interface 403. The various components in the electronic device are coupled together via a bus system 404. The communication interface 403 is used for information transmission with external devices. It is understood that the bus system 404 is used to implement communication between these components. In addition to a data bus, the bus system 404 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in… Figure 4 The general designated all buses as Bus System 404.

[0144] It is understood that the memory 402 in this embodiment can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. In some embodiments, the memory 402 stores the following elements: executable units or data structures, or subsets thereof, or extended sets thereof, operating systems, and applications. In this embodiment of the invention, the processor 401 executes the steps of the various embodiments of the method provided in this embodiment of the invention by calling the programs or instructions stored in the memory 402.

[0145] The method provided in this embodiment of the invention can be applied to processor 401, or implemented by processor 401. Processor 401 can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by integrated logic circuits in the hardware of processor 401 or by instructions in software form. The processor 401 can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor or any conventional processor.

[0146] The steps of the method provided in this embodiment of the invention can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software units in the decoding processor. The software units can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory 402, and processor 401 reads the information in memory 402 and combines it with its hardware to complete the steps of the method.

[0147] The electronic device may also include one or more physical components to execute instructions generated by the processor 401 when performing the methods provided in this embodiment of the invention. Different physical components may be located within the electronic device or outside the electronic device, such as in a cloud server. Each physical component, together with the processor 401 and the memory 402, works to realize the functions of the electronic device in this embodiment.

[0148] The terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0149] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention is not limited to the specific combination of the above-described technical features, but also includes other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the foregoing inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions as those in the present invention.

Claims

1. A method for intelligent prediction of deformation performance indicators of axially constrained concrete beams, characterized in that, include: S1. Obtain the key design parameters and key deformation performance index parameters of the concrete target beam, and form an initial sparse real dataset from the key design parameters and key deformation performance index parameters of the multiple concrete target beams. The key design parameters include the axial constraint characterization coefficient η, shear span ratio λ, and bending-shear ratio m, and the key deformation performance index parameters are the component displacement angle limits corresponding to the key performance points. S2. Preprocess the initial sparse real dataset, and then preprocess the real dataset D. real Divided into real training set D real_train Real verification set D real_val and the real test set D real_test ; S3. Construct a physical information neural network model, train the physical information neural network model using the initial sparse real dataset, and generate physically consistent synthetic data based on the trained physical information neural network model to form an expanded dataset D. syn ; S4, Based on the expanded dataset D syn The real training set D real_train and the real verification set D real_val A three-stage transfer learning strategy was used to train the prediction model. S5. Using the aforementioned real test set D real_test The trained prediction model is evaluated to obtain a qualified prediction model. S6. Input the key design parameters of the concrete beam to be predicted into the standard prediction model, output the key deformation performance index parameters of the concrete beam to be predicted, and determine the ordinary deformation performance index parameters of the concrete beam to be predicted based on the key deformation performance index parameters of the concrete beam to be predicted. The ordinary deformation performance index parameters are the component displacement angle limits corresponding to the ordinary performance points.

2. The intelligent prediction method for deformation performance indicators of axially constrained concrete beams according to claim 1, characterized in that, Obtain the axial constraint characterization coefficient η of the concrete target beam, including: For the actual structural environment of the target concrete beam, establish an equivalent substructure finite element model or an overall structural finite element model that includes itself, adjacent vertical members and connecting floor slabs. The equivalent axial constraint stiffness K at the beam end of the target concrete beam was obtained by performing mechanical analysis on the finite element model. a The formula η=K is used. a / bf c Calculate the dimensionless axial constraint characterization coefficient η; Where b is the beam cross-section width, f c This refers to the compressive strength of concrete.

3. The intelligent prediction method for deformation performance indicators of axially constrained concrete beams according to claim 2, characterized in that, Obtain the equivalent axial restraint stiffness K at the beam end of the target concrete beam. a ,include: Apply a load to the target concrete beam in the finite element model; Obtain the equivalent axial displacement δ at the beam end caused by the applied load on the target concrete beam. A and the end axial reaction force N c ; Using formula K a =N c / δ A Obtain the equivalent axial restraint stiffness K at the beam end of the target concrete beam. a .

4. The intelligent prediction method for deformation performance indicators of axially constrained concrete beams according to claim 1, characterized in that, Key deformation performance parameters of the target concrete beam were obtained, including: Identify three key performance points; The component displacement angle limits corresponding to each key performance point are obtained as θ1, θ5 and θ6, and the component displacement angle limits θ1, θ5 and θ6 are used as key deformation performance index parameters; Wherein, θ1 is the equivalent yield displacement angle limit corresponding to the component in an undamaged state, θ5 is the displacement angle limit corresponding to the bearing capacity decreasing to 80% corresponding to the component in a relatively severely damaged state, and θ6 is the displacement angle limit corresponding to the bearing capacity decreasing to 50% corresponding to the component in a severely damaged state.

5. The intelligent prediction method for deformation performance indicators of axially constrained concrete beams according to claim 4, characterized in that, The key deformation performance parameters of the concrete beam to be predicted include: the component displacement angle limits θ1, θ5, and θ6 corresponding to the three key performance points of the concrete beam to be predicted; the ordinary deformation performance parameters of the concrete beam to be predicted are determined based on the key deformation performance parameters, including: Based on the component displacement angle limits θ1, θ5, and θ6 corresponding to the component displacement angles of the concrete beam to be predicted, determine the component displacement angle limits θ2, θ3, and θ4 corresponding to the ordinary performance points of the concrete beam to be predicted, and obtain them using the following formula: θ2 = 0.25 × (θ5 - θ1) θ3 = 0.50 × (θ5 - θ1) θ4 = 0.75 × (θ5 - θ1) Wherein, θ2 is the displacement angle limit corresponding to the component under slight damage, θ3 is the displacement angle limit corresponding to the component under mild damage, and θ4 is the displacement angle limit corresponding to the component under moderate damage.

6. The intelligent prediction method for deformation performance indicators of axially constrained concrete beams according to claim 1, characterized in that, Based on the trained physical information neural network model, physically consistent synthetic data is generated to form an expanded dataset D. syn ,include: Based on the Latin hypercube sampling method, a large-scale virtual critical design parameter P is generated within a reasonable range of critical design parameters. syn Virtual key design parameter P syn Input the physical information into the neural network model and output the corresponding deformation performance prediction parameter R. syn To obtain the augmented dataset D syn ={P syn R syn } 7. The intelligent prediction method for deformation performance indicators of axially constrained concrete beams according to claim 6, characterized in that, The three-stage transfer learning strategy includes a pre-training stage, a fine-tuning stage, and a validation and early stopping stage. Based on the expanded dataset, a three-stage transfer learning strategy is used to train the prediction model, including: During the pre-training phase, the real training set D is used. real_train and augmented dataset D syn Merge and construct a hybrid training set D mix_train Using a mixed training set D mix_train Train the prediction model; During the fine-tuning phase, the real training set D is used. real_train Perform fine calibration on the prediction model; During the verification and early stopping phases, the real verification set D is used. real_val Validate the performance of the prediction model.

8. The intelligent prediction method for deformation performance indicators of axially constrained concrete beams according to claim 1, characterized in that, The construction of the physical information neural network model includes: embedding the deformation control equation of the concrete beam under the coupled action of compression, bending and shear into the loss function of the physical information neural network in the form of residual terms, so that the network output satisfies the physical law represented by the deformation control equation.

9. The intelligent prediction method for deformation performance indicators of axially constrained concrete beams according to claim 8, characterized in that, Using the real test set D real_test The trained prediction model is then evaluated to obtain a qualifying prediction model, including: During the testing process, multi-dimensional evaluation metrics are obtained, including prediction accuracy metrics, physical consistency metrics, and uncertainty quantification metrics. Determine whether the multi-dimensional evaluation indicators meet the standards; If the multi-dimensional evaluation metrics meet the standards, the trained prediction model is deemed to have met the standards; if the multi-dimensional evaluation metrics do not meet the standards, return to S3 to adjust the physical control equations and / or return to S4 to adjust the model architecture or training strategy to obtain a qualified prediction model.

10. An intelligent prediction system for the deformation performance index of axially constrained concrete beams, characterized in that, include: The acquisition module is used to acquire the key design parameters and key deformation performance index parameters of the concrete target beams, and the key design parameters and key deformation performance index parameters of the multiple concrete target beams form an initial sparse real dataset. The key design parameters include the axial constraint characterization coefficient η, shear span ratio λ, and bending-shear ratio m, and the key deformation performance index parameters are the component displacement angle limits corresponding to the key performance points. The preprocessing module is used to preprocess the initial sparse real dataset, and to preprocess the real dataset D. real Divided into real training set D real_train Real verification set D real_val and the real test set D real_test ; An expanded dataset generation module is used to construct a physical information neural network model, train the physical information neural network model using the initial sparse real dataset, and generate physically consistent synthetic data based on the trained physical information neural network model to form an expanded dataset D. syn ; Training module, used to train based on the augmented dataset D syn The real training set D real_train and the real verification set D real_val A three-stage transfer learning strategy was used to train the prediction model. Evaluation module, used to evaluate the real test set D real_test The trained prediction model is evaluated to obtain a qualified prediction model. The prediction module is used to input the key design parameters of the concrete beam to be predicted into the standard prediction model, output the key deformation performance index parameters of the concrete beam to be predicted, and determine the ordinary deformation performance index parameters of the concrete beam to be predicted based on the key deformation performance index parameters of the concrete beam to be predicted. The ordinary deformation performance index parameters are the component displacement angle limits corresponding to the ordinary performance points.

Images