An Inclination and Damage Detection Method for Ancient Brick and Stone Arch Bridges Based on Ensemble Learning

Abstract

The present invention discloses a method for detecting the inclination and damage of ancient masonry arch bridges based on ensemble learning. First, in the initial stage, the point cloud data recognition module of the ancient masonry arch bridge inclination and damage detection system is set up. The initial point cloud data of the ancient masonry arch bridge and its surrounding environment are collected by a 3D laser scanner, and a structural calculation model of the ancient masonry arch bridge is established. Then, the importance ranking of the components of the traditional masonry arch bridge is calculated by the elastic modulus change method. According to the ranking results, the initial point cloud data is preprocessed and segmented. Based on PointNet++, the segmented point cloud data is recognized, the neural network is trained and updated, and the optimal recognition model of the point cloud of the stone arch bridge components is obtained. Then, the inclination and damage calculation module is set up. The current point cloud data of the arch bridge structural components is segmented by the optimal recognition model, and the inclination angle of the components is obtained through vector regression training and the Sklansky algorithm. Then, the inclination and damage conditions of the stone arch bridge structural components are obtained. The damage detection method of the present invention adopts an automated and informatized intelligent judgment method, and ensures the accuracy of inclination and damage through the method of ensemble learning.

Description

Technical Field

[0001] The present invention belongs to the technical field of civil engineering, and particularly relates to a method for detecting the inclination and damage of ancient masonry arch bridges based on ensemble learning. Background Art

[0002] Architectural heritages have high historical, artistic and scientific values, and are important material carriers of the country's "cultural confidence" and "cultural power". However, many of these buildings have various types of damage and diseases. Under some natural disasters or strong external forces, it is very easy to cause damage to cultural relics. Among them, bridge cultural relics, especially ancient masonry arch bridges, have been reported to have multiple damage and collapse incidents in recent years. Therefore, the protection of these ancient masonry arch bridges will surely change from rescue protection to preventive protection, so as to predict risks in advance and minimize the probability of risks. The structural safety monitoring of ancient masonry arch bridges is the main technology for preventive protection, and an important damage index of ancient masonry arch bridges is the inclination index. How to detect the inclination and damage of ancient masonry arch bridges according to the collected environmental and structural information and test the information that meets the requirements of engineering analysis is a difficult problem currently faced. Therefore, there is an urgent need to develop a method for detecting inclination and damage based on ensemble learning for the protection of ancient masonry arch bridges.

[0003] Currently, the main methods for detecting bridge damage are: manual combination of bridge inspection vehicle detection equipment, unmanned aerial vehicle aerial photogrammetry technology, infrared thermal imaging detection technology, three-dimensional laser scanning technology, etc. However, a single low-precision detection technology or a subjective evaluation method will inevitably encounter the problem of distorted damage recognition. The following specific problems exist:

[0004] (a) For the ancient masonry arch bridges that currently need health monitoring, due to the accumulation of apparent damage, the apparent scanning information has discreteness and different characteristics from the smooth apparent detection of modern bridges;

[0005] (b) The importance of components is rarely considered in the finite element reverse modeling technology for identifying component segmentation based on point clouds, resulting in the final component segmentation result not meeting the requirements of structural calculation;

[0006] (c) Traditional inclination component recognition technologies have insufficient recognition ability and low robustness in the engineering problems of ancient masonry arch bridges. Summary of the Invention

[0007] To solve the above problems, the present invention discloses a method for detecting the inclination and damage of ancient masonry arch bridges based on ensemble learning, which adopts an automated and information-based intelligent judgment method, and ensures the accuracy of inclination and damage through the method of ensemble learning, and has the functions of being able to be redeveloped, being able to correct the accurate recognition range according to the instrument accuracy, and being reusable.

[0008] To achieve the above object, the technical solution of the present invention is as follows:

[0009] An ancient masonry arch bridge inclination and damage detection method based on ensemble learning, comprising the following steps:

[0010] a) Set the point cloud data recognition module of the ancient masonry arch bridge inclination and damage detection system in the initial stage:

[0011] First step, use a three-dimensional laser scanner to scan the ancient masonry arch bridge and its surrounding environment to obtain initial high-precision point cloud data. Select the target point O in a relatively stable and flat area on the bridge deck of the masonry arch bridge in the point cloud data as the coordinate origin, set the x-axis along the traffic direction of the bridge, the y-axis perpendicular to the traffic direction of the bridge, and the z-axis perpendicular to the bridge deck direction.

[0012] Second step, establish a structural calculation model of the ancient masonry arch bridge according to the point cloud data. The masonry arch bridge is mainly composed of structural components such as the bridge deck, piers, arches, side walls, abutments, internal fill, and railings. The energy method of reducing the elastic modulus is used to calculate the importance τ i of the components of the traditional masonry arch bridge. The calculation method for obtaining the percentage ranking of the importance of the arch bridge components in the structure according to the calculation results is as follows:

[0013]

[0014] In the formula, τ i is the component importance of component i, Sum represents summation, Sort represents sorting from large to small, and Cum represents accumulating the sorted array. When the accumulated result > 70%, extract the corresponding component number for data elimination, and thus obtain the preprocessed point cloud data.

[0015] Third step, perform data segmentation on the point cloud data, establish a sample database of the masonry arch bridge components according to the segmented point cloud data, and divide the training set and the validation set. Then, based on PointNet++, identify the segmented point cloud data, train and update the neural network to obtain the optimal recognition model of the point cloud of the stone arch bridge components.

[0016] b) Set the inclination and damage calculation module of the ancient masonry arch bridge inclination and damage detection system at time t0:

[0017] First step, use an unmanned aerial vehicle (UAV) oblique photography device to collect the current high-precision point cloud data of the ancient masonry arch bridge cultural relic and its surrounding environment on the water surface, and preprocess the point cloud data through the optimal recognition model to obtain multiple target current point cloud data after segmenting the components.

[0018] In the second step, extract the point cloud cross-section plane of a certain component, select the point cloud near the deformed area, project it onto the x-z or y-z plane, and perform linear fitting on the contour line on the target plane. The fitting result is obtained by performing vector regression training on the point cloud data according to the following expression:

[0019]

[0020] s.t.f(x i )-y i ≤ε+ζ i

[0021] y i -f(x i )≤ε+η i

[0022] η i ≥0,ζ i ≥0,i=1,2,...,n

[0023] In the formula, w and b are training parameters, β is the regularization coefficient, η i and ζ i are slack variables. In the formula, ε is the prediction residual of vector regression training, y i is the true value of the data point, f(x i ) is the predicted value, and n is the number of samples.

[0024] After obtaining the fitting line, the slope of the contour line of the target plane is recorded as the corresponding weight value obtained by vector regression training, and the inclination angle is denoted as α A . If the correlation coefficient R2>0 in the fitting result of vector regression training, proceed to the third step. Otherwise, repeat the second step.

[0025] In the third step, project the point cloud onto the x-z or y-z plane, use the Sklansky algorithm to obtain the convex hull of the point cloud, and use the rotating calipers algorithm to solve the minimum circumscribed rectangle of the convex hull to obtain the inclination angle of the minimum circumscribed rectangle in the component direction, denoted as α B .

[0026] In the fourth step, take the average value of α A and α B as the final inclination angle α 0 , which is the inclination and damage condition of the structural component of the stone arch bridge.

[0027] c) Set the inclination rate change recognition module of the ancient masonry arch bridge inclination and damage detection system at times t1, t2, t3…tn of regular detection:

[0028] The detection at time t1 is to enter the process in step b) to obtain the component inclination angle α 1 , and the inclination change rate β=(α1 -α 0 ) / α 0 The tilt damage change rate results at times t2, t3…tn are calculated according to the tilt change rate formula.

[0029] The beneficial effects of the present invention are:

[0030] During the use of the present invention, first, the initial point cloud data of the ancient masonry arch bridge and its surrounding environment is collected by a three-dimensional laser scanner, and the structural calculation model of the ancient masonry arch bridge is established, and high-precision point cloud data and high-precision structural calculation model of the ancient masonry arch bridge are obtained; then, the importance ranking of the components of the traditional masonry arch bridge is calculated by changing the elastic modulus method, and the initial point cloud data is preprocessed and the data is segmented according to the ranking result, and the segmented point cloud data is identified based on PointNet++, and the neural network is trained and updated to obtain the optimal recognition model of the stone arch bridge component point cloud, so that the component segmentation is more in line with the structural calculation and analysis requirements. Then, the current point cloud data of the arch bridge structural components are segmented by the optimal recognition model, and the inclination angle of the component is obtained by vector regression training and Sklansky algorithm. The inclination angle solution based on ensemble learning accurately identifies the inclination damage of the stone arch bridge structural components. The damage detection method of the present invention has the functions of being able to be secondary developed, being able to calibrate the precise recognition range according to the accuracy of the instrument, and being reusable, and adopts an automated and information-based intelligent judgment method, and ensures the accuracy of the inclination damage through an ensemble learning method. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a framework diagram of the present invention. DETAILED DESCRIPTION

[0032] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0033] like Figure 1 As shown, the method for detecting the inclination and damage of an ancient masonry arch bridge based on ensemble learning described in the present invention comprises the following steps:

[0034] a) Setting up a point cloud data recognition module 2 of an ancient masonry arch bridge tilt and damage detection system 1 in the initial stage;

[0035] b) setting the tilt damage calculation module 9 of the ancient masonry arch bridge tilt damage detection system 1 at time t0;

[0036] c) Setting up the inclination rate change recognition module 15 of the ancient masonry arch bridge inclination damage detection system 1 for regular detection at times t1, t2, t3 ... tn.

[0037] In the specific implementation process, the setting method of the point cloud data recognition module 2 described in step a) includes:

[0038] In the first step, use a three-dimensional laser scanner 3 to scan the ancient masonry arch bridge 4 and its surrounding environment to obtain initial high-precision point cloud data 5. Select the target point O in a relatively stable and flat area on the bridge deck of the masonry arch bridge in the point cloud data 5 as the coordinate origin, set the x-axis along the traffic direction of the bridge, the y-axis perpendicular to the traffic direction of the bridge, and the z-axis perpendicular to the bridge deck.

[0039] In the second step, establish a structural calculation model 6 of the ancient masonry arch bridge according to the point cloud data 5. The masonry arch bridge is mainly composed of structural components such as the bridge deck, bridge piers, arches, side walls, abutment walls, internal fill, and railings. The energy method of reducing the elastic modulus is used to calculate the importance τ of the components of the traditional masonry arch bridge. i For the calculation, the percentage ranking of the importance of the arch bridge components in the structure can be calculated using Equation (1):

[0040]

[0041] In the formula, τ i is the component importance of component i, Sum represents summation, Sort represents sorting from large to small, and Cum represents accumulating the sorted array. When the accumulated result > 70%, extract the corresponding component numbers for data elimination, and thus obtain the preprocessed point cloud data 7.

[0042] In the third step, perform data segmentation on the point cloud data 7, establish a sample database of the masonry arch bridge sub-components according to the segmented point cloud data, and divide the training set and the validation set. Then, based on PointNet++, identify the segmented point cloud data, train and update the neural network to obtain the optimal recognition model 8 of the point cloud of the stone arch bridge components.

[0043] In the specific implementation process, the setting method of the inclination and damage calculation module 9 described in step b) includes:

[0044] In the first step, use an unmanned aerial vehicle (UAV) oblique photography device 10 to collect the current high-precision point cloud data 11 of the ancient masonry arch bridge cultural relic on the water surface and its surrounding environment. Preprocess the point cloud data 11 through the optimal recognition model 8 to obtain multiple target current point cloud data 12 after segmenting the components.

[0045] In the second step, extract the point cloud data cross-section 13 of a certain component, select the point cloud data 14 near the deformed area, project it onto the x-z or y-z plane, and perform linear fitting on the contour line on the target plane. The fitting result is obtained by performing vector regression training on the point cloud data according to Equation (2):

[0046]

[0047] such that \(f(x\) i ) - y i \(\leq \varepsilon+\zeta\) i

[0048] y i - \(f(x\) i ) \(\leq \varepsilon+\eta\) i

[0049] \(\eta\) i \(\geq 0,\zeta\) i \(\geq 0,i = 1,2,\cdots,n\ (2)\)

[0050] where \(w\) and \(b\) are training parameters, \(\beta\) is the regularization coefficient, \(\eta\) i and \(\zeta\) i are slack variables, \(\varepsilon\) is the prediction residual of vector regression training, \(y\) i is the true value of the data point, \(f(x\) i ) is the predicted value, and \(n\) is the number of samples.

[0051] After obtaining the fitting line, the slope of the target plane contour line is recorded as the corresponding weight value obtained by vector regression training, and the inclination angle is denoted as \(\alpha\) A . If the correlation coefficient \(R^{2}>0\) in the fitting result of vector regression training, then go to the third step. Otherwise, repeat the second step.

[0052] In the third step, project the point cloud data onto the \(x - z\) or \(y - z\) plane, use the Sklansky algorithm to obtain the convex hull of the point cloud data, and use the rotating calipers algorithm to solve the minimum circumscribed rectangle of the convex hull to obtain the inclination angle of the minimum circumscribed rectangle in the component direction, denoted as \(\alpha\) B .

[0053] In the fourth step, take the average of \(\alpha\) A and \(\alpha\) B as the final inclination angle \(\alpha\) 0 , which is the inclination damage condition of the structural component of the stone arch bridge.

[0054] In the specific implementation process, the setting method of the inclination rate change recognition module 15 described in step c) includes:

[0055] The detection at time \(t_{1}\) is to enter the process in step b) to obtain the inclination angle \(\alpha\) 1 of the component, and the inclination change rate \(\beta=(\alpha\) 1 -\(\alpha\) 0 ) / \(\alpha\) 0 . The inclination damage change rate results at times \(t_{2},t_{3}\cdots t_{n}\) are calculated according to this inclination change rate formula.

[0056] It should be noted that the above content only illustrates the technical idea of the present invention, and the protection scope of the present invention cannot be limited thereby. For those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements all fall within the protection scope of the claims of the present invention.

Claims

1. A method for detecting the tilt and damage of an ancient masonry arch bridge based on ensemble learning, characterized in that: The following steps are involved: a) Setting up the point cloud data recognition module of the ancient masonry arch bridge tilt damage detection system in the initial stage; The first step is to use a 3D laser scanner to scan the ancient masonry arch bridge and its surrounding environment to obtain initial high-precision point cloud data. The target point O in a relatively stable and flat area on the bridge deck of the masonry arch bridge in the point cloud data is selected as the coordinate origin. The x-axis is set along the direction of travel of the bridge, the y-axis is perpendicular to the direction of travel of the bridge, and the z-axis is perpendicular to the bridge deck. In the second step, the structural calculation model of the ancient masonry arch bridge was established based on the point cloud data: the masonry arch bridge consists of a bridge deck, piers, arches, side walls, diamond walls, internal fill and railings. The energy method of reducing the elastic modulus was used to calculate the importance of the components of the traditional masonry arch bridge. i Calculation is performed and the importance percentage ranking of arch bridge components in the structure is obtained according to the calculation results using formula (1): t i ≥0,i=1,2,...,n (1) In the formula, τ i is the component importance of component i, Sum represents summation, Sort represents sorting from large to small, and Cum represents accumulation of the sorted array; when the accumulation result is > 70%, the corresponding component number is extracted for data elimination, thereby obtaining the preprocessed point cloud data; The third step is to segment the point cloud data, establish a sample database of masonry arch bridge components based on the segmented point cloud data, and divide it into training set and validation set; then, based on PointNet++, identify the segmented point cloud data, train and update the neural network, and obtain the optimal recognition model of the stone arch bridge component point cloud; b) Setting the tilt and damage calculation module of the ancient masonry arch bridge tilt and damage detection system at time t0; The first step is to use drone oblique photography equipment to collect high-precision point cloud data of the current status of the ancient masonry arch bridge relics and their surrounding environment on the water surface, and pre-process the point cloud data through the optimal recognition model to obtain multiple target current status point cloud data after segmentation of components; In the second step, the point cloud section plane of a component is extracted, and the point cloud data near the deformation area is selected and projected to the xz or yz plane. The contour line on the target plane is linearly fitted. The fitting result is obtained by vector regression training of the point cloud data according to formula (2): s.t.f(x i )-y i ≤ε+ζ i y i -f(x i )≤e+n i or i ≥0.ζ i ≥0,i=1,2,...,n (2) In the formula, w and b are training parameters, β is the regularization coefficient, η i and i is the slack variable, ε is the vector regression training prediction residual, y i is the true value of the data point, f(x i ) is the predicted value, n is the sample size; After obtaining the fitted straight line, the slope of the target plane contour line is recorded as the corresponding weight obtained by vector regression training, and the inclination angle is recorded as α A ; If the correlation coefficient R2>0 in the fitting result of the vector regression training, proceed to the third step; otherwise, repeat the second step; The third step is to project the point cloud data onto the xz or yz plane, use the Sklansky algorithm to obtain the convex hull of the point cloud data, and use the rotating caliper algorithm to solve the minimum circumscribed rectangle of the convex hull to obtain the inclination angle of the minimum circumscribed rectangle in the component direction, which is recorded as α. B ; The fourth step is to take α A and α B The average value of is the final inclination angle α0, which is the inclination and damage condition of the component of the stone arch bridge; c) Setting a tilt rate change recognition module for the ancient masonry arch bridge tilt damage detection system at the time t1, t2, t3…tn for regular detection; The detection at time t1 is the process to enter step b), and the obtained component inclination angle α1 and inclination change rate β=(α1-α0) / α0; the inclination damage change rate results at times t2, t3...tn are calculated according to the inclination change rate formula.

Images