Cable-stayed bridge cable precise positioning tensioning control device and control method thereof

Abstract

This application discloses a precise positioning and tensioning control device and method for cable-stayed bridge cables, comprising: acquiring and processing historical tensioning datasets and construction tensioning datasets corresponding to cable tensioning to generate a training sample library; inputting the training sample library into a preset machine learning network framework to generate a cable force change trend prediction model; collecting and inputting multi-source tensioning data into the cable force change trend prediction model to output cable force change trend data; generating cable force prediction deviation data and cable force change rate data based on the cable force change trend data; comparing the cable force prediction deviation data with a preset cable force deviation warning threshold, and simultaneously comparing the cable force change rate data with a preset cable force mutation threshold, and generating corresponding tension parameter adjustment commands based on the comparison results. This application can solve the problems of poor adaptability and insufficient prediction accuracy of existing general models through training and iterative optimization of the machine learning network framework.

Description

Technical Field

[0001] This application relates to the field of bridge cable construction technology, and in particular to a precise positioning and tensioning control device and control method for cable-stayed bridge cables. Background Technology

[0002] As one of the mainstream bridge types for long-span bridges, cable-stayed bridges rely on cable stays as their core load-bearing components. The precision of cable tension control during cable tensioning directly determines the bridge's final shape, overall stress distribution, and long-term service safety performance. This is especially true for long-span cable-stayed bridges with dense cable systems, which place extremely high demands on the precision of cable tension control and the stability of the tensioning process.

[0003] Currently, existing cable-stayed bridge tensioning construction generally adopts a single-cable independent tensioning mode, and the supporting control system adopts a passive closed-loop control mode of "tensioning-monitoring-adjustment". It can only correct the tensioning parameters after the cable force deviation has occurred, which has serious adjustment lag. When over-tensioning or under-tensioning occurs, not only is repeated tensioning and adjustment required, which greatly reduces construction efficiency, but it is also difficult to guarantee the accuracy of cable force control. Especially in the high-tensioning stage of long-span cable-stayed bridges, the delayed deviation adjustment can easily cause stress concentration in the bridge towers and beams, and even lead to structural deformation and safety hazards of exceeding the stress limit.

[0004] At the same time, most existing technologies for calculating and predicting cable force changes rely on static theoretical calculations based on classical structural mechanics formulas or simple parameter corrections based on construction experience, making it difficult to achieve accurate prediction of cable force changes.

[0005] However, the application of machine learning technology in the field of cable tension control has obvious shortcomings. Existing research only uses pure historical engineering data to train general prediction models, without combining the construction tension data of current construction projects to complete the model optimization. The general model has poor adaptability to the current structural characteristics of bridges and on-site construction conditions, and cannot output reliable prediction results of cable force change trends, making it difficult to meet the actual needs of high-precision tension control on engineering sites.

[0006] In summary, existing cable tensioning control technologies lack the ability to predict cable force change trends based on machine learning, and cannot achieve a shift from passive, lagging adjustments to proactive, predictive control. This has become a core bottleneck restricting the development of cable-stayed bridge tensioning construction technology. Summary of the Invention

[0007] This application aims to at least partially solve one of the technical problems in the aforementioned technologies.

[0008] To achieve the above objectives, the first aspect of this application proposes a precise positioning and tensioning control device and method for cable-stayed bridge cables, comprising the following steps: S100, acquiring historical tensioning datasets and construction tensioning datasets corresponding to the tensioning of the cable stays; S200, processing the historical tensioning datasets and the construction tensioning datasets to generate a training sample library; S300, inputting the training sample library into a preset machine learning network framework for training and iterative optimization to generate a cable force change trend prediction model; S400, collecting multi-source tensioning data corresponding to the currently tensioned cable stays; S500, processing the historical tensioning datasets and the construction tensioning datasets to generate a training sample library; S100, inputting the training sample library into a preset machine learning network framework for training and iterative optimization to generate a cable force change trend prediction model; S400, collecting multi-source tensioning data corresponding to the currently tensioned cable stays; S500, processing the training sample library into a training sample library; S101, acquiring historical tensioning datasets and construction tensioning datasets corresponding to the tensioning of the cable stays; S201, processing the training sample library into a training sample library; S202, processing the training sample library into a training sample library; S302, processing the training sample library into a training sample library; S303, processing the training sample library into a training sample library; S402, processing the training sample library into a training sample library; S403, processing the training sample library into a training sample library; S503, processing the training sample library into a training sample library; S404, processing the training sample library into a training sample library; S503, processing the training sample library into a training sample library; S404, processing the training sample library into a training sample library; S504, processing the training sample library into a training sample library; S604, processing the training sample library into a training sample library; S705, processing the training sample library into a training sample library; S605, processing the training sample library into a training sample library; S706, processing the training sample library into a Multi-source tensioning data is input into the cable force change trend prediction model, which outputs cable force change trend data corresponding to the tensioned cable within a preset time period; S600, the cable force change trend data is compared node by node with the preset cable force design target curve to generate cable force prediction deviation data, and the slope of the cable force change trend data is extracted to generate cable force change rate data; S700, the cable force prediction deviation data is compared with the preset cable force deviation warning threshold, and the cable force change rate data is compared with the preset cable force sudden change threshold, and the corresponding tension parameter adjustment command is generated based on the comparison results.

[0009] In addition, the precise positioning and tensioning control method for cable-stayed bridge cables proposed in this application may also have the following additional technical features:

[0010] As a further description of the above technical solution: step S100 specifically includes: S101, obtaining a preset number of historical tensioning datasets corresponding to the tensioning construction of similar cable-stayed bridges; S102, obtaining the construction tensioning datasets corresponding to the cable stays that have been tensioned in the current construction project.

[0011] As a further description of the above technical solution: S200 specifically includes: S201, performing missing value filling, outlier removal, and data normalization processing on the historical tension dataset and the real-time tension dataset in sequence, and extracting tension feature parameters that are strongly correlated with cable force changes; S202, dividing the processed dataset into a training set, a validation set, and a test set according to a preset ratio to generate a training sample library; wherein, the tension feature parameters include tension load, tension rate, cable force increment, ambient temperature, ambient humidity, and beam displacement data.

[0012] As a further description of the above technical solution: S300 specifically includes: S301, inputting the training set in the training sample library into a machine learning network framework built on a neural network; S302, with minimizing the cable force prediction deviation as the optimization objective, globally optimizing the initial weights and thresholds of the machine learning network framework; S303, iteratively verifying the model using the validation set in the training sample library and validating it using the test set, completing model training and optimization, and generating a cable force change trend prediction model.

[0013] As a further description of the above technical solution: S400 specifically includes: during the tensioning process of the stay cable, according to a preset sampling frequency, multi-source tensioning data corresponding to the current tensioned stay cable is collected in real time through cable force sensors, displacement sensors, load sensors deployed at the tensioning end, and environmental monitoring units deployed at the construction site; wherein, the multi-source tensioning data includes real-time cable force, real-time tensioning displacement, real-time tensioning load, and real-time environmental parameters.

[0014] As a further description of the above technical solution: Step S500 specifically includes: S501, performing normalization preprocessing on the multi-source tension data in the same manner as the construction process of the training sample library, and extracting tension feature parameters that match the dimensions of the training sample library; S502, validating the extracted tension feature parameters and removing invalid feature data that exceeds the preset feature threshold range; S503, inputting the valid tension feature parameters into the cable force change trend prediction model; S504, outputting cable force change trend data corresponding to equally spaced time nodes within a preset future time period through calculation by the cable force change trend prediction model.

[0015] As a further description of the above technical solution: Step S600 specifically includes: S601, calculating the difference between the cable force change trend data and the preset cable force design target curve node by node according to time nodes, generating single-node deviation data corresponding to each time node, and integrating all single-node deviation data to generate cable force prediction deviation data for the entire time window; S602, simultaneously performing first-order difference calculation on the cable force change trend data, extracting the change slope of each time node, and generating cable force change rate data.

[0016] As a further description of the above technical solution: step S700 specifically includes: S701, comparing the cable force prediction deviation data with a preset cable force deviation warning threshold, and simultaneously comparing the cable force change rate data with a preset cable force mutation threshold; S702, if the cable force prediction deviation data exceeds the cable force deviation warning threshold, then based on the future cable force change amplitude of the cable force change trend data, calculating the tension parameter adjustment quantization value, and generating a first tension parameter adjustment command to adjust the tensioning rate and load application amount of the current tensioning process; S703, if the cable force change rate data exceeds the cable force mutation threshold, then generating a second tension parameter adjustment command containing a pause or deceleration command.

[0017] To achieve the above objectives, a second aspect of this application proposes a precise positioning and tensioning control device for cable-stayed bridge cables, comprising a data acquisition unit, a processor, a model training unit, and a tensioning control unit. The data acquisition unit includes cable force sensors, displacement sensors, load sensors, and an environmental monitoring module. The data acquisition unit is used to acquire historical tensioning datasets, construction tensioning datasets, and multi-source tensioning data during the tensioning process. The model training unit is data-connected to the processor and is used to train and generate a cable force change trend prediction model based on a training sample library output by the processor. The output terminal of the processor is electrically connected to the input terminal of the tensioning control unit and is used to output tensioning parameter adjustment commands to the tensioning control unit. The output terminal of the tensioning control unit is electrically connected to a tensioning jack and is used to adjust the tensioning rate and load application during the tensioning process.

[0018] According to the cable-stayed bridge cable precise positioning and tensioning control device and its control method in this application, a training sample library is constructed by using historical tensioning datasets of similar cable-stayed bridges and construction tensioning datasets of the current project. Through training and iterative optimization of the machine learning network framework, the nonlinear variation law of cable force under multi-factor coupling during the tensioning process of the cable is accurately fitted, which solves the problems of poor adaptability and insufficient prediction accuracy of existing general models.

[0019] In addition, by inputting real-time multi-source tensioning data into the cable force change trend prediction model, the model outputs cable force change trend data within a preset time period in the future, predicting the future trend of cable force changes and potential deviations in advance. This fundamentally realizes the transformation from post-departure deviation correction to pre-emptive proactive prevention and control, effectively avoiding the problems of excessive or insufficient cable tensioning.

[0020] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0021] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0022] Figure 1 This is a flowchart illustrating a method for precise positioning and tensioning control of cable-stayed bridge cables according to an embodiment of this application;

[0023] Figure 2 This is a flowchart illustrating the construction of a cable force change trend prediction model according to an embodiment of this application;

[0024] Figure 3 This is a flowchart of the real-time tension cable force active prediction control according to an embodiment of this application. Detailed Implementation

[0025] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0026] The following description, in conjunction with the accompanying drawings, describes the precise positioning and tensioning control device and control method for cable-stayed bridge cables according to embodiments of this application.

[0027] like Figure 1 As shown, the cable-stayed bridge cable precise positioning and tensioning control device and its control method according to the embodiments of this application may include the following steps: S100, obtaining the historical tensioning dataset and construction tensioning dataset corresponding to the cable tensioning.

[0028] Specifically, it includes the following steps: S101, obtaining a preset number of historical tensioning datasets corresponding to the tensioning construction of similar cable-stayed bridges.

[0029] S102. Obtain the construction tensioning dataset corresponding to the stay cables that have been tensioned in the current construction project.

[0030] It should be noted that the same type of cable-stayed bridge refers to a completed cable-stayed bridge that is completely consistent with the current construction project in terms of structural system (such as double tower double cable plane, single tower single cable plane), span size (deviation not exceeding ±30%), and cable type (parallel wire cable / steel strand cable), to ensure the matching of the working conditions of the basic data; the preset quantity refers to no less than 5 cable-stayed bridges of the same type, and the effective tension sample size of a single bridge is no less than 2000 sets, to meet the minimum sample size requirement for training machine learning models.

[0031] The fields of the historical tensioning dataset and the construction tensioning dataset are consistent. Each sample includes core fields such as tensioning load (kN), tensioning displacement (mm), cable force (kN), ambient temperature (°C), ambient humidity (%RH), vertical displacement of the beam (mm), and tower deviation (mm). All fields are synchronously collected data at the same time point, and the sampling time interval is consistent to eliminate the impact of sampling frequency differences on subsequent data processing.

[0032] Understandably, historical tension datasets provide a general foundation for learning cable force variation patterns, addressing the problem of insufficient sample size in pure field data; construction tension datasets provide the model with personalized structural response characteristics for the current project, addressing the problem of poor adaptability between general models and field conditions. Together, they constitute a complete data foundation for training machine learning models.

[0033] S200. Process the historical tensioning dataset and the construction tensioning dataset to generate a training sample library.

[0034] It should be noted that the historical tensioning dataset and the construction tensioning dataset undergo completely identical preprocessing operations to ensure data standardization and consistency.

[0035] Specifically, the steps include: S201, performing missing value imputation (i.e., filling missing fields in the dataset using linear interpolation) on the historical tension dataset and the real-time tension dataset, outlier removal (i.e., using the 3σ criterion to identify and remove outliers; when a data point value exceeds the range of ±3 times the standard deviation of the field's mean, it is determined to be an outlier and removed to avoid abnormal data interfering with model training), data normalization (i.e., using the min-max normalization method to map all field values ​​to the [0, 1] interval to eliminate the influence of different units on model training), and extracting tension feature parameters strongly correlated with cable force changes (i.e., using the Pearson correlation coefficient method to calculate the linear correlation between each feature field and the cable force value, and extracting tension feature parameters strongly correlated with cable force changes).

[0036] Among them, the tensioning characteristic parameters include tensioning load, tensioning rate, cable force increment, ambient temperature, ambient humidity and beam displacement data.

[0037] S202. Divide the processed dataset into training set, validation set and test set according to the preset ratio to generate training sample library.

[0038] It should be noted that the entire dataset after preprocessing and feature extraction is randomly divided into training set, validation set and test set in a fixed ratio of 7:2:1. The training set is used for model parameter learning (accounting for 70%), the validation set is used for iterative verification and hyperparameter tuning during training (accounting for 20%), and the test set is used for accuracy verification of the model after training (accounting for 10%). The three datasets together constitute the complete training sample library.

[0039] S300. Input the training sample library into the preset machine learning network framework, perform training and iterative optimization, and generate a model for predicting the trend of cable force change.

[0040] Specifically, it includes the following steps: S301, inputting the training set from the training sample library into a machine learning network framework built on a neural network.

[0041] It should be noted that the default machine learning network framework is a 3-layer feedforward BP neural network, including an input layer, a hidden layer, and an output layer.

[0042] The number of neurons in the input layer is exactly the same as the number of strongly correlated tensile feature parameters extracted in step S200. Each neuron corresponds to an input feature field, ensuring that the input dimension matches the feature dimension of the training sample library in a 1:1 ratio with no dimensional bias.

[0043] Hidden Layer: A single hidden layer structure is adopted. The number of neurons is determined using an industry-standard, reproducible empirical formula, which is:

[0044]

[0045] in, This represents the number of neurons in the hidden layer. This represents the number of neurons in the input layer. This represents the number of neurons in the output layer. The constant is between 1 and 10. This scheme uses a fixed value of 8 to balance the network's fitting ability and generalization ability, and to avoid overfitting / underfitting.

[0046] The number of neurons in the output layer is fixed at 1, and the output value is the predicted value of the cable force at the corresponding time node, which corresponds completely to the label value (cable force value) in the training sample library.

[0047] In addition, the hidden layer uses the ReLU activation function and the output layer uses the linear activation function to ensure that the network can fit nonlinear laws and output continuous force prediction values. The initial learning rate is fixed at 0.001, and a learning rate decay strategy is used to reduce the learning rate to 90% of the original rate every 100 iterations to ensure rapid convergence in the early stage of training and stable optimization in the later stage of training.

[0048] Therefore, during execution, the feature parameter matrix of the training set in the training sample library is input into the BP neural network framework that has been built above, and the label value is the measured cable force value of the corresponding sample, thus completing the standardized input of the training data.

[0049] S302. With minimizing the prediction bias of the cable force as the optimization objective, the initial weights and thresholds of the machine learning network framework are globally optimized.

[0050] It should be noted that the core optimization objective is to minimize the cable force prediction deviation, which is quantified as minimizing the mean squared error (MSE) loss function. The loss function formula is as follows:

[0051]

[0052] in, This represents the model loss value (optimization objective value). The total number of samples in the training set; Predict the cable force value for the model of the i-th sample group; denoted as the measured cable force value (label value) of the i-th sample; the smaller the loss value, the smaller the cable force prediction deviation of the model and the higher the fitting accuracy.

[0053] An improved non-dominated sorting genetic algorithm (NSGA-II) is used to perform global optimization. This algorithm has strong global optimization ability and fast convergence speed, and can effectively avoid the BP neural network from getting trapped in local optima. The optimization object covers all trainable parameters of the network, including: connection weights from the input layer to the hidden layer, connection weights from the hidden layer to the output layer, hidden layer threshold, and output layer threshold.

[0054] The population size is fixed at 50, the maximum number of iterations is fixed at 200, the crossover probability is fixed at 0.7, and the mutation probability is fixed at 0.01. In each iteration, the fitness value is the loss function value, and the individual with the best fitness is selected until the maximum number of iterations is reached. The initial weights and thresholds of the global optimum are output to complete the global optimization.

[0055] S303. Iteratively verify the model using the validation set in the training sample library and validate it using the test set to complete model training and optimization, and generate a model for predicting cable force change trends.

[0056] It should be noted that the optimal initial weights and thresholds obtained from global optimization in step S302 are assigned to the BP neural network, and gradient descent is used for iterative training. After each full training round (i.e., 1 epoch), the validation set in the training sample library is input into the model, the validation set loss value is calculated, and the generalization ability of the model is monitored in real time.

[0057] If the validation set loss value does not decrease for 10 consecutive iterations, training is immediately terminated, and the current optimal model parameters are retained to avoid the model overfitting the training set data, which could lead to a decrease in the actual prediction accuracy in the field.

[0058] After the iteration terminates, the training sample set is input into the trained model, and the mean absolute percentage error is used as the accuracy evaluation index. The calculation formula is as follows:

[0059]

[0060] Where k is the total number of samples in the test set; Predict the cable force value for the i-th test sample using the model; Let be the measured cable force value of the i-th test sample.

[0061] When the MAPE of the test set is less than or equal to 1%, the model is deemed to meet the accuracy requirements for engineering applications. The model training and optimization are then completed, and the final model for predicting cable force change trends is generated. If the MAPE of the test set is greater than 1%, the hyperparameters such as the number of hidden layer neurons and the initial learning rate are adjusted, and the training process of S301-S303 is re-executed until the accuracy requirements are met.

[0062] S400: Collect multi-source tensioning data corresponding to the currently tensioned cable-stayed cable.

[0063] Specifically, during the tensioning process of the stay cables, according to the preset sampling frequency, multi-source tensioning data corresponding to the current tensioned stay cables are collected in real time through cable force sensors, displacement sensors, load sensors deployed at the tensioning end, and environmental monitoring units deployed at the construction site. The multi-source tensioning data includes real-time cable force, real-time tensioning displacement, real-time tensioning load, and real-time environmental parameters.

[0064] As one possible scenario, the cable force sensor is a magnetic flux cable force sensor, deployed at the tensioning end anchor of the currently tensioned cable to collect real-time cable force data; the displacement sensor is a wire displacement sensor, deployed at the piston rod end of the tensioning jack to collect real-time tension displacement data; the load sensor is a through-hole pressure sensor, deployed between the tensioning jack and the anchor plate to collect real-time tension load data; the environmental monitoring unit is an integrated temperature and humidity sensor, deployed at the bridge deck location of the currently tensioned beam segment (avoiding direct sunlight and heat source interference) to collect real-time ambient temperature and humidity data.

[0065] S500: Input multi-source tensioning data into the cable force change trend prediction model, and output the cable force change trend data corresponding to the tensioned cable in the future preset time period.

[0066] Step S500 specifically includes:

[0067] S501. Perform normalization preprocessing on the multi-source tension data in the same way as the training sample library construction process, and extract tension feature parameters that match the dimensions of the training sample library.

[0068] It should be noted that this step uses the same min-max normalization method as step S200, and the extreme values ​​of the corresponding fields used in step S200 when constructing the training sample library are the same. This ensures that the numerical distribution of the real-time input data is completely consistent with the distribution of the sample data during model training, thus avoiding a decrease in model prediction accuracy caused by data distribution shift.

[0069] In the preprocessed multi-source tension data, feature parameters that are completely consistent with the number, fields, and order of strongly correlated tension feature parameters in the training sample library in step S200 are extracted to form a feature vector that matches the training sample library in 1:1 dimension. This ensures that the feature vector can be directly input into the cable force change trend prediction model without any dimension mismatch problem.

[0070] S502. Verify the validity of the extracted tension feature parameters and remove invalid feature data that exceeds the preset feature threshold range.

[0071] It should be noted that the preset feature threshold range for each feature field is the interval between the minimum and maximum values ​​of the corresponding feature field in the training sample library in step S200, which is completely consistent with the feature value range of the training sample library; feature data that exceeds this interval is determined to be invalid feature data that deviates significantly from the distribution of the model training data.

[0072] In addition, for each feature value in the feature vector extracted in step S501, it is compared one by one with the preset feature threshold range of the corresponding field. If all feature values ​​are within the threshold range, it is determined that the verification is qualified; if any feature value exceeds the threshold range, it is determined that the set of feature data is invalid.

[0073] If the data is determined to be invalid, the data set is immediately removed, and the acquisition operation in step S400 and the preprocessing and feature extraction operation in step S501 are re-executed until the valid feature parameters are obtained.

[0074] S503. Input the qualified tension characteristic parameters into the cable force change trend prediction model.

[0075] Understandably, the feature vectors that pass the S502 verification and perfectly match the dimensions of the training sample library are input into the input layer of the cable force change trend prediction model generated by S300. Each element of the feature vector corresponds one-to-one with each neuron in the input layer, completing the model input and ensuring that the input data and network structure are perfectly matched without any input misalignment.

[0076] S504. Through the calculation of the cable force change trend prediction model, output the cable force change trend data corresponding to the time nodes at equal intervals within the future preset time period.

[0077] It should be noted that the cable force change trend prediction model is based on the input feature vector. The network weights and thresholds trained in step S300 are used to perform forward propagation calculations, which pass through the input layer, hidden layer, and output layer in sequence, and finally output the cable force prediction value at the corresponding time node. The calculation process is entirely based on the trained network parameters without any additional human intervention.

[0078] When setting the preset time period in the future, it can be flexibly set according to the control requirements of the tensioning construction stage. The conventional tensioning stage is fixed at 10s, while the high-tension tensioning stage can be reduced to 5s, balancing the timeliness of the prediction results and the control accuracy. The time interval is completely consistent with the sampling frequency and is fixed at 1s. That is, when the preset time period is 10s, the cable force prediction values ​​for the next 10 time nodes are output.

[0079] It is understandable that the cable force change trend data is an ordered sequence of predicted cable force values ​​at equally spaced time nodes within a preset future time period, expressed as:

[0080]

[0081] Where n is the total number of time points within the preset future time period. This is the predicted cable force value for the kth second in the future.

[0082] S600: The cable force change trend data is compared node by node with the preset cable force design target curve to generate cable force prediction deviation data. At the same time, the slope of the cable force change trend data is extracted to generate cable force change rate data.

[0083] S700: Compare the cable force prediction deviation data with the preset cable force deviation warning threshold, and compare the cable force change rate data with the preset cable force sudden change threshold, and generate the corresponding tension parameter adjustment command based on the comparison results.

[0084] For example, taking a double-tower, double-cable-stayed prestressed concrete cable-stayed bridge as an implementation case, the core parameters of the bridge are: main span 450m, side span 200m, a total of 128 cable stays, the cable body adopts φ15.24 low-relaxation steel strands, the maximum design tension of a single cable is 4200kN, and the design requirement is that the cable force control accuracy is ≤±1%.

[0085] First, obtain a preset number of historical tensioning datasets corresponding to the tensioning construction of similar cable-stayed bridges. Specifically, through the domestic bridge engineering archive database, select and obtain complete tensioning construction data for 6 similar cable-stayed bridges. All 6 bridges are double-tower, double-cable-stayed structures with main spans ranging from 380m to 520m (the deviation from the 450m main span of this project does not exceed ±30%), and the cables are all steel strands.

[0086] The effective tension sample size for a single bridge is 2,500 sets, and the total effective sample size for the six bridges is 15,000 sets. Each sample set has completely unified fields, including seven core fields: tension load (kN), tension displacement (mm), cable force (kN), ambient temperature (°C), ambient humidity (%RH), vertical displacement of the beam (mm), and bridge tower offset (mm).

[0087] Next, obtain the construction tensioning dataset corresponding to the stay cables that have been tensioned in the current construction project. That is, the tensioning construction of the first 8 stay cables in this project has been completed. Collect the closed-loop data of the entire process from the start of tensioning to the locking of cable force for these 8 cables as the construction tensioning dataset.

[0088] A total of 5,000 valid samples were collected. The fields and sampling frequency were completely consistent with the historical tensioning dataset. Each data set included the input parameters and final cable force results for the entire tensioning process, providing the model with personalized data specific to this project.

[0089] Then, the historical tensioning dataset and the construction tensioning dataset were processed to generate a training sample library. Specifically, missing value imputation, outlier removal, and data normalization were performed on 15,000 sets of historical tensioning datasets and 5,000 sets of construction tensioning datasets. The Pearson correlation coefficient method was used to calculate the correlation between each field and the cable force value. Finally, for example, tensioning load was 0.92, tensioning rate was 0.88, cable force increment was 0.95, ambient temperature was 0.71, ambient humidity was 0.63, and beam displacement was 0.68. All features |r| were ≥0.6 and were identified as strongly correlated tensioning feature parameters, thus completing feature extraction.

[0090] The 20,000 preprocessed and feature-extracted complete datasets are randomly divided in a fixed ratio of 7:2:1: 14,000 sets are used for training, which is used for model parameter learning; 4,000 sets are used for iterative verification and hyperparameter tuning during training; and 2,000 sets are used for model accuracy verification after training. These three datasets together constitute a standardized training sample library for the cable force prediction model.

[0091] Next, the training sample library is input into the preset machine learning network framework for training and iterative optimization to generate a model for predicting the trend of cable force changes.

[0092] Based on the six strongly correlated feature parameters extracted, a three-layer feedforward BP neural network was constructed.

[0093] The number of neurons in the input layer is 6, which matches the number of strongly correlated feature parameters in a 1:1 ratio. Each neuron corresponds to one feature field.

[0094] The hidden layer is a single hidden layer structure, according to the formula calculate, =6、 =1、 =8, therefore... ≈10.64, rounded down to 11 hidden layer neurons.

[0095] The number of neurons in the output layer is 1, and the output value is the predicted cable force value at the corresponding time node, which corresponds to the sample label value (measured cable force value).

[0096] The 6-dimensional feature parameter matrix of 14,000 samples in the training set is input into the network, and the label value is the measured cable force value of the corresponding sample to complete the standardized input.

[0097] Then, minimizing the loss function is used as the optimization objective. For example, if a sample predicts a cable force of 3500kN and the actual cable force is 3480kN, substituting this into the formula yields a loss value of 400 for that sample. The smaller the average loss value across the entire training set, the higher the model accuracy. The NSGA-II algorithm is used to perform global optimization, with the optimization targets being the network connection weights and thresholds. Fixed parameters are: population size 50, maximum number of iterations 200, crossover probability 0.7, and mutation probability 0.01. In each iteration, the loss function value is used as the fitness to select the optimal individual. After 200 iterations, the globally optimal initial weights and thresholds are output.

[0098] The globally optimal initial weights and thresholds are assigned to the BP neural network, and iterative training is performed using gradient descent. After each round of full training, 4000 sets of validation data are input into the model to calculate the loss value. If the validation set loss value does not decrease for 10 consecutive rounds, training is immediately terminated, and the current optimal parameters are retained to avoid overfitting. 2000 sets of test data are input into the trained model, and the mean absolute percentage error (MAPE) is calculated. For example, if the final test set MAPE is 0.82%, which meets the accuracy requirement of ≤1%, the model is deemed qualified, and a cable force change trend prediction model is generated. If the MAPE is >1%, the number of hidden layer neurons and the learning rate are adjusted, and the model is retrained until the requirements are met.

[0099] Then, multi-source tensioning data corresponding to the current tensioned cable are collected. Taking the current tensioning object as the 9th cable of this project as an example, it is currently in the loading stage before the 1200kN tensioning load is held.

[0100] According to the plan requirements, the data acquisition equipment was deployed, including magnetic flux cable force sensors, wire displacement sensors, and through-core load sensors installed at the tensioning end, and integrated temperature and humidity sensors installed on the bridge deck.

[0101] After tensioning is started, multi-source tensioning data is collected. Taking the effective data collected at the 10th second of tensioning as an example, the multi-source tensioning data are: real-time cable force 1200kN, real-time tensioning displacement 120mm, real-time tensioning load 1250kN, real-time ambient temperature 23.5℃, and real-time ambient humidity 58%RH.

[0102] Next, the multi-source tensioning data is input into the cable force change trend prediction model, which outputs the cable force change trend data corresponding to the tensioned cable in the future preset time period.

[0103] The collected multi-source tension data is normalized, and the maximum and minimum values ​​of each field are exactly the same as those in the training sample library. For example, if the real-time tension load is 1250kN, the maximum value of this field in the training sample library is 5000kN and the minimum value is 0kN. After normalization, the value is 0.25. Six-dimensional feature parameters that are exactly the same as those in the training sample library in terms of quantity, fields, and order are extracted to form a 1:1 matching feature vector, ensuring that it can be directly input into the model.

[0104] The preset threshold range for each feature field is completely consistent with the numerical range of the corresponding field in the training sample library. For example, the tension load threshold is 0-5000kN. The 1250kN collected this time is within the range, and all feature values ​​meet the threshold requirements, so the verification is qualified.

[0105] If a set of collected tension loads is 6000kN, which exceeds the threshold range of 0-5000kN, it is determined to be invalid data and immediately discarded. The data collection and preprocessing are then repeated until qualified data is obtained.

[0106] Then, the verified 6-dimensional feature vector is input into the input layer of the cable force change trend prediction model. The 6 feature values ​​correspond one-to-one with the 6 input neurons, thus completing the model input.

[0107] The model performs forward propagation calculations based on the input feature vector and the trained weights and thresholds, without human intervention. Currently, it is in the normal tensioning stage, with a preset future time period of 10 seconds and a time interval of 1 second. Finally, it outputs the cable force change trend data for the next 10 time nodes. For example, the 10 output trend data are: 1204KN, 1209KN, 1214KN, 1219KN, 1224KN, 1229KN, 1234KN, 1239KN, 1244KN, and 1249KN.

[0108] Then, the cable force change trend data is compared node by node with the preset cable force design target curve to generate cable force prediction deviation data. At the same time, the slope of the cable force change trend data is extracted to generate cable force change rate data. The cable force prediction deviation data is compared with the preset cable force deviation warning threshold, and the cable force change rate data is compared with the preset cable force sudden change threshold. Based on the comparison results, the corresponding tension parameter adjustment command is generated.

[0109] In one embodiment of this application, step S600 specifically includes: S601, calculating the difference between the cable force change trend data and the preset cable force design target curve according to time nodes, generating single-node deviation data corresponding to each time node, and integrating all single-node deviation data to generate cable force prediction deviation data for the entire time window.

[0110] It should be noted that the current tensioning of the cable stays is an ordered sequence of theoretical design cable force values ​​corresponding to each time node in the entire tensioning process; the time interval and number of time nodes of this curve are consistent with the time interval and number of time nodes of the cable force change trend data output in step S500; the time nodes correspond one-to-one with the equally spaced time nodes of the cable force change trend data in step S500.

[0111] Specifically, the cable force change trend data and the cable force design target curve are mapped one-to-one at the same time node, and the difference is calculated node by node to generate single-node deviation data. The calculation formula is as follows: ,in, This represents the single-node deviation data corresponding to the k-th time point; a positive value indicates that the predicted cable force is greater than the design target value, and a negative value indicates that the predicted cable force is less than the design target value. The predicted cable force value corresponding to the kth time node in the cable force change trend data output in step S500; The target value of the design cable force at the k-th time node in the cable force design target curve.

[0112] Arrange the single-node deviation data corresponding to all time points in chronological order to generate a one-dimensional ordered sequence with the same length as the cable force change trend data. This sequence is the cable force prediction deviation data for the entire time window, expressed as:

[0113]

[0114] Where n is the total number of time nodes within the preset time period in the future, which is consistent with the total number of nodes of the cable force change trend data in step S500.

[0115] Understandably, by comparing each node, the trend of cable force change is transformed into a quantified degree of cable force deviation, which accurately predicts the deviation between cable force and design target over the entire future time period. This provides a precise quantitative basis for adjusting tensioning parameters in advance, completely different from the passive mode of existing technologies that can only identify deviations that have already occurred.

[0116] S602. Simultaneously, perform first-order difference calculation on the cable force change trend data, extract the change slope at each time node, and generate cable force change rate data.

[0117] It should be noted that the calculation formula is as follows:

[0118]

[0119] in, The rate of change of cable force at the k-th time point is represented by a positive value, which indicates that the cable force is increasing, and a negative value indicates that the cable force is decreasing. The larger the absolute value, the faster the cable force changes and the higher the risk of sudden change. This is the predicted cable force value at the (k+1)th time node in the cable force change trend data. Δt represents the predicted cable force value at the k-th time node in the cable force change trend data; Δt is the time interval between adjacent time nodes.

[0120] The cable force change rates at all time points are arranged in chronological order to generate a one-dimensional ordered sequence, which is the cable force change rate data. The expression is:

[0121]

[0122] Understandably, by using first-order difference calculations to accurately capture the rate of change and abrupt change trends of cable force, the safety risks of sudden increases / decreases in cable force can be identified in advance. This is especially suitable for the safety control requirements of high-tensioning stages of long-span cable-stayed bridges, complementing the deviation data in step S601 and achieving dual-dimensional prediction.

[0123] For clarity, in the embodiments of this application, step S700 specifically includes: S701, comparing the cable force prediction deviation data with the preset cable force deviation warning threshold, and comparing the cable force change rate data with the preset cable force sudden change threshold.

[0124] It should be noted that, based on the cable force control accuracy requirements in the "Design Specifications for Highway Cable-Stayed Bridges" and combined with the allowable deviation of cable force specified in the project design documents, the fixed value is ±1% of the design target cable force value at the current tensioning stage; for example, when the design target cable force at the current stage is 4200kN, the cable force deviation warning threshold is ±42kN.

[0125] Based on the cable tensioning safety control requirements in the "Technical Specifications for Highway Bridge and Culvert Construction", and combined with the elastic modulus of the cable and the response speed of the hydraulic system of the tensioning jack, the preset cable force mutation threshold is set at 0.5 kN / s. When the cable force change rate exceeds this value, it is determined that there is an abnormal cable force mutation, which may cause a safety risk of structural stress exceeding the limit.

[0126] The two comparisons are performed simultaneously, without any order. The comparison rule is uniformly the "maximum value comparison principle", which is as follows: take the maximum absolute value of all single-node deviation data in the cable force prediction deviation data and compare it with the cable force deviation warning threshold; if the maximum value exceeds the cable force deviation warning threshold, it is determined that the cable force prediction deviation data exceeds the warning threshold.

[0127] Take the maximum absolute value of all rate values ​​in the cable force change rate data and compare it with the cable force mutation threshold; if the maximum value exceeds the cable force mutation threshold, it is determined that the cable force change rate data exceeds the mutation threshold.

[0128] By using fixed maximum value comparison rules, it is ensured that as long as there is a risk of cable tension exceeding the tolerance or sudden change at any time node within a preset time period in the future, it can be accurately identified, avoiding the problem of missed judgment in node-by-node comparison. At the same time, synchronous comparison ensures the timeliness of control decisions and meets the needs of real-time tension control.

[0129] S702. If the cable force prediction deviation data exceeds the cable force deviation warning threshold, then based on the future cable force change amplitude of the cable force change trend data, calculate the tension parameter adjustment quantification value, generate the first tension parameter adjustment command, and adjust the tension rate and load application amount of the current tensioning process.

[0130] It should be noted that, based on the future cable force change range (i.e., the maximum absolute value of the cable force prediction deviation data) based on the cable force change trend data, the quantitative values ​​for tension rate adjustment and load application adjustment are calculated separately to ensure that the adjustment amount is accurately matched with the predicted deviation range, and to avoid over-adjustment or under-adjustment.

[0131] Formula for calculating the quantitative value of tension rate adjustment:

[0132]

[0133] in, The quantitative value is adjusted to the tensioning rate; the negative sign indicates that the adjustment direction is opposite to the deviation direction, that is, if the predicted cable force is too large, the tensioning rate is reduced, and if the predicted cable force is too small, the tensioning rate is increased. This represents the steady-state tensioning rate currently being applied during the tensioning process. It represents the maximum absolute value of all single-node deviation data in the cable force prediction deviation data; This represents the design target cable force value for the current tensioning stage.

[0134] Formula for calculating the quantitative value of load application adjustment:

[0135]

[0136] in, Adjust the quantitative value for the load application amount; the negative sign indicates that the adjustment direction is opposite to the deviation direction, that is, if the predicted cable force is too large, the target load application amount is reduced, and if the predicted cable force is too small, the target load application amount is increased. This represents the maximum value of all single-node deviation data in the cable force prediction deviation data.

[0137] Based on the calculated quantification values of the tensioning rate adjustment and the load application adjustment, generate the first tensioning parameter adjustment instruction, which clearly includes the adjusted target tensioning rate and the target load application amount. Send this instruction to the tensioning hydraulic control unit, and through adjusting the oil inlet rate and the oil supply pressure of the tensioning jack, realize the closed-loop adjustment of the tensioning rate and the load application amount, and eliminate in advance the risk of cable force deviation predicted.

[0138] S703. If the cable force change rate data exceeds the cable force mutation threshold, generate the second tensioning parameter adjustment instruction including a pause or speed reduction instruction.

[0139] It should be noted that based on the amplitude by which the cable force change rate exceeds the threshold, hierarchical control is executed. The instructions are divided into two levels: the speed reduction instruction and the pause tensioning instruction. The specific rules are as follows:

[0140] First-level trigger (mildly exceeding the threshold): The maximum absolute value of the cable force change rate data exceeds the cable force mutation threshold but does not exceed 2 times the threshold. Generate the second tensioning parameter adjustment instruction including a speed reduction instruction, and the instruction clearly requires immediately reducing the current tensioning rate to the preset safe tensioning rate (fixed at 0.2 kN / s), and simultaneously locking the maximum load application amount to avoid exacerbating the cable force mutation.

[0141] Second-level trigger (severely exceeding the threshold): The maximum absolute value of the cable force change rate data exceeds 2 times the cable force mutation threshold or more. Generate the second tensioning parameter adjustment instruction including a pause tensioning instruction, and the instruction clearly requires the tensioning hydraulic control unit to immediately close the oil inlet valve of the jack, stop oil supply, lock the current tensioning state, prohibit continued loading, and only resume tensioning after finding out the cause of the cable force mutation and eliminating the risk.

[0142] If the first tensioning parameter adjustment instruction in step S702 and the second tensioning parameter adjustment instruction in step S703 are triggered simultaneously, the second tensioning parameter adjustment instruction shall be preferentially executed to avoid structural safety risks caused by adjusting the cable force accuracy and ensure the essential safety of the tensioning construction.

[0143] A precise positioning tensioning control device for the stay cables of a cable-stayed bridge, characterized in that it includes a data acquisition unit, a processor, a model training unit and a tensioning control unit.

[0144] Among them, the data acquisition unit includes a cable force sensor, a displacement sensor, a load sensor and an environmental monitoring module. The data acquisition unit is used to collect the historical tensioning data set, the construction tensioning data set and the multi-source tensioning data during the tensioning process corresponding to the stay cable tensioning.

[0145] The model training unit is data-connected to the processor and is used to train and generate a cable force change trend prediction model based on the training sample library output by the processor.

[0146] The processor's output is electrically connected to the input of the tension control unit, and is used to output tension parameter adjustment commands to the tension control unit.

[0147] The output of the tension control unit is electrically connected to the tension jack and is used to adjust the tensioning rate and load application during the tensioning process.

[0148] It should be noted that this device is the hardware carrier for realizing the aforementioned method of precise positioning and active control of cable tensioning. It corresponds one-to-one with the entire process steps of the method claims. The core is to form a complete closed loop of "data acquisition - model training - decision calculation - execution control" through the collaboration of multiple units, so as to realize the prediction of cable force change trend and active control of the tensioning process based on machine learning. This solves the core problems of passive and lagging adjustment of existing tensioning equipment, insufficient cable force control accuracy, and difficulty in preventing and controlling structural safety risks.

[0149] In summary, the cable-stayed bridge cable-stayed cable precise positioning and tensioning control device and its control method according to the embodiments of this application construct a training sample library by using historical tensioning datasets of similar cable-stayed bridges and construction tensioning datasets of the current project. Through training and iterative optimization of the machine learning network framework, the nonlinear variation law of cable force under multi-factor coupling during the tensioning process of the cable-stayed bridge is accurately fitted, solving the problems of poor adaptability and insufficient prediction accuracy of existing general models.

[0150] By inputting real-time multi-source tensioning data into the cable force change trend prediction model, the model outputs cable force change trend data within a preset time period in the future. This allows for the prediction of future cable force changes and potential deviations, fundamentally shifting from post-event deviation correction to pre-event proactive prevention and control, effectively avoiding the problems of excessive or insufficient cable tensioning.

[0151] Based on cable force change trend data, cable force prediction deviation data and cable force change rate data are generated simultaneously. Tensioning parameter adjustments are triggered by threshold comparison. At the same time, based on the future change range of cable force change trend, the adjustment values ​​of tensioning rate and load application amount are accurately quantified, realizing active control of the tensioning process. This can effectively improve the control accuracy of single cable force, significantly improve the uniformity of cable force distribution after bridge completion, and ensure the overall long-term stress performance of the bridge.

[0152] Based on machine learning-based trend prediction control, this invention simultaneously improves construction efficiency and structural safety. By predicting and proactively controlling cable force deviations in advance, it significantly reduces the number of repeated adjustments and iterations after tensioning, shortens the single-cable tensioning cycle and the overall construction period, and significantly improves construction efficiency. Furthermore, by monitoring and predicting the rate of cable force change, it can identify the risk of sudden cable force changes in advance, especially during the high-tensioning stage of long-span cable-stayed bridges, effectively reducing the safety hazards of stress concentration in bridge towers and beams, thus balancing construction efficiency and structural safety.

[0153] In the description of this specification, 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 indicated technical features. 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.

[0154] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0155] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for precise positioning and tension control of stay cables in a cable-stayed bridge, characterized in that, Includes the following steps: S100. Obtain the historical tensioning dataset and construction tensioning dataset corresponding to the stay cable tensioning; S200. Process the historical tensioning dataset and the construction tensioning dataset to generate a training sample library; S300. Input the training sample library into a preset machine learning network framework, perform training and iterative optimization, and generate a model for predicting the trend of cable force change. S400: Collect multi-source tensioning data corresponding to the currently tensioned cable-stayed cable; S500: Input the multi-source tensioning data into the cable force change trend prediction model, and output the cable force change trend data corresponding to the tensioned cable in a future preset time period; S600. The cable force change trend data is compared node by node with the preset cable force design target curve to generate cable force prediction deviation data. At the same time, the slope of the cable force change trend data is extracted to generate cable force change rate data. S700: Compare the predicted cable force deviation data with the preset cable force deviation warning threshold, and compare the cable force change rate data with the preset cable force sudden change threshold, and generate the corresponding tension parameter adjustment command based on the comparison result.

2. The method for precise positioning and tension control of cable-stayed bridge cables according to claim 1, characterized in that, Step S100 specifically includes: S101. Obtain a preset number of historical tensioning datasets corresponding to the tensioning construction of similar cable-stayed bridges. S102. Obtain the construction tensioning dataset corresponding to the stay cables that have been tensioned in the current construction project.

3. The method for precise positioning and tension control of cable-stayed bridge cables according to claim 1, characterized in that, S200 specifically includes: S201. The historical tensioning dataset and the real-time tensioning dataset are sequentially processed by missing value filling, outlier removal, and data normalization to extract tensioning feature parameters that are strongly correlated with cable force changes. S202. Divide the processed dataset into training set, validation set and test set according to the preset ratio to generate a training sample library; The tensioning characteristic parameters include tensioning load, tensioning rate, cable force increment, ambient temperature, ambient humidity, and beam displacement data.

4. The method for precise positioning and tension control of cable-stayed bridge cables according to claim 3, characterized in that, The S300 specifically includes: S301. Input the training set in the training sample library into a machine learning network framework built on a neural network. S302. With minimizing the force prediction deviation as the optimization objective, the initial weights and thresholds of the machine learning network framework are globally optimized. S303. Iteratively verify the model using the validation set in the training sample library and validate it using the test set to complete model training and optimization, and generate a model for predicting cable force change trends.

5. The method for precise positioning and tension control of cable-stayed bridge cables according to claim 1, characterized in that, The S400 specifically includes: During the tensioning process of the stay cable, multi-source tensioning data corresponding to the current tensioned stay cable is collected in real time according to the preset sampling frequency through cable force sensors, displacement sensors, load sensors deployed at the tensioning end, and environmental monitoring units deployed at the construction site. The multi-source tensioning data includes real-time cable force, real-time tensioning displacement, real-time tensioning load, and real-time environmental parameters.

6. The method for precise positioning and tension control of cable-stayed bridge cables according to claim 1, characterized in that, Step S500 specifically includes: S501. Perform normalization preprocessing on the multi-source tension data in the same way as the training sample library construction process, and extract tension feature parameters that match the dimensions of the training sample library. S502. Verify the validity of the extracted tension feature parameters and remove invalid feature data that exceeds the preset feature threshold range. S503. Input the qualified tension characteristic parameters into the cable force change trend prediction model; S504. Based on the calculation of the cable force change trend prediction model, output the cable force change trend data corresponding to the time nodes at equal intervals within a preset time period in the future.

7. The method for precise positioning and tension control of cable-stayed bridge cables according to claim 1, characterized in that, Step S600 specifically includes: S601. The difference between the cable force change trend data and the preset cable force design target curve is calculated node by node according to time nodes to generate single node deviation data corresponding to each time node, and all single node deviation data are integrated to generate cable force prediction deviation data for the whole time window. S602. Simultaneously, perform first-order difference calculation on the cable force change trend data, extract the change slope at each time node, and generate cable force change rate data.

8. The method for precise positioning and tension control of cable-stayed bridge cables according to claim 1, characterized in that, Step S700 specifically includes: S701. Compare the cable force prediction deviation data with the preset cable force deviation warning threshold, and at the same time compare the cable force change rate data with the preset cable force sudden change threshold. S702. If the cable force prediction deviation data exceeds the cable force deviation warning threshold, then based on the future cable force change amplitude of the cable force change trend data, calculate the tension parameter adjustment quantification value and generate the first tension parameter adjustment command to adjust the tension rate and load application amount of the current tensioning process. S703. If the cable force change rate data exceeds the cable force mutation threshold, a second tensioning parameter adjustment command containing a pause or deceleration command is generated.

9. A precise positioning and tensioning control device for cable-stayed bridge stay cables, characterized in that, It includes a data acquisition unit, a processor, a model training unit, and a tension control unit, among which, The data acquisition unit includes a cable force sensor, a displacement sensor, a load sensor, and an environmental monitoring module. The data acquisition unit is used to collect historical tension datasets, construction tension datasets, and multi-source tension data during the tensioning process of the stay cables. The model training unit is connected to the processor and is used to train and generate a model for predicting the trend of cable force change based on the training sample library output by the processor. The output terminal of the processor is electrically connected to the input terminal of the tension control unit, and is used to output tension parameter adjustment commands to the tension control unit; The output of the tension control unit is electrically connected to the tension jack and is used to adjust the tensioning rate and load application during the tensioning process.

Images