A fast calculation method for global strain field of bridges

Abstract

The present invention discloses a method for rapidly calculating the global strain field of a bridge, which relates to the field of bridge structure health monitoring. The method comprises: firstly, generating stress simulation data of sampling positions of the entire bridge based on finite element simulation; then, performing supervised training on a calculation model based on the stress simulation data of sampling positions of the entire bridge; finally, using the real-time measurement data of the bridge stress sensor as input, and performing rapid calculation of the stress of the entire bridge using the trained calculation model. The present invention realizes the establishment of a calculation workflow for a finite element model of a bridge under the action of random bridge deck loads with full automatic processing throughout the entire process.

Description

Technical Field

[0001] The present invention relates to the field of bridge structure health monitoring, and in particular to a method for quickly calculating the global strain field of a bridge. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and may not constitute prior art.

[0003] In the process of bridge structure health monitoring, strain sensors are currently often used to monitor the stress and strain conditions of the control section. However, due to the limited number of sensors, all these monitoring data cannot fully reflect the stress and strain conditions of all sections of the entire bridge structure.

[0004] To obtain the global stress and strain field distribution of a bridge, the current method is to iterate the measured data from strain sensors into the original bridge finite element model to obtain the stress and strain status of the entire bridge structure. However, this method is an offline operation and generally has high computational costs (it takes a long time under ordinary computing terminal configurations), making it impossible to effectively understand the overall stress and strain response status of the bridge in real time. Summary of the Invention

[0005] In order to effectively solve the problem of obtaining real-time information on the global strain field distribution of a bridge using a limited amount of monitoring data from key sections, the present invention proposes a method for rapidly calculating the global strain field of the bridge. By using machine learning methods, a computational framework system for rapidly and in real time calculating the global strain field of the entire bridge has been developed, achieving a technical improvement in the global applicability and timeliness of bridge structure health monitoring technology.

[0006] The technical solutions of the present invention are as follows:

[0007] A fast calculation method for the global strain field of a bridge, comprising:

[0008] Step S1: generating stress simulation data of full-bridge sampling positions based on finite element simulation;

[0009] Step S2: Based on the stress simulation data of the full-bridge sampling position, the computational model is supervisedly trained;

[0010] Step S3: Using the real-time measurement data of the bridge stress sensor as input, the trained calculation model is used to quickly calculate the stress of the entire bridge.

[0011] Furthermore, the step S1 includes:

[0012] Step S11: establishing a finite element model according to the bridge drawings;

[0013] Step S12: Obtain stress simulation data of sampling positions of the entire bridge from the finite element model through a specific random sampling method.

[0014] Furthermore, the step S12 includes:

[0015] Various random load conditions are applied to the finite element model. By randomly adding realistic white noise pressure to the bridge pavement, various possible situations are simulated and applied to the finite element model. On this basis, a large amount of simulation experimental strain data is generated, collectively referred to as the stress simulation data of the sampling positions of the entire bridge.

[0016] Furthermore, the computational model is a multi-layer perceptron, which is used to fit the analytical numerical solution of the strain field that conforms to the actual bridge stress partial differential equation, and then on this basis, the real-time solution of the bridge global strain field stress is realized.

[0017] Furthermore, the multi-layer perceptron includes:

[0018]

[0019] in:

[0020] n is the value of all points on the entire bridge calculated by finite element method simulation;

[0021] m is the value of the corresponding input point after removing the verification point;

[0022] F is a neural network;

[0023] θ is the network parameter being trained.

[0024] Furthermore, it also includes:

[0025] Step S4: Verify point selection based on variance analysis and physics-informed heuristic self-checking.

[0026] Furthermore, the step S4 includes:

[0027] Step S41: Calculating residuals based on the measured values of the verification points screened by the combined variance and the output values of the verification points output by the calculation model;

[0028] Step S42: Compare the residual with the threshold to determine whether the calculation model is wrong, and then issue an alarm in time.

[0029] Furthermore, the step S42 includes:

[0030] When the residual is higher than the threshold, an alarm is triggered;

[0031] When the residual is lower than the threshold, the calculation model is judged to be normal.

[0032] Compared with the existing technology, the beneficial effects of the present invention are:

[0033] A method for rapidly calculating the global strain field of a bridge comprises: step S1: generating stress simulation data at sampling positions across the entire bridge based on finite element simulation; step S2: performing supervised training on a calculation model based on the stress simulation data at sampling positions across the entire bridge; step S3: using real-time measurement data from bridge stress sensors as input, and rapidly calculating the stress of the entire bridge using a trained calculation model; the method is designed to establish a complete tool chain from cloud-based data generation and model training to real-time reasoning and analysis at the bridge end; and the efficient reasoning capability of the trained neural network is applied to achieve real-time and rapid calculation of the stress of the entire bridge body in the global strain field of the bridge; thus achieving the establishment of a calculation workflow for a finite element model of a bridge under random bridge deck loads that is fully automatically processed throughout the entire process. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a flowchart of a method for rapid calculation of the global strain field of a bridge;

[0035] Figure 2 This is a schematic diagram of a single hidden layer multilayer perceptron network;

[0036] Figure 3 This is a flow chart of the result self-checking method;

[0037] Figure 4 The figure is a schematic diagram of the cross-sectional strain field stress generated by the calculation method proposed in the present invention. DETAILED DESCRIPTION

[0038] It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus comprising the element.

[0039] The features and performance of the present invention are further described in detail below with reference to the embodiments.

[0040] Example 1

[0041] See also Figure 1 , a fast calculation method for the global strain field of a bridge, specifically comprising the following steps:

[0042] Step S1: Generate stress simulation data at sampling positions across the entire bridge based on finite element simulation. In this embodiment, the core is to rapidly calculate the global strain field of the bridge using the finite element method, which is a common numerical method for solving partial differential equations.

[0043] Step S2: Based on the stress simulation data of the full-bridge sampling position, the computational model is supervisedly trained;

[0044] Step S3: Using the real-time measurement data of the bridge stress sensor as input, the trained calculation model is used to quickly calculate the stress of the entire bridge. Preferably, the trained calculation model can be deployed on an edge computing unit that receives the actual monitoring data of the bridge stress sensor to achieve real-time solution and analysis of the strain and stress of the entire bridge. Specifically, Figure 4 A schematic diagram of the cross-sectional strain field stress generated by the computational model is given.

[0045] In this embodiment, specifically, step S1 includes:

[0046] Step S11: establishing a finite element model according to the bridge drawings; it should be noted that the finite element model is the CAD model structure of the known bridge;

[0047] Step S12: Obtain stress simulation data of sampling positions of the entire bridge from the finite element model through a specific random sampling method.

[0048] In this embodiment, specifically, step S12 includes:

[0049] Various random load conditions are applied to the finite element model. By randomly adding realistic white noise pressure to the bridge pavement, various possible situations are simulated and applied to the finite element model. On this basis, a large amount of simulation experimental strain data is generated, collectively referred to as the stress simulation data of the full-bridge sampling positions; this stress simulation data of the full-bridge sampling positions is used for the next step of supervised training of the calculation model.

[0050] In this embodiment, specifically, the computing model is a multi-layer perceptron (specifically, Figure 2Figure 3 shows a schematic diagram of a single-hidden-layer multilayer perceptron network. The multilayer perceptron is used to fit the analytical numerical solution of the strain field that conforms to the actual bridge stress partial differential equation, and then on this basis, the real-time solution of the bridge's global strain field stress is realized. It should be noted that the multilayer perceptron is a typical neural network structure with the ability to fit arbitrary multidimensional continuous functions. The predicted value output can be obtained by inputting the stress simulation data of the sampling positions of the entire bridge into the multilayer perceptron, and the sum of the mean square errors (SSE) between the predicted value and the "true value" (output label) generated by the finite element method is continuously reduced through the gradient method to realize the solution of the neural network parameters.

[0051] In this embodiment, specifically, the multilayer perceptron includes:

[0052]

[0053] in:

[0054] n is the value of all points on the entire bridge calculated by finite element method simulation;

[0055] m is the value of the corresponding input point after removing the verification point;

[0056] F is a neural network;

[0057] θ is the network parameter being trained.

[0058] In the above formula, we use subscripts to represent the dimensions of the vectors. It is worth noting that this patent does not explicitly incorporate the domain and boundary conditions of the partial differential equation when designing the loss function. Instead, it directly pre-sets the conditions during the simulation data generation phase to generate a large amount of training data, allowing the network to learn these constraints from the training data in a completely data-driven manner.

[0059] In order to prevent errors in online reasoning calculations due to insufficient model generalization ability, in this embodiment, please refer to Figure 3 , also includes:

[0060] Step S4: Verify point selection based on variance analysis and physics-informed heuristic self-checking.

[0061] In this embodiment, specifically, step S4 includes:

[0062] Step S41: Calculating residuals based on the measured values of the verification points screened by the combined variance and the output values of the verification points output by the calculation model;

[0063] Step S42: Compare the residual with the threshold to determine whether the calculation model is wrong, and then issue an alarm in time;

[0064] Specifically, several sensor-deployed points are reserved as validation points. These are not used as model input data, but rather to verify the solution. The selection of validation points is based on empirical experience with bridge stresses and the variance of actual measurements. Representative points with large, dramatic, and consistent variations (such as the base at midspan) are selected. After the neural network model is deployed and online inference begins, the residual difference between the model output and the values at the validation points is compared with a threshold to determine if the neural network model is faulty, triggering a timely alarm.

[0065] In this embodiment, specifically, step S42 includes:

[0066] When the residual is higher than the threshold, an alarm is triggered;

[0067] When the residual is lower than the threshold, the calculation model is judged to be normal.

[0068] The above-described embodiments merely represent specific implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of protection of the present application. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the technical concept of the present application, and all such variations and improvements fall within the scope of protection of the present application.

[0069] This background section is provided to generally present the context of the invention, and the work of the presently named inventors, the work to the extent described in this background section, and aspects of the description in this section that did not constitute prior art at the time of filing are neither explicitly nor implicitly admitted to be prior art to the present invention.

Claims

1. A method for rapid calculation of the global strain field of a bridge, characterized by: include: Step S1: generating stress simulation data of full-bridge sampling positions based on finite element simulation; Step S2: Based on the stress simulation data of the full-bridge sampling position, the computational model is supervisedly trained; Step S3: Using the real-time measurement data of the bridge stress sensor as input, the trained calculation model is used to quickly calculate the stress of the entire bridge; The computational model is a multi-layer perceptron, which is used to fit the analytical numerical solution of the strain field that conforms to the actual bridge stress partial differential equation, and then realize the real-time solution of the bridge global strain field stress on this basis; The multi-layer perceptron comprises: in: The numerical values of all points on the entire bridge calculated by finite element method simulation; The value of the corresponding input point after removing the verification point; For neural networks; The network parameters to be trained.

2. A method for rapid calculation of the global strain field of a bridge according to claim 1, characterized in that: The step S1 comprises: Step S11: establishing a finite element model according to the bridge drawings; Step S12: obtaining stress simulation data of sampling positions of the entire bridge from the finite element model by a specific random sampling method; The specific random sampling method in step S12 includes: Various random load conditions are applied to the finite element model. By randomly adding realistic white noise pressure to the bridge pavement, various possible situations are simulated and applied to the finite element model. On this basis, a large amount of simulation experimental strain data is generated, collectively referred to as the stress simulation data of the sampling positions of the entire bridge.

3. A rapid calculation method for the global strain field of a bridge according to claim 1, characterized in that: Also includes: Step S4: Validate site selection based on variance analysis and physics-informed heuristic self-checking.

4. A method for rapid calculation of the global strain field of a bridge according to claim 3, characterized in that: The step S4 comprises: Step S41: Calculating residuals based on the measured values of the verification points screened by the combined variance and the output values of the verification points output by the calculation model; Step S42: Compare the residual with the threshold to determine whether the calculation model is wrong, and then issue an alarm in time.

5. A method for rapid calculation of the global strain field of a bridge according to claim 4, characterized in that: The step S42 includes: When the residual is higher than the threshold, an alarm is triggered; When the residual is lower than the threshold, the calculation model is judged to be normal.

Images