Data interaction method, system, equipment and medium for engineering test and numerical simulation

Abstract

The present invention relates to the field of engineering computing technology, and specifically to a data interaction method, system, equipment and medium for engineering experiments and numerical simulations. The method comprises: building a physical engineering model of corresponding level according to an experimental object, and selecting numerical simulation software corresponding to the physical engineering model; determining the interaction nodes corresponding to each level of the physical engineering model and the numerical simulation software; constructing a dynamic constitutive model of the experimental object, and establishing a deep learning model to optimize the dynamic constitutive model, and deriving an optimal dynamic constitutive model; applying the optimal dynamic constitutive model to physical engineering model experiments and numerical simulation software simulations, and determining the experimental step size and experimental time, and simultaneously conducting physical engineering model experiments and numerical simulation software simulations; and respectively collecting engineering test data and numerical simulation data of interaction nodes at each level of the physical engineering model and the numerical simulation software to perform data interaction.

Description

Technical Field

[0001] The present invention relates to the technical field of engineering experiments, and in particular to a data interaction method, system, equipment and medium for engineering experiments and numerical simulations. Background Art

[0002] In the model test research of geotechnical engineering problems, the main research methods currently include physical model test, numerical simulation, field test, etc. All of the above research methods have their own shortcomings.

[0003] For physical engineering experiments, the physical processes in geotechnical engineering can be realistically simulated, and the reliability of the results can be improved through experimental repeatability verification and stability testing. However, the cost of the experiment is high, requiring a lot of time, resources and equipment, and is limited by the site. The model cannot be infinitely large, and considering the size effect of the model, it is impossible to fully replicate the actual engineering conditions. Some complex phenomena or difficult-to-measure parameters may not be accurately reproduced.

[0004] For numerical simulation, by simulating the dynamic behaviors and interactions that occur in complex geotechnical engineering, results can be obtained in a shorter time. Although the model parameters and conditions can be flexibly modified, parameter sensitivity analysis and optimization can be performed. However, compared with the real engineering scenario, the accuracy of the model is limited by the quality of the input data and the rationality of the model assumptions. In addition, the selection and calibration of model parameters may be subjective, and the calculation error is large.

[0005] For field tests, research can be conducted at the actual engineering site to provide data that is closest to the actual engineering. However, there are also some problems, such as the experimental conditions are difficult to control and are affected by environmental and external factors. The cost of field tests is high and requires a lot of time and human resources. In addition, some field tests cannot cover engineering behaviors under specific conditions. Summary of the invention

[0006] In order to solve the above problems, the present invention provides a data interaction method for engineering test and numerical simulation in a first aspect, comprising:

[0007] Establish grading standards based on experimental complexity, experimental data accuracy requirements, and experimental environment, build physical engineering models of corresponding levels according to experimental objects, and select numerical simulation software corresponding to physical engineering models;

[0008] Divide the levels of data interaction between the physical engineering model and the numerical simulation software, and determine the interaction nodes corresponding to each level of the physical engineering model and the numerical simulation software;

[0009] By collecting relevant data of the experimental object in the physical engineering model, synchronizing it to the numerical simulation software to build a dynamic constitutive model of the experimental object, and establishing a deep learning model to optimize the dynamic constitutive model, the optimal dynamic constitutive model is derived;

[0010] Apply the optimal dynamic constitutive model to the physical engineering model test and numerical simulation software simulation, and determine the experimental step length and experimental time, and conduct physical engineering model tests and numerical simulation software simulation at the same time;

[0011] The engineering test data and numerical simulation data of the interaction nodes at each level of the physical engineering model and the numerical simulation software are collected respectively, and the engineering test data of the physical engineering model is input into the corresponding interaction nodes of the numerical simulation software. At the same time, the physical engineering model receives the numerical simulation data output by the corresponding interaction nodes of the numerical simulation software, and the data interaction is carried out synchronously in real time.

[0012] In some implementations of the first aspect, the collecting of relevant data of the experimental object in the physical engineering model also includes collecting interface load, stress, strain and displacement data of the experimental object in the physical engineering model, producing a sample data set after preprocessing, and training the deep learning model through the sample data set.

[0013] The specific operations of the pre-processing are:

[0014] Perform data cleaning: process missing values ​​and outliers in the data to eliminate noise and outliers and ensure the integrity and reliability of the data;

[0015] Data labeling: Process the cleaned data, extract key features of the data and label them;

[0016] Perform data segmentation: Divide the data into training data and test data according to the preset ratio.

[0017] The specific operation of establishing the deep learning model to optimize the dynamic constitutive model is:

[0018] Divide the sample data set for deep learning model training;

[0019] Test and evaluate trained deep learning models;

[0020] Based on the evaluation results of the deep learning model, the parameters of the dynamic constitutive model are optimized;

[0021] The optimized dynamic constitutive model is verified based on the validation set divided by the sample data set.

[0022] In some implementations of the first aspect, the deep learning model adopts a Faster R-CNN model.

[0023] The data interaction method of the engineering test and numerical simulation also includes constructing a deep learning framework TensorFlow for distributed training during the process of optimizing the dynamic constitutive model by the deep learning model, and the specific method is as follows:

[0024] Prepare a pre-trained deep learning model and load it in the deep learning framework TensorFlow;

[0025] Set the cluster configuration for distributed training, including setting the master node and worker node information, communication protocol, and port;

[0026] Select a distributed training strategy based on the cluster configuration for distributed training;

[0027] Prepare data sets for training and use data parallel strategies to perform distributed training on deep learning models;

[0028] Collect the accuracy and loss value of the model during the training process, evaluate the deep learning model, and obtain a trained deep learning model.

[0029] In some implementations of the first aspect, the data interaction also includes stopping the data interaction and terminating the experiment based on the experimental step size and experimental time, or based on the destruction of the dynamic constitutive model of the experimental object, and judging whether the experimental step size and the experimental time reach certain values. If so, stopping the data interaction and terminating the experiment; if not, judging whether the dynamic constitutive model is destroyed, if so, stopping the data interaction and terminating the experiment; if otherwise, continuing to execute the data interaction.

[0030] The second aspect provides a data interaction system for engineering experiments and numerical simulations, including:

[0031] The initial module establishes a grading standard based on the complexity of the experiment, the accuracy requirements of the experimental data, and the experimental environment, builds a physical engineering model of the corresponding level according to the experimental object, and selects the numerical simulation software of the corresponding physical engineering model;

[0032] A node determination module divides the levels of data interaction between the physical engineering model and the numerical simulation software, and determines the interaction nodes corresponding to each level of the physical engineering model and the numerical simulation software;

[0033] The model building module collects relevant data of the experimental object in the physical engineering model, synchronizes it to the numerical simulation software to build a dynamic constitutive model of the experimental object, and establishes a deep learning model to optimize the dynamic constitutive model and derive the optimal dynamic constitutive model;

[0034] The test simulation module applies the optimal dynamic constitutive model to the physical engineering model test and numerical simulation software simulation, determines the experimental step length and experimental time, and conducts the physical engineering model test and numerical simulation software simulation at the same time;

[0035] The interactive module collects the engineering test data and numerical simulation data of the interactive nodes at each level of the physical engineering model and the numerical simulation software, inputs the engineering test data of the physical engineering model into the corresponding interactive nodes of the numerical simulation software, and at the same time, the physical engineering model receives the numerical simulation data output by the corresponding interactive nodes of the numerical simulation software, and performs data interaction in real time and synchronously.

[0036] The third aspect provides a data interaction device for engineering experiments and numerical simulations, comprising a processor and a memory, wherein the processor implements the data interaction method for engineering experiments and numerical simulations as described above when executing program data stored in the memory.

[0037] A fourth aspect provides a readable storage medium for storing control program data, wherein the control program data, when executed by a processor, implements the data interaction method between engineering experiments and numerical simulations as described above.

[0038] The beneficial effects are as follows: the present invention provides a data interaction method for engineering experiments and numerical simulations, which classifies large and complex engineering experiments by selecting physical engineering models and numerical simulation software, adopts multi-layer physical engineering models and numerical simulation software to be carried out simultaneously, transmits experimental data step by step, and breaks through the model size limitation; at the same time, deep learning technology is used to optimize the dynamic constitutive model, and the results of physical engineering experiments and numerical simulation experiments are output synchronously, and the engineering test data and numerical simulation data are analyzed and predicted, so as to realize the effective combination of field experiments and numerical simulations, and make the experimental results more efficient and accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 Flow chart of the data interaction method between engineering experiments and numerical simulations. DETAILED DESCRIPTION

[0040] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings.

[0041] See also Figure 1 This embodiment provides a data interaction method for engineering experiments and numerical simulations, and the specific implementation steps are as follows:

[0042] S1. Establish grading standards based on experimental complexity, experimental data accuracy requirements, and experimental environment, build physical engineering models of corresponding levels according to experimental objects, and select numerical simulation software corresponding to physical engineering models;

[0043] In S1, a grading standard is established based on the complexity of the experiment, the accuracy requirements of the experimental data, and the experimental environment. The complexity of the experiment is divided into simple, medium, and difficult; the accuracy requirements of the experimental data are divided into low, medium, and high; and the experimental environment is divided into conventional, special, and extreme.

[0044] Establishing a grading standard based on the complexity of the experiment, the accuracy requirements of the experimental data and the experimental environment. The grading standard includes primary standard, intermediate standard and advanced standard. The primary standard represents a basic experiment with simple experimental complexity and carried out in a conventional environment, and has a low requirement for the accuracy of the experimental data.

[0045] The intermediate standard means that the experiment is of medium complexity and needs to be carried out under conventional or special conditions, with high requirements for the accuracy of experimental data;

[0046] The advanced standard indicates that the experiment is difficult and complex, needs to be carried out in extreme environments, and requires extremely high accuracy of experimental data.

[0047] Evaluate the experimental objects and experimental data accuracy requirements to determine the grading standards, build physical engineering models of corresponding levels for engineering experiments, and select numerical simulation software corresponding to the physical engineering models for numerical simulation.

[0048] S2, dividing the levels of data interaction between the physical engineering model and the numerical simulation software, and determining the interaction nodes corresponding to each level of the physical engineering model and the numerical simulation software;

[0049] The experimental data of the physical engineering model is used as the initial value and boundary condition for the simulation of the numerical simulation software, and the results output by the numerical simulation software are fed back to the physical engineering model to achieve dynamic interaction.

[0050] According to the level of the constructed physical engineering model, the output nodes of the physical engineering model and the numerical simulation software are hierarchically divided. The interactive nodes are the boundary nodes of the experimental box in the physical engineering model or the interface nodes between the structure and the soil. This enables large-scale engineering experiments to transmit experimental data in a hierarchical manner, thereby improving the accuracy of engineering experimental data.

[0051] S3. Collect relevant data of the experimental object in the physical engineering model, synchronize it to the numerical simulation software to build a dynamic constitutive model of the experimental object, and establish a deep learning model to optimize the dynamic constitutive model to derive the optimal dynamic constitutive model;

[0052] This application constructs a dynamic constitutive model to describe the relationship between stress and strain of the experimental object material. Through the data acquisition device of the physical engineering model, the interface load, stress, strain and displacement data of the experimental object in the physical engineering model are collected, and after preprocessing, a sample data set for training the model is produced. The preprocessing operation is:

[0053] Perform data cleaning: process missing values ​​and outliers in experimental data to eliminate noise and outliers and ensure the integrity and reliability of experimental data; missing values ​​refer to the lack of specific characteristic values ​​and information of certain data points, and data samples with missing values ​​are deleted through data cleaning. Outliers refer to values ​​that do not match most data points in the data, which may be caused by collection errors, equipment failures or other abnormal situations; outliers are replaced with reasonable values ​​by deleting or correcting outliers.

[0054] Data annotation: Process the cleaned data, extract key features of the data and annotate them, including the bounding box and category label of the experimental object;

[0055] Perform data segmentation: Divide the data into training data and test data according to the preset ratio.

[0056] At the same time, a deep learning model is established and trained with a sample data set. Specifically, the deep learning model adopts the Faster R-CNN model, which includes a feature extraction network, a region candidate network, a domain of interest pooling, and a target recognition network. The feature extraction network is used to extract image features; the region candidate network is used to generate candidate target regions, detect the extracted image features to determine the target position; the domain of interest pooling is used to collect the coordinates of the candidate target regions in the region candidate network, extract the target, and transmit it to the target recognition network for recognition, classification, and regression.

[0057] The Faster R-CNN model is trained using the sample data set, and the Faster R-CNN model is used to optimize the parameters of the dynamic constitutive model. The specific operations are as follows:

[0058] S3.1, divide the sample data set for deep learning model training;

[0059] The sample data set is divided into training set, test set and validation set, and normalized and data augmented, and the training set is used to train the deep learning model.

[0060] S3.2. Test and evaluate the trained deep learning model;

[0061] The trained deep learning model is tested using a test set to calculate the accuracy, recall, and precision of the model, compare the error between the predicted data and the experimental data of the deep learning model, and evaluate the performance and effect of the deep learning model.

[0062] S3.3. Optimize the parameters of the dynamic constitutive model based on the evaluation results of the deep learning model;

[0063] S3.4. Verify the optimized dynamic constitutive model based on the validation set divided by the sample data set.

[0064] At the same time, the Faster R-CNN model used in this application to optimize the parameters of the dynamic constitutive model takes a long time and consumes a lot of resources. By building a deep learning framework TensorFlow to implement distributed training, the performance and effect of the Faster R-CNN model on the optimization of the parameters of the dynamic constitutive model can be further improved.

[0065] The operations of constructing the deep learning framework TensorFlow for distributed training are specifically as follows:

[0066] Prepare the pre-trained Faster R-CNN model and load it in the deep learning framework TensorFlow;

[0067] Set the cluster configuration for distributed training, including setting the master node and worker node information, communication protocol, and port;

[0068] Select a distributed training strategy based on the cluster configuration for distributed training;

[0069] Prepare the dataset for training and use data parallel strategy to perform distributed training on Faster R-CNN model;

[0070] The accuracy and loss value of the model during the training process are collected, and the Faster R-CNN model is evaluated to obtain the trained Faster R-CNN model.

[0071] According to the evaluation results of the trained Faster R-CNN model, the parameters of the dynamic constitutive model are adjusted and optimized to obtain the optimal dynamic constitutive model and export it.

[0072] S4. Apply the optimal dynamic constitutive model to the physical engineering model test and numerical simulation software simulation, determine the experimental step length and experimental time, and conduct the physical engineering model test and numerical simulation software simulation at the same time;

[0073] S5. Collect engineering test data and numerical simulation data from interaction nodes at each level of the physical engineering model and the numerical simulation software respectively, input the engineering test data of the physical engineering model into the corresponding interaction node of the numerical simulation software, and at the same time, the physical engineering model receives the numerical simulation data output by the corresponding interaction node of the numerical simulation software, and performs data interaction synchronously in real time.

[0074] By exchanging the engineering test data of the physical engineering model with the computational simulation data of the numerical simulation software, the comparison and verification of the physical engineering test and the numerical simulation results can be achieved to ensure the accuracy and reliability of the physical engineering model test.

[0075] At the same time, the data interaction is also based on the experimental step length and experimental time, or according to the destruction of the dynamic constitutive model of the experimental object, the data interaction is stopped to terminate the experiment, and the specific operations are:

[0076] Determine whether the experimental step length and experimental time have reached the determined value. If so, stop the data interaction and terminate the experiment; if not, determine whether the dynamic constitutive model is destroyed. If so, stop the data interaction and terminate the experiment; if not, continue to execute data interaction.

[0077] In addition, the present application also provides a data interaction system for engineering experiments and numerical simulations, including:

[0078] The initial module establishes a grading standard based on the complexity of the experiment, the accuracy requirements of the experimental data, and the experimental environment, builds a physical engineering model of the corresponding level according to the experimental object, and selects the numerical simulation software of the corresponding physical engineering model;

[0079] A node determination module divides the levels of data interaction between the physical engineering model and the numerical simulation software, and determines the interaction nodes corresponding to each level of the physical engineering model and the numerical simulation software;

[0080] The training optimization module constructs the dynamic constitutive model of the experimental object, collects relevant data of the experimental object in the physical engineering model, and establishes a deep learning model to optimize the dynamic constitutive model and derive the optimal dynamic constitutive model;

[0081] The test simulation module imports the optimal dynamic constitutive model into the numerical simulation software, determines the experimental step length and experimental time, and simultaneously conducts physical engineering model tests and numerical simulation software simulations;

[0082] The interactive module collects the engineering test data and numerical simulation data of the interactive nodes at each level of the physical engineering model and the numerical simulation software, inputs the engineering test data of the physical engineering model into the corresponding interactive nodes of the numerical simulation software, and at the same time, the physical engineering model receives the numerical simulation data output by the corresponding interactive nodes of the numerical simulation software to conduct data interaction.

[0083] Secondly, a data interaction device for engineering experiments and numerical simulations is also provided, comprising a processor and a memory, wherein the processor implements the data interaction method for engineering experiments and numerical simulations as described above when executing program data stored in the memory.

[0084] Finally, a readable storage medium is provided for storing control program data, wherein the control program data, when executed by a processor, implements the data interaction method between engineering experiments and numerical simulations as described above.

Claims

1. A data interaction method for engineering experiments and numerical simulations, characterized in that: include: Establish grading standards based on experimental complexity, experimental data accuracy requirements, and experimental environment, build physical engineering models of corresponding levels according to experimental objects, and select numerical simulation software corresponding to physical engineering models; Divide the levels of data interaction between the physical engineering model and the numerical simulation software, and determine the interaction nodes corresponding to each level of the physical engineering model and the numerical simulation software; By collecting relevant data of the experimental object in the physical engineering model, synchronizing it to the numerical simulation software to build a dynamic constitutive model of the experimental object, and establishing a deep learning model to optimize the dynamic constitutive model, and deriving the optimal dynamic constitutive model; the deep learning model adopts the Faster R-CNN model; In the process of optimizing the dynamic constitutive model by the deep learning model, the deep learning framework TensorFlow is constructed for distributed training. The specific method is as follows: Prepare a pre-trained deep learning model and load it in the deep learning framework TensorFlow; Set the cluster configuration for distributed training, including setting the master node and worker node information, communication protocol, and port; Select a distributed training strategy based on the cluster configuration for distributed training; Prepare data sets for training and use data parallel strategies to perform distributed training on deep learning models; Collect the accuracy and loss value of the model during the training process, evaluate the deep learning model, and obtain a trained deep learning model; Apply the optimal dynamic constitutive model to the physical engineering model test and numerical simulation software simulation, and determine the experimental step length and experimental time, and conduct physical engineering model tests and numerical simulation software simulation at the same time; The engineering test data and numerical simulation data of the interaction nodes at each level of the physical engineering model and the numerical simulation software are collected respectively, and the engineering test data of the physical engineering model is input into the corresponding interaction nodes of the numerical simulation software. At the same time, the physical engineering model receives the numerical simulation data output by the corresponding interaction nodes of the numerical simulation software, and the data interaction is carried out synchronously in real time.

2. The data interaction method according to claim 1, characterized in that: The method of collecting relevant data of the experimental object in the physical engineering model also includes collecting interface load, stress, strain and displacement data of the experimental object in the physical engineering model, making a sample data set after preprocessing, and training the deep learning model through the sample data set.

3. The data interaction method according to claim 2, characterized in that: The specific operation of establishing the deep learning model to optimize the dynamic constitutive model is: Divide the sample data set for deep learning model training; Test and evaluate trained deep learning models; Based on the evaluation results of the deep learning model, the parameters of the dynamic constitutive model are optimized; The optimized dynamic constitutive model is verified based on the validation set divided by the sample data set.

4. The data interaction method according to claim 1, characterized in that: The data interaction also includes stopping the data interaction and terminating the experiment based on the experiment step and the experiment time, or according to the destruction of the dynamic constitutive model of the experimental object, and judging whether the experiment step and the experiment time reach a certain value. If so, stopping the data interaction and terminating the experiment; if not, judging whether the dynamic constitutive model is destroyed, if so, stopping the data interaction and terminating the experiment; if not, continuing to execute the data interaction.

5. The data interaction method according to claim 2, characterized in that: The specific operations of the pre-processing are: Perform data cleaning: process missing values ​​and outliers in the data to eliminate noise and outliers and ensure the integrity and reliability of the data; Data labeling: Process the cleaned data, extract key features of the data and label them; Perform data segmentation: Divide the data into training data and test data according to the preset ratio.

6. A data interaction system for engineering experiments and numerical simulations, characterized in that: include: The initial module establishes a grading standard based on the complexity of the experiment, the accuracy requirements of the experimental data, and the experimental environment, builds a physical engineering model of the corresponding level according to the experimental object, and selects the numerical simulation software of the corresponding physical engineering model; A node determination module divides the levels of data interaction between the physical engineering model and the numerical simulation software, and determines the interaction nodes corresponding to each level of the physical engineering model and the numerical simulation software; The model building module collects relevant data of the experimental object in the physical engineering model, synchronizes it to the numerical simulation software to build a dynamic constitutive model of the experimental object, and establishes a deep learning model to optimize the dynamic constitutive model and derive the optimal dynamic constitutive model; the deep learning model adopts the Faster R-CNN model; In the process of optimizing the dynamic constitutive model by the deep learning model, the deep learning framework TensorFlow is constructed for distributed training, which is specifically as follows: Prepare a pre-trained deep learning model and load it in the deep learning framework TensorFlow; Set the cluster configuration for distributed training, including setting the master node and worker node information, communication protocol, and port; Select a distributed training strategy based on the cluster configuration for distributed training; Prepare data sets for training and use data parallel strategies to perform distributed training on deep learning models; Collect the accuracy and loss value of the model during the training process, evaluate the deep learning model, and obtain a trained deep learning model; The test simulation module applies the optimal dynamic constitutive model to the physical engineering model test and numerical simulation software simulation, determines the experimental step length and experimental time, and conducts the physical engineering model test and numerical simulation software simulation at the same time; The interactive module collects the engineering test data and numerical simulation data of the interactive nodes at each level of the physical engineering model and the numerical simulation software, inputs the engineering test data of the physical engineering model into the corresponding interactive nodes of the numerical simulation software, and at the same time, the physical engineering model receives the numerical simulation data output by the corresponding interactive nodes of the numerical simulation software, and performs data interaction in real time and synchronously.

7. A data interaction device for engineering experiments and numerical simulations, characterized in that: It comprises a processor and a memory, wherein when the processor executes the program data stored in the memory, the data interaction method of the engineering test and numerical simulation as described in any one of claims 1 to 5 is implemented.

8. A readable storage medium, characterized in that: Used to store control program data, wherein the control program data, when executed by a processor, implements the data interaction method for engineering test and numerical simulation as described in any one of claims 1 to 5.

Images