A method, equipment and medium for active control of overturning prevention of single-column pier bridges

Abstract

The application discloses a single-column pier bridge anti-overturning active control method, equipment and medium, method steps: deploying lightning visual displacement monitoring system and hydraulic drive system; obtaining lightning visual multi-modal fusion data, and calculating beam transverse overturning angle; constructing a digital twin for bridge structure overturning state prediction, and predicting bridge overturning state at multiple future time points; constructing an intelligent agent, and controlling the hydraulic drive system to perform anti-overturning action based on the predicted bridge overturning state. The equipment and medium are used to implement the above method. The application has the beneficial effects that: for the traditional single structure displacement detection method, the application proposes a structure displacement monitoring method based on lightning visual heterogeneous fusion, realizes interference factor identification and weighted fusion in different environments, and ensures the reliability of the monitored structure displacement data.

Description

Technical Field

[0001] This application relates to the field of bridge safety monitoring technology, and in particular to an active control method, equipment and medium for preventing overturning of a single-column pier bridge. Background Technology

[0002] Currently, the most common modification measures for preventing overturning of single-column pier bridges are widening the cap beam at the pier top or adding steel brackets with clamps. Essentially, this changes single-point support to multi-point support, relying on physical barriers to limit the overturning of the beam. However, a gap must be maintained between the pre-installed lateral restraint blocks and the beam body to allow for thermal expansion and contraction. When overturning occurs, the beam accelerates and impacts the restraint blocks, generating a huge impact force that can easily lead to localized crushing of the concrete or yielding of the steel bracket material. Furthermore, the increasing number of overloaded vehicles and repeated collisions over a long period can cause the device to fail. The aforementioned methods for preventing overturning of single-column pier bridges still fall under the category of passive overturning prevention. However, with technological advancements, structural response prediction methods based on bridge displacement may be used as an active overturning prevention measure for single-column pier bridges.

[0003] However, existing non-contact structural displacement measurement technologies mainly rely on visual DIC and radar subsystems. Visual DIC calculates structural displacement by tracking pixel changes and physical transformations of a target pattern in the camera frame, and has been widely used in bridge deflection monitoring, but it is susceptible to ambient light. In recent years, long-range DIC based on UAVs or fixed cameras has become a hot topic; radar subsystem monitoring technology mainly uses the phase difference of electromagnetic waves to measure minute displacements, but it is quite sensitive to electromagnetic interference.

[0004] Traditional bridge structural response prediction relies on the finite element method (FEM), which solves the structural physical equations to obtain the target structural response, such as displacement, velocity, and acceleration. In recent years, deep learning time-series prediction models, such as LSTM and Transformer, have emerged in the field of structural health monitoring, replacing the FEM model for structural response analysis. These models train neural networks using limited historical simulation data to fit the input-output relationship; however, high-precision FEM analysis is extremely slow, leading to data-driven predictions that often violate fundamental mechanical principles and suffer from poor engineering reliability. Summary of the Invention

[0005] One objective of this application is to provide an active control method for preventing overturning of single-column pier bridges that can solve at least one of the defects in the aforementioned background technology.

[0006] Another objective of this application is to provide an electronic device capable of implementing an active control method for preventing overturning of single-column pier bridges that addresses at least one of the deficiencies in the aforementioned background technology.

[0007] Another object of this application is to provide a computer-readable storage medium capable of implementing an active control method for preventing overturning of a single-column pier bridge that addresses at least one of the deficiencies in the aforementioned background art.

[0008] To achieve at least one of the above objectives, the technical solution adopted in this application is: an active control method for preventing overturning of a single-column pier bridge, comprising the following steps:

[0009] S100: Deploy a radar-based displacement monitoring system to monitor the displacement on both sides of the bottom of the main beam at the support of the single-column pier bridge, and install a hydraulic drive system at the support position of the single-column pier bridge to provide anti-overturning moment.

[0010] S200: Based on the deployed radar-based displacement monitoring system, acquire radar-based multimodal fusion data with adaptive modal weights based on environmental changes, and calculate the lateral overturning angle of the beam.

[0011] S300: Construct a digital twin and introduce physical constraints including moment balance and non-negative support reactions. By combining the acquired beam lateral overturning angle data with structural displacement data, predict the bridge overturning state at multiple future moments.

[0012] S400: Constructs an intelligent agent to take over control of the hydraulic drive system and controls the hydraulic drive system to perform anti-overturning actions based on structural response closed loop and force control servo closed loop based on predicted bridge overturning state.

[0013] Preferably, the radar-visual displacement monitoring system includes a radar subsystem and a vision subsystem, both of which monitor the displacement on both sides of the single-column pier bridge. Deviation calculations are performed on the structural displacement data monitored by the radar and vision subsystems respectively. If the deviation of a sampling point exceeds a set threshold, the data for that sampling point is removed. When weighted and fused to obtain radar-visual multimodal fusion data, if there is poor lighting, rain or fog, or low image contrast, the weight of the structural displacement data monitored by the vision subsystem is reduced. If there is strong electromagnetic interference or severe multipath effect in the environment, the weight of the structural displacement data monitored by the radar subsystem is reduced.

[0014] Preferably, when constructing radar-visual multimodal fusion data, the visual confidence score and the reciprocal of the radar confidence score calculated based on environmental perception are mapped to the diagonal elements R in the Kalman filter observation noise covariance matrix. vis and R rad The Kalman filter algorithm is based on the real-time calculated element R. vis With R radThe Kalman gain K is automatically updated. Real-time histogram analysis and sharpness evaluation are performed on the images acquired by the vision subsystem. If the overall grayscale mean of the image is lower than the first threshold representing darkness or higher than the second threshold representing direct strong light, or if the Laplacian gradient of the image is lower than the third threshold, it is determined to be a visual disturbance condition, thereby reducing the visual confidence. If the multipath clutter density of unstructured objects is lower than the set fourth threshold, or if the echo intensity variance of the target corner reflector exceeds the fifth threshold, it is determined to be a radar disturbance condition, thereby reducing the radar confidence.

[0015] Preferably, for the visual subsystem, an image quality factor Q is constructed to characterize visual confidence by calculating the Laplacian gradient variance and brightness deviation of the image. ing For the radar subsystem, the signal-to-noise ratio (SNR) of the target echo is extracted in real time to characterize the radar confidence level; when constructing the Kalman filter observation noise covariance matrix, the element R... vis and R rad The expression is as follows:

[0016] ;

[0017] ;

[0018] Among them, R base-vis and R base-rad Let SNR represent the visual fundamental noise covariance and the radar fundamental noise covariance, respectively. α represents the noise amplification factor, β represents the attenuation rate factor, and SNR represents the noise level coefficient. th denoted by , where represents the radar signal-to-noise ratio threshold, and k represents the power-law exponent.

[0019] Preferably, step S300 includes the following processes: constructing a finite element model of the single-column pier bridge, simulating displacement-support reaction force data pairs generated under various load conditions, and using these data pairs to train the displacement-response force mapping relationship of the digital twin based on the Mamba-2 architecture; the loss function of the digital twin includes a data-driven loss, a mechanical equilibrium constraint loss based on the lateral overturning angle of the beam, and a boundary constraint loss based on the non-negativity of the support reaction force; during the training process of the digital twin, the weight coefficients of the data-driven loss, mechanical equilibrium constraint loss, and boundary constraint loss are adaptively adjusted according to the gradient of the network parameters of the digital twin or the loss descent rate of each loss; after the digital twin completes training, the support reaction force representing the overturning state is output according to the real-time working conditions of the single-column pier bridge. If the output support reaction force tends to 0, it indicates that the support is at critical disengagement and active anti-overturning intervention is required through the hydraulic drive system.

[0020] Preferably, the confidence verification of the critical disengagement state of the support based on the output of the digital twin includes the following process: based on the displacement data collected by the radar displacement monitoring system, combined with the geometric parameters of the bridge section, the theoretical estimated reaction force R of the support is derived in real time through the rigid body static equilibrium equation. est ; Calculate the support reaction force R output by the digital twin. pred Compared with the theoretically estimated reaction force R est The normalized residual ε; introducing the confidence score function P conf When the digital twin prediction support is in a critical state of vacancy, the confidence level is determined by the normalized residual ε, and the confidence score function P conf The expression is as follows:

[0021] ;

[0022] The following conditions must be met for hydraulic drive systems to be integrated:

[0023] (R pred <R limit )∩(P conf >P threshold );

[0024] Where γ represents the steepness coefficient, ε th R represents the maximum allowable deviation threshold. limit P represents the safety threshold of the support reaction force. threshold This represents the confidence threshold.

[0025] Preferably, in step S400, an agent is constructed based on a deep reinforcement learning algorithm, and a reward function r that comprehensively considers safety, comfort, and energy consumption is introduced into the agent's policy generation. t In a virtual environment constructed using a digital twin, various extreme working conditions are simulated to train the agent to balance the overturning of a single-column pier bridge with minimal force, so that the reward function r t Maximize; reward function r t The expression is as follows:

[0026] ;

[0027] Where ω1, ω2, and ω3 represent weight coefficients, ReLU(·) represents the activation function, and R0 limit R represents the safety threshold of the support reaction force. pred A represents the support reaction force predicted by the digital twin. t This indicates the opening degree of the servo valve in the hydraulic drive system. This indicates the opening speed of the servo valve in the hydraulic drive system.

[0028] Preferably, in step S400, for the structural response closed loop, the control effect is evaluated by comparing the structural displacement monitoring data before and after the hydraulic drive system's action, and then fed back to the intelligent agent to optimize the control strategy for the next moment; for the force control servo closed loop, the actual output force of the hydraulic drive system is fed back in real time by the sensor and compared with the instructions of the intelligent agent. Based on the comparison result, the servo valve of the hydraulic drive system is controlled to make real-time fine-tuning of the opening to ensure execution accuracy.

[0029] An electronic device includes a processor and a memory; the memory is used to store a computer program, and the processor is used to execute the computer program to implement the above-described active control method for preventing overturning of a single-column pier bridge.

[0030] A computer-readable storage medium storing a computer program; when the computer program is executed by a processor, it implements the above-described active control method for preventing overturning of a single-column pier bridge.

[0031] Compared with the prior art, the beneficial effects of this application are as follows:

[0032] (1) In view of the traditional method of detecting structural displacement by a single structure, this application proposes a structural displacement monitoring method based on the heterogeneous fusion of radar vision, which realizes the identification and weighted fusion of interference factors in different environments, and ensures the reliability of the monitored structural displacement data.

[0033] (2) A digital twin based on physical perception was constructed, which can avoid the possibility that traditional data-driven models may predict results that violate the common sense of mechanics.

[0034] (3) An active hydraulic anti-overturning control strategy based on an agent is proposed. Through training with a multi-objective reward function, the agent learns the "soft landing" strategy—intervening at the millisecond level before overturning occurs to stabilize the beam with minimal impact force. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the overall working steps of this application.

[0036] Figure 2 This is a schematic diagram of the structure in this application where the artificial image target and millimeter-wave radar are mounted on a fixed bracket.

[0037] Figure 3 This is a schematic diagram of the layout structure of the radar subsystem in this application.

[0038] Figure 4 This is a schematic diagram of the layout structure of the vision subsystem in this application.

[0039] In the figure: fixed bracket 100, artificial image target 201, millimeter wave radar 301, corner reflector 302, industrial camera 202. Detailed Implementation

[0040] The present application will now be further described in conjunction with specific embodiments. It should be noted that, in the description of this specification, the use of terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicates that the specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms should not be construed as necessarily referring to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. In addition, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0041] In the description of this application, it should be noted that the terms "center", "lateral", "longitudinal", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", etc., which indicate the orientation and positional relationship based on the orientation or positional relationship shown in the accompanying drawings, are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and should not be construed as limiting the specific protection scope of this application.

[0042] It should be noted that the terms "first," "second," etc., in the specification and claims of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0043] In this application, unless otherwise expressly specified and limited, the terms "installation," "connection," "joining," and "fixing," etc., should be interpreted broadly. For example, they can refer to a connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0044] In this application, unless otherwise expressly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature being directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature being directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0045] The terms “comprising” and “having”, and any variations thereof, in the specification and claims of this application are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product, or device.

[0046] One aspect of this application provides an active control method for preventing overturning of a single-column pier bridge, such as... Figure 1 As shown, one preferred embodiment includes the following steps:

[0047] S100: Deploy a radar-based displacement monitoring system to monitor the displacement on both sides of the bottom of the main beam at the support of the single-column pier bridge, and install a hydraulic drive system at the support position of the single-column pier bridge to provide anti-overturning moment.

[0048] Understandably, the radar-visual position monitoring system can simultaneously monitor the displacement of both sides of the main beam bottom at the support of a single-column pier bridge using both radar and visual detection; the combination of radar and visual detection improves the accuracy of lateral displacement monitoring for single-column pier bridges. The hydraulic drive system is used to actively provide resisting torque to achieve proactive anti-overturning control of the single-column pier bridge when it shows a tendency to overturn. Compared to traditional methods using single structural displacement detection, this application proposes a structural displacement monitoring method based on radar-visual heterogeneous fusion, which realizes the identification and weighted fusion of interference factors under different environments, ensuring the reliability of the monitored structural displacement data.

[0049] S200: Based on the deployed radar-visual displacement monitoring system, acquire radar-visual multimodal fusion data with adaptive modal weights based on environmental changes, and calculate the lateral overturning angle of the beam.

[0050] Understandably, radar and visual inspection operate on different principles, resulting in varying degrees of resistance to interference in different environments. Therefore, when environmental changes occur, the data from radar and visual inspection can be adaptively weighted based on the current environment to ensure accurate calculation of the lateral tilt angle of the bridge beam. The lateral tilt angle of the bridge beam can be used to reflect the tilting state of a single-column pier bridge; that is, when the lateral tilt angle exceeds a certain threshold, it can be determined that the single-column pier bridge may be on the verge of overturning.

[0051] S300: Construct a digital twin and introduce physical constraints including moment balance and non-negative support reactions. By combining the acquired beam lateral overturning angle data with structural displacement data, predict the bridge overturning state at multiple future moments.

[0052] Understandably, the traditional basis for judging the overturning state of a bridge is that the bearing reaction force reaches zero. However, it is difficult to insert force sensors under the bearings of existing bridges. In traditional methods for predicting bearing reaction forces, the models are basically based on pure data-driven predictions, which may lead to predictions that violate common sense about forces. In the technical solution of this application, by constructing a digital twin based on physical perception, the AI ​​model is forced to comply with Newtonian mechanics, thus ensuring the engineering credibility of the prediction results.

[0053] S400: Constructs an intelligent agent to take over control of the hydraulic drive system and controls the hydraulic drive system to perform anti-overturning actions based on structural response closed loop and force control servo closed loop based on predicted bridge overturning state.

[0054] Understandably, traditional anti-overturning reinforcement measures are all passive measures for single-column pier bridges, and the reinforcement methods for single-column pier bridges are all rigid. This means that when a single-column pier bridge has an overturning tendency, it may impact the rigid fixing components, causing damage to the fixing components. In the technical solution of this application, through the design of a hydraulic drive system, active control of the anti-overturning action of the single-column pier bridge can be achieved. That is, based on the overturning prediction results of the single-column pier bridge, the hydraulic drive system can be actively controlled to extend, thereby providing the single-column pier bridge with a torque to resist overturning; and based on the dual closed-loop feedback of structural response closed loop and force control servo closed loop, the stability of the anti-overturning action of the hydraulic drive system is ensured.

[0055] In this embodiment, the radar-visual displacement monitoring system deployed in step S100 mainly includes a radar subsystem and a vision subsystem. The radar subsystem uses a millimeter-wave radar 301 to monitor the displacement of both sides of the bottom of the main beam at the support of the single-column pier bridge, while the vision subsystem uses an industrial camera 202 to monitor the displacement of both sides of the bottom of the main beam at the support of the single-column pier bridge.

[0056] Specifically, such as Figures 2 to 4As shown, the radar subsystem includes a millimeter-wave radar 301 and a corner reflector 302. The millimeter-wave radar 301 is installed at the bridge structure monitoring point, specifically at the bottom of the main beam near the pier support. The corner reflector 302 is installed below the millimeter-wave radar 301 and fixed by a steel pipe erected on the ground. The vision subsystem includes an industrial camera 202, a supplementary light, and an artificial image target 201. The industrial camera 202 and the supplementary light are installed on the pier opposite the bridge structure monitoring point. The artificial image target 201 is installed in the same position as the millimeter-wave radar 301. The artificial image target 201 consists of a steel target surface and a fixed bracket 100, which are installed on the concrete at the bottom of the bridge using expansion bolts. A specific artificial image is affixed to the front of the target surface, and angle steel is vertically welded to the back for mounting the millimeter-wave radar 301. The millimeter-wave radar 301 and the angle steel behind the target surface are connected by a universal joint base.

[0057] It is understood that the specific structures and working principles of the millimeter-wave radar 301 and the industrial camera 202 are well known to those skilled in the art, and therefore will not be repeated here. For the millimeter-wave radar 301, the displacement monitoring firmware is burned into the core board chip to calculate the structural displacement, and the edge terminal then directly acquires the calculated vertical displacement of the structure through the serial port; for the industrial camera 202, the original target image is acquired through the network port and transmitted to the edge terminal for structural displacement calculation.

[0058] It is important to understand that the main cause of the collapse of the single-column pier bridge was the presence of heavy vehicles traveling on one side of the bridge, causing the local load moment to exceed the set upper limit. To ensure the accuracy of the radar displacement monitoring system's displacement identification of the single-column pier bridge, the system is activated when the target heavy vehicle is about to reach the target position on the bridge. This allows the system to begin collecting structural displacement data of the single-column pier bridge via millimeter-wave radar 301 and industrial camera 202.

[0059] During the acquisition of structural displacement data, deviation calculations can be performed in real time on the structural displacement data acquired by the millimeter-wave radar 301 and the industrial camera 202 at each sampling point. If the deviation at a certain sampling point exceeds a set threshold, it indicates that the structural displacement data at that sampling point is abnormal, and the data at that sampling point needs to be removed to reduce abnormal data. The specific value of the threshold for judging data anomalies can be selected according to the actual needs of those skilled in the art. For example, the threshold can be set to 10% to 20% of the average value of the structural displacement data acquired by the millimeter-wave radar 301 and the industrial camera 202.

[0060] After the millimeter-wave radar 301 and industrial camera 202 complete the acquisition of structural displacement data, when performing weighted fusion to obtain multimodal fusion data for radar vision, if the lighting is poor, or there is rain or fog, or the image contrast is low, the weight of the structural displacement data acquired by industrial camera 202 can be reduced, while the weight of the structural displacement data acquired by millimeter-wave radar 301 can be increased; conversely, the weight of the structural displacement data acquired by millimeter-wave radar 301 can be reduced, while the weight of the structural displacement data acquired by industrial camera 202 can be increased. For ease of understanding, the specific fusion process based on environmental changes during the multimodal fusion of radar vision data will be described in detail below.

[0061] In this embodiment, when constructing the radar-visual multimodal fusion data in step S200, the visual confidence score and the reciprocal of the radar confidence score calculated based on environmental perception can be mapped to the diagonal element R in the Kalman filter observation noise covariance matrix, respectively. vis and R rad Therefore, the Kalman filter algorithm is used to calculate the element R in real time. vis With R rad Automatically update the Kalman gain K to accurately output the displacements on both sides of the bottom of the main beam at the supports of a single-column pier bridge under the current environment. For example, when the visual perception of the industrial camera 202 is interfered with by rain or fog, the weight of the visual confidence will decrease, causing the corresponding element R to... vis The Kalman gain K is automatically increased, thus the weight of visual data is automatically reduced during calculation, making the system output more dependent on the observations of the millimeter-wave radar 301, and vice versa.

[0062] Specifically, for the perception of the visual environment, real-time histogram analysis and sharpness evaluation can be performed on the images captured by the industrial camera 202. If the overall grayscale mean of the image is lower than the first threshold representing darkness or higher than the second threshold representing direct strong light, or if the Laplacian gradient of the image is lower than the third threshold, it is determined to be a visually disturbed condition, thereby reducing the visual confidence level. The specific values ​​of the first, second, and third thresholds are constants known to those skilled in the art, and therefore will not be described in detail here.

[0063] Regarding radar environment perception, when the millimeter-wave radar 301 receives radar signals, if the multipath clutter density of unstructured objects is lower than a set fourth threshold, or the echo intensity variance of the target corner reflector 302 exceeds a fifth threshold, the radar is determined to be in a disturbed operating condition, thereby reducing the radar confidence level. The specific values ​​of the fourth and fifth thresholds are constants known to those skilled in the art, and therefore will not be described in detail here.

[0064] In this embodiment, for ease of understanding, the following will explain how element R is obtained through radar confidence and visual confidence respectively. vis With Rrad The specific process is described in detail. For the industrial camera 202 of the vision subsystem, an image quality factor Q, which characterizes visual confidence, is constructed by calculating the Laplacian gradient variance and brightness deviation of the image. ing For the millimeter-wave radar 301 of the radar subsystem, the signal-to-noise ratio (SNR) of the target echo is extracted in real time to characterize the radar confidence level.

[0065] When constructing the observation noise covariance matrix of the Kalman filter, element R vis This can be viewed as visual observation noise, element R rad This can be considered as radar observation noise. Visual observation noise R vis and radar observation noise R rad The expression is as follows:

[0066] .

[0067] .

[0068] Among them, R base-vis and R base-rad Let SNR represent the visual fundamental noise covariance and the radar fundamental noise covariance, respectively. α represents the noise amplification factor, β represents the attenuation rate factor, and SNR represents the noise level coefficient. th denoted by , where represents the radar signal-to-noise ratio threshold, and k represents the power-law exponent.

[0069] Based on the above expression, it can be seen that when encountering rain, fog, or darkness, the image quality factor Q is affected. ing During descent, visual observation noise R vis It grows exponentially; when electromagnetic interference causes the signal-to-noise ratio to fall below the threshold SNR th At that time, radar observation noise R rad The power law increases as the signal-to-noise ratio decreases. That is, when the noise calculated by one of the visual and radar observation devices is much greater than the noise of the other device, the algorithm will automatically reduce the fusion weight corresponding to the noisier device to near zero, ensuring the displacement recognition accuracy of single-column pier bridges.

[0070] It is important to know that the displacement on both sides of the bottom of the main beam at the support of a single-column pier bridge, as monitored in real time by the Ravis displacement monitoring system, can be denoted as d. L (t) and d R (t); where upward displacement is positive and downward displacement is negative. Based on the obtained displacement, the lateral overturning angle θ(t) of the beam and the amount of support detachment and upward tilt can be calculated. The precursor to the overturning of a single-column pier bridge is that one side sinks under pressure, while the other side's support detaches and tilts upwards. The lateral overturning angle θ(t) of the beam and the amount of upward tilting due to support detachment are key indicators. The specific expression is:

[0071] θ(t)=arctan[(d L (t)-d R (t)) / B]; .

[0072] Where B represents the distance between the measurement points on both sides of the bottom of the main beam at the support of a single-column pier bridge.

[0073] In this embodiment, a digital twin based on the Mamba-2 architecture is constructed in step S300, which can predict the support reactions of a single-column pier bridge. The input X of the digital twin... t =[d L (t), d R The output of the digital twin is the support reaction force R. x =[R L R R ]; where R L and R R These represent the support reactions on the left and right sides of the support, respectively. For ease of understanding, the specific construction process of the digital twin based on the Mamba-2 architecture will be described in detail below.

[0074] (1) Model pre-training.

[0075] A high-precision finite element model of a single-column pier bridge can be constructed. Then, various load conditions can be simulated based on the finite element model, generating corresponding displacement-support reaction force data pairs. Specific simulated load conditions include single-vehicle eccentric loading, dual-vehicle eccentric loading, 200% overload, and support aging failure. Based on the obtained data pairs, the displacement-response force mapping relationship is trained on a digital twin based on the Mamba-2 architecture.

[0076] (2) Add physical constraints to the loss function of the digital twin to ensure that the output of the model conforms to the laws of mechanics.

[0077] The overall loss function includes the data-driven loss L. Data Loss L based on mechanical equilibrium constraint term of beam lateral overturning angle Equilibrium And the boundary constraint term loss L based on the non-negativity of support reaction force Boundary .

[0078] For data-driven loss L Data Its expression is as follows:

[0079] .

[0080] Where N represents the total number of samples, This represents the support reaction force calculated by the finite element method in the x-th sample.

[0081] For the loss L of the mechanical equilibrium constraint term Equilibrium For a single-column pier bridge, moment balance must be satisfied in the transverse section. We can assume the bridge section rotates about the center O of the single-column pier, neglecting minute elastic deformation, and approximate it as a rigid body, thus obtaining the moment balance equation:

[0082] .

[0083] Among them, B L and B R P represents the length of the lever arm from each side of the support to the center of the single-column pier. load This indicates an external eccentric load, such as a vehicle load; 'e' represents the eccentricity of the external load; and 'I' represents the moment of inertia of the bridge superstructure cross-section about the center of rotation. This represents the angular acceleration obtained by the second derivative of the lateral overturning angle of the beam.

[0084] Based on the above torque balance equations, the loss L of the mechanical equilibrium constraint term can be obtained. Equilibrium The expression is:

[0085] .

[0086] For the boundary constraint term loss L Boundary Generally, bridge bearings are subjected to compression but not tension. Therefore, the bearing reaction force predicted by the digital twin must be greater than 0. If it is less than 0, constraints should be applied. The boundary constraint term loss L... Boundary The expression is:

[0087] .

[0088] Where ReLU(·) represents the activation function; that is, if the support reaction force predicted by the digital twin is greater than 0, the ReLU output is 0, and the boundary constraint term loss L Boundary The value is 0; if the predicted support reaction force is less than 0, the ReLU output will be positive, resulting in a huge boundary constraint term loss L. Boundary value.

[0089] The total loss function L during the training process of the digital twin Total Data-driven loss L Data Loss L of mechanical equilibrium constraint term Equilibrium and boundary constraint term loss L Boundary We obtain the result by weighting; the specific expression is:

[0090] L Total =λ1·L Data +λ2·L Equilibrium +λ3·L Boundary Where λ1, λ2, and λ3 are the corresponding weight coefficients.

[0091] It is important to know that during the training of the digital twin, the weight coefficients of the data-driven loss, the mechanical equilibrium constraint loss, and the boundary constraint loss are adaptively adjusted according to the gradient of the network parameters or the rate of loss descent of each loss term. For ease of understanding, a detailed description will be provided below.

[0092] (3) Weight allocation of loss function.

[0093] At the initial stage of training, dimensional normalization and initial weight setting are performed based on mechanical experience: Since the order of magnitude difference between data-driven loss and mechanical equilibrium constraint loss is huge, the losses can be dimensionless first (divided by the characteristic peak value of each physical quantity); then, based on mechanical analysis experience, the primary task of the digital twin in the initial stage is to fit the geometric shape, followed by satisfying physical laws; therefore, the initial weight ratio is set to λ1:λ2:λ3=1:0.1:0.1. That is, in the initial stage of training the digital twin, pure data-driven terms should dominate to quickly converge to the underlying features; the torque balance term serves as a regularization term; the boundary term (support reaction force > 0) is only triggered when a violation occurs, and its initial weight should not be too large to avoid gradient oscillation.

[0094] Real-time fine-tuning is performed during gradient-normalized training: the gradient norm of each loss term with respect to the parameters of the shared network layers is calculated; if the gradient of a physical constraint (such as torque balance) is too small, it means that the network is "ignoring" physical laws, and the algorithm will automatically increase the weight of that term. At the end of a single training cycle, the relative rate of decrease of each loss term is calculated; if the rate of decrease of the physical constraint term is significantly slower than that of the data term, the value of the weight coefficient λ2 is increased through an adaptive formula. As training progresses, the weight coefficient λ1 gradually stabilizes, while the weight coefficients λ2 and λ3 adaptively increase, forcing the digital twin to strictly approximate Newton's laws of mechanics in its output results, ensuring that it can output support reactions that conform to physical common sense even on unlabeled test data.

[0095] (4) Fine-tuning of measured data.

[0096] The digital twin model is fine-tuned using displacement data (safety status) collected during daily operations to eliminate errors between simulation and real-world data.

[0097] In this embodiment, after training the digital twin based on the Mamba-2 architecture, it can be used as a virtual sensor to output the overturning state of the bridge in real time. That is, after the digital twin completes training, it outputs virtual support reaction forces representing the overturning state in real time based on the real-time working conditions of the single-column pier bridge. If the output support reaction force tends to 0, it indicates that the support is at critical disengagement, and active anti-overturning intervention is required through the hydraulic drive system.

[0098] It's important to understand that when the digital twin outputs a support reaction force approaching zero, meaning the support of a single-column pier bridge is in a critical state of detachment, several scenarios are possible. One is that the support is indeed in a critical state of detachment; the other is that the data output by the digital twin is abnormal. Therefore, to ensure the accuracy of the digital twin's output results, the results can be verified using bridge structure displacement data collected by the radar-based displacement detection system. For ease of understanding, a detailed description will follow.

[0099] In this embodiment, the confidence verification of the critical disengagement state of the support based on the output of the digital twin includes the following process:

[0100] (1) Constructing a simplified physical reverse-engineering observer: that is, using displacement data collected by the radar-based displacement monitoring system, combined with the geometric parameters of the bridge section, the theoretical estimated reaction force R of the support is derived in real time through the rigid body static equilibrium equation. est The specific expression is as follows:

[0101] .

[0102] Where G represents the bridge dead load, B represents the distance between the measurement points on both sides of the bottom of the main beam at the support, and L represents the distance between the measurement points on both sides of the support. arm The lever arm represents the center of gravity under dead load relative to the center of rotation of a single-column pier bridge, and I represents the moment of inertia of the bridge superstructure cross section about the center of rotation. This represents the angular acceleration of the beam during lateral rotation.

[0103] (2) Verify the differences between the outputs of the digital twin and the physical reverse observer: calculate the support reaction force R output by the digital twin in real time. pred Compared with the theoretically estimated reaction force R est The normalized residual ε is expressed as follows:

[0104] .

[0105] (3) Constructing the critical state confidence function: Introducing a confidence scoring function P in the form of Sigmoid. conf When the digital twin prediction support is in a critical state of vacancy, the confidence level is determined by the normalized residual ε, and the confidence score function P conf The expression is as follows:

[0106] .

[0107] Where γ represents the steepness coefficient, ε th This indicates the maximum permissible deviation threshold.

[0108] (4) Control logic of the control strategy: The intervention of the hydraulic drive system needs to meet the following conditions:

[0109] (R pred <R limit )∩(P conf >P threshold ).

[0110] Among them, R limit P represents the safety threshold of the support reaction force. threshold This represents the confidence threshold.

[0111] In this embodiment, the prerequisite for installing the hydraulic drive system is that the single-column pier bridge has already been fitted with a widened cap beam or a steel bracket. The hydraulic drive system includes a pair of hydraulic actuators, such as hydraulic servo cylinders, as well as an accumulator group and an electro-hydraulic controlled servo valve. The hydraulic actuators adopt a double-rod design, capable of both raising and lowering. The two hydraulic actuators are installed on both sides of the support and spaced a certain distance apart to ensure sufficient anti-overturning moment is generated. The accumulator group is used to store high-pressure oil, which can release huge amounts of energy instantaneously in milliseconds. The servo valve is a high-frequency response valve used to receive instructions from the algorithm and precisely control the oil pressure. When the digital twin predicts that the virtual support reaction force on one side of the single-column pier bridge tends to 0, the agent can control the servo valve to open, thereby outputting the oil stored in the accumulator group to the corresponding hydraulic actuator, causing the actuator's push rod to extend to contact the beam and apply the calculated anti-overturning moment to the beam.

[0112] It is important to note that when implementing active anti-overturning control for single-column pier bridges using a hydraulic drive system, the hydraulic actuator's push rod should actively extend and flexibly contact the beam, avoiding rigid contact. This effectively extends the structural lifespan of the beam. For the flexible contact of the hydraulic drive system, an intelligent agent can act based on the predicted trends of a digital twin, offsetting the hardware's action time lag. Specifically, during the policy generation process in the intelligent agent, a reward function r that comprehensively considers safety, comfort, and energy consumption is introduced. t By outputting the reward function r t The optimal top thrust strategy, which maximizes the thrust, enables refined operation. For ease of understanding, the refined control process of the agent will be described in detail below.

[0113] In this embodiment, during step S400, an intelligent agent is constructed based on a deep reinforcement learning algorithm. The input to the intelligent agent is provided by the radar displacement monitoring system and the digital twin, mainly including the current structural state of the single-column pier bridge, the support reaction force predicted by the digital twin, and the current state of the hydraulic drive system. The output of the intelligent agent is the opening control signal of the servo valve, which directly determines the magnitude of the anti-overturning torque output by the hydraulic actuator. Reward function r tIn the process of policy output by the intelligent agent, the changes are based on the support reaction force predicted by the digital twin; that is, if the support reaction force predicted by the digital twin drops below the safety threshold, a heavy penalty is imposed; at the same time, the intelligent agent is required to control the hydraulic drive system as "fuel-efficiently" and "smoothly" as possible, avoiding sudden pushes that could damage the beam; the reward function r t The expression is as follows:

[0114] .

[0115] Where ω1, ω2, and ω3 all represent weighting coefficients, and R limit R represents the safety threshold of the support reaction force. pred A represents the support reaction force predicted by the digital twin. t This indicates the opening degree of the servo valve in the hydraulic drive system. This indicates the opening speed of the servo valve in the hydraulic drive system.

[0116] Specifically, in the actual operation of the intelligent agent, it needs to be trained first through the output environment of the digital twin. This involves simulating various extreme working conditions in a virtual environment constructed by the digital twin, training the agent to counteract these simulated extreme conditions and achieve balance against the overturning of a single-column pier bridge with minimal force. To further ensure safety, a rule filter is set between the agent's output and the hydraulic drive system. This means that if the anti-overturning moment output by the agent exceeds the structure's bearing limit, it needs to be forcibly cut off to prevent damage to the beam.

[0117] It is important to note that during the normal operation of a single-column pier bridge, the intelligent agent only provides suggestions and does not control the hydraulic drive system to verify the rationality of its decisions. Only when the digital twin predicts that the support reaction force of the single-column pier bridge tends to 0, that is, when the bridge is about to overturn, will the intelligent agent take over control and use the hydraulic drive system to make millimeter-level fine-tuning interventions.

[0118] In this embodiment, to further achieve refined operation of the intelligent agent, a dual closed-loop structure is required to provide feedback during the agent's control process. The structural response closure serves as the outer loop feedback to ensure the single-column pier bridge's posture truly returns to a safe state; the force control servo closed loop serves as the inner loop feedback to ensure the hardware execution accuracy of the hydraulic drive system and eliminate hardware errors.

[0119] Specifically, for the structural response closed loop, the control effect is evaluated by comparing structural displacement monitoring data before and after the hydraulic drive system's action, and then fed back to the agent to optimize the control strategy for the next moment. For example, if the agent outputs a strategy of applying hydraulic pressure at the current moment, and the visual displacement monitoring system detects a new structural displacement at the next moment, if the imbalance of the single-column pier bridge decreases, it indicates that the agent's control is effective, and the agent continues to maintain the strategy of applying hydraulic pressure at the next moment. If the imbalance of the single-column pier bridge increases or remains unchanged, it indicates that there are unknown factors, and the agent must immediately calculate a new control strategy. For the force-controlled servo closed loop, the actual output force of the hydraulic drive system is fed back in real time by sensors and compared with the agent's command. Based on the comparison result, the servo valve of the hydraulic drive system is controlled to make real-time fine adjustments to the opening to ensure execution accuracy.

[0120] Another aspect of this application provides an electronic device, in one preferred embodiment of which includes a processor and a memory; the memory is used to store a computer program, and the processor is used to execute the computer program to implement the above-described active control method for preventing overturning of a single-column pier bridge.

[0121] Another aspect of this application provides a computer-readable storage medium, in a preferred embodiment of which a computer program is stored on the storage medium; when the computer program is executed by a processor, the above-described active control method for preventing overturning of a single-column pier bridge is implemented.

[0122] The basic principles, main features, and advantages of this application have been described above. Those skilled in the art should understand that this application is not limited to the above embodiments. The embodiments and descriptions in the specification are merely the principles of this application. Various changes and modifications can be made to this application without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claims. The scope of protection claimed by this application is defined by the appended claims and their equivalents.

Claims

1. A method for active control of overturning of a single column pier bridge, characterized by, The method comprises the following steps: S100: deploying a radar and vision displacement monitoring system to monitor the displacement of the two sides of the main beam bottom at the support of the single-column pier bridge, and installing a hydraulic drive system for providing an anti-overturning moment at the support position of the single-column pier bridge; wherein the radar and vision displacement monitoring system comprises a radar subsystem and a vision subsystem, and both the radar subsystem and the vision subsystem monitor the displacement of the two sides of the single-column pier bridge; S200: obtaining radar and vision multi-modal fusion data based on modal weight self-adaption according to the deployed radar and vision displacement monitoring system, and calculating the beam lateral overturning angle; S300: constructing a digital twin and introducing physical constraints including moment balance and non-negative support reaction, predicting the bridge overturning state at multiple future time points by combining the obtained beam lateral overturning angle data with the structural displacement data; S400: constructing an intelligent agent for taking over the control right of the hydraulic drive system, and controlling the hydraulic drive system to perform anti-overturning actions based on structural response closed loop and force control servo closed loop based on the predicted bridge overturning state; The construction and training of the digital twin of step S300 comprises the following process: A finite element model of the single-column pier bridge is constructed to simulate the displacement-support reaction force data pairs generated under multiple load cases, which are used to train the mapping relationship between displacement and reaction force of the digital twin based on the Mamba-2 architecture; The loss function of the digital twin includes a data-driven item loss, a mechanical balance constraint item loss based on the beam lateral overturning angle, and a boundary constraint item loss based on the non-negative characteristic of the support reaction force; During the training process of the digital twin, the weight coefficients of the data-driven item loss, the mechanical balance constraint item loss, and the boundary constraint item loss are adaptively adjusted according to the network parameter gradient or loss reduction rate of the digital twin; The collected displacement data generated during daily operation is used to fine-tune the digital twin, eliminating the error between simulation data and real data.

2. The method of claim 1, wherein the method is characterized by: The structural displacement data monitored by the radar subsystem and the vision subsystem are subjected to deviation calculation, and if the deviation of a certain sampling point exceeds the set threshold, the data of the sampling point is removed; When the structural displacement data monitored by the radar subsystem and the vision subsystem are subjected to weighted fusion to obtain radar and vision multi-modal fusion data, if the light difference, rain and fog weather, or low image contrast, the weight of the structural displacement data monitored by the vision subsystem is reduced, and if there is strong electromagnetic interference or serious multipath effect in the environment, the weight of the structural displacement data monitored by the radar subsystem is reduced.

3. The method of claim 2, wherein the control law is given by ###0001### where K is a gain matrix, x is the state vector, and x is the state vector of the previous time step. When constructing the radar-visual multimodal fusion data, the visual confidence score and the reciprocal of the radar confidence score calculated based on environmental perception are mapped to the diagonal elements R in the Kalman filter observation noise covariance matrix. vis and R rad The Kalman filter algorithm is based on the real-time calculated element R. vis With R rad Automatically update the Kalman gain K; Real-time histogram analysis and definition evaluation are performed on the images collected by the vision subsystem, and if the overall image gray mean value is lower than the first threshold value representing night or higher than the second threshold value representing strong light direct radiation, or the image Laplacian gradient is lower than the third threshold value, it is determined that the vision is disturbed, and the vision confidence is reduced; If the density of non-structure body multipath clutter is lower than the set fourth threshold value, or the echo intensity variance of the target corner reflector exceeds the fifth threshold value, it is determined that the radar is disturbed, and the radar confidence is reduced.

4. The method of claim 3, wherein the control law is given by ###0002### For the visual subsystem, an image quality factor Q is constructed to represent the visual confidence by computing the Laplacian gradient variance and the luminance bias of the image ing ; For the radar subsystem, the signal-to-noise ratio (SNR) of the target echo is extracted in real time to represent the radar confidence. In constructing the Kalman filter observation noise covariance matrix, the elements R vis and R rad are expressed as follows: ； ； where R base-vis and R base-rad represent the visual and radar fundamental noise covariances, respectively, a represents a noise amplification coefficient, β represents a decay velocity coefficient, SNR th represents a radar signal-to-noise ratio threshold, and k represents a power law exponent.

5. The monopole pier bridge anti-overturning active control method according to any one of claims 1-4, characterized in that, After the digital twin is trained, the support reaction force representing the overturning state is output according to the real-time working condition of the single-column pier bridge. If the output support reaction force tends to 0, it indicates that the support is in a critical emptying state, and active anti-overturning intervention needs to be performed through the hydraulic driving system.

6. The monopole pier bridge anti-overturning active control method according to claim 5, characterized in that, The confidence verification of the critical emptying state of the support based on the output of the digital twin includes the following processes: Based on the displacement data collected by the radar displacement monitoring system, combined with the geometric parameters of the bridge section, the current theoretical estimated counterforce R of the support is real-time backstepped through the rigid body static equilibrium equation est ; computing a support reaction force R output by the digital twin pred with a normalized residual error ε of the theoretically estimated reaction force R est ​ A confidence score function P is introduced conf When the digital twin predicts that the support is in a critical voiding state, the confidence is determined by the normalized residual error ε, and the confidence score function P conf The expression of P is as follows: ； The intervention of the hydraulic driving system needs to meet the following conditions: (R pred <R limit )∩(P conf >P threshold ); where γ represents a steepness coefficient, ε th represents a maximum deviation threshold value allowed, R limit represents a safety threshold value of the support reaction force, P threshold represents a confidence threshold value.

7. The monopole pier bridge anti-overturning active control method according to claim 1, wherein In step S400, an agent is constructed based on a deep reinforcement learning algorithm, and a reward function r considering safety, comfort and energy consumption is introduced to the policy generation of the agent t ; various extreme working conditions are simulated in the virtual environment constructed by the digital twin, the agent is trained to balance the overturning of the single-column pier bridge with the minimum force, so that the reward function r t is maximized; the expression of the reward function r t is as follows: ； where ω1, ω2, ω3 represent weight coefficients, ReLU(·) represents an activation function, R limit represents a safety threshold of the support reaction force, R pred represents a support reaction force predicted by the digital twin, A t represents the servo valve opening degree of the hydraulic drive system, represents the servo valve opening speed of the hydraulic drive system.

8. The monopole pier bridge anti-overturning active control method of claim 1, wherein, In step S400, for the structure response closed loop, the control effect is evaluated by comparing the structure displacement monitoring data before and after the action of the hydraulic driving system, and then fed back to the intelligent agent to optimize the control strategy at the next moment; For the force control servo closed loop, the actual output force of the hydraulic driving system is fed back in real time through the sensor, and compared with the instruction of the intelligent agent. According to the comparison result, the servo valve of the hydraulic driving system is controlled for real-time opening degree fine adjustment to ensure the execution precision.

9. An electronic device, comprising: The storage medium stores a computer program; when the computer program is executed by the processor, the active control method for preventing the single-column pier bridge from overturning according to any one of claims 1-8 is realized.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program; when the computer program is executed by the processor, the active control method for preventing the single-column pier bridge from overturning according to any one of claims 1-8 is realized.

Images