Arch rib linear back calculation method based on multi-point strain and deep neural network

Abstract

The invention discloses an arch rib linear back calculation method based on multipoint strain and a deep neural network, and the method comprises the steps: firstly building an arch bridge finite element analysis model, which comprises the simulation of geometric nonlinearity and material nonlinearity, and the definition of a construction stage; and defining load loading modes and change ranges in different stress states needing to be considered, and determining a load combination through an orthogonal test design. And then performing finite element analysis to calculate multi-point strain and line shape of the main arch rib of the arch bridge under different load combinations, assembling a learning sample by taking the strain as input and the line shape as output, training, verifying and testing the LSTM by using the learning sample, establishing a mapping relation between the strain and the line shape, and further realizing rapid back calculation of the line shape of the main arch rib through the strain. According to the method, the LSTM is used for establishing the mapping relation between the multi-point strain and the line shape of the arch bridge main arch, the line shape of the main arch can be inversely calculated according to the newly monitored multi-point strain value of the main arch, and the problem that a traditional inverse calculation method is not suitable for arch bridge line shape calculation is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of arch bridge construction monitoring, and in particular to an arch rib linear backcalculation method based on multi-point strain and deep neural network. Background Art

[0002] With the development of large-span arch bridges, quality control and safety issues in the construction process have become increasingly prominent. In engineering construction, the main arch line shape can intuitively reflect the working status and applicability of the arch bridge. Excessive deformation of the main arch will lead to unfavorable structural stress, causing structural damage, and will also affect the subsequent installation of arch rib segments, hangers, and arch columns. It may even cause cracks in the arch building and collapse of the bridge. Therefore, timely and accurate grasp of the changes in the main arch line shape is crucial to ensuring the construction accuracy and safety of the bridge, and is an indispensable part of bridge engineering construction monitoring. Arch rib line shape monitoring research has important engineering significance for ensuring the safety, economy and reliability of bridge structures.

[0003] At present, the commonly used method for calculating the deflection curve of a simply supported beam is mainly the conjugate beam method. The conjugate beam method uses conjugate supports to replace actual supports, thereby generating the so-called "conjugate beam". The modified conjugate beam method uses the distribution of strain to achieve the distribution of equivalent loads on the conjugate beam. At the same time, it can also solve the load of the real beam and establish the relationship between strain and line shape, so as to accurately grasp the line shape changes of the simply supported beam. However, this method will be limited when applied to the line shape monitoring of the main arch of the arch bridge for the following reasons:

[0004] (1) Different stress characteristics: Under normal working conditions, simply supported beams are mainly subjected to bending, and the strain changes linearly in the height direction of the beam section. However, arches are mainly subjected to compression and supplemented by bending, and the strain in the height direction of the arch section can be uniformly distributed, linearly distributed, or distributed in other forms. The modified conjugate beam theory is only applicable to pure bending members and is not applicable to linear calculations of arch structures;

[0005] (2) Different structural shapes: A simply supported beam is a linear structure, and its deflection is the accumulation of the cross-sectional rotation angle or curvature, that is, there is a relatively clear corresponding relationship between deflection and strain. However, an arch is a curved structure, and its linear shape is not only related to the strain, but also to the shape of the arch itself, that is, there is a complex influence relationship between linear shape and strain.

[0006] (3) Different structural effects: The mechanical behavior of a simply supported beam is relatively simple. Under normal working conditions, it can be considered as a linear structure when subjected to bending. However, when an arch is deformed under the coupling of force, compression and bending, a more significant geometric nonlinear effect occurs. It is more difficult to invert the linear shape of the arch from the strain.

[0007] In view of this, a method for backcalculation of arch rib linear shape based on multi-point strain and deep neural network is needed. Summary of the invention

[0008] In view of the problem in the prior art that the traditional conjugate beam method for calculating the deflection curve can only be applied to pure curved members and linear structures, but not to arch structures, the present invention provides an arch rib linear inverse calculation method based on multi-point strain and deep neural network, which can apply different load combinations to the arch structure, calculate the strain of the upper and lower edges of the arch and the deformation of the arch, assemble learning samples with strain as input and linear changes as output, use LSTM to learn these samples, and then master the complex mapping relationship between the strain and deformation of the arch structure, and finally input a set of new arch strains to inversely calculate the arch deformation. It is not only applicable to arch structures in a closed state, but also to arch structures in a cantilever state. The specific technical solution is as follows:

[0009] A method for back-calculating the linear shape of arch ribs of a long-span arch bridge based on multi-point strains comprises the following steps:

[0010] S1 Establish a finite element analysis model for an arch bridge, taking into account the geometric nonlinearity including large displacement effect and beam-column effect, the nonlinearity of steel bars and concrete materials, and defining the construction stages of the arch bridge according to the actual construction process;

[0011] S2 simulates the loads that may occur during construction and operation in the finite element model of the arch bridge, defines the loading position and fluctuation range of the load according to the actual project, and uses orthogonal test design to combine the loads within this range. The purpose of using "orthogonal test design" is to use as few samples as possible to describe the behavior characteristics of the finite element model and reduce the workload of finite element calculations;

[0012] S3 applies different load combinations to the arch rib to calculate different arch rib linear shapes and strains. The arch rib eighth or sixteenth point is used as the key section to extract the arch rib linear shape and upper and lower edge strain values. The arch rib upper and lower edge multi-point strain is used as input and the arch rib linear shape change is used as output. Learning samples are assembled in pairs, and all learning samples are divided into training set, validation set, and test set in a ratio of 8:1:1.

[0013] S4 uses the TensorFlow platform to build an LSTM network model, uses learning samples to train and verify the network, uses MSE and MAE indicators to evaluate the convergence speed and stability of network training, and optimizes network training and verification through early termination and hyperparameter fine-tuning;

[0014] S5 inputs the strains of the upper and lower edges of the arch ribs in the test set into the trained LSTM model, predicts the corresponding arch rib linear changes, compares the predicted linear changes with the finite element calculation results, and uses RMSE and R 2 Evaluate the prediction error and verify the accuracy of the LSTM model;

[0015] S6 sets the RMSE accuracy requirement to less than 2 and sets R2 The accuracy requirement is greater than 0.99. When the LSTM accuracy does not meet the requirement, steps S4 and S5 are repeated until an LSTM model with sufficient accuracy is finally determined.

[0016] Preferably, in step S1, a finite element model of a large-span arch bridge is established which takes into account the construction process, geometric nonlinearity, and material nonlinearity in detail.

[0017] Preferably, in step S4, the early termination and model hyperparameter fine-tuning strategies are as follows:

[0018] (1) Early termination: During the training process, monitor the loss change on the LSTM model validation set. Once the loss no longer continues to decrease, stop training to avoid overfitting.

[0019] (2) Hyperparameter fine-tuning: When the LSTM model training results are not ideal, increase or decrease the number of LSTM layers and the number of units in each LSTM layer so that the LSTM network can quickly and stably complete the learning of data samples.

[0020] The LSTM includes an input layer, an LSTM layer, a flattening layer, a fully connected layer and an output layer. The number of input layers is 1, and the number of units is consistent with the number of points for extracting the strain of the arch ribs. The number of output layers is 1, and the number of units is consistent with the number of points for extracting the deformation. The number of LSTM layers is optimally determined between 2 and 6, and the number of units in each layer is optimally determined between 4 and 32. The number of fully connected layers is optimally determined between 1 and 4, and the number of units in each layer is optimally determined between 4 and 32. A flattening layer is set between the LSTM layer and the fully connected layer. The fully connected layer adopts the LeakyReLU activation function, and the slope α is 0.1. The optimizer adopts Adam, and the learning rate is optimally determined in the range of 0.001 to 0.01.

[0021] Preferably, in step S4, the loss function uses MSE and the evaluation function uses MAE to track the learning performance of LSTM. The calculation formulas of the two functions are as follows:

[0022] MSE calculates the square of the difference between the model's predictions and the actual values:

[0023]

[0024] MAE calculates the absolute value of the difference between the model's predicted value and the actual value:

[0025]

[0026] In the above formula, n is the number of samples, y is i is the ith actual value, is the corresponding predicted value.

[0027] Preferably, in step S5, the root mean square error RMSE and the coefficient of determination R are selected. 2 To evaluate the test performance of LSTM, the calculation formulas of the two evaluation indicators are as follows:

[0028] The calculation formula of RMSE is as follows:

[0029]

[0030] RMSE is the root mean square error between the predicted value and the true value. The smaller the value, the better. The limit value of the present invention is set to 2.

[0031] R 2 The calculation formula is as follows:

[0032]

[0033] In the above formula, represents the mean of the actual data, R 2 It is used to evaluate the degree of fit between model prediction and actual data. The closer the value is to 1, the better the model prediction effect is. The limit value is set to 0.99 in the present invention.

[0034] If the prediction result of the LSTM model is significantly different from the deflection change obtained in the finite element model, repeat steps S4 and S5 to perform network optimization training until the LSTM prediction accuracy meets the requirements.

[0035] After obtaining the LSTM model that meets the accuracy requirements, a new set of arch rib strain values ​​can be input into the model to quickly backcalculate the corresponding arch rib linear changes. The arch rib linear changes are added to the original arch rib linear shape to obtain the absolute linear shape of the arch rib.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] In view of the problem that the traditional conjugate beam method for calculating the deflection curve shape can only be applied to pure bending members and linear structures, but not to arch structures, the present invention proposes a method for back-calculating the arch rib line shape of a large-span arch bridge based on multi-point strain. The present invention applies different load combinations to the arch structure, calculates the strains of the upper and lower edges of the arch and the deformation of the arch, assembles learning samples with strain as input and linear changes as output, uses LSTM to learn these samples, and then grasps the complex mapping relationship between the strain and deformation of the arch structure. Finally, a new set of arch strains is input to back-calculate the arch deformation. The technology proposed in the present invention is not only applicable to arch structures in a closed state, but also to arch structures in a cantilever state. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for the specific embodiments or the description of the prior art. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual scale.

[0039] Figure 1 It is the technical roadmap of the present invention;

[0040] Figure 2 This is the finite element model diagram of a long-span arch bridge;

[0041] Figure 3 This is the diagram of the arch bridge in the maximum cantilever state;

[0042] Figure 4 This is a picture of the arch bridge in the closed state;

[0043] Figure 5 is the best LSTM model diagram;

[0044] Figure 6 This is the change graph of the network training loss for the maximum cantilever arch structure;

[0045] Figure 7 This is the graph showing the changes in the evaluation indicators for the network training of the maximum cantilever arch structure;

[0046] Figure 8 This is the strain diagram of the maximum cantilever arch structure test sample 1;

[0047] Fig. 9 This is the strain diagram of the maximum cantilever arch structure test sample 2;

[0048] Fig.10 This is a comparison diagram between the actual and predicted line shapes of the maximum cantilever arch structure test sample 1;

[0049] Fig.11 This is a comparison diagram between the actual and predicted line shapes of the maximum cantilever arch structure test sample 2;

[0050] Fig.12 This is the graph showing the changes in the network training loss for the closed arch structure;

[0051] Fig.13 This is the graph showing the changes in evaluation indicators for the network training of the closure arch structure;

[0052] Fig.14 This is the strain diagram of the test sample 1 of the closed arch structure;

[0053] Fig.15 This is the strain diagram of the test sample 2 of the closed arch structure;

[0054] Fig.16This is a comparison diagram between the actual and predicted alignments of the joint arch structure test sample 1;

[0055] Fig.17 This is a comparison chart between the actual and predicted alignments of the joint arch structure test sample 2. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0058] It should also be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0059] It should be further understood that the term "and / or" used in the present description and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0060] Combine the following Figure 1-Figure 17 , the embodiments of the present invention are further described.

[0061] like Figure 1 As shown, a method for back-calculating the arch rib linear shape of a long-span arch bridge based on multi-point strain includes the following steps:

[0062] 1. Establishment of finite element model: Midas Civil is used to establish a finite element model considering the construction stage. Figure 2 As shown, the arch structure model of the maximum cantilever state is split as follows Figure 3 As shown in Figure 2, the arch structure model in the closure state is as follows: Figure 4 As shown;

[0063] Figure 2 It is a finite element model diagram considering the construction method and process. Figure 3 The model is in the maximum cantilever state. Figure 4The model is in a closed state.

[0064] The finite element modeling part is not the focus. This part is already very mature and can be implemented using a variety of software, so I will not go into details here.

[0065] 2. Application of load combination: Carry out orthogonal test design of load combination. The mechanical behaviors of arch structures in cantilever state and closure state are different, so they are considered separately:

[0066] Cantilever arch structure model, apply concentrated loads at the cable positions of arch rib construction, determine the theoretical values ​​of cable loads at these points during construction, determine the concentrated load fluctuation range and fluctuation level, and use orthogonal test design method to design multiple groups of fluctuating concentrated load combinations; (case bridge, with 12 cables, fluctuation range [-10%, +10%], fluctuation level set to 8, generate 100 samples)

[0067] The closed arch structure model applies concentrated loads to the columns on the arch ribs in the closed state, determines the theoretical load values ​​of these columns during the operation of the completed bridge, determines the fluctuation range and level of concentrated loads, and uses the orthogonal experimental design method to design multiple groups of fluctuating concentrated load combinations; (the case bridge has 11 columns on the arch, the fluctuation range is [-20%, +20%], the fluctuation level is set to 9, and 100 samples are generated;)

[0068] 3. Sample preparation: The load combination samples determined by orthogonal test design are calculated by finite element model, and the strain and linear changes of the arch ribs under different concentrated load combinations are calculated (the points for extracting strain and linear shapes can be the eighth or sixteenth points of the arch ribs, there is no fixed requirement). For arch ribs with more complex cross-sections, it is difficult to solve the position of the neutral axis, and a large number of uniform strains at the upper and lower edges of the arch rib structure are required, and the linear shapes at the corresponding positions are extracted at the same time.

[0069] 100 sets of load combinations can be used to calculate 100 sets of structural data, with the multi-point strains of the upper and lower edges of the arch rib as input and the arch rib line shape as output to form learning samples. All learning samples are divided into training set, validation set, and test set in a ratio of 8:1:1;

[0070] 4. Training and verification of LSTM model: Use TensorFlow platform to build LSTM network model, use learning samples to train and verify the network, and optimize the training and verification of the network through early termination and hyperparameter fine-tuning strategies. Finally, use multiple evaluation indicators to evaluate the stability and prediction accuracy of the model;

[0071] The best LSTM model is Figure 5As shown in the figure, the model includes 1 input layer, 1 output layer, 2 LSTM layers with 16 and 8 units respectively, 1 flattening layer, and 1 fully connected layer with 4 units. The optimizer uses Adam, and the learning rate lr = 0.001. The activation function of the fully connected layer is LeakyReLU, and the slope α = 0.1.

[0072] 5. Testing of LSTM model: After the LSTM model is trained, the corresponding arch structure line shape can be quickly obtained by inputting the new arch structure strain into LSTM.

[0073] Application examples:

[0074] Establish the Midas finite element model of Jinsha River Bridge Figure 2 As shown, the model bridge has a calculated span of 500 meters, an arch axis coefficient of m = 2.5; the arch rib rise f = 105m, and the rise-span ratio f / L = 1 / 4.76; the arch rib inclination is 2.73°; ​​the arch rib center distances at the arch crown and arch foot are 22m and 32m respectively; the arch crown truss height is 11m, and the arch foot truss height is 16m. The main arch chord is made of Q370q steel and poured with C70 high-performance concrete. A single main arch rib is divided into 25 segments and is hoisted by inclined cable. The construction process is symmetrical installation on both sides.

[0075] (1) Maximum cantilever arch model

[0076] a. Sample preparation

[0077] Based on the orthogonal experimental design method, multiple groups of fluctuating cable force load combinations are designed for the maximum cantilever arch structure to induce different forms of deformation of the arch rib. 12 groups of cable force loads are used as 12 factors. The fluctuation range is set to [-10%, +10%] on the basis of the theoretical cable force, and the fluctuation level is set to 8. A total of 100 test samples are generated.

[0078] b. Sample assembly

[0079] The sixteen points of the arch rib are selected as the control section. The strain and line shape of the upper and lower edges of the control section are calculated through the finite element model. The strain is used as input and the corresponding line shape change is used as output to assemble a sample database. A total of 100 samples are obtained, of which 80 samples are used as training sets, 10 samples are used as validation sets, and the remaining 10 samples are used as test sets to evaluate the accuracy of the model. The number of samples captured for each training is batch_size, which is 4, and the number of iterations is epochs, which is 500. During training, the model hyperparameters are adjusted by tracking the changes in loss and accuracy.

[0080] c. Model training and validation

[0081] The MSE and MAE changes of the maximum cantilever arch model training are as follows Figure 6 and Figure 7As shown in the figure, the blue curve shows the changes on the training set, and the orange curve shows the changes on the validation set. Due to the early termination strategy to prevent overfitting, it can be seen that after about 100 epochs, the network loss tends to stabilize and no longer decreases significantly, which indicates that the model training has been completed.

[0082] d. Use and evaluation of mapping models

[0083] For the maximum cantilever arch model, 2 groups of test results were selected from 10 groups of test samples for detailed presentation. Figure 8 and Fig. 9 They represent the input of sample 1 and sample 2 respectively. The first 15 points in the figure correspond to the strain at the upper edge of the control point, while the last 15 points correspond to the strain at the lower edge of the control point. Therefore, it can be clearly observed from the figure that there is a significant difference in the strain before and after the input.

[0084] Fig.10 and Fig.11 The comparison between the model's predicted linear shape and the actual linear shape under the corresponding input is shown. In these two figures, the blue curve represents the actual linear change, while the orange curve represents the linear change predicted by the model. By comparing the linear shapes, it can be observed that the model's predicted results are very close to the actual linear change.

[0085] Tables 1 and 2 present the evaluation indices between the predicted and actual alignments of the two test samples of the maximum cantilever arch model. These indices correspond to Fig.10 and Fig.11 Results of the midline comparison.

[0086] For sample 1, the evaluation indices are 1.71930 and 0.99704, and for sample 2, the indices are 1.80790 and 0.99609. The R 2 It is already very close to 1, indicating that extremely high calculation accuracy has been achieved.

[0087] Table 1 Test sample 1 indicators

[0088]

[0089]

[0090] Table 2 Test sample 2 indicators

[0091] RMSE <![CDATA[R 2 ]]> 1.80790 0.99609

[0092] (2) Joint arch structure model

[0093] a. Sample preparation

[0094] The joint arch structure model uses the orthogonal experimental design method to design multiple groups of fluctuating concentrated load combinations according to the positions of the 11 arch columns to induce different forms of deformation of the arch ribs. The fluctuation range of the 11 groups of concentrated loads is [-20%, +20%], the fluctuation level is set to 9, and a total of 100 test samples are generated.

[0095] b. Sample assembly

[0096] The sixteen points of the arch rib are selected as the control section. The strain and line shape of the upper and lower edges of the arch rib control section are calculated through the finite element model. The upper and lower edge strains of the control section are combined as input, and the corresponding arch rib line shape changes are output to form a learning sample database. There are a total of 100 samples in the closed arch structure model, of which 80 samples are used to construct the training set, 10 samples are used to construct the validation set, and the remaining 10 groups of samples are used to construct the test set. The batch_size of the number of samples captured for each training is 4, and the number of epochs is 500. During the training, the network hyperparameters are adjusted by tracking the changes in MSE and MAE.

[0097] c. Model training and validation

[0098] The loss changes and evaluation function changes of the joint arch structure model training are as follows: Fig.12 and Fig.13 As shown, it can be seen that after about 122 epochs, the MSE and MAE changes of the model tend to be stable.

[0099] d. Use and evaluation of mapping models

[0100] For the joint arch structure model, 2 groups of test results were selected from 10 groups of test samples for detailed display. Fig.14 and Fig.15 Represent the input conditions of sample 1 and sample 2 respectively.

[0101] Fig.16 and Fig.17 The comparison between the predicted line shape of the model and the actual line shape under the corresponding input is shown.

[0102] Tables 3 and 4 present the evaluation indices between the predicted and actual alignments of the two test samples of the joint arch structure model. These indices correspond to Fig.16 and Fig.17 Results of the midline comparison.

[0103] For sample 1, the indices are 0.44695 and 0.99515 respectively. For sample 2, the indices are 0.31402 and 0.99942 respectively.

[0104] Table 3 Test sample 1 indicators

[0105] RMSE <![CDATA[R 2 ]]> 0.44695 0.99515

[0106] Table 4 Test sample 2 indicators

[0107] RMSE <![CDATA[R 2 ]]> 0.31402 0.99942

[0108] A comprehensive investigation of the linear inversion results of the maximum cantilever arch model and the closed arch model shows that the LSTM mapping model can capture the inherent correlation and mutual influence of the arch rib strain. The comprehensive inversion of the arch rib linear shape can realize the inversion of the linear shapes of two different arch models, and the inversion results have high accuracy. The LSTM-based arch rib inversion model overcomes the defects of the modified conjugate beam calculation method.

[0109] In summary, the present invention aims at the problem that the traditional conjugate beam method for calculating the deflection curve shape can only be applied to pure bending members and linear structures, but not to arch structures, and proposes a method for back-calculating the arch rib line shape of a large-span arch bridge based on multi-point strain. The present invention applies different load combinations to the arch structure, calculates the strains of the upper and lower edges of the arch and the deformation of the arch, assembles learning samples with strain as input and linear changes as output, uses LSTM to learn these samples, and then grasps the complex mapping relationship between the strain and deformation of the arch structure. Finally, a set of new arch strains are input to back-calculate the arch deformation. The technology proposed in the present invention is not only applicable to arch structures in a closed state, but also to arch structures in a cantilever state.

[0110] Those of ordinary skill in the art will appreciate that the units of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition of each example has been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0111] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical function division, and there may be other division methods in actual implementation, for example, multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored, etc.

[0112] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0113] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-0nlyMemory), random access memory (RAM, RandomAccessMemory), mobile hard disk, magnetic disk or optical disk, etc., which can store program code.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and specification of the present invention.

Claims

1. A method for backcalculating the arch rib linear shape based on multi-point strain and deep neural network, characterized in that: The following steps are involved: Establish a finite element analysis model of the arch bridge, and simulate the loads that may occur under construction and operation conditions in the finite element model of the arch bridge; Define the loading position and fluctuation range of the load, and combine the loads within the fluctuation range. Applying different load combinations to the arch ribs can calculate different arch rib linear shapes and strains. Taking the multi-point strains of the upper and lower edges of the arch rib as input and the linear change of the arch rib as output, the learning samples are assembled in pairs, and all the learning samples are divided into training set, validation set and test set. The neural network model is trained and optimized based on the divided training set, validation set and test set. A new set of arch rib strain values ​​is input into the trained neural network model, and the corresponding arch rib linear shape changes are obtained by back-calculation. The arch rib linear shape changes are added to the original arch rib linear shape to obtain the absolute linear shape of the arch rib.

2. The arch rib linear shape back-calculation method based on multi-point strain and deep neural network according to claim 1 is characterized in that: In the process of establishing the finite element analysis model of the arch bridge, the geometric nonlinearity of large displacement effect and beam-column effect, as well as the nonlinearity of steel and concrete materials are considered, and the construction stages of the arch bridge are defined according to the actual construction process.

3. The arch rib linear shape back-calculation method based on multi-point strain and deep neural network according to claim 1 is characterized in that: The specific process of load combination is: define the loading position and fluctuation range of the load according to the actual project, and then use orthogonal test design to carry out load combination within the fluctuation range.

4. The arch rib linear shape back-calculation method based on multi-point strain and deep neural network according to claim 1 is characterized in that: The arch rib linear shape and the strain values ​​of the upper and lower edges are extracted with the eighth or sixteenth point of the arch rib as the key section.

5. The arch rib linear shape back-calculation method based on multi-point strain and deep neural network according to claim 1 is characterized in that: The training process of the neural network model is as follows: Use TensorFlow platform to build LSTM network model, use learning samples to train and verify the network, use MSE and MAE indicators to evaluate the convergence speed and stability of network training, and optimize network training and verification through early termination and hyperparameter fine-tuning; The strains of the upper and lower edges of the arch ribs in the test set were input into the trained LSTM model to predict the corresponding arch rib linear changes. The predicted linear changes were compared with the finite element calculation results, and the RMSE and R2 were used to evaluate the prediction error to verify the accuracy of the LSTM model. The accuracy requirement of RMSE is set to be less than 2, and the accuracy requirement of R2 is set to be greater than 0.

99. When the LSTM accuracy does not meet the requirements, steps S4 and S5 are repeated until an LSTM model with sufficient accuracy is finally determined.

6. The arch rib linear shape back-calculation method based on multi-point strain and deep neural network according to claim 5 is characterized in that: The early termination and model hyperparameter fine-tuning strategies are as follows: Early termination: During the training process, monitor the loss change on the LSTM model validation set. Once the loss no longer continues to decrease, stop training to avoid overfitting. Hyperparameter fine-tuning: When the LSTM model training results are not ideal, increase or decrease the number of LSTM layers, and increase or decrease the number of units in each LSTM layer, so that the LSTM network can quickly and stably complete the learning of data samples.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute the arch rib linear backcalculation method based on multi-point strain and deep neural network as described in any one of claims 1 to 6.

8. A processor, characterized in that: The processor is used to run a program, wherein when the program is run, the arch rib linear shape backcalculation method based on multi-point strain and deep neural network described in any one of claims 1 to 6 is executed.

Images