Displacement prediction method applied to high and large formwork supporting structure

Abstract

The invention provides a displacement prediction method applied to a tall and large formwork supporting structure, and the method comprises the steps: S1, building a visual displacement monitoring system for each monitoring node on the tall and large formwork supporting structure at a construction site, the visual displacement monitoring system comprises a calibration object for calibrating each monitoring node and a monitoring camera for acquiring a real-time displacement monitoring picture generated by each monitoring node; s2, acquiring displacement response data of each monitoring node through a visual monitoring system; step S3, constructing a CNN (Convolutional Networks)-BiLSTM (BiLSTM-Adaboost) structure response prediction model; and S4, inputting the displacement response data into the CNN-BiLSTM-Adaboost structure response prediction model for training, and performing prediction by using the CNN-BiLSTM-Adaboost structure response prediction model to obtain the displacement amount of the tall and large template support structure in the future time period. The method has the beneficial effects that the displacement amount of the high and large formwork supporting structure in the future time period can be accurately predicted, and sufficient emergency processing time is reserved for an engineering site.

Description

Technical Field

[0001] The invention relates to the technical field of displacement prediction of high formwork structures, and in particular to a displacement prediction method applied to high formwork support structures. Background Art

[0002] In recent years, my country's construction industry has developed in the direction of large spans and large spaces. Tall formwork support structures (referred to as high formwork) have been widely used in the construction process. Due to the high height and poor stability of high formwork structures, their safety is difficult to guarantee. Improper high formwork design and insufficient understanding of the working status of the formwork support system during concrete pouring are the main causes of collapse. Therefore, real-time monitoring of high formwork structures and rapid response prediction can timely grasp the stress state of the structure, take emergency measures in advance, prevent structural instability or collapse, and ensure the safety of construction workers.

[0003] At present, domestic and foreign scholars' research on high formwork is mostly focused on the ultimate bearing capacity, semi-rigidity of nodes, stress and deformation, etc. Some have studied the structural instability mechanism and failure mode through static destruction tests of full-scale formwork frames; some have studied the stable bearing capacity of formwork frames and proposed a simplified calculation method; some have conducted experimental research on the nodes of fastener-type formwork frames and concluded that the vertical and horizontal bar connection nodes are semi-rigid connections. In terms of overall structural safety, real-time monitoring of high formwork is an important means to ensure construction safety. Most of the existing IoT perception and early warning systems are based on contact sensors, but such sensors require manual climbing operations during installation, detection, maintenance, and disassembly, which poses a great risk to operational safety. Therefore, it is necessary to develop a remote, non-contact monitoring and early warning system suitable for the complex environment of construction sites.

[0004] Computer vision technology, as a typical non-contact monitoring method, is widely used in the field of civil engineering. For example, using consumer-grade cameras as acquisition equipment and optical flow technology to achieve motion tracking of the target, the dynamic characteristics of the structure can be accurately identified. The results show that it has high measurement accuracy in the frequency domain. A new method for identifying construction site hazards that combines computer vision and ontology models has also been proposed. When a hazard occurs, it can automatically and accurately identify the hazard source while complying with safety regulations. For high-support formwork structures, a set of intelligent real-time monitoring methods for construction based on computer vision and finite elements has been established to achieve remote, real-time, and multi-point monitoring of the supporting structure. A high-support formwork structure displacement monitoring method based on online cameras has also been proposed, which solves the problem of light path occlusion caused by complex environments in on-site monitoring and has been successfully applied to actual formwork projects. Although the above research has made certain progress in real-time monitoring of high-support formwork structures, it has not yet explored in depth the early prediction of structural response.

[0005] In summary, for high-support formwork structures, the existing response warning method only triggers an alarm when the displacement exceeds the limit. It cannot accurately predict the response trend and cannot reserve sufficient emergency response time for the construction site. Summary of the invention

[0006] The technical problem to be solved by the present invention is to accurately predict the displacement in future time periods and reserve sufficient emergency processing time for the engineering site. In order to overcome the defects of the above-mentioned prior art (or related technology), the present invention provides a displacement prediction method applied to tall formwork support structures.

[0007] The present invention provides a displacement prediction method applied to a tall formwork support structure, comprising the following steps: Step S1, building a visual displacement monitoring system for each monitoring node on the tall formwork support structure at the construction site, wherein the visual displacement monitoring system includes a calibration object for calibrating each monitoring node and a monitoring camera for obtaining a real-time displacement monitoring image generated by each monitoring node; Step S2, obtaining displacement response data of each monitoring node through the visual monitoring system; Step S3, constructing a CNN-BiLSTM-Adaboost structural response prediction model; Step S4, inputting the displacement response data into a CNN-BiLSTM-Adaboost structural response prediction model for training, and using the CNN-BiLSTM-Adaboost structural response prediction model to predict the displacement of the tall template support structure in a future period.

[0008] Compared with the prior art, the displacement prediction method for a tall formwork support structure of the present invention has the following advantages: The present invention takes a tall formwork support structure as a monitoring object, sets monitoring cameras for different monitoring nodes of the tall formwork support structure, and monitors the monitoring nodes in different areas. A visual displacement monitoring system is built through step S1, displacement response data is collected through step S2, a CNN-BiLSTM-Adaboost structural response prediction model is constructed through step S3, and the CNN-BiLSTM-Adaboost structural response prediction model is trained and the displacement of the tall formwork support structure in the future time period is predicted through step S4. The displacement of the tall formwork support structure after the future time period can be accurately predicted, and the displacement trend of the tall formwork support structure can be predicted in advance, so as to reserve sufficient emergency processing time for the engineering site.

[0009] In a possible implementation, in step S3, the constructed CNN-BiLSTM-Adaboost structural response prediction model integrates a CNN model, a BiLSTM model and an Adaboost algorithm.

[0010] In a possible implementation, the CNN model is used to extract image features from the displacement response data, capture local features through a convolutional layer, reduce the spatial dimension through a pooling layer, and output the extracted image features to the BiLSTM model.

[0011] In a possible implementation, the BiLSTM model includes two LSTM models, and the BiLSTM model is used to perform sequence modeling on the image features. The sequence dynamics of the image features are captured and outputted in a forward and reverse manner through the two LSTM models.

[0012] In a possible implementation, the Adaboost algorithm constructs a strong classifier by integrating multiple weak classifiers, and iteratively adjusts the weights of the weak classifiers to output a prediction result as the displacement of the tall template support structure in the future period.

[0013] In a possible implementation, the displacement response data includes peak displacements in different time periods. In step S4, for each time period, the peak displacement of the time period is used as input, and the peak displacement of the next time period after the time period is used as output to train the CNN-BiLSTM-Adaboost structural response prediction model.

[0014] In a possible implementation manner, the future time period is the next hour.

[0015] In a possible implementation, the step S1 of building a visual displacement monitoring system for monitoring nodes on the high formwork at the construction site specifically includes: Step S11, installing the calibration object on the pole corresponding to the monitoring node; Step S12, determining the image resolution Res of the monitoring camera and the measurement accuracy of the monitoring nodes in the construction site , in mm, and based on the calculation formula Calculate the field of view FOV of the surveillance camera, where: Indicates the measurement accuracy of the visual displacement monitoring system, in pixels; Step S13, combining the field of view FOV, size and focal length of the surveillance camera , based on the calculation formula Calculate the working distance of the surveillance camera , where The scale factor for mapping the image captured by the surveillance camera to the coordinates of the construction site: In the formula, is the actual distance between the monitoring camera and the monitoring node, is the pixel distance of the surveillance camera.

[0016] In a possible implementation, step S2 includes: Step S21, collecting an image of the monitoring node before entering the visual monitoring work through a monitoring camera as a reference image, calibrating the position of the monitoring node in the reference image through a rectangular search box as a feature area, and using the feature area as a predefined template; Step S22, using the same rectangular search frame as in step S21 to move pixel by pixel on the image collected during the visual monitoring work, and calculating the correlation of each pixel in order to find the coordinate position of the feature area in the image; Step S23, inversely transform the coordinate positions of the feature areas of the images at different times found in step S22 to determine the coordinate positions of the corresponding monitoring nodes on the high formwork at the construction site. In a possible implementation manner, the calculation formula for calculating the correlation of each pixel in step S22 is: In the formula, Represents the correlation coefficient of the pixel in the i-th row and j-th column (-1< <1), m and n represent the length and width of the rectangular search box respectively, Represents pixel information on a predefined template. Represents the average grayscale of the predefined template, It is represented by the pixel information of the xth row and yth column on the rectangular search box at sampling time t. Represents the average grayscale of the rectangular search box at sampling time t; The region closest to 1 is taken as the feature region. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a flow chart of the steps of the present invention; Figure 2 A visual displacement monitoring system is built for two monitoring nodes at a construction site in a specific embodiment of the present invention; Figure 3 This is a real-time monitoring image of displacement collected by a ball camera in a specific embodiment of the present invention; Figure 4 It is a schematic diagram of the CNN structure of the present invention; Figure 5It is a structural schematic diagram of the LSTM model of the present invention; Figure 6 It is a schematic diagram of the CNN-BiLSTM-Adaboost structure of the present invention; Figure 7 It is a schematic diagram of the horizontal displacement training set (RMSE=0.056283) of the upright pole 1 of the present invention; Figure 8 It is a schematic diagram of the horizontal displacement test set (RMSE=0.088748) of the upright pole 1 of the present invention; Fig. 9 It is a schematic diagram of the vertical displacement training set (RMSE=0.040511) of the upright pole 1 of the present invention; Fig.10 It is a schematic diagram of the vertical displacement test set (RMSE=0.041525) of the upright pole 1 of the present invention; Fig.11 It is a regression diagram of the vertical pole 1 of the present invention - a schematic diagram of a horizontal displacement training set; Fig.12 It is a regression diagram of the vertical pole 1 of the present invention - a schematic diagram of a horizontal displacement test set; Fig.13 It is a regression diagram of the vertical pole 1 of the present invention - a schematic diagram of a horizontal displacement training set; Fig.14 It is a regression diagram of the vertical pole 1 of the present invention - a schematic diagram of a horizontal displacement test set; Fig.15 It is a schematic diagram of the horizontal displacement prediction error of the upright pole 1 of the present invention; Fig.16 It is a schematic diagram of the vertical displacement prediction error of the upright pole 1 of the present invention. DETAILED DESCRIPTION

[0018] First, those skilled in the art should understand that these implementations are only used to explain the technical principles of the embodiments of the present invention, and are not intended to limit the protection scope of the embodiments of the present invention. Those skilled in the art can make adjustments to them as needed to adapt to specific application scenarios.

[0019] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] See also Figure 1 The embodiment of the present invention discloses a displacement prediction method applied to a tall formwork support structure, comprising the following steps: Step S1, a visual displacement monitoring system is set up for each monitoring node on the tall formwork support structure at the construction site. The visual displacement monitoring system includes a calibration object for calibrating each monitoring node and a monitoring camera for obtaining a real-time displacement monitoring image generated by each monitoring node, specifically including: Step S11, determine the image resolution Res of the monitoring camera and the measurement accuracy of the monitoring nodes in the construction site , in mm, and based on the calculation formula Calculate the field of view FOV of the surveillance camera, where: Indicates the measurement accuracy of the visual displacement monitoring system, in pixels; Step S12, combining the field of view FOV, size and focal length of the surveillance camera , based on the calculation formula Calculate the working distance of the surveillance camera , where The scale factor for mapping the image captured by the surveillance camera to the coordinates of the construction site: In the formula, is the actual distance between the monitoring camera and the monitoring node, is the pixel distance of the surveillance camera; Step S2, obtaining the displacement response data of each monitoring node through the visual monitoring system, specifically includes: Step S21, taking the image of the monitoring node captured by the monitoring camera before entering the visual monitoring work as a reference image, calibrating the position of the monitoring node in the reference image through a rectangular search box as a feature area, and taking the feature area as a predefined template; Step S22, after the visual monitoring system enters the visual monitoring work, the monitoring camera collects monitoring images according to the preset monitoring frequency, and uses the same rectangular search box as step S21 to move pixel by pixel on the image collected during the visual monitoring work, and calculates the correlation of each pixel to find the coordinate position of the feature area in the monitoring image; the calculation formula of the correlation of each pixel is: In the formula, Represents the correlation coefficient of the pixel in the i-th row and j-th column (-1< <1), m and n represent the length and width of the rectangular search box respectively, Represents pixel information on a predefined template. Represents the average grayscale of the predefined template, It is represented by the pixel information of the xth row and yth column on the rectangular search box at sampling time t. Represents the average grayscale of the rectangular search box at sampling time t; The area closest to 1 is taken as the feature area; Step S23, inversely transforming the coordinate positions of the feature areas of the images at different times found in step S22 to determine the coordinate positions of the corresponding monitoring nodes on the high formwork in the construction site; Step S3, constructing a CNN-BiLSTM-Adaboost structural response prediction model; Step S4, inputting the displacement response data into the CNN-BiLSTM-Adaboost structural response prediction model for training, and using the CNN-BiLSTM-Adaboost structural response prediction model to predict the displacement of the tall template support structure in the future period.

[0021] Continue to see Figure 1 In step S22, the method of calculating the correlation pixel by pixel using a rectangular search box can only obtain integer pixel coordinates, and the measurement accuracy is limited. Therefore, in this specific embodiment, it is necessary to use sub-pixel technology to further improve the measurement accuracy, and use a double parabola fitting method to make the displacement measurement reach sub-pixel accuracy. The double parabola fitting method is a prior art and will not be described in detail here.

[0022] Continue to see Figure 1 In step S23, in order to solve the problem of insufficient recognition accuracy and low robustness in adverse scenarios such as incomplete outdoor monitoring markers and blurred images, this specific embodiment uses the YOLOX algorithm to identify occluded markers and the SRGAN network to improve image resolution. At the same time, a non-displacement reference point is selected near the structure, and the relative displacement between the point and the structure is used as the measured displacement to reduce the error in the recognition algorithm. The technical contents about the YOLOX algorithm, the SRGAN network and the non-displacement reference point are prior arts and will not be elaborated here.

[0023] Continue to see Figure 1 The image resolution of the ball camera's real-time capture is 1920×1088, the visual measurement system needs to achieve a measurement accuracy of 1mm, and the pixel recognition accuracy of the conservatively estimated algorithm is 0.5pixel. The maximum field of view of the ball camera is 3.84m×2.18m, the ball camera sensor size is 1 / 2.8 (5.0mm×2.8mm after conversion), and the focal length is variable between 4.8 and 110mm. Combining the above parameters, it is calculated that the appropriate working distance of the ball camera is about 3.7m~84.0m.

[0024] Continue to see Figure 1 In this specific embodiment, the facade of the public building to be measured at the construction site is directly opposite to the project department and the straight-line distance is 40m, which is within the reasonable working range of the ball camera. Therefore, the ball camera is installed in the project department for monitoring, and two elevation points are selected for displacement monitoring. The specific layout is as follows: Figure 2 As shown, select the corner and its adjacent poles on the left and right, such as Figure 3The rods 1-3 shown are used as monitoring objects, and two ball cameras are used to monitor the areas separated by 8 meters in the height direction. Among them, the monitoring node 1 monitored by ball camera 1 is located between 15.4-16.9 meters, and the monitoring node 2 monitored by ball camera 2 is located at 7.9 meters.

[0025] Continue to see Figure 1 The present invention uses a hybrid deep learning method to train visual monitoring displacement response data to achieve short-term prediction of structural displacement, which helps to provide early warning and leave sufficient emergency response time for on-site staff. In engineering monitoring practice, for temporary structures with greater safety risks, due to the lack of short-term prediction function, the alarm will only be triggered when the actual displacement exceeds the warning threshold, resulting in insufficient emergency response time when facing potential accidents. In order to solve this problem, the present invention adopts an adaptive enhanced ensemble learning algorithm (i.e., CNN-BiLSTM-Adaboost) of a hybrid deep learning model to perform short-term prediction of the displacement response of tall template support structures.

[0026] Continue to see Figure 1 The CNN-BiLSTM-Adaboost structural response prediction model integrates convolutional neural network (CNN), bidirectional long short-term memory network (BiLSTM) and adaptive boosting algorithm (Adaboost), and has been widely used in image recognition and natural language processing. The key to this algorithm is that it combines CNN spatial feature extraction, BiLSTM sequence modeling and Adaboost integrated learning, enabling the model to skillfully handle complex pattern recognition tasks. The following is a brief overview of these three components: (1) Figure 4 The CNN shown in Figure 1 is good at processing image data. It extracts image features through convolutional layers, reduces spatial dimensions through pooling layers, and performs classification or regression tasks through fully connected layers. (2) LSTM models, such as Figure 5 As shown in , LSTM is a special recurrent neural network (RNN) architecture that can learn long-term dependencies. BiLSTM consists of two LSTM models that capture both forward and reverse sequence information; (3) Figure 6 As shown in Figure 1, Adaboost is an integrated learning algorithm that builds a strong classifier by merging multiple weak classifiers. It improves the overall performance by iteratively adjusting the weights of weak classifiers.

[0027] Continue to see Figure 1,The implementation of the CNN-BiLSTM-Adaboost algorithm includes several steps: first, CNN is used to extract image features from input data (such as real-time images or text sequences). Its convolutional layer captures local features and the pooling layer reduces the spatial dimension. Next, the extracted image features are input into BiLSTM, which processes sequence data and captures sequence dynamics through the LSTM model forward and backward; then, the BiLSTM output is used as the input of Adaboost to improve the classification performance by integrating multiple weak classifiers. Adaboost iteratively adjusts the weights of weak classifiers during model training to improve classification accuracy. After the model training is completed, the data to be predicted is input into the trained CNN-BiLSTM-Adaboost structural response prediction model. By applying CNN for feature extraction, BiLSTM for sequence modeling, and Adaboost integrated learning method in sequence, the displacement response is finally predicted.

[0028] Continue to see Figure 1 The present invention uses the CNN-BiLSTM-Adaboost structure to make short-term predictions on the displacement of tall formwork support structures, and uses the displacement data under actual construction conditions obtained through visual monitoring as training samples. Through discussions with construction personnel and considering the process operation time, the time required for early warning is determined. Due to the inconsistency of daily processes on construction sites, the warning time cannot be more than one day in advance. However, if the warning time is too short, there will not be enough time for on-site emergency response. Therefore, the present invention uses computer vision monitoring results as the main data source, divides the working day into time periods in hours, selects the maximum displacement value in each time period, takes the peak displacement of the current time period as the model input, and the peak displacement of the subsequent time period as the model output to construct a data set. A total of 600 groups of sample data are selected, 420 groups (70%) of sample data are selected as training sets, and the remaining 180 groups (30%) are selected as test sets. The model is trained and tested, which can predict the potential damage risk of high-supported formwork structures one hour in advance.

[0029] Continue to see Figure 1 Due to space limitations, the present invention cannot include all prediction results. Figure 5-Figure 14 The horizontal and vertical displacements of the pole 1 captured by the camera 1 are shown as Figure 7 , Figure 8 , Fig. 9 , Fig.10 As shown in Figure 2, the algorithm exhibits accurate short-term prediction capability and robustness, with the RMSE of both the training and test data sets remaining within 0.1 mm, and the regression plots also showing strong correlation, as shown in Figure 2. Fig.11 , Fig.12 , Fig.13 , Fig.14 As shown in Figure 2, the horizontal and vertical displacement prediction errors are kept within 0.2 mm and 0.1 mm, respectively. Fig.15 , Fig.16 As shown, with the enhancement of visual monitoring data and model analysis, the iterative update of the algorithm will further improve the accuracy of structural displacement prediction.

[0030] In the description of the present invention, the description with reference to the terms "one embodiment", "some embodiments", "in the present embodiment", "specific example", or "some examples" etc. means that the specific features, mechanisms, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, mechanisms, materials or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0031] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A displacement prediction method applied to a tall formwork support structure, characterized in that: The following steps are involved: Step S1, building a visual displacement monitoring system for each monitoring node on the tall formwork support structure at the construction site, wherein the visual displacement monitoring system includes a calibration object for calibrating each monitoring node and a monitoring camera for obtaining a real-time displacement monitoring image generated by each monitoring node; Step S2, obtaining displacement response data of each monitoring node through the visual monitoring system; Step S3, constructing a CNN-BiLSTM-Adaboost structural response prediction model; Step S4, inputting the displacement response data into a CNN-BiLSTM-Adaboost structural response prediction model for training, and using the CNN-BiLSTM-Adaboost structural response prediction model to predict the displacement of the tall template support structure in a future period.

2. The displacement prediction method according to claim 1, characterized in that: In step S3, the constructed CNN-BiLSTM-Adaboost structural response prediction model integrates the CNN model, the BiLSTM model and the Adaboost algorithm.

3. The displacement prediction method according to claim 2, characterized in that: The CNN model is used to extract image features from the displacement response data, capture local features through a convolutional layer, reduce the spatial dimension through a pooling layer, and output the extracted image features to the BiLSTM model.

4. The displacement prediction method according to claim 3, characterized in that: The BiLSTM model includes two LSTM models, and the BiLSTM model is used to perform sequence modeling on the image features. The sequence dynamics of the image features are captured and outputted through the two LSTM models in the forward and reverse directions.

5. The displacement prediction method according to claim 2, characterized in that: The Adaboost algorithm constructs a strong classifier by integrating multiple weak classifiers, and outputs a prediction result as the displacement of the tall template support structure in the future period by iteratively adjusting the weights of the weak classifiers.

6. The displacement prediction method according to claim 1, characterized in that: The displacement response data includes peak displacements in different time periods. In step S4, for each time period, the peak displacement of the time period is used as input, and the peak displacement of the next time period after the time period is used as output to train the CNN-BiLSTM-Adaboost structural response prediction model.

7. The displacement prediction method according to claim 1, characterized in that: In step S4, the future period is the next hour.

8. The displacement prediction method according to claim 1, characterized in that: The step S1 of building a visual displacement monitoring system for monitoring nodes on the high formwork at the construction site specifically includes: Step S11, installing the calibration object on the pole corresponding to the monitoring node; Step S12, determining the image resolution Res of the monitoring camera and the measurement accuracy of the monitoring nodes in the construction site , in mm, and based on the calculation formula Calculate the field of view FOV of the surveillance camera, where: Indicates the measurement accuracy of the visual displacement monitoring system, in pixels; Step S13, combining the field of view FOV, size and focal length of the surveillance camera , based on the calculation formula Calculate the working distance of the surveillance camera , where The scale factor for mapping the image captured by the surveillance camera to the coordinates of the construction site: In the formula, is the actual distance between the monitoring camera and the monitoring node, is the pixel distance of the surveillance camera.

9. The displacement prediction method according to claim 1, characterized in that: The step S2 comprises: Step S21, collecting an image of the monitoring node before entering the visual monitoring work through a monitoring camera as a reference image, calibrating the position of the monitoring node in the reference image through a rectangular search box as a feature area, and using the feature area as a predefined template; Step S22, using the same rectangular search frame as in step S21 to move pixel by pixel on the image collected during the visual monitoring work, and calculating the correlation of each pixel in order to find the coordinate position of the feature area in the image; Step S23, inversely transforming the coordinate positions of the feature areas of the images at different times found in step S22 to determine the coordinate positions of the corresponding monitoring nodes on the high formwork in the construction site.

10. The displacement prediction method according to claim 9, characterized in that: The calculation formula for calculating the correlation of each pixel in step S22 is: In the formula, Represents the correlation coefficient of the pixel in the i-th row and j-th column (-1< <1), m and n represent the length and width of the rectangular search box respectively, Represents pixel information on a predefined template. Represents the average grayscale of the predefined template, It is represented by the pixel information of the xth row and yth column on the rectangular search box at sampling time t. Represents the average grayscale of the rectangular search box at sampling time t; The region closest to 1 is taken as the feature region.

Images