A method for online extraction of vehicle-induced over-bridge impulse signals from bridge structural response monitoring data

Abstract

The application discloses a method for extracting a vehicle-induced over-bridge pulse signal from bridge structure effect monitoring data online, adopts a sliding window to acquire a data segment online, pre-processes the data segment, carries out segmentation through a robust segmentation algorithm based on clustering stability, determines the optimal over-bridge pulse signal and the segmentation point of the no-load signal by carrying out multiple hierarchical clustering under different cluster numbers and automatically searching for a stable region where the segmentation result remains unchanged, and finally optimizes the segmentation result and outputs the same. The method has the advantages of no training, simple parameter setting, self-adaption to different bridges, sensors and noise working conditions, accurate and stable online extraction of the over-bridge pulse signal, and high engineering application value.

Description

Technical Field

[0001] This invention relates to the field of bridge structural health monitoring technology, and in particular to a method for online extraction of vehicle-induced bridge crossing pulse signals from bridge structural effect monitoring data. Background Technology

[0002] Bridge structural health monitoring systems generate massive amounts of continuous time-series monitoring data during operation, such as strain, displacement, and acceleration. In order to efficiently analyze the structural response of bridges under traffic loads, a key preliminary step is to accurately extract the response signals triggered by vehicles or trains passing through the continuous data stream, namely the "bridge crossing pulse signals".

[0003] The existing technology has the following main shortcomings: 1) Threshold method: This method distinguishes bridge-crossing pulse signals from background noise by manually setting one or more threshold values ​​for signal amplitude or energy. Although simple, the threshold setting heavily relies on engineering experience and has poor adaptability to different bridges, sensors, and environmental noise levels, making it prone to misjudgment or missed judgment. 2) Heuristic optimization algorithms: These algorithms use genetic algorithms, particle swarm optimization, etc., to search for and optimize the split points. These methods are computationally expensive, lack stability, and are prone to getting trapped in local optima. 3) Machine learning and deep learning methods: Temporal segmentation using methods such as Long Short-Term Memory (LSTM) networks and Hidden Markov Models (HMM). The performance of these methods is highly dependent on large-scale, high-quality labeled training datasets, which are expensive to obtain in practical engineering. Furthermore, the trained models have limited generalization ability to new working conditions, making it difficult to meet the ever-changing needs of online monitoring.

[0004] In summary, existing technologies lack a method for extracting bridge pulse signals that requires no large amount of training data, no complex parameter tuning, adapts to different working conditions, and is processed online. This is the area that this application aims to improve. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method for online extraction of vehicle-induced bridge crossing pulse signals from bridge structural effect monitoring data. This method requires no preset threshold, no sample training, has simple parameter settings, and is adaptive to different bridge, sensor, and noise conditions, thereby achieving accurate and stable online extraction of bridge crossing pulse signals from bridge time-series monitoring data.

[0006] To address the above technical problems, this invention provides a method for online extraction of vehicle-induced bridge crossing pulse signals from bridge structural effect monitoring data, comprising the following steps: Step S1: Data acquisition and preprocessing; Step S11: Deploy sensors on the bridge to collect real-time monitoring data of the bridge during vehicle / train passage; Step S12: Use a sliding window to obtain a fixed-length data segment from the continuously acquired raw time-series monitoring data stream as buffer data to be processed; Step S13: Preprocess the data in the buffer. Preprocessing includes, but is not limited to, filtering, detrending, signal envelope detection, or resampling to enhance signal characteristics and eliminate noise interference. Step S2: Pulse signal segmentation based on iterative clustering; This step is the core of the invention. By iteratively performing hierarchical clustering and analyzing the stability of its results, adaptive signal segmentation is achieved. Step S21: Within a preset cluster number range, perform multiple agglomerative hierarchical clustering operations on the data points in the buffer. Each clustering operation corresponds to a cluster number K, dividing all data points into K clusters. Step S22: For each K clusters obtained from clustering, based on preset physical characteristics, including signal energy, first-order rate of change of signal amplitude, and second-order rate of change of signal amplitude, the clusters are merged. The cluster with the smallest feature value is defined as the "empty signal cluster", and the remaining K-1 clusters are merged into the "pulse signal cluster", thus forming a binary classification segmentation result. Step S23: Identify stable regions where the binary classification segmentation results remain unchanged during the continuous change of the cluster number K, use the segmentation results corresponding to the stable regions as the segmentation scheme for the current buffer data, and determine the start and end points of the bridge pulse signal; Step S3: Segmentation result optimization and online output; Step S31: For highway bridges, a "coarse-to-fine" search strategy is used to optimize the segmentation points; the single pulse signal segment initially segmented in step S2 and its adjacent empty signal segment are merged, and step S2 is repeated on the merged data segment to obtain a more accurate segmentation boundary. Step S32: Remove pseudo-pulse segments; By analyzing the duration and fluctuation of all segmented pulse segments, density clustering method is used to identify and remove pseudo-pulse segments with short duration and small fluctuation. Step S33: Output all bridge pulse signal segments extracted in the current buffer, and move the sliding window to process new data, thereby realizing online processing of the entire data stream.

[0007] The beneficial effects of this invention are as follows: 1) Strong adaptability and robustness: This invention does not require pre-training of the model or setting a fixed segmentation threshold. It automatically determines the optimal segmentation by finding the "stable" state of the clustering results, and can adapt to different bridge types, sensor conditions, traffic loads and noise levels. 2) Online processing capability: The sliding window mechanism is adopted to distribute the computing tasks into the streaming processing process, which meets the real-time requirements of online monitoring of engineering projects; 3) High accuracy and stability: The segmentation algorithm based on cluster stability avoids the problem of traditional clustering requiring preset cluster number or distance threshold, and also overcomes the convergence problem of heuristic algorithms, and can stably and accurately identify signal boundaries; 4) Simple parameter setting: The core parameters of the whole method have clear physical meanings and are determined through simple trial calculations in the early stage. No human intervention is required afterward, thus realizing intelligent processing. Attached Figure Description

[0008] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart of a method for online extraction of vehicle-induced bridge crossing pulse signals according to a specific embodiment of the present invention; Figure 2 This is a flowchart of the segmentation algorithm in a specific embodiment of the present invention. Detailed Implementation

[0009] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0010] like Figure 1 As shown, this invention provides a method for online extraction of vehicle-induced bridge-crossing pulse signals from bridge structural effect monitoring data. This method is applicable to automatically and in real-time extracting signal segments generated when vehicles or trains pass through continuous data collected by various bridge monitoring sensors, such as strain gauges, accelerometers, and displacement gauges. The method includes the following specific steps: Step S1: Data acquisition and preprocessing; Sensors are deployed on the bridge to continuously collect real-time monitoring data. Online processing is performed using a sliding window approach. A fixed-size buffer is set to accommodate the length of 2 to 3 typical bridge-crossing pulse signals. Data continuously flows into the buffer, and a processing cycle is triggered when the buffer is full. To improve the accuracy of subsequent segmentation, the data segments in the current buffer are preprocessed as follows: 1) For quasi-static strain and displacement signals, digital filters are used to remove high-frequency noise and low-frequency trend terms; 2) For dynamic acceleration signals, in addition to filtering, envelope detection is also required to reduce the oscillation frequency of the signal and highlight the trend of energy change; 3) In order to unify the buffer length, signals with arbitrary sampling rates are unified to an appropriate frequency through resampling.

[0011] Step S2: Pulse signal segmentation based on iterative clustering; This is the core step in extracting the bridge pulse signal. First, a feature vector for clustering is calculated for each data point in the buffer. To balance physical meaning and computational efficiency, the signal's energy and amplitude change rate can be selected as features. Second, a stability-based iterative clustering process is performed, such as... Figure 2 As shown.

[0012] 1) Iterative Clustering: Define a range for the number of clusters to be examined, and gradually increase the number of clusters K starting from 2. For each value of K, perform agglomerative hierarchical clustering on all data points in the buffer based on their feature vectors, dividing them into K clusters; 2) Binary Classification and Merging: Calculate the feature centroids of K clusters. Label the cluster with the smallest feature as "empty signal," and merge the remaining K-1 clusters, labeling them as "pulse signal." For any value of K, this results in a segmentation that divides all data points into "empty" and "pulse" classes. 3) Stable Region Identification: Record the binary classification segmentation result corresponding to each K value. Observe the change in the segmentation result as K increases. When K reaches a certain value, it enters a "stable phase," at which point the segmentation result no longer changes with the increase of K. This invention automatically identifies this "stable phase" by setting a stability threshold M, that is, finding a cluster number K such that the binary classification segmentation results are completely identical in the M consecutive iterations from KM to K. During the iteration process, once a stable region is found, the iteration process is terminated, and the segmentation result is used as the temporal segmentation scheme for this buffer, thereby avoiding the problem of preset cluster number or threshold.

[0013] Step S3: Segmentation result optimization and online output; 1) "Coarse-to-fine" search: Applicable to highway bridges. First, perform the iterative clustering algorithm described above once to obtain the "coarse" segmentation result. Then, merge each identified pulse signal segment with its temporally adjacent idle signal segment into a new, smaller data segment. Perform step S2 again on this new data segment to perform a "fine" search, thereby more accurately locating the start and end boundaries of the pulse signal; 2) Pseudo-pulse removal: During signal acquisition, sensor jumps or electromagnetic interference may generate short-lived pseudo-pulses. By analyzing the "duration" and "standard deviation (fluctuation level)" of all segmented pulse segments, the DBSCAN density clustering algorithm is used to remove pulse segments with short duration and small fluctuations. 3) Output and Sliding: After completing all the above steps, output the final determined bridge pulse signal segment in the buffer for subsequent structural analysis. Then, move the sliding window forward by one step (the step size is set to accommodate the duration of three typical bridge pulses) and wait to process the next batch of data.

[0014] By cyclically executing the above steps, this invention achieves online, adaptive, and high-precision extraction of bridge crossing pulse signals from bridge monitoring data streams, solving the problems of existing technologies that rely on manual thresholds or pre-trained models and have poor adaptability.

[0015] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for online extraction of vehicle-induced bridge crossing pulse signals from bridge structural effect monitoring data, characterized in that: Includes the following steps: Step S1: Use a sliding window to segment and preprocess the continuously collected time-series monitoring data to obtain the data in the buffer. Step S2, segmenting the data in the buffer to distinguish between bridge pulse signal segments and idle signal segments, the segmentation step including: (i) For the data in the buffer, gradually increase the preset number of clusters K and perform multiple hierarchical clustering to obtain K clusters; (ii) Based on the magnitude of the preset features, the K clusters are merged into an empty signal cluster and a pulse signal cluster to form a binary classification segmentation result; (iii) Identify the region where the binary classification segmentation result remains stable during the change of the number of clusters K, and use the segmentation result in this region as the temporal segmentation scheme; Step S3: Output the extracted bridge pulse signal and move the sliding window to process new timing monitoring data.

2. The method for online extraction of vehicle-induced bridge crossing pulse signals from bridge structural effect monitoring data according to claim 1, characterized in that: Before step S1, a preprocessing step is included for the raw time-series monitoring data, which includes at least two of the following: filtering, signal envelope detection, and resampling.

3. The method for online extraction of vehicle-induced bridge crossing pulse signals from bridge structural effect monitoring data according to claim 1, characterized in that: The stable region identification in step S2(iii) is achieved by finding a cluster number K such that the binary classification segmentation result remains unchanged within the cluster number range from KM to K, where M is a preset stability threshold.

4. The method for online extraction of vehicle-induced bridge crossing pulse signals from bridge structural effect monitoring data according to claim 1, characterized in that: The preset features used for classification in step S2(ii) include signal energy, first-order rate of change of signal amplitude, and second-order rate of change of signal amplitude.

5. The method for online extraction of vehicle-induced bridge crossing pulse signals from bridge structural effect monitoring data according to claim 1, characterized in that: After step S2 and before step S3, there is a step of optimizing the segmentation result using a coarse-to-fine search strategy. The strategy includes merging the initially segmented pulse signal segments and adjacent empty signal segments, and performing secondary segmentation on them.

6. The method for online extraction of vehicle-induced bridge crossing pulse signals from bridge structural effect monitoring data according to claim 1, characterized in that: After step S2 and before step S3, a step of removing pseudo-pulse segments is also included.

7. The method for online extraction of vehicle-induced bridge crossing pulse signals from bridge structural effect monitoring data according to claim 6, characterized in that: The step of eliminating pseudo-pulse segments is achieved by analyzing the duration and fluctuation of the pulse segments and using a density-based clustering method.

Images