Domain-Adaptive Intelligent Identification and Early Control Method, Device, and Storage Medium for Vortex-Induced Vibration of Bridge Main Girder
By extracting the characteristics of vortex-induced vibration of bridges using a domain-adaptive method, training an identification model, and establishing a strategy library, the problem of identifying and controlling vortex-induced vibration of long-span bridges was solved, achieving real-time identification and efficient early control.
Patent Information
- Application Number
- CN202410435547.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-11
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2044-04-11
AI Technical Summary
Existing technologies struggle to identify vortex-induced vibrations on long-span bridges lacking historical vortex-induced vibration data, and there is a lack of effective intelligent active/semi-active early control methods.
A domain-adaptive approach is adopted to extract vortex-induced vibration features by acquiring sensor data, train a classifier using the equilibrium assignment adaptation method, establish an identification model, and perform early control based on a policy library.
It enables real-time identification of vortex-induced vibration without requiring historical data of the target bridge, and is applicable to multiple bridges, improving control efficiency and accuracy.
Smart Images

Figure CN118395253B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of structural health monitoring and intelligent control, and in particular to a method, device and storage medium for intelligent identification and early control of vortex-induced vibration of bridge main beams based on domain adaptation. Background Technology
[0002] Machine learning algorithms can sensitively identify vortex-induced vibrations in the large amount of historical data accumulated over many years in structural health monitoring (SHM) systems, and are gradually becoming an effective method for vortex-induced vibration identification. However, current identification methods still have certain limitations, mainly for the following reasons: (1) Historical vortex-induced vibration data of the target bridge is needed to determine the model or parameters in the vortex-induced vibration identification algorithm, but many long-span bridges have not experienced vortex-induced vibrations in the past or the vortex-induced vibration data has not been recorded, so it is difficult to determine the model or parameters in the algorithm; (2) Some clustering-based vortex-induced vibration identification algorithms need to be combined with manual judgment to identify vortex-induced vibrations in the large amount of historical data accumulated over several years in the SHM system, and it is difficult to achieve real-time identification of vortex-induced vibrations in the SHM system; (3) Vortex-induced vibration identification algorithms are difficult to apply to multiple bridges of the same type at the same time, and identification algorithms need to be constructed separately for each long-span bridge.
[0003] In addition, vortex-induced vibration is long-lasting and highly hazardous, requiring control and intervention. However, current control methods have certain limitations. The main reason is that current research on various aerodynamic measures for controlling vortex-induced vibration of bridges still relies on wind tunnel tests to analyze specific bridges, without explaining the underlying principles. Furthermore, most measures are still passive control measures, with a single and inefficient control mode, and a lack of targeted intelligent active / semi-active online early control methods. Summary of the Invention
[0004] The purpose of this invention is to provide a method, device, and storage medium for intelligent identification and early control of vortex-induced vibration of bridge main beams based on domain adaptation.
[0005] The objective of this invention can be achieved through the following technical solutions:
[0006] A domain-adaptive intelligent identification and early control method for vortex-induced vibration of bridge main beams includes:
[0007] Step S1: Acquire the acceleration data of the target bridge collected by multiple sensors within the first time interval, and extract the vortex-induced vibration characteristics of the target bridge.
[0008] Step S2: Adaptively align the source domain and target domain based on the domain, where the target domain is the vortex-induced vibration characteristics of the target bridge, and the source domain is the vortex-induced vibration characteristics of other bridges that have undergone vortex-induced vibration in the same way.
[0009] Step S3: Use the aligned source domain to train the recognition model;
[0010] Step S4: Input the vortex-induced vibration characteristics of the target bridge into the trained recognition model to obtain the vortex-induced vibration state of the target bridge;
[0011] Step S5: When the obtained vortex-induced vibration state is vortex-induced vibration, a control scheme is matched in the strategy library based on the bridge vibration characteristics, and early control is performed based on the matched control scheme.
[0012] The vortex-induced vibration characteristics include statistical characteristics and frequency domain characteristics. The statistical characteristics include the root square amplitude of acceleration R. ms Standard deviation V and Hopkins statistic H,
[0013] The root-square magnitude of the acceleration R ms The mathematical expression is:
[0014]
[0015] Where: n is the number of signal sampling points, y i For the acceleration data of the i-th sampling point,
[0016] The mathematical expression for the standard deviation V is:
[0017]
[0018] in: This represents the average value of the vibration signal.
[0019] The process of obtaining the Hopkins statistic H includes:
[0020] Step S1-1-1: Randomly select a portion of all signal sampling points as random sampling points;
[0021] Step S1-1-2: For each random sampling point, find the point closest to the random sampling point in the sample space, and take the distance between the found nearest point and the random sampling point as the minimum distance corresponding to the random sampling point;
[0022] Step S1-1-3: Randomly generate multiple virtual points within the possible range of the samples, wherein the number of virtual points is the same as the number of random sampling points;
[0023] Step S1-1-4: In the sample space, find the nearest point to each virtual point, and take the distance between the found point and the virtual point as the minimum distance corresponding to the virtual point;
[0024] Step S1-1-5: Obtain the Hopkins statistic H:
[0025]
[0026] Where: K is the number of random sampling points, x k For the minimum distance corresponding to the k-th random sampling point, l k This is the minimum distance corresponding to the k-th virtual point.
[0027] The frequency domain characteristics include the dominant frequency PSD amplitude P, the energy concentration factor R, and the relative change in the dominant frequency D. w The energy concentration factor R is the ratio of the amplitude of the main frequency PSD to the amplitude of the second main frequency PSD.
[0028] The relative change in the main frequency D w The acquisition process specifically includes:
[0029] Step S1-2-1: Divide the acceleration data within the first time interval into two segments;
[0030] Step S1-2-2: Divide the difference between the main frequency of the previous data segment and the main frequency of the next data segment by the average of the main frequencies of the previous and next data segments, and use the result as the relative change in main frequency D. w .
[0031] The length of the first time interval is ten minutes.
[0032] In step S2, feature transformation and transfer are performed using the balanced assignment adaptation method. A classifier is trained using source domain data and labels. The trained classifier is then directly applied to target domain data to obtain pseudo-labels for the target domain.
[0033] The process of establishing the strategy library includes:
[0034] Multimodal vortex-induced vibration analysis was performed on the target bridge to obtain the bridge vibration characteristics and the degree of damage to the bridge under different modes of vortex-induced vibration.
[0035] Based on the existing active and semi-active controllers for bridges, numerical simulation analysis is used to optimize the control parameters for the vibration characteristics of each bridge, thus forming the optimal control strategy under this working condition.
[0036] Step S5 specifically includes:
[0037] Step S5-1: When the obtained vortex-induced vibration state is vortex-induced vibration, extract the real-time bridge vibration characteristics, wherein the bridge vibration characteristics include vortex-induced vibration frequency domain and time domain amplitude parameters.
[0038] Step S5-2: Based on the obtained real-time bridge vibration characteristics, match control schemes in the strategy library, and perform early control based on the matched control schemes.
[0039] A domain-adaptive intelligent identification and early control device for vortex-induced vibration of bridge main beams includes a memory, a processor, and a program stored in the memory. When the processor executes the program, it implements the method described above.
[0040] A storage medium having a program stored thereon, which, when executed, implements the method described above.
[0041] Compared with existing technologies, this invention has the following advantages: it eliminates the need for target bridge vortex-induced vibration data, enabling real-time or historical dataset identification of the initial stage of vortex-induced vibration, and is applicable to multiple long-span bridges. The control method in this invention can automatically perform active / semi-active precise control of vortex-induced vibration in its early stages, significantly improving control efficiency and accuracy. Attached Figure Description
[0042] Figure 1 This is a flowchart illustrating the main steps of the present invention;
[0043] Figure 2 This is a schematic diagram of the technical route of the present invention. Detailed Implementation
[0044] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.
[0045] A domain-adaptive intelligent identification and early control method for vortex-induced vibration of bridge main beams, such as... Figure 1 and Figure 2 As shown, it includes:
[0046] Step S1: Acquire the acceleration data of the target bridge collected by multiple sensors within the first time interval, and extract the vortex-induced vibration characteristics of the target bridge. Specifically, the signal is continuously collected, but this application uses a sliding window to analyze the signal for a period of time (such as ten minutes, which is also the width of the window) to determine whether vortex-induced vibration has occurred. Then, after sliding through a short period of time (such as 1 minute), the signal of the window width time is captured again for analysis to determine whether vortex-induced vibration has occurred.
[0047] Specifically, in this embodiment, a monitoring dataset based on n sensors is constructed, and an m×n dimensional variable is defined, where m represents the dataset length, i.e., the number of samplings; X ji Let represent the j-th sampling point of the i-th sensor; X is an n-dimensional random variable composed of n sensors, denoted as:
[0048]
[0049] X i =(X 1i ,X 2i , ...X mi ) T
[0050] Where: X i Let the data set be a one-dimensional variable for the i-th sensor. Then, by processing the data in the order of sensor 1 to n, we can transform it into a one-dimensional variable with m × n data points.
[0051] D = (X1, X2, ... X n ) T
[0052] Where: D represents the preprocessed dataset.
[0053] Let the frequency of data acquisition be f. s The time interval for collecting data is Δt = 1 / f s The total acquisition time for each sensor is T = n × Δt, and the data visualization time is defined as t;
[0054] The preprocessed dataset D, consisting of n sensors, is divided into... Each data segment is divided into several data points, and each sensor data point is further divided into... A data segment;
[0055] The preprocessed dataset D obtained above includes N data segments; the six features in the preprocessed dataset D are calculated using the Tsfresh built-in functions.
[0056] Eddy-induced vibration characteristics include statistical characteristics and frequency domain characteristics. Statistical characteristics include the root square amplitude of acceleration R. ms Standard deviation V and Hopkins statistic H,
[0057] Root square magnitude of acceleration R ms The mathematical expression is:
[0058]
[0059] Where: n is the number of signal sampling points, y i For the acceleration data of the i-th sampling point,
[0060] The mathematical expression for the standard deviation V is:
[0061]
[0062] in: This represents the average value of the vibration signal.
[0063] The process of obtaining the Hopkins statistic H includes:
[0064] Step S1-1-1: Randomly select a portion of all signal sampling points as random sampling points;
[0065] Step S1-1-2: For each random sampling point, find the point closest to the random sampling point in the sample space, and take the distance between the found nearest point and the random sampling point as the minimum distance corresponding to the random sampling point. Here, the sample space is the acceleration data of all sampling points.
[0066] Step S1-1-3: Randomly generate multiple virtual points within the possible range of the samples, wherein the number of virtual points is the same as the number of random sampling points;
[0067] Step S1-1-4: In the sample space, find the nearest point to each virtual point, and take the distance between the found point and the virtual point as the minimum distance corresponding to the virtual point;
[0068] Step S1-1-5: Obtain the Hopkins statistic H:
[0069]
[0070] Where: K is the number of random sampling points, x k For the minimum distance corresponding to the k-th random sampling point, l k This is the minimum distance corresponding to the k-th virtual point.
[0071] Frequency domain characteristics include the dominant frequency PSD amplitude P, the energy concentration factor R, and the relative change in dominant frequency D. w The energy concentration factor R is the ratio of the amplitude of the primary frequency PSD to the amplitude of the second primary frequency PSD.
[0072] Furthermore, frequency lock-in occurs during vortex-induced vibration, meaning that the large-amplitude vibration of the bridge, in turn, affects the frequency of vortex shedding, so that variations in wind speed within a certain range do not change the frequency of the bridge's vortex-induced vibration. Therefore, this application introduces the relative change in dominant frequency, D. w The relative change in the main frequency D w The acquisition process specifically includes:
[0073] Step S1-2-1: Divide the acceleration data within the first time interval into two segments;
[0074] Step S1-2-2: Divide the difference between the main frequency of the previous data segment and the main frequency of the next data segment by the average of the main frequencies of the previous and next data segments, and use the result as the relative change in main frequency D. w .
[0075] In this embodiment, the length of the first time interval is ten minutes.
[0076] Finally, by calculating φ statistical features for each of the N data segments using a function, the feature dataset is denoted as:
[0077] F = (F1, F2, ..., F φ )
[0078] Where F φ Let F represent the φ-th feature column of the preprocessed dataset D, and let F represent the N×φ-dimensional feature data of dataset D.
[0079] Step S2: Adaptively align the source domain and target domain based on the domain, where the target domain is the vortex-induced vibration characteristics of the target bridge, and the source domain is the vortex-induced vibration characteristics of other bridges that have undergone vortex-induced vibration in the same way.
[0080] In this method, feature transformation and transfer are performed using the balanced assignment adaptation method. A classifier is trained using source domain data and labels. The trained classifier is then directly applied to target domain data to obtain pseudo-labels for the target domain.
[0081] Balanced Distribution Adaptation (BDA) is a feature transformation transfer method that improves upon two transfer learning methods, Transfer Component Analysis (TCA) and Joint Distribution Adaptation (JDA), to achieve better results.
[0082] BDA introduces a balancing factor μ to further discuss the impact of different proportions of conditional and marginal distributions in the final optimization objective on the accuracy of the classification task. The results show that when the source and target domain datasets differ significantly, the marginal distribution should dominate, and μ approaching 0 yields better results; when the datasets are relatively similar, the conditional distribution dominates, and μ approaches 1. The corrected optimization objective is shown in the following equation.
[0083]
[0084]
[0085] BDA proposed using the A-distance index, i.e., formula (9), to estimate the balance factor μ and achieved good results, in which... and , where are the samples and labels of the source and target domains, respectively; h is a linear classifier trained using the source and target domain data; and ∈(h) is the error rate of this linear classifier in the source and target domains. The balance factor can be estimated using equation (10), where d M =d A , and These are the sets of c-th class samples and labels, representing the source and target domains, respectively.
[0086]
[0087]
[0088] BDA first trains a classifier using source domain data and labels, and then applies this classifier directly to the target domain data to obtain pseudo-labels for the target domain.
[0089] Step S3: Use the aligned source domain to train the recognition model;
[0090] Step S4: Input the vortex-induced vibration characteristics of the target bridge into the trained recognition model to obtain the vortex-induced vibration state of the target bridge;
[0091] The process of building a strategy library includes:
[0092] Multimodal vortex-induced vibration analysis was performed on the target bridge to obtain the bridge vibration characteristics and the degree of damage to the bridge under different modes of vortex-induced vibration.
[0093] Based on the existing active and semi-active controllers for bridges, numerical simulation analysis is used to optimize the control parameters for the vibration characteristics of each bridge, thus forming the optimal control strategy under this working condition.
[0094] Step S5: When the obtained vortex-induced vibration state is vortex-induced vibration, a control scheme is matched in the strategy library based on the bridge vibration characteristics, and early control is performed based on the matched control scheme, specifically including:
[0095] Step S5-1: When the obtained vortex-induced vibration state is vortex-induced vibration, extract the real-time bridge vibration characteristics, which include the frequency domain and time domain amplitude parameters of vortex-induced vibration.
[0096] Step S5-2: Based on the obtained real-time bridge vibration characteristics, match control schemes in the strategy library, and perform early control based on the matched control schemes.
[0097] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
Claims
1. A method for bridge girder vortex-induced vibration intelligent identification and early control based on domain adaptation, characterized in that, include: Step S1: Acquire the acceleration data of the target bridge collected by multiple sensors within the first time interval, and extract the vortex-induced vibration characteristics of the target bridge. Step S2: Adaptively align the source domain and target domain based on the domain, where the target domain is the vortex-induced vibration characteristics of the target bridge, and the source domain is the vortex-induced vibration characteristics of other bridges that have undergone vortex-induced vibration in the same way. Step S3: Use the aligned source domain to train the recognition model; Step S4: Input the vortex-induced vibration characteristics of the target bridge into the trained recognition model to obtain the vortex-induced vibration state of the target bridge; Step S5: When the obtained vortex-induced vibration state is vortex-induced vibration, a control scheme is matched in the strategy library based on the bridge vibration characteristics, and early control is performed based on the matched control scheme. In step S2, feature transformation and transfer are performed using the balanced assignment adaptation method. A classifier is trained using source domain data and labels. The trained classifier is then directly applied to target domain data to obtain pseudo-labels for the target domain. The balanced allocation adaptation method introduces a balance factor μ It is found that the edge distribution dominates when the difference between the source and target domains is large, in which case μ is close to 0; the conditional distribution dominates when the data sets are similar, in which case μ is close to 1. Balanced allocation adaptation method to balance factor μ Estimation: Where: d A The A-distance index, The error rate of the linear classifier in the source and target domains. Let c be the set of samples and labels of class c in the source domain. Let c be the set of samples and labels of the c-th class in the target domain.
2. The intelligent identification and early control method for vortex-induced vibration of bridge main beams based on domain adaptation as described in claim 1, characterized in that, The vortex-induced vibration characteristics include statistical characteristics and frequency domain characteristics. The statistical characteristics include the root square amplitude of acceleration. R ms Standard deviation V Hopkins statistic H , The root square magnitude of acceleration R ms The mathematical expression is: in: n The number of signal sampling points. y i For the first i Acceleration data at each sampling point The standard deviation V The mathematical expression is: in: This represents the average value of the vibration signal.
3. The intelligent identification and early control method for vortex-induced vibration of bridge main beams based on domain adaptation as described in claim 2, characterized in that, The Hopkins statistic H The acquisition process includes: Step S1-1-1: Randomly select a portion of all signal sampling points as random sampling points; Step S1-1-2: For each random sampling point, find the point closest to the random sampling point in the sample space, and take the distance between the found nearest point and the random sampling point as the minimum distance corresponding to the random sampling point; Step S1-1-3: Randomly generate multiple virtual points within the possible range of the samples, wherein the number of virtual points is the same as the number of random sampling points; Step S1-1-4: In the sample space, find the nearest point to each virtual point, and take the distance between the found point and the virtual point as the minimum distance corresponding to the virtual point; Step S1-1-5: Obtain the Hopkins statistic. H : in: K The number of random sampling points. x k For the corresponding to the first k The minimum distance between random sampling points l k For the corresponding to the first k The minimum distance between virtual points.
4. The intelligent identification and early control method for vortex-induced vibration of bridge main beams based on domain adaptation according to claim 2, characterized in that, The frequency domain features include the dominant frequency PSD amplitude. P Energy Concentration Coefficient R The relative change in frequency D w The energy concentration factor R is the amplitude of the main frequency PSD. P The ratio of the second main frequency PSD amplitude The main frequency relative change amount D w The acquisition process specifically includes: Step S1-2-1: Divide the acceleration data within the first time interval into two segments; Step S1-2-2: The difference between the main frequency of the previous data and the main frequency of the subsequent data is divided by the average of the main frequencies of the previous data and the subsequent data, and the result is taken as the main frequency relative variation D w .
5. The intelligent identification and early control method for vortex-induced vibration of bridge main beams based on domain adaptation according to claim 1, characterized in that, The length of the first time interval is ten minutes.
6. The intelligent identification and early control method for vortex-induced vibration of bridge main beams based on domain adaptation according to claim 1, characterized in that, The process of establishing the strategy library includes: Multimodal vortex-induced vibration analysis was performed on the target bridge to obtain the bridge vibration characteristics and the degree of damage to the bridge under different modes of vortex-induced vibration. Based on the existing active and semi-active controllers for bridges, numerical simulation analysis is used to optimize the control parameters for the vibration characteristics of each bridge, thus forming the optimal control strategy under this working condition.
7. The intelligent identification and early control method for vortex-induced vibration of bridge main beams based on domain adaptation as described in claim 1, characterized in that, Step S5 specifically includes: Step S5-1: When the obtained vortex-induced vibration state is vortex-induced vibration, extract the real-time bridge vibration characteristics, wherein the bridge vibration characteristics include vortex-induced vibration frequency domain and time domain amplitude parameters. Step S5-2: Based on the obtained real-time bridge vibration characteristics, match control schemes in the strategy library, and perform early control based on the matched control schemes.
8. A domain-adaptive intelligent identification and early control device for vortex-induced vibration of bridge main girder, comprising a memory, a processor, and a program stored in the memory, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1-7.
9. A storage medium having a program stored thereon, characterized in that, When the program is executed, it implements the method as described in any one of claims 1-7.
