Bridge and tunnel geological disaster monitoring method and system based on unmanned aerial vehicle

Abstract

The invention provides a bridge and tunnel geological disaster monitoring method and system based on an unmanned aerial vehicle, and relates to the technical field of geological disaster monitoring, and the method comprises the steps: obtaining the data of a sensor carried by the unmanned aerial vehicle; performing space-time alignment and confidence coefficient weighting on the three-dimensional point cloud data, the image data and the flight attitude data through a space-time weighted covariance fusion model to obtain fusion deformation feature data; inputting the data into a wavelet-Volterra series coupling prediction algorithm, performing nonlinear deformation prediction on the key part, and outputting a future displacement prediction value; generating a geological disaster risk probability graph according to the predicted value and a preset risk threshold value; and based on the probability graph, dynamically optimizing the flight speed, height and flight path of the unmanned aerial vehicle through a Bayesian adaptive flight path planning module, and outputting a real-time monitoring path instruction. Through multi-modal fusion and dynamic path planning, the geological disaster monitoring precision and the cruising ability of the unmanned aerial vehicle are improved, and the method is suitable for safety protection of bridges and tunnels in complex terrains.

Description

Technical Field

[0001] The present invention relates to the technical field of geological disaster monitoring, and particularly to a method and system for monitoring geological disasters of bridges and tunnels based on unmanned aerial vehicles (UAVs). Background Art

[0002] In the field of infrastructure monitoring, especially the geological disaster monitoring of bridges and tunnels, traditional methods mainly rely on manual inspections and fixed sensors. Although these methods can provide information on structural health monitoring to a certain extent, manual inspections usually can only cover limited areas and are affected by factors such as weather and traffic, making it difficult to ensure the continuity and timeliness of monitoring. The layout and quantity of fixed sensors often cannot comprehensively reflect the health status of the structure, resulting in the locality and one-sidedness of data. High cost: The maintenance costs of manual inspections and traditional sensors are relatively high, especially in remote or inaccessible areas.

[0003] With the rapid development of unmanned aerial vehicle (UAV) technology, UAV-based monitoring methods have gradually become an effective alternative. UAVs can carry a variety of sensors, such as inertial navigation units (IMUs), light detection and ranging (LiDAR), and high-definition cameras. UAVs can cover large areas in a short time and quickly obtain high-resolution geographical and environmental data; by combining IMU, LiDAR, and visual image data, UAVs can provide more comprehensive monitoring information, enhance the reliability and accuracy of data, and UAVs can achieve real-time data transmission and processing, timely detect potential geological disaster risks, and improve the emergency response ability.

[0004] However, although UAV technology shows great potential in geological disaster monitoring, the existing technology still faces some challenges:

[0005] (1) There are differences in the data obtained by different sensors in terms of time and space. How to effectively fuse the data to improve the monitoring accuracy is a key issue.

[0006] (2) When bridges and tunnels are affected by external factors, complex non-linear deformations may occur, and traditional prediction models are difficult to accurately capture these changes.

[0007] (3) When UAVs perform monitoring tasks, how to optimize the flight path to reduce energy consumption while ensuring the monitoring quality is still an urgent problem to be solved.

[0008] In summary, the UAV-based method for monitoring geological disasters of bridges and tunnels has significant advantages, but further research and innovation are still needed in aspects such as data fusion, deformation prediction, and energy consumption optimization. Summary of the Invention

[0009] To overcome the deficiencies of the prior art, the objective of the present invention is to provide a method and system for monitoring geological disasters of bridges and tunnels based on unmanned aerial vehicles (UAVs). Through multimodal fusion, nonlinear prediction, and dynamic path planning, the accuracy of geological disaster monitoring and the endurance of UAVs are significantly improved, which is applicable to the safety protection of bridges and tunnels in complex terrains.

[0010] To achieve the above objective, the present invention provides the following solutions:

[0011] A method for monitoring geological disasters of bridges and tunnels based on UAVs, comprising:

[0012] Obtaining data collected by sensors carried by the UAV; the sensors include a lidar, a visible light camera, and an inertial measurement unit; the lidar is used to collect three-dimensional point cloud data of the surfaces of the target bridge and tunnel; the visible light camera is used to capture image data of the surfaces of the target bridge and tunnel; the inertial measurement unit is used to record the flight attitude data of the UAV; the UAV performs monitoring tasks according to a preset monitoring path;

[0013] Performing spatio-temporal alignment and confidence weighting of the three-dimensional point cloud data, the image data, and the flight attitude data through a spatio-temporal weighted covariance fusion model to obtain fused deformation feature data;

[0014] Inputting the fused deformation feature data into a wavelet-Volterra series coupling prediction algorithm to perform nonlinear deformation prediction on key parts of the target bridge and tunnel, and outputting displacement prediction values for a future time window;

[0015] Generating a geological disaster risk probability map according to the displacement prediction values and a preset risk threshold;

[0016] Based on the geological disaster risk probability map, dynamically optimizing the flight speed, altitude, and flight path of the UAV through a Bayesian adaptive flight trajectory planning module to output a monitoring path command for real-time adjustment.

[0017] Preferably, performing spatio-temporal alignment and confidence weighting of the three-dimensional point cloud data, the image data, and the flight attitude data through a spatio-temporal weighted covariance fusion model to obtain fused deformation feature data, including:

[0018] Obtaining historical error data of each of the lidar, the visible light camera, and the inertial measurement unit, and calculating the information entropy value of the historical error data;

[0019] Dynamically allocating weights according to the information entropy value to obtain dynamic weights;

[0020] Based on the timestamp of the three-dimensional point cloud data, performing millisecond-level synchronization of the image data through cubic spline interpolation to obtain a spatio-temporally aligned multimodal data set;

[0021] Perform Mahalanobis distance normalization on the multi-modal data set using the covariance matrix to eliminate the dimensional difference and obtain the standardized sensor data;

[0022] Weightedly fuse the standardized sensor data according to the dynamic weight and superimpose the time drift compensation term to obtain the deformation feature data.

[0023] Preferably, the calculation formula for the deformation feature data is:

[0024]

[0025] where n is the number of sensors, is the deformation feature data at time t, ω i is the dynamic weight, C i is the covariance matrix representing the data of the i-th sensor, S i,t is the state data obtained from the i-th sensor at time t, μ i is the mean value of the data of the i-th sensor, K t is the time drift compensation term, Δt is the time interval; the calculation formula for the dynamic weight is: where H i is the information entropy value of the data of the i-th sensor, H i =-∑p(e i )logp(e i ), p(e i ) is the historical error data of the i-th sensor.

[0026] Preferably, input the fused deformation feature data into the wavelet-Volterra series coupling prediction algorithm to perform non-linear deformation prediction on the key parts of the target bridge and tunnel, and output the displacement prediction value for the future time window, including:

[0027] Obtain the displacement sequence D in the fused deformation feature data t , and perform wavelet packet decomposition through the Morlet wavelet basis function to obtain the high-frequency noise component and the low-frequency trend component;

[0028] Construct a second-order Volterra prediction model for the low-frequency trend component;

[0029] When the prediction residual , trigger the Kalman filter-particle filter hybrid correction algorithm and output the corrected displacement prediction value; where σ 残差 is the standard deviation of the displacement prediction residual.

[0030] Preferably, the expression of the Morlet wavelet basis function is:

[0031]

[0032] Among them, ψ(t) is the Morlet wavelet basis function, ω 0 is the wavelet center frequency, and π -1 / 4 is the normalization coefficient.

[0033] Preferably, the expression of the second-order Volterra prediction model is:

[0034]

[0035] Among them, α k is the linear weight coefficient of the wavelet component, representing the linear contribution of the low-frequency trend component, and β m is the quadratic term weight coefficient of the Volterra kernel, representing the contribution of the nonlinear dynamic effect. is the future displacement value predicted by the second-order Volterra prediction model, k is the expansion order representing the linear wavelet component, m is the expansion order representing the nonlinear Volterra kernel, h(τ) is the Volterra kernel function, which is identified online by the recursive least squares method to obtain the nonlinear deformation prediction value, D t-τ is the historical displacement, and τ is the time delay variable represented in the Volterra prediction model, that is, the delay step between the current time t and the historical time point.

[0036] Preferably, based on the geological disaster risk probability map, the flight speed, altitude, and flight path of the UAV are dynamically optimized through the Bayesian adaptive flight trajectory planning module to output a monitoring path command adjusted in real time, including:

[0037] Divide the monitoring area of the geological disaster risk probability map into grids of 0.5m×0.5m, and calculate the real-time risk value of each grid; the calculation formula of the real-time risk value is: Among them, is the real-time strain rate, representing the instantaneous deformation rate of the geological body or structure, and P hist (x,y) represents the Bayesian posterior probability based on historical data, x represents the abscissa value of the coordinates in the monitoring area, and y represents the ordinate value of the coordinates in the monitoring area;

[0038] Generate a candidate waypoint set {P i (x i ,y i ,h i )} according to the real-time risk value, and obtain the UAV aerodynamic parameters and the camera frame rate f cam ; among them, the UAV aerodynamic parameters include the first parameter k 1and the second parameter k 2 ; k 1 = 0.05, k 2 = 1.2;

[0039] Construct the objective function;

[0040] Solve the optimal solution of the objective function through the Monte Carlo tree search algorithm, and output the flight speed v, altitude h and waypoint sequence;

[0041] When the real-time risk value Risk(x, y)>0.8, forcibly switch the flight mode of the corresponding area to spiral scanning, and output the spiral radius r = 1.2h and the angular velocity ω = 0.3v / r.

[0042] Preferably, the expression of the objective function is:

[0043]

[0044] A bridge and tunnel geological disaster monitoring system based on an unmanned aerial vehicle, comprising:

[0045] A data acquisition module for acquiring data collected by sensors carried by the unmanned aerial vehicle; the sensors include a lidar, a visible light camera and an inertial measurement unit; the lidar is used to collect three-dimensional point cloud data of the surfaces of the target bridge and tunnel; the visible light camera is used to capture image data of the surfaces of the target bridge and tunnel; the inertial measurement unit is used to record the flight attitude data of the unmanned aerial vehicle; the unmanned aerial vehicle performs monitoring tasks according to a preset monitoring path;

[0046] A feature fusion module for performing spatio-temporal alignment and confidence weighting of the three-dimensional point cloud data, the image data and the flight attitude data through a spatio-temporal weighted covariance fusion model to obtain fused deformation feature data;

[0047] A deformation prediction module for inputting the fused deformation feature data into a wavelet-Volterra series coupling prediction algorithm to perform non-linear deformation prediction on key parts of the target bridge and tunnel, and outputting displacement prediction values for a future time window;

[0048] A probability map generation module for generating a geological disaster risk probability map according to the displacement prediction value and a preset risk threshold;

[0049] An optimization module for dynamically optimizing the flight speed, altitude and flight path of the unmanned aerial vehicle based on the geological disaster risk probability map through a Bayesian adaptive flight trajectory planning module to output a monitoring path command for real-time adjustment.

[0050] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0051] The present invention provides a method and system for monitoring geological disasters of bridges and tunnels based on unmanned aerial vehicles. The method includes: acquiring data collected by sensors carried by the unmanned aerial vehicle; the sensors include a lidar, a visible light camera, and an inertial measurement unit; the lidar is used to collect three-dimensional point cloud data of the surfaces of the target bridge and tunnel; the visible light camera is used to capture image data of the surfaces of the target bridge and tunnel; the inertial measurement unit is used to record the flight attitude data of the unmanned aerial vehicle; the unmanned aerial vehicle performs a monitoring task according to a preset monitoring path; performing spatio-temporal alignment and confidence weighting of the three-dimensional point cloud data, the image data, and the flight attitude data through a spatio-temporal weighted covariance fusion model to obtain fused deformation feature data; inputting the fused deformation feature data into a wavelet-Volterra series coupling prediction algorithm to perform non-linear deformation prediction on key parts of the target bridge and tunnel, and outputting displacement prediction values for a future time window; generating a geological disaster risk probability map according to the displacement prediction values and a preset risk threshold; dynamically optimizing the flight speed, height, and flight path of the unmanned aerial vehicle based on the geological disaster risk probability map through a Bayesian adaptive flight trajectory planning module to output a monitoring path command for real-time adjustment. The present invention significantly improves the accuracy of geological disaster monitoring and the endurance of the unmanned aerial vehicle through multi-modal fusion, non-linear prediction, and dynamic path planning, and is applicable to the safety protection of bridges and tunnels in complex terrains. Description of the Drawings

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0053] Figure 1 It is a flowchart of the method provided by the embodiment of the present invention;

[0054] Figure 2 It is a schematic structural diagram of the system provided by the embodiment of the present invention. Detailed Embodiments

[0055] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0056] To make the above objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0057] Figure 1 The flowchart of the method provided by the embodiment of the present invention is as Figure 1 shown. The present invention provides a method for monitoring geological disasters of bridges and tunnels based on unmanned aerial vehicles (UAVs), including:

[0058] Step 100: Obtain the data collected by the sensors carried by the UAV; the sensors include lidar, visible light cameras, and inertial measurement units; the lidar is used to collect three-dimensional point cloud data of the surfaces of the target bridges and tunnels; the visible light cameras are used to capture image data of the surfaces of the target bridges and tunnels; the inertial measurement units are used to record the flight attitude data of the UAV; the UAV performs monitoring tasks according to a preset monitoring path;

[0059] Step 200: Perform spatio-temporal alignment and confidence weighting of the three-dimensional point cloud data, image data, and flight attitude data through a spatio-temporal weighted covariance fusion model to obtain fused deformation feature data;

[0060] Step 300: Input the fused deformation feature data into a wavelet-Volterra series coupled prediction algorithm to perform non-linear deformation prediction on the key parts of the target bridges and tunnels, and output displacement prediction values for a future time window;

[0061] Step 400: Generate a geological disaster risk probability map based on the displacement prediction values and a preset risk threshold;

[0062] Step 500: Dynamically optimize the flight speed, altitude, and flight path of the UAV based on the geological disaster risk probability map through a Bayesian adaptive flight trajectory planning module to output a monitoring path command for real-time adjustment.

[0063] Specifically, step 100 of this embodiment includes:

[0064] Step 101: Before performing the monitoring task, this embodiment first needs to formulate a reasonable monitoring path according to the specific location and structural characteristics of the target bridges and tunnels. This embodiment uses geographic information system (GIS) software for analysis to determine the optimal flight route and altitude to ensure that the UAV can comprehensively cover the target area. The monitoring path should take into account factors such as obstacles, flight altitude, speed, and the field of view of the sensors to ensure that the required data can be effectively collected during the flight. In addition, the path planning also needs to consider weather conditions and flight safety to ensure the stability and safety of the UAV during the mission.

[0065] Step 102: When the drone flies along the preset monitoring path, the sensors carried in this embodiment will start data acquisition. The Light Detection and Ranging (LiDAR) emits laser beams and receives the reflected signals to generate three-dimensional point cloud data of the surfaces of the target bridge and tunnel. These point cloud data can accurately reflect the geometric shape and surface features of the structure. Meanwhile, the visible light camera will capture high-resolution image data, capturing the surface details of the bridge and tunnel, such as cracks, corrosion, and other potential damages. The Inertial Measurement Unit (IMU) records the flight attitude data of the drone in real time, including acceleration, angular velocity, and heading information. These data are crucial for the spatio-temporal alignment of the subsequent point cloud data and image data.

[0066] Preferably, through a spatio-temporal weighted covariance fusion model, spatio-temporal alignment and confidence weighting of the three-dimensional point cloud data, the image data, and the flight attitude data are performed to obtain fused deformation feature data, including:

[0067] Obtain the historical error data of each of the LiDAR, the visible light camera, and the Inertial Measurement Unit, and calculate the information entropy value of the historical error data;

[0068] Dynamically allocate weights according to the information entropy value to obtain dynamic weights;

[0069] Based on the timestamp of the three-dimensional point cloud data, perform millisecond-level synchronization on the image data through cubic spline interpolation to obtain a spatio-temporally aligned multi-modal data set;

[0070] Use the covariance matrix to perform Mahalanobis distance normalization on the multi-modal data set to eliminate the dimension difference and obtain standardized sensor data;

[0071] Weight and fuse the standardized sensor data according to the dynamic weights, and superimpose the time drift compensation term to obtain the deformation feature data.

[0072] Preferably, the calculation formula of the deformation feature data is:

[0073]

[0074] where n is the number of sensors, is the deformation feature data at time t, ω i is the dynamic weight, C i is the covariance matrix representing the data of the i-th sensor, S i,t is the state data obtained from the i-th sensor at time t, μ i is the mean of the data of the i-th sensor, K t is the time drift compensation term, Δt is the time interval; the calculation formula of the dynamic weight is: where Hi is the information entropy value of the i-th sensor data, H i = -∑p(e i ) log p(e i ), where p(e i ) is the historical error data of the i-th sensor.

[0075] Specifically, in order to achieve the spatio-temporal alignment and confidence weighting of 3D point cloud data, image data, and flight attitude data based on the spatio-temporal weighted covariance fusion model, this embodiment first obtains the historical error data of each sensor and calculates its information entropy value. These historical error data reflect the accuracy fluctuations of the sensors during long-term use. By calculating the information entropy value, the reliability and uncertainty of each sensor data can be quantitatively evaluated. The higher the information entropy value, the greater the uncertainty of the sensor data, and vice versa. Based on these information entropy values, weights are dynamically assigned to ensure that more attention is paid to the sensor data with better historical performance during the data fusion process, thereby improving the credibility of the fusion result.

[0076] Optionally, this embodiment uses the timestamp of the 3D point cloud data to synchronize the image data at the millisecond level through cubic spline interpolation. This process ensures the temporal consistency of different sensors, enabling the 3D point cloud data and image data obtained at the same moment to accurately correspond. Through this high-precision spatio-temporal alignment, the instantaneous state of the structure can be effectively captured, providing a reliable data basis for subsequent analysis. This step of this embodiment is an important link in realizing multi-modal data sets, ensuring the compatibility and comparability between different types of data.

[0077] Furthermore, after completing the data synchronization, this embodiment uses the covariance matrix to perform Mahalanobis distance normalization on the spatio-temporally aligned multi-modal data set to eliminate the differences between different dimensions. The normalization process enables the data of different sensors to be compared under the same standard, avoiding data fusion errors caused by different dimensions. The standardized sensor data will be more consistent, laying a foundation for subsequent weighted fusion.

[0078] Even further, this embodiment weights and fuses the standardized sensor data according to dynamic weights and superimposes a time drift compensation term to obtain the final deformation feature data. In the specific calculation formula, the dynamic weights are adjusted according to the information entropy values of the sensors, and the covariance matrix is used to quantify the uncertainty of the data, ensuring the accuracy and reliability of the final result. Through this series of processes, the deformation feature data obtained in this embodiment can not only accurately reflect the state changes of bridges and tunnels during monitoring, but also provide strong data support for subsequent structural health monitoring and risk assessment, significantly improving the accuracy and efficiency of monitoring.

[0079] Preferably, input the fused deformation feature data into the wavelet-Volterra series coupling prediction algorithm to perform non-linear deformation prediction on the key parts of the target bridge and tunnel, and output the displacement prediction values in the future time window, including:

[0080] Obtain the displacement sequence D in the fused deformation feature data t , perform wavelet packet decomposition through the Morlet wavelet basis function to obtain the high-frequency noise component and the low-frequency trend component;

[0081] Construct a second-order Volterra prediction model for the low-frequency trend component;

[0082] When the prediction residual , trigger the Kalman filter-particle filter hybrid correction algorithm and output the corrected displacement prediction value; where σ 残差 is the standard deviation of the displacement prediction residual.

[0083] Preferably, the expression of the Morlet wavelet basis function is:

[0084]

[0085] where ψ(t) is the Morlet wavelet basis function, ω 0 is the wavelet center frequency, π -1 / 4 is the normalization coefficient.

[0086] Preferably, the expression of the second-order Volterra prediction model is:

[0087]

[0088] where α k is the linear weight coefficient of the wavelet component, representing the linear contribution of the low-frequency trend component, β m is the quadratic term weight coefficient of the Volterra kernel, representing the contribution of the non-linear dynamic effect, is the future displacement value predicted by the second-order Volterra prediction model, k is the expansion order representing the linear wavelet component, m is the expansion order representing the non-linear Volterra kernel, h(τ) is the Volterra kernel function, which is identified online by the recursive least squares method to obtain the non-linear deformation prediction value, D t-τ is the historical displacement, τ is the time delay variable represented in the Volterra prediction model, that is, the delay step between the current time t and the historical time point.

[0089] Optionally, in order to predict the non-linear deformation of key parts of the target bridge and tunnel, in this embodiment, the fused deformation feature data is first input into the wavelet-Volterra series coupling prediction algorithm. The first step of this algorithm is to extract the displacement sequence from the fused deformation feature data and perform wavelet packet decomposition using the Morlet wavelet basis function. This process decomposes the displacement sequence into high-frequency noise components and low-frequency trend components, enabling us to more clearly identify the important trends and noises in the data. The expression of the Morlet wavelet basis function provides the time-frequency analysis ability of the signal and can effectively capture the instantaneous changes in the signal.

[0090] Specifically, in this embodiment, a second-order Volterra prediction model is constructed for the low-frequency trend component. This model can handle the non-linear features in the signal and describe the dynamic behavior of the system through linear and non-linear terms. Specifically, the expression of the second-order Volterra model contains the linear weight coefficient and quadratic term weight coefficient of the wavelet component, which respectively characterize the linear contribution of the low-frequency trend component and the contribution of the non-linear dynamic effect. This modeling method enables us to make full use of the low-frequency trend information to predict future displacement changes.

[0091] Optionally, when making predictions, this embodiment needs to calculate the prediction residuals, which involves analyzing the difference between the model output and the actual observed values. The standard deviation of the prediction residuals provides an evaluation of the model prediction accuracy and helps us judge the reliability of the prediction results. By using the recursive least squares method to online identify the Volterra kernel function, the model parameters can be further optimized to make the prediction results more accurate. Historical displacement data plays a key role in the model. As the input historical information, it can effectively enhance the model's prediction ability for future displacements.

[0092] Furthermore, using the constructed second-order Volterra prediction model, this embodiment can output the displacement prediction values for the future time window. These prediction values will provide an important basis for the structural health monitoring of bridges and tunnels, helping relevant personnel to timely identify potential risks and abnormal changes, and thus take corresponding maintenance and reinforcement measures. Through this non-linear deformation prediction method, not only can the response speed to the state changes of bridges and tunnels be improved, but also the intelligent level of the monitoring system can be effectively enhanced, ensuring the safety and stability of the infrastructure.

[0093] Exemplarily, the non-linear dynamic effect in this embodiment refers to the fact that in the system response, the relationship between the input and the output is no longer a simple linear relationship, but exhibits complex and non-linear characteristics. This effect usually occurs when a structure or material is subjected to external loads, environmental changes or other influences, resulting in its behavior being unable to be accurately described by a linear model. For example, in the monitoring of bridges and tunnels, as the load increases or the material ages, the structure may exhibit non-linear deformations, changes in vibration modes or fatigue damage and other phenomena. These non-linear characteristics make it impossible for traditional linear prediction models to effectively capture and predict the true dynamic behavior of the structure. Therefore, in this embodiment, a more complex non-linear model (such as the Volterra model) is adopted to accurately describe and predict these dynamic effects.

[0094] As an alternative implementation, in this embodiment, by statistically analyzing historical monitoring data, the displacement change patterns of the structure under different conditions are identified, and risk thresholds are set based on these patterns. The threshold reflects the maximum acceptable displacement of the structure under normal use conditions. Exceeding this threshold may indicate a potential geological disaster risk. Using machine learning or statistical models, historical data can be trained to generate a risk assessment model, so as to automatically calculate the real-time risk value when new data arrives and compare it with the set threshold.

[0095] When generating the geological disaster risk probability map, this embodiment adopts the method of probability density function (PDF), combines the displacement prediction values of each monitoring point with their corresponding risk thresholds, and calculates the risk probability of each monitoring point. This process can map the monitoring results to the entire monitoring area through interpolation methods to form a continuous risk probability map. Through visualization means, the risk probability map is presented to decision-makers to facilitate the rapid identification of high-risk areas and take corresponding monitoring and early warning measures. Considering various factors (such as historical data, environmental conditions, structural characteristics, etc.), the threshold should be determined by a scientific and reasonable method to ensure the sensitivity and accuracy of the monitoring system.

[0096] Preferably, based on the geological disaster risk probability map, the flight speed, altitude and flight path of the UAV are dynamically optimized through the Bayesian adaptive flight trajectory planning module to output real-time adjusted monitoring path instructions, including:

[0097] Divide the monitoring area of the geological disaster risk probability map into grids of 0.5m×0.5m, and calculate the real-time risk value of each grid; the calculation formula of the real-time risk value is: Where, is the real-time strain rate, which characterizes the instantaneous deformation rate of the geological body or structure, P hist(x, y) represents the Bayesian posterior probability based on historical data, where x represents the abscissa value of the coordinates within the monitoring area, and y represents the ordinate value of the coordinates within the monitoring area;

[0098] Generate a candidate waypoint set {P i (x i , y i , h i )} according to the real-time risk value, and obtain the aerodynamic parameters of the UAV and the camera frame rate f cam ; where the aerodynamic parameters of the UAV include the first parameter k 1 and the second parameter k 2 ; k 1 = 0.05, k 2 = 1.2;

[0099] Construct the objective function;

[0100] Solve the optimal solution of the objective function through the Monte Carlo tree search algorithm, and output the flight speed v, altitude h and waypoint sequence;

[0101] When the real-time risk value Risk(x, y) > 0.8, force the flight mode of the corresponding area to be switched to spiral scanning, and output the spiral radius r = 1.2h and the angular velocity ω = 0.3v / r.

[0102] Preferably, the expression of the objective function is:

[0103]

[0104] Optionally, in this embodiment, the monitoring area is divided into multiple grids, and the real-time risk value of each grid is calculated. The calculation formula of the real-time risk value combines the real-time strain rate and the Bayesian posterior probability based on historical data, and can reflect the instantaneous deformation rate of the geological body or structure at a specific location. This embodiment ensures a comprehensive assessment of the potential risks within the monitoring area, enabling the UAV to conduct more intensive monitoring in high-risk areas. Then, this embodiment generates a candidate waypoint set according to the calculated real-time risk value, and obtains the aerodynamic parameters (such as flight speed and altitude) of the UAV and the camera frame rate. These aerodynamic parameters are crucial for the flight performance and monitoring efficiency of the UAV. By constructing the objective function, this function will comprehensively consider the real-time risk value, the aerodynamic characteristics of the UAV, and the requirements of the monitoring task to optimize the flight path and monitoring efficiency. The design of the objective function aims to maximize the monitoring effect while minimizing the flight time and energy consumption.

[0105] Furthermore, in this embodiment, the Monte Carlo tree search algorithm is used to solve the optimal solution of the objective function, and the flight speed, altitude, and waypoint sequence that are adjusted in real time are output. When the real-time risk value exceeds the set threshold, the system will forcibly switch the flight mode of the UAV to spiral scanning to ensure full coverage of high-risk areas. In this mode, the system will output the spiral radius and angular velocity to optimize the monitoring effect. This embodiment not only improves the adaptability of the UAV in complex environments but also enhances the real-time response ability to geological disaster risks, ensuring the effectiveness and safety of the monitoring task.

[0106] Optionally, this embodiment further includes an exception handling process:

[0107] (1) Obtain the pitch angle θ in the data of the real-time inertial measurement unit, and calculate the coordinate offset Δx of the lidar sensor = L·(sinθ y , where L is the length of the mounting arm and θ y -sinθ 0 )(θ 0 is the calibration angle);

[0108] (2) Perform coordinate transformation on the lidar data according to the offset Δx to obtain the compensated three-dimensional structure data;

[0109] (3) Perform affine transformation matching on the compensated three-dimensional structure data and the visual image, and output a fusion monitoring result with an X-Y plane projection error ≤ 3 pixels.

[0110] Corresponding to the above method, this embodiment also provides a bridge and tunnel geological disaster monitoring system based on a UAV, including:

[0111] A data acquisition module for acquiring data collected by sensors carried by the UAV; the sensors include a lidar, a visible light camera, and an inertial measurement unit; the lidar is used to collect three-dimensional point cloud data on the surfaces of the target bridge and tunnel; the visible light camera is used to capture image data on the surfaces of the target bridge and tunnel; the inertial measurement unit is used to record the flight attitude data of the UAV; the UAV performs a monitoring task according to a preset monitoring path;

[0112] A feature fusion module for performing spatio-temporal alignment and confidence weighting of the three-dimensional point cloud data, the image data, and the flight attitude data through a spatio-temporal weighted covariance fusion model to obtain fused deformation feature data;

[0113] A deformation prediction module for inputting the fused deformation feature data into a wavelet-Volterra series coupling prediction algorithm to perform non-linear deformation prediction on key parts of the target bridge and tunnel, and outputting displacement prediction values for a future time window;

[0114] A probability map generation module, configured to generate a geological disaster risk probability map based on the displacement prediction value and a preset risk threshold;

[0115] An optimization module, configured to dynamically optimize the flight speed, altitude, and flight path of the UAV based on the geological disaster risk probability map through a Bayesian adaptive flight path planning module, so as to output a monitoring path instruction with real-time adjustment.

[0116] The beneficial effects of the present invention are as follows:

[0117] (1) Through the multi-modal data fusion technology, the present invention effectively integrates the data of lidar, visible light camera, and inertial measurement unit, and can significantly improve the monitoring accuracy of structures such as bridges and tunnels. The fused deformation feature data can more comprehensively reflect the true state of the structure, reduce the errors and uncertainties that may be brought by a single sensor, and thus improve the reliability of the monitoring results.

[0118] (2) Based on the dynamic optimization of the flight path planning of the geological disaster risk probability map, the present invention enables the UAV to adjust the flight speed, altitude, and flight path according to the real-time risk value. This real-time response ability ensures that the UAV can conduct more intensive monitoring in high-risk areas, timely capture potential structural changes and risks, and enhances the flexibility and adaptability of the monitoring system.

[0119] (3) The present invention adopts a wavelet-Volterra series coupling prediction algorithm to predict the non-linear deformation of key parts, and can effectively identify and predict the dynamic behavior of the structure under different loads and environmental conditions. This prediction ability provides forward-looking information for structural health monitoring, helps relevant personnel take maintenance and reinforcement measures in a timely manner, and reduces potential risks.

[0120] (4) By optimizing the flight path and monitoring parameters of the UAV through the Monte Carlo tree search algorithm, the present invention can minimize the flight time and energy consumption while ensuring the monitoring effect. This optimization not only improves the monitoring efficiency, but also reduces the operation cost, making the resource allocation more reasonable.

[0121] (5) The comprehensive monitoring results of the present invention provide a scientific basis for structural health assessment and risk management, help decision-makers timely identify potential safety hazards, and formulate corresponding countermeasures. This data-driven decision support system can effectively improve the safety of infrastructure and ensure public safety.

[0122] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The same or similar parts among the various embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0123] In this article, specific examples are used to illustrate the principles and implementation modes of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation modes and application scopes. To sum up, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A method for monitoring geological disasters in bridges and tunnels based on drones, characterized in that: include: Obtain data collected by sensors carried by drones; The sensor includes a laser radar, a visible light camera and an inertial measurement unit; the laser radar is used to collect three-dimensional point cloud data of the target bridge and tunnel surface; the visible light camera is used to capture image data of the target bridge and tunnel surface; the inertial measurement unit is used to record the flight attitude data of the drone; the drone performs monitoring tasks according to a preset monitoring path; Performing spatiotemporal alignment and confidence weighting of the three-dimensional point cloud data, the image data, and the flight attitude data through a spatiotemporal weighted covariance fusion model to obtain fused deformation feature data; Inputting the fused deformation feature data into the wavelet-Volterra series coupling prediction algorithm to perform nonlinear deformation prediction on key parts of target bridges and tunnels, and outputting displacement prediction values ​​in future time windows; Generate a geological disaster risk probability map based on the displacement prediction value and a preset risk threshold; Based on the geological disaster risk probability map, the flight speed, altitude and track of the UAV are dynamically optimized through a Bayesian adaptive flight trajectory planning module to output real-time adjusted monitoring path instructions.

2. The method for monitoring geological disasters of bridges and tunnels based on drones according to claim 1 is characterized in that: The spatiotemporal alignment and confidence weighting of the three-dimensional point cloud data, the image data and the flight attitude data are performed through a spatiotemporal weighted covariance fusion model to obtain fused deformation feature data, including: Obtaining historical error data of each of the laser radar, the visible light camera, and the inertial measurement unit, and calculating an information entropy value of the historical error data; Dynamically assign weights according to the information entropy value to obtain dynamic weights; Based on the timestamp of the three-dimensional point cloud data, the image data is synchronized at millisecond level by cubic spline interpolation to obtain a multimodal data set aligned in time and space; Using a covariance matrix to perform Mahalanobis distance normalization on the multimodal data set to eliminate dimensional differences and obtain standardized sensor data; The standardized sensor data is weightedly fused according to the dynamic weights, and a time drift compensation term is superimposed to obtain the deformation feature data.

3. The method for monitoring geological disasters of bridges and tunnels based on drones according to claim 2 is characterized in that: The calculation formula of the deformation feature data is: Where n is the number of sensors, is the deformation feature data at time t, ω i is the dynamic weight, C i is the covariance matrix representing the i-th sensor data, S i,t is the state data obtained from the i-th sensor at time t, μ i is the mean value of the i-th sensor data, K t is the time drift compensation term, Δt is the time interval; the calculation formula of dynamic weight is: Among them, H i is the information entropy value of the i-th sensor data, H i =-∑p(e i )logp(e i ), p(e i ) is the historical error data of the ith sensor.

4. The method for monitoring geological disasters of bridges and tunnels based on drones according to claim 1, characterized in that: The fused deformation feature data is input into the wavelet-Volterra series coupling prediction algorithm to perform nonlinear deformation prediction on key parts of target bridges and tunnels, and output displacement prediction values ​​in future time windows, including: Get the displacement sequence D in the fused deformation feature data t , wavelet packet decomposition is performed through Morlet wavelet basis function to obtain high-frequency noise components and low-frequency trend components; Constructing a second-order Volterra prediction model for the low-frequency trend component; When the prediction residual When , the Kalman filter-particle filter hybrid correction algorithm is triggered to output the corrected displacement prediction value; where σ 残差 is the standard deviation of the displacement prediction residuals.

5. The method for monitoring geological disasters of bridges and tunnels based on unmanned aerial vehicles according to claim 4 is characterized in that: The expression of the Morlet wavelet basis function is: Among them, ψ(t) is the Morlet wavelet basis function, ω0 is the wavelet center frequency, π -1 / 4 is the normalization coefficient.

6. The method for monitoring geological disasters of bridges and tunnels based on drones according to claim 4 is characterized in that: The expression of the second-order Volterra prediction model is: Among them, α k is the linear weight coefficient of the wavelet component, representing the linear contribution of the low-frequency trend component, β m is the weight coefficient of the quadratic term of the Volterra kernel, which represents the contribution of nonlinear dynamic effects, is the future displacement value predicted by the second-order Volterra prediction model, k is the expansion order of the linear wavelet component, m is the expansion order of the nonlinear Volterra kernel, h(τ) is the Volterra kernel function, and the nonlinear deformation prediction value is obtained by online identification using the recursive least squares method. t-τ is the historical displacement, τ is the time lag variable represented in the Volterra prediction model, that is, the delay step between the current time t and the historical time point.

7. The method for monitoring geological disasters of bridges and tunnels based on drones according to claim 1, characterized in that: Based on the geological disaster risk probability map, the flight speed, altitude and track of the UAV are dynamically optimized through the Bayesian adaptive flight trajectory planning module to output real-time adjusted monitoring path instructions, including: The monitoring area of ​​the geological disaster risk probability map is divided into 0.5m×0.5m grids, and the real-time risk value of each grid is calculated; the calculation formula of the real-time risk value is: in, is the real-time strain rate, which represents the instantaneous deformation rate of the geological body or structure, P hist (x, y) represents the Bayesian posterior probability based on historical data, x represents the horizontal coordinate value of the coordinate in the monitoring area, and y represents the vertical coordinate value of the coordinate in the monitoring area; Generate a candidate waypoint set according to the real-time risk value i (x i ,y i ,h i )}, and obtain the drone aerodynamic parameters and camera frame rate f cam ; The aerodynamic parameters of the drone include a first parameter k1 and a second parameter k2; k1 = 0.05, k2 = 1.2; Construct the objective function; The objective function is solved for the optimal solution by using a Monte Carlo tree search algorithm, and the flight speed v, altitude h and waypoint sequence are output; When the real-time risk value Risk(x,y)>0.8, the flight mode of the corresponding area is forced to switch to spiral scanning, and the output spiral radius r=1.2f and angular velocity ω=0.3v / r.

8. The method for monitoring geological disasters of bridges and tunnels based on unmanned aerial vehicles according to claim 7, characterized in that: The expression of the objective function is:

9. A bridge and tunnel geological disaster monitoring system based on drones, characterized in that: include: A data acquisition module is used to obtain data collected by sensors carried by the drone; The sensor includes a laser radar, a visible light camera and an inertial measurement unit; the laser radar is used to collect three-dimensional point cloud data of the target bridge and tunnel surface; the visible light camera is used to capture image data of the target bridge and tunnel surface; the inertial measurement unit is used to record the flight attitude data of the drone; the drone performs monitoring tasks according to a preset monitoring path; A feature fusion module, used for performing spatiotemporal alignment and confidence weighting of the three-dimensional point cloud data, the image data and the flight attitude data through a spatiotemporal weighted covariance fusion model to obtain fused deformation feature data; A deformation prediction module is used to input the fused deformation feature data into a wavelet-Volterra series coupling prediction algorithm to perform nonlinear deformation prediction on key parts of target bridges and tunnels and output displacement prediction values ​​in future time windows; A probability map generation module, used to generate a geological disaster risk probability map according to the displacement prediction value and a preset risk threshold; The optimization module is used to dynamically optimize the flight speed, altitude and track of the UAV based on the geological disaster risk probability map through the Bayesian adaptive flight trajectory planning module to output real-time adjusted monitoring path instructions.

Images