Intelligent vibration suppression device for vortex-induced vibration of split type box girder and control method of intelligent vibration suppression device

Abstract

The invention discloses an intelligent vibration suppression device for split type box girder vortex-induced vibration and a control method, and relates to the field of bridge vortex-induced vibration control. The device comprises a monitoring module, a control module, an adjustable air blocking grid execution module and a power supply module. The monitoring module collects vibration data of a wind field and a structure through a distributed sensor array; the control module is based on an ARM architecture, is internally provided with a pre-trained deep neural network, processes data and generates a ventilation rate adjusting instruction; the execution module dynamically adjusts the ventilation rate of the groove in the center of the box girder through a rotatable rotor wing; the power supply module provides stable power supply. According to the control method, active vibration suppression is realized through data acquisition, parameter calculation, neural network decision, adjustment execution and closed-loop verification. An active intelligent regulation and control strategy is adopted, wind field changes can be dynamically adapted, vortex-induced vibration is remarkably restrained, the safety and durability of a bridge are improved, installation is convenient, and the operation and maintenance cost is low.

Description

Technical Field

[0001] This invention belongs to the field of vortex-induced vibration control for bridges. Specifically, it relates to an intelligent vibration suppression device and control method for vortex-induced vibration of a split box girder. Background Technology

[0002] Compared to traditional truss-type main girders and closed single-box girders, the application of split-type box girder bridge sections can further break through the ultimate span limit, improving the flutter stability and ultimate design wind speed of long-span bridges. However, the complex flow field characteristics at the central slot make it susceptible to vortex-induced vibration. When airflow passes through the slot, it forms vortices that directly collide with the box girder surface, leading to a decrease in flow field stability, inducing large-amplitude vibrations at low wind speeds, and consequently causing structural fatigue damage and traffic safety hazards. Traditional vibration suppression measures, such as fixed wind deflectors or dampers, have limited effectiveness and lack dynamic adaptability, making them difficult to cope with changing wind field conditions. Summary of the Invention

[0003] Based on the above background, the purpose of this invention is to propose an intelligent vibration suppression device for vortex-induced vibration of split box girder bridges. This device combines a dynamically adjustable central windbreak grid system with an adaptive control algorithm based on real-time monitoring data, thereby achieving active and precise suppression of vortex-induced vibration. It overcomes the limitations of traditional passive measures and provides an efficient solution for the safety and durability of split box girder bridges.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows: an intelligent vibration suppression device for split-type box girder vortex-induced vibration, comprising a monitoring module, a control module, an adjustable windbreak grille execution module and a power supply module, wherein the power supply module supplies power to the entire system, and the modules work together to achieve active suppression of vortex-induced vibration;

[0005] The monitoring module employs a distributed sensor array, with six three-dimensional ultrasonic anemometers and nine unidirectional accelerometers installed at 1 / 4, 1 / 2, and 3 / 4 of the main span of the box girder. These sensors collect wind field parameters and structural vibration response data. The three-dimensional ultrasonic anemometers have a wind speed measurement resolution of 0.01 m / s, an accuracy of ±1%, and an effective range of 0~40 m / s; and a wind direction resolution of 0.1° and an accuracy of ±2%.

[0006] The control module is based on an embedded microcontroller with an ARM architecture. It has a built-in deep neural network pre-trained with wind tunnel test and CFD simulation data, integrates real-time spectrum analysis function, processes acceleration signals through fast Fourier transform algorithm to extract vibration characteristic frequencies, and can also receive data from monitoring module and calculate average wind speed, turbulence intensity and gust coefficient.

[0007] The adjustable windbreak grille execution module is installed in the central slot of the box girder and includes four fixed grille plates. Each fixed grille plate is equipped with 10 rotatable rotors driven by embedded micro servo motors. The rotors can rotate precisely within the range of 0° to 90°. The execution module has a built-in position sensor to monitor the rotor rotation angle and air permeability in real time.

[0008] Furthermore, the input layer data of the deep neural network includes average wind speed, triaxial turbulence intensity, triaxial gust coefficient, and vibration frequency. The hidden layer has 1 layer, 128 units, and the activation function is Tansig. The output layer activation function is Tansig. The training iterations are 1000 times. The Adam optimizer is used with a learning rate of 0.001. It also has an online learning mechanism, triggered every 24 hours or when the vibration amplitude changes by ≥10%.

[0009] Furthermore, in the adjustable windshield grille execution module, the fixed grille width is 480mm, the spacing between adjacent grille panels is 800mm, the width of a single rotor is 200mm and the length is 400mm, and the air permeability can be dynamically changed within the range of 48% to 62% under controlled conditions by adjusting the rotation of the rotor.

[0010] Furthermore, all sensors in the monitoring module transmit the collected data to the control module wirelessly. The sensor arrangement meets the following requirements: 9 unidirectional acceleration sensors are set at 3 measuring points on each of the 1 / 4, 1 / 2, and 3 / 4 sections of the main span, with 2 on the west side measuring vertical and lateral acceleration respectively, and 1 on the east side measuring vertical acceleration; 6 three-dimensional ultrasonic anemometers are installed on both sides of the above sections, 6 meters above the bridge deck, with the bridge axis direction as the 0° reference, and the wind direction angle increasing in a clockwise direction.

[0011] The present invention also provides a method for controlling vortex-induced vibration of a split box girder based on the intelligent vibration damping device described above, comprising the following steps:

[0012] S1: The monitoring module collects real-time data on wind speed, wind direction, and structural acceleration, and transmits it to the control module;

[0013] S2: The control module processes the received data, calculates the triaxial turbulence intensity according to formula (1), calculates the triaxial gust coefficient according to formula (2), and obtains the vibration characteristic frequency by analyzing the acceleration signal with a time interval of 30 minutes using the fast Fourier transform algorithm; among which, the formula for calculating the triaxial turbulence intensity is as shown in (1):

[0014]

[0015] In the formula, , These represent the turbulence intensities along the wind direction, across the wind direction, and vertically. The standard deviation of the three-dimensional pulsating wind speed; The average wind speed over 10 minutes;

[0016] The formula for calculating the three-directional gust coefficient is shown in (2):

[0017]

[0018] In the formula, , , These are the gust coefficients for downwind, crosswind, and vertical directions, respectively. Three-way gust wind speed, gust duration It lasts for 3 seconds;

[0019] S3: The control module inputs the key parameters obtained in step S2 into the deep neural network to generate the optimal ventilation rate adjustment command. The ventilation rate R is calculated as shown in (3):

[0020]

[0021] In the formula, b1 is the width of the fixed grating plate; To fix the number of grating plates; The number of rotors that are shut down during the gap; Width of a single rotor; d is the length of a single rotor; d is the width of the central slot in the box girder; l is the clearance length.

[0022] S4: The adjustable windshield grille execution module receives the adjustment command and drives the rotor to rotate to the target angle. The position sensor transmits the real-time monitored rotor angle and air permeability back to the control module.

[0023] S5: The control module determines whether the adjustment meets the standard, which is that the angle error is ≤ ±1° and the air permeability deviation is ≤ ±2%. If the standard is not met, a secondary adjustment command is triggered every 5 seconds, with a maximum of 3 attempts, forming a feedback closed loop.

[0024] Furthermore, the CFD simulation data covers Reynolds numbers. With the Strauha number St=0.1~0.3, a nonlinear mapping relationship from multi-dimensional wind field characteristics to the optimal wind permeability was established.

[0025] Furthermore, the online learning mechanism uses real-time data from the most recent 30 minutes for each update, with a sample size of 1000. It employs gradient descent to iterate the model parameters for 5 iterations and uses an early stopping mechanism to verify that the loss has not decreased for 3 consecutive iterations, thus avoiding overfitting.

[0026] Advantages and beneficial effects of the present invention:

[0027] This invention breaks through the limitations of traditional passive vibration suppression measures by combining a dynamically adjustable windbreak grille system with an adaptive control algorithm based on a deep neural network. It can respond to changes in wind field and bridge vibration in real time, and interfere with the formation and development of vortices by precisely adjusting the air permeability, thereby reducing the lift coefficient of the box girder by more than 80%, significantly weakening the intensity of vortex-induced vibration, and effectively avoiding structural fatigue damage.

[0028] Precise and comprehensive monitoring, rapid control response: The system adopts a distributed sensor array, with 6 three-dimensional ultrasonic anemometers and 9 unidirectional accelerometers to collect wind field and structural vibration data from all directions, resulting in high measurement accuracy and resolution. The entire system takes ≤10 seconds from data acquisition to execution of adjustment. With the help of a feedback closed-loop mechanism (angle error ≤±1°, air permeability deviation ≤±2%), the system ensures the real-time performance and accuracy of vibration suppression control.

[0029] It has strong adaptability and can adapt to complex working conditions: the deep neural network is pre-trained with wind tunnel test and CFD simulation data, covering a wide range of Reynolds number and Strauha number, and has an online learning mechanism, which can dynamically optimize control parameters according to the real-time environment; the ventilation rate can be flexibly adjusted within the range of 48%~62%, which can cope with the changing wind field conditions and adapt to different operating scenarios.

[0030] Easy and feasible to install, with outstanding engineering value: The modular design of the device allows the adjustable windbreak grille execution module to be directly installed in the central slot of the box girder without requiring major modifications to the main structure of the bridge, resulting in low installation and construction difficulty; its active vibration suppression strategy effectively improves the safety and durability of the bridge, providing an efficient solution for vortex-induced vibration control of large-span split box girder bridges, with broad prospects for engineering applications. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of the wind condition monitoring module layout according to Embodiment 1 of the present invention.

[0032] Figure 2 This is a schematic diagram of a control module process according to Embodiment 1 of the present invention.

[0033] Figure 3 The diagram shows the overall arrangement of the adjustable windbreak grille execution module installed in the central slot of the box girder according to Embodiment 1 of the present invention; (a) uncontrolled state, (b) controlled state.

[0034] Figure 4 The diagram shows a partial structure of the adjustable windshield grille execution module in different control states according to Embodiment 1 of the present invention: (a) uncontrolled state, (b) controlled state (medium ventilation rate), and (c) controlled state (minimum ventilation rate).

[0035] Figure 5The following are cross-sectional vorticity diagrams of the adjustable windshield grille execution module under different control states according to Embodiment 1 of the present invention: (a) uncontrolled state, (b) controlled state (only the side wing plate near the box girder is open), and (c) controlled state (all wing plates are open).

[0036] Figure 6 This is a diagram showing the cross-sectional lift coefficient of the adjustable windshield grille execution module under different control states according to Embodiment 1 of the present invention.

[0037] Figure 7 The following are cross-sectional time-averaged streamlines and time-averaged turbulent kinetic energy diagrams of the adjustable windshield grid execution module under different control states according to Embodiment 1 of the present invention: (a) uncontrolled state, (b) controlled state (only the side wing plate near the box girder is open), (c) controlled state (all wing plates are open). Detailed Implementation

[0038] The invention will be further described below with reference to the accompanying drawings.

[0039] Example 1

[0040] An intelligent vibration suppression device for split-type box girder vortex-induced vibration includes a monitoring module, a control module, an adjustable windbreak grille execution module, and a power supply module. The power supply module supplies power to the entire system. All modules work together to actively suppress vortex-induced vibration. For example, Figure 1 As shown, the monitoring module employs a distributed sensor array layout. AC represents the accelerometer, and UA represents the three-dimensional ultrasonic anemometer, which are used to collect real-time data on structural vibration response and wind field parameters, respectively. Nine force-balanced single-axis accelerometers are located at sections 1 / 4, 1 / 2, and 3 / 4 of the main span, with three measuring points at each section. Two are located on the west side to measure vertical and lateral acceleration, respectively; one is located on the east side to measure vertical acceleration. Six three-dimensional ultrasonic anemometers are installed on both sides of sections 1 / 4, 1 / 2, and 3 / 4 of the main span, approximately 6 meters above the bridge deck. The anemometers use the bridge axis as the 0° reference, with wind direction angles increasing clockwise. The wind speed measurement resolution is 0.01 m / s, with an accuracy of ±1%, and an effective range of 0–40 m / s; the wind direction resolution is 0.1°, with an accuracy of ±2%. All sensors transmit real-time wind speed, wind direction, and acceleration data to the control module via wireless transmission technology, thereby ensuring comprehensive monitoring of structural vibration and wind field conditions.

[0041] The control module is based on an ARM-based embedded microcontroller, with a built-in deep neural network pre-trained using wind tunnel tests and CFD simulation data. It integrates real-time spectrum analysis, processes acceleration signals using a fast Fourier transform algorithm to extract vibration characteristic frequencies, and can also receive data from the monitoring module and calculate average wind speed, turbulence intensity, and gust coefficient. Figure 2The diagram illustrates the workflow of the control module. This module employs an embedded microcontroller based on the ARM architecture and integrates an adaptive control algorithm. First, the module receives wind speed data from the wind condition monitoring unit and acceleration signals from the structural response monitoring unit, and calculates parameters such as average wind speed, turbulence intensity, gust coefficient, and vibration frequency. The formula for calculating the three-dimensional turbulence intensity is shown in (1):

[0042]

[0043] In the formula, , , These represent the turbulence intensities along the wind direction, across the wind direction, and vertically. The standard deviation of the three-dimensional pulsating wind speed; The average wind speed is 10 minutes.

[0044] The formula for calculating the gust coefficient is shown in (2):

[0045]

[0046] In the formula, , , These are the gust coefficients for downwind, crosswind, and vertical directions, respectively. Three-way gust wind speed, gust duration Take 3 seconds.

[0047] The control module integrates real-time spectrum analysis, utilizing a Fast Fourier Transform (FFT) algorithm to perform spectrum analysis on 30 minutes of acceleration data and extract the characteristic frequencies of vortex-induced vibration. These parameters collectively constitute the input feature vector of the neural network. The deep neural network is pre-trained using large-scale wind tunnel test data and computational fluid dynamics (CFD) simulation data covering Reynolds numbers... The Strouhal number St = 0.1~0.3 was used to establish a nonlinear mapping relationship from multi-dimensional wind field characteristics to the optimal permeability. The input layer data included average wind speed, triaxial turbulence intensity, triaxial gust coefficient, vibration frequency, etc. The number of hidden layers was 1, the number of units was 128, and the activation function of the hidden layer was Tansig. The activation function of the output layer was Tansig, the number of training iterations was 1000, the Adam optimizer was used, and the learning rate was 0.001. In actual operation, the network was continuously optimized through an online learning mechanism. The triggering condition was every 24 hours or the vibration amplitude change was ≥10%. Each update used the most recent 30 minutes of real-time data (about 1000 samples), and the gradient descent iterative model parameters were used (5 iterations). The early stopping mechanism (verification that the loss did not decrease for 3 consecutive times) was used to avoid overfitting. The final output was the optimal permeability R, and its calculation formula is shown in (3):

[0048]

[0049] Where b1 is the width of the fixed grating plate; To fix the number of grating plates; The number of rotors that are shut down during the gap; Width of a single rotor; d is the length of a single rotor; d is the width of the central slot in the box girder; l is the gap length.

[0050] like Figure 3 As shown, the adjustable windshield grille actuator module is installed in the central slot of the box girder, wherein, Figure 3 (a) is in an uncontrolled state, with all rotor blades retracted and attached to the surface of the fixed grid plate. Figure 3 (b) Show the structure after adding the adjustable wind deflector grid execution module. Ten rotatable rotors are evenly installed on each fixed grid plate. The rotors are driven by embedded micro servo motors and can rotate precisely within the range of 0° to 90° to dynamically adjust the air permeability, thereby optimizing the flow field structure and suppressing vortex-induced vibration.

[0051] like Figure 4 As shown, this is a partial structural diagram (unit: mm) of the adjustable windshield grille actuator module under different control states. Figure 4 (a) is a partial schematic diagram of the uncontrolled state, in which the width of the fixed grid plate is 480mm, the spacing between adjacent grid plates is 800mm, all rotors are fully retracted and attached to the surface of the fixed grid plate, forming a reference aerodynamic state with a maximum air permeability of 68%. Figure 4 (b) is a partial schematic diagram of the controlled state (medium air permeability, cross-gap mode). A micro servo motor selectively drives some rotors to rotate and close, while the remaining rotors remain deployed, forming a non-grid-shaped cross-ventilation gap. The airflow channels are staggered, achieving precise local air permeability control and effectively interfering with vortex formation and shear layer interaction. The width of a single rotor is 200 mm and the length is 400 mm. Figure 4 (c) is a partial schematic diagram of the controlled state (minimum air permeability), where all rotors rotate synchronously to 90° and fully deploy, reducing the air permeability to 48%.

[0052] Through high-precision numerical simulation, the surrounding flow field of the split-type double box girder structure under three working conditions—uncontrolled operation, partial rotor opening (medium air permeability), and full rotor opening (minimum air permeability)—was reproduced and visualized for analysis. Figure 5As shown, the results indicate that the gradual deployment of the rotor significantly suppressed the vortex amplitude within the gaps between the grating plates, reflecting that the presence of the adjustable rotor effectively weakened the vortex structure strength between the fixed grating plates. Meanwhile, under uncontrolled conditions, the vortex shedding location was relatively close to the downstream edge of the box girder; however, under controlled conditions (especially with all rotors fully deployed), the vortex shedding location shifted significantly further downstream, farther from the downstream edge of the box girder. This suggests that the adjustable windbreak grating effectively prolonged the vortex formation path and reduced wake instability by interfering with the local flow field.

[0053] like Figure 6 As shown, the time history curves of the lift coefficient of the split box girder are displayed under three conditions: uncontrolled operation, partial rotor opening (medium air permeability), and full rotor opening (minimum air permeability). The comparison shows that compared to the uncontrolled condition, the average lift coefficient does not change much when the rotor is partially open, but the pulsation amplitude is significantly reduced. When the rotor is fully open, the average lift coefficient further decreases compared to the uncontrolled and partially open conditions (from approximately 0.04 to approximately 0.05), with a decrease exceeding 80%. Furthermore, within the same time range, the number of periodic changes in the aerodynamic forces of the box girder under controlled conditions is significantly less than under uncontrolled conditions. This indicates that the adjustable wind deflector effectively delays the vortex shedding phenomenon in the flow field, significantly reducing the vortex shedding frequency.

[0054] like Figure 7As shown, the time-averaged streamlines and time-averaged turbulent kinetic energy diagrams of the flow field around the split-type double box girder are presented under three operating conditions: uncontrolled operation, partial rotor opening (medium permeability), and full rotor opening (minimum permeability). The results indicate that, from the perspective of the gap flow field, there are obvious small-scale vortex structures between the grid plates under the uncontrolled operation condition. These small vortices may lead to an increase in local stress on the grid plates and amplify the unsteady load on the downstream box girder surface. When the rotor is partially open, the small vortex structures between the grid plates are partially suppressed, and the stability of the gap flow field is improved. When the rotor is fully open, the small vortex structures are further weakened and almost disappear, and the gap flow field tends to be smooth. From the perspective of wake turbulent kinetic energy distribution, a significant turbulent kinetic energy peak region appears on the lower surface of the downstream box girder under uncontrolled conditions, and the high turbulent kinetic energy region at the tail is close to the box girder surface, resulting in a strong turbulent pulsation effect. When the rotor is partially open, the turbulent kinetic energy peak region at the lower part of the downstream box girder has basically disappeared, and the wake peak position has shifted further back, farther from the box girder surface. When the rotor is fully open, the turbulent kinetic energy peak region moves to the farthest wake position, the peak intensity further decreases, and the high turbulent kinetic energy region completely separates from the box girder surface. The above phenomenon can be explained by the adjustable rotor dynamically changing the air permeability, interfering with the shear layer separation and vortex generation process between the grating plates, while extending the wake vortex formation path and weakening the vortex collision and turbulent energy transfer mechanism. The results fully verify the effectiveness of the control measures of the present invention, which not only significantly reduces the turbulent load and vibration excitation on the box girder surface, but also makes the overall aerodynamic characteristics approach a stable streamlined structure, achieving active and reliable suppression of vortex-induced vibration.

[0055] Through the above design, the adjustable windshield grid actuator module achieves dynamic adjustment of the air permeability from 48% to 62% under controlled conditions. Within this parameter range, the device exhibits excellent suppression capability for vortex-induced vibration of the double box girder, and its aerodynamic performance and control effect are substantially improved.

[0056] When vibration suppression control is not required, this device can be manually or automatically switched to standby mode via the controller, shutting down the monitoring and drive modules to save energy. The device also integrates remote status monitoring and fault diagnosis functions. By analyzing parameters such as the operating current, stroke resistance, and adjustment accuracy of the actuator in real time, it intelligently identifies fault states such as rotor rotation jamming and drive mechanism abnormalities, thereby reducing system maintenance difficulty and operating costs. All structural components are made of high-strength, corrosion-resistant materials to ensure durability. The entire process from sensor data acquisition to the adjustable windshield actuator completing adjustment takes ≤10 seconds, ensuring real-time performance. The power module provides stable power to the entire system and has comprehensive overvoltage, overcurrent, and short-circuit protection functions. It adopts a highly efficient and energy-saving design and supports remote status monitoring, enabling maintenance personnel to grasp system parameters in real time, quickly locate and troubleshoot faults. Furthermore, in extreme weather conditions, it can switch to the bridge's main power interface to ensure device reliability.

[0057] To achieve a feedback closed-loop mechanism, the adjustable windshield grille actuator module incorporates a position sensor (such as a Hall sensor) to monitor the actual rotor rotation angle and air permeability in real time, transmitting this data back to the control module via a wireless or wired interface (such as a CAN bus). The control module determines whether the adjustment is adequate based on the transmitted data (standards: angle error ≤ ±1°, air permeability deviation ≤ ±2%). If the standards are not met, a secondary adjustment command is triggered (e.g., checking every 5 seconds until the standards are met or the maximum number of attempts (3) is reached). This closed loop ensures system stability and accuracy.

Claims

1. An intelligent vibration suppression device for split-type box girder vortex-induced vibration, comprising a monitoring module, a control module, an adjustable windbreak grille execution module, and a power supply module, wherein the power supply module supplies power to the entire system, and the modules work together to actively suppress vortex-induced vibration; characterized in that: The monitoring module employs a distributed sensor array, with six three-dimensional ultrasonic anemometers and nine unidirectional accelerometers installed at 1 / 4, 1 / 2, and 3 / 4 of the main span of the box girder. These sensors collect wind field parameters and structural vibration response data. The three-dimensional ultrasonic anemometers have a wind speed measurement resolution of 0.01 m / s, an accuracy of ±1%, and an effective range of 0~40 m / s; and a wind direction resolution of 0.1° and an accuracy of ±2%. The control module is based on an embedded microcontroller with an ARM architecture. It has a built-in deep neural network pre-trained with wind tunnel test and CFD simulation data, integrates real-time spectrum analysis function, processes acceleration signals through fast Fourier transform algorithm to extract vibration characteristic frequencies, and can also receive data from monitoring module and calculate average wind speed, turbulence intensity and gust coefficient. The adjustable windbreak grille execution module is installed in the central slot of the box girder and includes four fixed grille plates. Each fixed grille plate is equipped with 10 rotatable rotors driven by embedded micro servo motors. The rotors can rotate precisely within the range of 0° to 90°. The execution module has a built-in position sensor to monitor the rotor rotation angle and air permeability in real time.

2. The intelligent vibration suppression device for split-type box girder vortex-induced vibration according to claim 1, characterized in that, The input layer data of the deep neural network includes average wind speed, triaxial turbulence intensity, triaxial gust coefficient, and vibration frequency. The hidden layer has 1 layer, 128 units, and the activation function is Tansig. The output layer activation function is Tansig. The training iterations are 1000 times. The Adam optimizer is used with a learning rate of 0.

001. It also has an online learning mechanism, which is triggered every 24 hours or when the vibration amplitude changes by ≥10%.

3. The intelligent vibration suppression device for vortex-induced vibration of a split-type box girder according to claim 1, characterized in that, In the adjustable windbreak grille execution module, the fixed grille width is 480mm, the spacing between adjacent grille panels is 800mm, the width of a single rotor is 200mm and the length is 400mm. The air permeability can be dynamically changed within the range of 48% to 62% under controlled conditions by adjusting the rotation of the rotor.

4. The intelligent vibration suppression device for vortex-induced vibration of a split box girder according to claim 1, characterized in that, All sensors in the monitoring module transmit the collected data to the control module wirelessly. The sensor arrangement meets the following requirements: 9 unidirectional acceleration sensors are set at 3 measuring points on each of the 1 / 4, 1 / 2, and 3 / 4 sections of the main span, with 2 on the west side measuring vertical and lateral acceleration respectively, and 1 on the east side measuring vertical acceleration; 6 three-dimensional ultrasonic anemometers are installed on both sides of the above sections, 6 meters above the bridge deck, with the bridge axis direction as the 0° reference, and the wind direction angle increasing in a clockwise direction.

5. A split box girder vortex-induced vibration control method based on the intelligent vibration suppression device according to any one of claims 1-4, characterized in that, Includes the following steps: S1: The monitoring module collects real-time data on wind speed, wind direction, and structural acceleration, and transmits it to the control module; S2: The control module processes the received data, calculates the triaxial turbulence intensity according to formula (1), calculates the triaxial gust coefficient according to formula (2), and obtains the vibration characteristic frequency by analyzing the acceleration signal with a time interval of 30 minutes using the fast Fourier transform algorithm; wherein, the formula for calculating the triaxial turbulence intensity is as shown in (1): wherein , are the along-wind, cross-wind and vertical turbulence intensities, respectively; is the three-dimensional standard deviation of the fluctuating wind speed; is the 10 min average wind speed; The formula for calculating the three-directional gust coefficient is shown in (2): In the formula, , , are the in-wind, cross-wind, and vertical gust coefficients, respectively; is the three-direction gust wind speed, and the gust time interval is 3s; S3: The control module inputs the key parameters obtained in step S2 into the deep neural network to generate the optimal ventilation rate adjustment command. The ventilation rate R is calculated as shown in (3): In the formula, b1 is the width of the fixed grid plate; is the number of fixed grid plates; is the number of closed rotors in the gap; is the width of a single rotor; is the length of a single rotor; d is the width of the central slot of the box girder; and l is the length of the gap. S4: The adjustable windshield grille execution module receives the adjustment command and drives the rotor to rotate to the target angle. The position sensor transmits the real-time monitored rotor angle and air permeability back to the control module. S5: The control module determines whether the adjustment meets the standard, which is that the angle error is ≤ ±1° and the air permeability deviation is ≤ ±2%. If the standard is not met, a secondary adjustment command is triggered every 5 seconds, with a maximum of 3 attempts, forming a feedback closed loop.

6. The method of claim 5, wherein the method further comprises: The CFD simulation data covers Reynolds numbers , Strouhal numbers St=0.1~0.3, and a nonlinear mapping relationship between the multi-dimensional wind field characteristics and the optimal wind permeability is established.

7. The method of claim 5, wherein the method further comprises: The online learning mechanism uses real-time data from the most recent 30 minutes for each update, with a sample size of 1000. It employs gradient descent to iterate the model parameters for 5 iterations and uses an early stopping mechanism to verify that the loss has not decreased for 3 consecutive iterations, thus avoiding overfitting.

Images