A creep prediction method for bamboo-wood composites based on rheological topology mapping

Abstract

The application relates to a creep prediction method for bamboo-wood composites based on rheological topological mapping, which comprises the following steps: S1, obtaining microstructure scanning data and environmental parameters of the bamboo-wood composites, and constructing a structure tensor field containing fiber direction information based on the microstructure scanning data and the environmental parameters; S2, constructing a rheological topological mapping network, the input of the rheological topological mapping network being the structure tensor field, the output layer of the rheological topological mapping network being configured to output specific rheological model parameters, and introducing a thermodynamic monotonicity constraint term and a parameter non-negativity constraint term to construct a loss function, the loss function being used for training the rheological topological mapping network; S3, obtaining monitoring data of sparse measuring points in an engineering site, and reversely correcting the rheological model parameters through a Markov chain Monte Carlo sampling algorithm; and S4, substituting the corrected rheological model parameters into a rheological formula to obtain a creep strain.

Description

Technical Field

[0001] This application relates to the field of creep prediction technology for bamboo and wood composite materials, and in particular to a creep prediction method for bamboo and wood composite materials based on rheological topology mapping. Background Technology

[0002] Bamboo-wood composite materials (such as inorganic adhesive composite bamboo beams and reconstituted bamboo) are biomass viscoelastic materials. They exhibit significant creep deformation under long-term loads and are highly sensitive to changes in environmental temperature and humidity.

[0003] Existing technologies have the following main drawbacks in predicting creep in bamboo and wood materials:

[0004] 1. Pure mechanistic models such as Findley and Burgers have limitations. They rely on idealized assumptions and cannot accurately characterize the complex stress field inside bamboo and wood materials caused by initial defects (knots, cracks) and non-uniformity. In addition, the model parameters are fixed and cannot adapt to environmental changes.

[0005] 2. Purely data-driven models such as Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN) are at risk of physical failure, meaning that existing deep learning methods lack physical constraints. With small sample training data, they are prone to making predictions that violate the laws of thermodynamics (such as spontaneous decrease in strain under constant load), and in long-period extrapolation predictions, the error accumulates and diverges over time.

[0006] 3. Poor environmental adaptability: In actual engineering, temperature and humidity vary greatly. Once the existing model is trained, the parameters are fixed and cannot be dynamically corrected using sparse monitoring data from the engineering site to adapt to the new environment, resulting in a significant decrease in prediction accuracy.

[0007] Therefore, there is an urgent need for a high-precision creep prediction method that can both ensure physical reliability and utilize sparse data for adaptive updates. Summary of the Invention

[0008] Therefore, it is necessary to provide a creep prediction method for bamboo-wood composite materials based on rheological topology mapping, including:

[0009] S1: Obtain microstructure scanning data and environmental parameters of bamboo-wood composite materials, and construct a structural tensor field containing fiber orientation information based on the microstructure scanning data and environmental parameters.

[0010] S2: Construct a rheological topology mapping network. The input of the rheological topology mapping network is the structural tensor field. The output layer of the rheological topology mapping network is configured to output specific rheological model parameters. A loss function is constructed by introducing thermodynamic monotonicity constraint terms and parameter nonnegativity constraint terms. The loss function is used to train the rheological topology mapping network.

[0011] S3: Obtain monitoring data from sparse measuring points at the engineering site, and use the Markov chain Monte Carlo sampling algorithm to reverse-correct the parameters of the rheological model;

[0012] S4: Substitute the modified rheological model parameters into the rheological formula to obtain the creep strain.

[0013] Preferably, in S1, the microstructure scanning data includes the local grayscale field of the bamboo-wood composite material, and the environmental parameters include stress tensor, temperature, and humidity;

[0014] The structural tensor is calculated based on the local grayscale field of bamboo-wood composite materials. The calculation formula is as follows:

[0015] ;

[0016] in, Let i represent the structure tensor at the i-th point in voxel space. Let represent the local grayscale field at the i-th point in voxel space. This represents the local density gradient or grayscale gradient at the i-th point in voxel space. Indicates transpose;

[0017] Eigenvalue decomposition is performed on the structural tensor to extract the principal fiber direction, radial constraint direction, and tangential direction of the bamboo-wood composite material.

[0018] The structural tensor field is obtained by combining the stress tensor, temperature, humidity, principal fiber direction, radial constraint direction, and tangential direction.

[0019] Preferably, the rheological topology mapping network includes:

[0020] Input layer, used to obtain the structure tensor field;

[0021] Graph neural networks or multilayer perceptrons are used as backbone networks to map structural tensor fields into rheological model parameters.

[0022] The output layer is used to output specific rheological model parameters.

[0023] The analytical calculation layer is used to define the rheological formula.

[0024] Preferably, the rheological model includes the generalized Kelvin-Voigt model.

[0025] Preferably, the rheological model parameters include modulus and viscosity.

[0026] Preferably, the rheological formula is expressed as follows:

[0027] ;

[0028] in, This represents the creep strain at time t. Represents the stress tensor. This represents the modulus of the first Kelvin-Voigt element. This represents the modulus of the k-th Kelvin-Voigt unit. Indicates the number of Kelvin-Voigt units. This represents the viscosity of the k-th Kelvin-Voigt unit.

[0029] Preferably, the expression for the loss function is:

[0030] ;

[0031] ;

[0032] in, Represents the loss function. This represents the mean square error between the predicted creep strain and the actual strain. Indicates the first weight. This represents the thermodynamic monotonicity constraint term. Represents the ReLU activation function. Indicates creep strain. Indicates time t. Indicates partial derivative, Indicates the second weight. This indicates a non-negativity constraint term, which is used to constrain the modulus and viscosity to be positive.

[0033] Preferably, the gradient descent method is used to train the rheomorphic topology mapping network based on the loss function, and the network parameters of the trained rheomorphic topology mapping network are updated.

[0034] Preferably, S3 includes:

[0035] Acquire monitoring data from sparse measuring points at the engineering site, wherein the monitoring data is the displacement of sparse measuring points on the bamboo-wood composite material;

[0036] A forward model is constructed based on the rheological model parameters. The expression of the forward model is:

[0037] ;

[0038] in, This represents the actual rheological model parameters of the k-th Kelvin-Voigt element. This represents the local defect correction factor. This represents the predicted rheological model parameters for the k-th Kelvin-Voigt unit;

[0039] A likelihood function is constructed based on the monitoring data and the rheological model parameters of the corresponding points;

[0040] The likelihood function is solved using the Markov chain Monte Carlo sampling algorithm to obtain parameter samples that follow a posterior distribution, and the statistics of the parameter samples are calculated.

[0041] Substituting the statistics of the parameter samples into the actual rheological model parameters in the forward model yields the corrected rheological model parameters.

[0042] Preferably, the expression for the likelihood function is:

[0043] ;

[0044] in, Displacement of sparse measuring points With rheological model parameters The likelihood function between them Indicates direct proportion. This represents the predicted displacement at the corresponding point under rheological model parameter B. Describing the L2 norm, This represents the variance of the measurement noise.

[0045] Beneficial effects: This method breaks through the physical failure bottleneck of pure data-driven models in long-term prediction, significantly reduces the model's dependence on long-term monitoring data, and has the ability to adapt to the environment under sparse data. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a flowchart of the creep prediction method for bamboo and wood composite materials based on rheological topology mapping in the embodiments of this application.

[0048] Figure 2 This is a schematic diagram of the architecture of the rheotropic topology mapping network in the embodiments of this application.

[0049] Figure 3 This is a comparison chart of long-cycle creep predictions in the embodiments of this application.

[0050] Figure 4 This is a comparison chart showing the physical consistency verification under noise interference in the embodiments of this application.

[0051] Figure 5 This is a schematic diagram of the parameter adaptive convergence process under sudden environmental changes in an embodiment of this application. Detailed Implementation

[0052] To make the above-mentioned objectives, features, and advantages of this application more apparent and understandable, the specific embodiments of this application are described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of this application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0053] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0054] Example 1

[0055] like Figure 1 As shown, this embodiment provides a creep prediction method for bamboo-wood composite materials based on rheological topology mapping, including:

[0056] S1: Obtain microstructure scanning data and environmental parameters of bamboo-wood composite materials, and construct a structural tensor field containing fiber orientation information based on the microstructure scanning data and environmental parameters.

[0057] In this embodiment, the microstructure scanning data includes the local grayscale field (characterizing density) of the bamboo-wood composite material, and the environmental parameters include stress tensor, temperature, and humidity.

[0058] Specifically, the process of constructing the structure tensor field includes:

[0059] The structural tensor is calculated based on the local grayscale field of bamboo-wood composite materials. The calculation formula is as follows:

[0060] ;

[0061] in, Let i represent the structure tensor at the i-th point in voxel space. Let represent the local grayscale field at the i-th point in voxel space. This represents the local density gradient or grayscale gradient at the i-th point in voxel space. Indicates transpose;

[0062] Eigenvalue decomposition is performed on the structural tensor to extract the principal fiber direction, radial constraint direction, and tangential direction of the bamboo-wood composite material.

[0063] The structural tensor field is obtained by combining the stress tensor, temperature, humidity, principal fiber direction, radial constraint direction, and tangential direction.

[0064] S2: Construct a rheological topology mapping network. The input of the rheological topology mapping network is the structural tensor field. The output layer of the rheological topology mapping network is configured to output specific rheological model parameters. A loss function is constructed by introducing thermodynamic monotonicity constraints and parameter nonnegativity constraints. The loss function is used to train the rheological topology mapping network.

[0065] In this embodiment, a Rheological Topology Mapping Network (RTMN) is proposed, which is a physical information-guided parameter mapping network structure constructed for the task of predicting rheological parameters of bamboo and wood composite materials. The output of this network is the rheological model parameters, which are used to characterize the creep behavior of the material under long-term loading.

[0066] In this embodiment, the core lies in the design of the output layer of the Rheological Topology Mapping Network (RTMN). The last layer of the network is not a regular neuron, but a non-trainable, physically constitutive, analytical computation layer, preventing the network from directly predicting strain. Instead, it predicts the parameters of the rheological model and then calculates the strain using rheological formulas.

[0067] Specifically, such as Figure 2 As shown, the rheological topology mapping network includes:

[0068] Input layer, used to obtain the structure tensor field;

[0069] Graph neural networks or multilayer perceptrons are used as the backbone network, i.e. Figure 2 Hidden layers in the model extract local features to map the structural tensor field to rheological model parameters;

[0070] The output layer is used to output specific rheological model parameters. The number of nodes in the output layer strictly corresponds to the number of parameters in the preset rheological model.

[0071] The analytical computation layer is Figure 2 The physical constitutive layer in the model is used to define the rheological formula. This layer does not participate in training but only performs deterministic physical calculations to ensure that the predicted creep strain necessarily conforms to the rheological form.

[0072] Furthermore, the rheological models include, but are not limited to, the generalized Kelvin-Voigt model, and the rheological model parameters include modulus and viscosity.

[0073] Furthermore, the rheological formula can be expressed as:

[0074] ;

[0075] in, This represents the creep strain at time t. Represents the stress tensor. This represents the modulus of the first Kelvin-Voigt unit. This represents the modulus of the k-th Kelvin-Voigt unit. Indicates the number of Kelvin-Voigt units. This represents the viscosity of the k-th Kelvin-Voigt unit.

[0076] In this example, to ensure that the prediction results conform to the laws of thermodynamics under conditions of scarce training data and noise interference, this invention designs a composite physical loss function. By introducing physical constraints, overfitting that may be caused by data sparsity can be addressed. The expression for the loss function is:

[0077] ;

[0078] ;

[0079] in, Represents the loss function. This represents the mean square error between the predicted creep strain and the actual strain. Indicates the first weight. This represents the thermodynamic monotonicity constraint term, based on the second law of thermodynamics, representing the creep strain rate of a viscoelastic material under constant load. It must be non-negative. That is, creep is irreversible under constant load, i.e., the strain rate... , This represents the ReLU activation function, which penalizes time points that violate the thermodynamic monotonicity constraint. If the predicted strain rate is less than 0, the ReLU activation function will produce a positive penalty value, forcing the network to correct itself. Indicates creep strain. Indicates time t. Indicates partial derivative, Indicates the second weight. This indicates a non-negativity constraint term, which is used to constrain the modulus and viscosity to be positive.

[0080] The rheotropic topology mapping network is trained using the gradient descent method based on the loss function, and the network parameters of the trained rheotropic topology mapping network are then updated.

[0081] S3: Obtain monitoring data from sparse measuring points at the engineering site, and use the Markov chain Monte Carlo sampling algorithm to reverse-correct the rheological model parameters.

[0082] In this embodiment, when environmental conditions (such as humidity) change, the rheological model parameters are inverted using the extremely sparse monitoring data set at the engineering site.

[0083] Specifically, the steps include:

[0084] Acquire monitoring data from sparse measuring points at the engineering site, wherein the monitoring data is the displacement of sparse measuring points on the bamboo-wood composite material;

[0085] A forward model is constructed based on the rheological model parameters. The expression of the forward model is:

[0086] ;

[0087] in, This represents the actual rheological model parameters of the k-th Kelvin-Voigt element. This represents the local defect correction factor. This represents the predicted rheological model parameters for the k-th Kelvin-Voigt unit;

[0088] Based on the monitoring data and the rheological model parameters of the corresponding points, a likelihood function is constructed. The expression of the likelihood function is:

[0089] ;

[0090] in, Displacement of sparse measuring points With rheological model parameters The likelihood function between them Indicates direct proportion. This represents the predicted displacement at the corresponding point under rheological model parameter B. Represents the L2 norm. This represents the variance of the measurement noise.

[0091] The likelihood function is solved using the Markov Chain Monte Carlo (MCMC) sampling algorithm to obtain parameter samples that follow a posterior distribution, and the statistics of the parameter samples are calculated.

[0092] Substituting the statistics of the parameter samples into the actual rheological model parameters in the forward model yields the corrected rheological model parameters.

[0093] The Bayesian MCMC correction algorithm in this embodiment does not require full-field data. As long as the errors of key measurement points converge, the material degradation of unmeasured areas can be inferred through the physical correlation of the network (neighborhood aggregation of GNN).

[0094] S4: Substitute the modified rheological model parameters into the rheological formula to obtain the creep strain.

[0095] The method provided in this embodiment breaks through the physical failure bottleneck of pure data-driven models in long-term prediction, significantly reduces the model's dependence on long-term monitoring data, and has the ability to adapt to the environment under sparse data.

[0096] Example 2

[0097] To verify the effectiveness of the present invention, this embodiment uses the actual physical experimental parameters of inorganic adhesive composite bamboo beams as a benchmark.

[0098] Reference source: Zhao Mingyun. Study on the load-carrying effect of inorganic adhesive composite bamboo beams under bending [D]. Shandong Jianzhu University, 2024.

[0099] Material parameters: The standard value of the short-term bending strength of the inorganic adhesive composite bamboo beam is set at 35.87 MPa, and the elastic modulus is approximately 10 GPa.

[0100] Physical baseline model: The Findley power-law model is adopted. As a rheological formula, where... The initial creep strain is defined. Based on fitted data from literature measurements, a time exponent is set. The creep amplitude coefficient m is dynamically adjusted according to the stress level.

[0101] Experiment 1: Extrapolation and prediction of long-term creep trends using small sample data;

[0102] 1. Experimental setup:

[0103] Simulated engineering site monitoring scenario: Only the deformation data of inorganic glue composite bamboo beam during the initial loading period of 0-500 hours can be obtained as the training set. The model is required to predict the long-term deformation trend of the beam in the next 500-2500 hours as the test set.

[0104] Loading condition: A constant bending load is applied, with a stress level of 30% of the ultimate strength, approximately 10.8 MPa.

[0105] Comparison Algorithms:

[0106] Control group: Standard LSTM Long Short-Term Memory network, representing existing pure data-driven methods.

[0107] The present invention relates to a rheological topology mapping network RTMN: a deep neural network embedded with a Findley rheological topology layer.

[0108] 2. Experimental Results:

[0109] Table 1 below shows a comparison of the prediction accuracy of the two models for future deformation after the training data cutoff point (500h).

[0110] Table 1. Comparison of prediction results between LSTM and RTMN networks

[0111] Time (hours) Actual deformation value (mm) [Based on Findley benchmark] Current LSTM prediction values ​​(mm) RTMN prediction value (mm) of this invention LSTM relative error RTMN relative error 500 (Training deadline) 2.10 2.09 2.10 0.4% 0.0% 1000 2.18 2.12 2.19 2.7% 0.4% 1500 2.23 2.14 2.24 4.0% 0.4% 2000 2.27 2.15 2.28 5.3% 0.4% 2500 2.30 2.16 2.31 6.1% 0.4%

[0112] 3. Results Analysis:

[0113] From Table 1, Figure 3 It is known that existing LSTM networks, lacking physical mechanism support, cannot accurately capture the unique "power-law growth" long-tail effect of bamboo and wood creep, leading to significant "underfitting" in long-term predictions, with a cumulative error of 6.1% over 2500 hours. In contrast, the RTMN method of this invention, due to hard-coding the rheological topology in the network output layer, ensures that its prediction curve closely matches the actual physical trajectory even in extrapolation regions lacking training data, with a maximum relative error of only 0.4%. This demonstrates that this invention possesses extremely high accuracy and stability in small-sample, long-period prediction scenarios.

[0114] Experiment 2: Verification of physical consistency under strong noise interference;

[0115] 1. Experimental setup:

[0116] To simulate real-world engineering scenarios where sensors are subjected to electromagnetic interference or environmental vibration, ±5% Gaussian random noise is superimposed onto real creep data as model input. The goal is to verify whether the model will output erroneous predictions that violate the second law of thermodynamics (such as the spontaneous decrease of creep strain over time).

[0117] 2. Experimental Results:

[0118] Table 2 below shows a comparison of the prediction accuracy of future deformation with and without constraints after extracting key time segments.

[0119] Table 2 Comparison of prediction results between unconstrained DNN and RTMN networks

[0120] ;

[0121] 3. Results Analysis:

[0122] From Table 2, Figure 4It can be seen that at t=102h, the sensor data experienced an abnormally sharp drop due to noise, from 1.92 to 1.78. Traditional unconstrained deep neural network (DNN) models blindly fit the noise, predicting a decrease in strain, resulting in a negative calculated instantaneous strain rate. This violates the physical principle of irreversible creep under constant load. The method of this invention benefits from the thermodynamic monotonicity constraint term built into the loss function. It can automatically identify and filter out noise fluctuations that violate the laws of physics, and force the output to increase monotonically by 1.86mm, thus ensuring the physical reliability of the prediction results.

[0123] Experiment 3: Sparse Bayesian adaptive update under environmental changes;

[0124] 1. Experimental setup:

[0125] Bamboo and wood materials are highly sensitive to humidity. According to literature, inorganic adhesive composite bamboo beams exhibit significantly stronger creep effects in fluctuating indoor humidity environments (condition B) than in constant temperature and humidity environments (condition A).

[0126] Initial state: The model has been pre-trained on "constant temperature and humidity" data (parameter m) init = 0.148).

[0127] Operating Condition Switching: The model is deployed to an "indoor fluctuating humidity" environment, where the actual material creep amplitude parameter has changed to m. true ≈0.230.

[0128] Data conditions: Only 3 sparse measurement points (hours 100, 300, and 500) under operating condition B were acquired.

[0129] Task: Use the Markov Chain Monte Carlo Sampling Algorithm (MCMC) of this invention to reverse-correct model parameters.

[0130] 2. Experimental Results:

[0131] Table 3 below shows the impact of MCMC correction on model environment adaptation;

[0132] Table 3. Impact of MCMC Correction on Prediction Error

[0133] Iteration phase Model parameter state (m) 1000-hour forecast error (RMSE) Status Description initial state 0.148 (pre-training value) 18.5% The model severely underestimates deformation and is not adapted to the new environment. Iteration 1 0.192 8.2% Bayesian engines begin to perceive environmental differences Iteration 3 0.225 3.4% The parameters quickly approximate the true value. Iteration 5 0.231 (convergence value) 2.1% Model completes adaptive calibration Truth Reference 0.230 (measured value from literature) 0.0% -

[0134] 3. Results Analysis:

[0135] From Table 3, Figure 5It is known that pre-trained models alone cannot adapt to environmental changes, with an initial error as high as 18.5%. Using the sparse Bayesian online update method proposed in this invention, the system can correct the key physical parameter m to near its true value within 5 iterations using only data from 3 sparse measurement points, rapidly reducing the prediction error to 2.1%. This demonstrates that the system possesses strong environmental adaptability, greatly reducing the reliance on recalibrated data in engineering projects.

[0136] The creep prediction method for bamboo-wood composite materials based on rheological topology mapping provided in this embodiment has the following beneficial effects:

[0137] 1. Overcoming the physical failure bottleneck of pure data-driven models in long-term prediction: By using a rheological topology mapping network (RTMN), the creep constitutive equation is embedded in the network output layer, and thermodynamic monotonicity constraints are introduced. Experiments show that even under conditions with 5% random noise, this invention can still ensure that the output strain rate remains non-negative. This eliminates physical fallacies and ensures the safety of engineering applications.

[0138] 2. Significantly reduces the model's dependence on long-term monitoring data and solves the small sample problem: Traditional methods require thousands of hours of continuous monitoring data to train a reliable model. Simulation experiments based on inorganic adhesive composite bamboo beams show that by using only the first 500 hours of short-term data for training, this invention can accurately predict the deformation trend for the next 2000 hours, with a relative error controlled within 0.4%, while the conventional LSTM model under the same conditions has an error as high as 6.1%.

[0139] 3. Environmental Adaptability under Sparse Data: Addressing the challenges of bamboo and wood materials' high sensitivity to temperature and humidity and the sparseness of monitoring points at engineering sites, this invention constructs a sparse Bayesian targeted update mechanism. Simulation verification shows that after a sudden change in environmental conditions, only 5 iterations using 3 sparse monitoring points are needed to reduce the model prediction error from 18.5% to 2.1%, achieving a true digital twin for structural health monitoring.

[0140] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0141] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for predicting creep in bamboo-wood composite materials based on rheological topology mapping, characterized in that, include: S1: Obtain microstructure scanning data and environmental parameters of bamboo-wood composite materials, and construct a structural tensor field containing fiber orientation information based on the microstructure scanning data and environmental parameters. S2: Construct a rheological topology mapping network. The input of the rheological topology mapping network is the structural tensor field. The output layer of the rheological topology mapping network is configured to output rheological model parameters. A loss function is constructed by introducing thermodynamic monotonicity constraints and parameter nonnegativity constraints. This loss function is used to train the rheological topology mapping network. The rheological model parameters include modulus and viscosity. The expression of the loss function is: ； ； in, Represents the loss function. This represents the mean square error between the predicted creep strain and the actual strain. Indicates the first weight. This represents the thermodynamic monotonicity constraint term, indicating that creep is irreversible under constant load, i.e., strain rate. , This represents the ReLU activation function, which penalizes time points that violate the thermodynamic monotonicity constraint. If the predicted strain rate is less than 0, the ReLU activation function will produce a positive penalty value, forcing the network to correct itself. Indicates creep strain. Indicates time t. Indicates partial derivative, Indicates the second weight. This indicates a non-negativity constraint term, which is used to constrain the modulus and viscosity to be positive. S3: Obtain monitoring data from sparse measuring points at the engineering site, and use the Markov chain Monte Carlo sampling algorithm to reverse-correct the parameters of the rheological model; S4: Substitute the modified rheological model parameters into the rheological formula to obtain the creep strain.

2. The creep prediction method for bamboo-wood composite materials based on rheological topology mapping according to claim 1, characterized in that, In S1, the microstructure scanning data includes the local grayscale field of the bamboo-wood composite material, and the environmental parameters include stress tensor, temperature, and humidity. The structural tensor is calculated based on the local grayscale field of bamboo-wood composite materials. The calculation formula is as follows: ； in, Let i represent the structure tensor at the i-th point in voxel space. Let represent the local grayscale field at the i-th point in voxel space. This represents the local density gradient or grayscale gradient at the i-th point in voxel space. Indicates transpose; Eigenvalue decomposition is performed on the structural tensor to extract the principal fiber direction, radial constraint direction, and tangential direction of the bamboo-wood composite material. The structural tensor field is obtained by combining the stress tensor, temperature, humidity, principal fiber direction, radial constraint direction, and tangential direction.

3. The creep prediction method for bamboo-wood composite materials based on rheological topology mapping according to claim 1, characterized in that, Rheological topology mapping networks include: Input layer, used to obtain the structure tensor field; Graph neural networks or multilayer perceptrons are used as backbone networks to map structural tensor fields into rheological model parameters. The output layer is used to output the rheological model parameters; The analytical calculation layer is used to define the rheological formula.

4. The creep prediction method for bamboo-wood composite materials based on rheological topology mapping according to claim 1, characterized in that, Rheological models include the generalized Kelvin-Voigt model.

5. The method for predicting creep of bamboo-wood composite materials based on rheological topology mapping according to claim 4, characterized in that, The rheological formula is expressed as follows: ； in, This represents the creep strain at time t. Represents the stress tensor. This represents the modulus of the first Kelvin-Voigt unit. This represents the modulus of the k-th Kelvin-Voigt unit. Indicates the number of Kelvin-Voigt units. This represents the viscosity of the k-th Kelvin-Voigt unit.

6. The method for predicting creep of bamboo-wood composite materials based on rheological topology mapping according to claim 1, characterized in that, The rheotropic topology mapping network is trained using the gradient descent method based on the loss function, and the network parameters of the trained rheotropic topology mapping network are then updated.

7. The creep prediction method for bamboo-wood composite materials based on rheological topology mapping according to claim 1, characterized in that, S3 include: Acquire monitoring data from sparse measuring points at the engineering site, wherein the monitoring data is the displacement of sparse measuring points on the bamboo-wood composite material; A forward model is constructed based on the rheological model parameters. The expression of the forward model is: ； in, This represents the actual rheological model parameters of the k-th Kelvin-Voigt element. This represents the local defect correction factor. This represents the predicted rheological model parameters for the k-th Kelvin-Voigt unit; A likelihood function is constructed based on the monitoring data and the rheological model parameters of the corresponding points; The likelihood function is solved using the Markov chain Monte Carlo sampling algorithm to obtain parameter samples that follow a posterior distribution, and the statistics of the parameter samples are calculated. Substituting the statistics of the parameter samples into the actual rheological model parameters in the forward model yields the corrected rheological model parameters.

8. The method for predicting creep of bamboo-wood composite materials based on rheological topology mapping according to claim 7, characterized in that, The expression for the likelihood function is: ； in, Displacement of sparse measuring points With rheological model parameters The likelihood function between them Indicates direct proportion. This represents the predicted displacement at the corresponding point under rheological model parameter B. Represents the L2 norm. This represents the variance of the measurement noise.

Images