An adaptive online clipping method for bridge high-frequency vehicle-induced response

Abstract

The application discloses a kind of bridge high-frequency vehicle actuation response's adaptive online intercepting method, it is related to bridge health monitoring technical field, first using smooth iteration idea total strain data is initially segmented into pseudo temperature strain data section and pseudo vehicle-induced strain data section, subsequently vehicle-induced strain data section in pseudo temperature strain data section is correctly classified, finally according to the short time range characteristics of temperature strain data section in pseudo vehicle-induced strain data section, temperature strain data section is eliminated from pseudo vehicle-induced strain data section by range-based screening method;This method can significantly reduce the amount of vehicle-induced strain data, effectively reduce the burden of data storage and transmission, show strong anti-noise ability under high-frequency noise environment, can maintain high segmentation accuracy, the accuracy of vehicle-induced strain extraction is significantly improved, the storage and real-time processing problem in large data monitoring is solved, and effective support is provided for the sustainable operation of bridge health monitoring system.

Description

Technical Field

[0001] This invention belongs to the field of bridge health monitoring technology, specifically relating to a data processing method that can adaptively extract vehicle-induced strain data segments in real time. Background Technology

[0002] In recent years, lightweight monitoring of small- and medium-span bridges has become a development trend. Compared to long-span bridges, small- and medium-span bridges experience more frequent stress changes due to their unique structural characteristics, thus requiring higher frequency strain monitoring and resulting in a significant increase in monitoring data. Vehicle-induced strain can be used to identify localized damage such as bridge cracks, fatigue, and corrosion, and after eliminating the influence of temperature, it can more accurately reflect the stress state of the bridge. In addition, vehicle-induced strain can be used to assess the impact of different traffic loads on bridges, providing a basis for bridge design and traffic management. In high-frequency monitoring, vehicle-induced strain is the most valuable data for analysis, while the effect of strain at other times is limited. Furthermore, temperature-induced strain is often much larger than vehicle-induced strain, sometimes even ten times higher, which may mask the true impact of vehicle loads. Therefore, reducing data redundancy, improving the efficiency of high-frequency monitoring data processing and storage, and accurately extracting vehicle-induced strain are key to lightweight monitoring of small- and medium-span bridges.

[0003] To effectively eliminate temperature strain, researchers have proposed various methods over the past few decades, mainly falling into two categories. One category is based on mathematical models of the relationship between temperature and strain. This method analyzes the correlation between temperature and strain, calculates the strain caused by temperature changes using measured temperature data, and removes the influence of temperature from the measured strain, thus achieving separation of temperature strain from strain caused by vehicle loads. For example, some researchers use regression models combined with distributed temperature data to predict thermal response; others use discretized thermal loads and multivariate linear modeling to efficiently separate the temperature effect in the response of cable-stayed bridges, solving the problems of complex calculations and reliance on long-term data in traditional methods; still others have proposed improving the Bayesian dynamic linear model by incorporating an autoregressive component, building upon the traditional model's consideration of trend, seasonal, and regression components. This improved Bayesian dynamic linear model can capture the relationship between temperature and temperature-induced strain, and it can simultaneously adapt to stationary and non-stationary time series data, describing the time-varying structural strain response by invoking different hidden components. However, although statistical models are convenient for separating temperature effects, they are often complicated and sometimes even have accuracy deviations because they require the collection of massive amounts of temperature and strain data and the precise construction of functional relationships between parameters.

[0004] Another type of approach is based on signal decomposition techniques, such as wavelet transform and mode decomposition, which effectively separate dynamic and static strains through multi-scale signal decomposition. For example, some researchers have used wavelet transform to identify temperature-induced beam deflection and thus achieve response separation; others have used wavelet transform to decompose displacement data into three sub-components: temperature fluctuation, quasi-static effects, and dynamic factors, thereby separating the thermally induced displacement component; still others have proposed an enhanced empirical mode decomposition method combining empirical mode decomposition, Hilbert marginal spectrum, and Gaussian mixture model clustering, achieving real-time decoupling of temperature and vehicle strain with high intelligence and separation accuracy; some researchers have even solved the problems of mode aliasing and endpoint effects through signal reconstruction and proposed a temperature-induced deflection separation method, significantly improving separation accuracy. Existing separation methods based on empirical mode decomposition can already achieve automatic separation of vehicle-induced strain and temperature strain, but pseudo-eigenmode functions are generated during the empirical mode decomposition process, increasing the difficulty of dynamic-static response separation.

[0005] However, with the continuous development of bridge health monitoring technology, the amount of data generated by monitoring systems has increased dramatically, especially in large-scale bridge monitoring systems. This massive data volume presents serious challenges to the storage, processing, and analysis of health monitoring data. Long-term monitoring generates huge amounts of raw strain data, and how to efficiently store, manage, and back up this data has become a problem to be solved. To address this challenge, compressed sensing has emerged. It effectively reduces the burden of data storage and transmission by reducing the amount of sampled data while retaining the main characteristics of the signal. Some researchers have explored the application of compressed sensing in vibration data compression in bridge structural health monitoring, simulated compressed sensing sensor algorithms, and verified its data compression capabilities with actual data. Other researchers have applied compressed sensing to long-term monitoring with wireless sensors, demonstrating that this technology can reduce power consumption and data transmission volume. Some researchers have even proposed a bridge health monitoring method based on compressed sensing, which reduces the amount of data collected and improves signal reconstruction accuracy by optimizing the observation matrix. While compressed sensing can effectively reduce the amount of data collected, its drawback lies in the need for a complex signal reconstruction process, which may affect real-time performance, especially potentially causing delays in computationally demanding applications. Furthermore, compressed sensing relies on the sparsity of the signal. For some complex or noisy signals, it may not be possible to fully recover the original information, which may result in the loss of some details in the signal and affect the reconstruction accuracy.

[0006] In view of the above, this paper proposes an adaptive online truncation method for the high-frequency vehicle actuation response of bridges to solve the above problems. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this study proposes an adaptive online interception method for high-frequency vehicle-induced strain in bridges. This method aims to reduce the amount of data and effectively extract vehicle-induced strain. It is real-time, adaptive, and easy to embed into existing monitoring systems. In high-frequency monitoring data environments, it can achieve efficient real-time extraction of vehicle-induced strain by employing adaptive iteration and data segment post-processing techniques.

[0008] An adaptive online truncation method for high-frequency vehicle actuation response of bridges, characterized by comprising the following steps:

[0009] S1: First, based on the characteristic of large short-term fluctuations in vehicle-induced strain, the total strain data is initially divided into pseudo-temperature strain data segment and pseudo-vehicle-induced strain data segment using the idea of ​​smooth iteration.

[0010] S2: Subsequently, based on the length characteristics of the fragmented vehicle strain data segments, the vehicle strain data segments in the pseudo temperature strain data segments in S1 are correctly classified by the threshold-based length segmentation method.

[0011] S3: Finally, based on the short-time range characteristics of the temperature strain data segment in the pseudo-vehicle strain data segment, the temperature strain data segment is removed from the pseudo-vehicle strain data segment in S1 by using a range-based screening method.

[0012] The beneficial effects of this technical solution are as follows:

[0013] (1) This method uses the smoothed mean to iteratively segment the strain data, initially separating the pseudo temperature strain data segment and the pseudo vehicle strain data segment. Since the pseudo temperature strain data segment contains fragmented vehicle strain data segments and the pseudo vehicle strain data segment contains temperature strain data segments with small ranges, the vehicle strain data in the pseudo temperature strain data segment and the pseudo vehicle strain data segment are extracted respectively. Finally, the online real-time extraction of high-frequency sampled vehicle strain data is successfully realized. It can run stably within the set time window and meet the technical requirements of the bridge health monitoring system for real-time response.

[0014] (2) In terms of data volume optimization, this method compresses the original data volume by more than 50% through effective data processing strategies, which significantly reduces storage and transmission costs, and is especially suitable for the long-term operation of high-frequency lightweight monitoring systems for medium and short span bridges.

[0015] (3) This method is designed to address interference issues in complex engineering environments. It can maintain stable cutting capabilities even under noise interference in the range of 60 to 50 dB. Even under conditions of significant high-frequency noise interference, it can still effectively distinguish between vehicle-induced strain and temperature strain components, demonstrating excellent anti-interference capabilities and environmental adaptability.

[0016] (4) This method can also accurately extract the strain response caused by vehicle load and compress the amount of data in the monitoring system in real time in actual bridge monitoring. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall process of the segmentation method of the present invention;

[0018] Figure 2 This is a schematic diagram of the geometric structure of the T-beam;

[0019] Figure 3 This is a schematic diagram of the temperature variation curve throughout the year in an embodiment of the present invention;

[0020] Figure 4 This is a schematic diagram of the vehicle weight change curve and the resulting strain when the vehicle passes over the T-beam in an embodiment of the present invention;

[0021] Figure 5 This is a schematic diagram of the temperature strain curve throughout the year in an embodiment of the present invention;

[0022] Figure 6 This is a schematic diagram of the vehicle-induced strain curves throughout the year in an embodiment of the present invention;

[0023] Figure 7 This is a schematic diagram of the total strain curve with noise throughout the year in an embodiment of the present invention;

[0024] Figure 8 This is a schematic diagram of the Allan variance calculation results of this invention;

[0025] Figure 9 This is a schematic diagram illustrating the initial smoothed mean and the results of iterations 1, 2, and 6 in an embodiment of the present invention;

[0026] Figure 10 This is a schematic diagram of the pseudo-temperature strain data segment after re-classification of vehicle-induced strain in an embodiment of the present invention;

[0027] Figure 11 This is a schematic diagram of the pseudo-vehicle-induced strain data segment after removing temperature strain in an embodiment of the present invention;

[0028] Figure 12 This is a schematic diagram of the cutting results within a 15-minute time period in an embodiment of the present invention;

[0029] Figure 13 This is a schematic diagram of the strain segmentation results throughout the year in an embodiment of the present invention;

[0030] Figure 14 This is a schematic diagram of the strain portion cutting results for various noise values ​​throughout the year in an embodiment of the present invention;

[0031] Figure 15 This is a schematic diagram of the sensor measurement point layout in an embodiment of the present invention;

[0032] Figure 16 This is a schematic diagram of the MR-2-S strain map at the measuring point in an embodiment of the present invention;

[0033] Figure 17 This is a schematic diagram of the final iteration result in an embodiment of the present invention;

[0034] Figure 18 This is a schematic diagram of pseudo-temperature strain data and pseudo-vehicle-induced strain data in an embodiment of the present invention;

[0035] Figure 19 This is a schematic diagram of the final cutting result temperature strain data and vehicle-induced strain data in an embodiment of the present invention;

[0036] Figure 20 This is a schematic diagram of the strain cutting effect at measuring point MR-2-S in an embodiment of the present invention;

[0037] Figure 21 This is a schematic diagram of the zero-strain result at measuring point MR-2-S in an embodiment of the present invention. Detailed Implementation

[0038] The foregoing and other technical contents, features and effects of the present invention are described below with reference to the appendix to this application. Figures 1-21 As will be clearly shown in the detailed description of the embodiments, the methods mentioned in the following embodiments are all based on the accompanying drawings.

[0039] Example 1: Establishing a method for real-time acquisition of vehicle-induced strain, such as... Figure 1 The overall process consists of three main steps: First, leveraging the large short-term fluctuations in vehicle-induced strain, a smooth iteration approach is used to initially segment the total strain data into pseudo-temperature strain data segments and pseudo-vehicle-induced strain data segments. Theoretically, the temperature strain data segments in the pseudo-temperature strain data segments are caused solely by temperature changes, while the vehicle-induced strain data segments in the pseudo-vehicle-induced strain data segments are caused by the combined effects of vehicle load and temperature. However, in practice, the resulting pseudo-temperature strain data segments contain fragmented vehicle-induced strain data segments, while the pseudo-vehicle-induced strain data segments are interspersed with temperature strain data segments with small short-term ranges. Subsequently, based on the length characteristics of the fragmented vehicle-induced strain data segments, a threshold-based length segmentation method is used to correctly classify the vehicle-induced strain data segments in the pseudo-temperature strain data segments. Finally, based on the short-term range characteristics of the temperature strain data segments in the pseudo-vehicle-induced strain data segments, a range-based filtering method is used to remove temperature strain data segments from the pseudo-vehicle-induced strain data segments.

[0040] Example 2, based on Example 1, involves preliminary separation of vehicle-induced strain. Since the total strain data collected by the bridge health monitoring system over a certain period includes both vehicle-induced strain and temperature strain, and vehicle-induced strain exhibits significant fluctuations over a short time, while temperature strain changes relatively smoothly, when the smoothing mean window size is appropriate, the difference between vehicle-induced strain and the smoothed mean will be significantly greater than the difference between temperature strain and the smoothed mean. Based on this principle, by setting an appropriate difference discrimination threshold, vehicle-induced strain and temperature strain in the strain data can be initially separated. The specific steps are as follows:

[0041] Step 1: Based on the short-term fluctuation characteristics of vehicle-induced strain and temperature strain, and combining Allan variance to balance the smoothing effect of the signal with the preservation of short-term fluctuations, the window size m of the smoothing mean is adaptively determined, and the time period [t] is... i ,t i+1 Strain data within ] Perform smoothing and mean processing to obtain the smoothed mean.

[0042] For time period [t] i ,t i+1 The k-th strain data in the j-th iteration, i = 1, 2, 3, ..., n, j = 1, 2, 3, ..., N, k = 1, 2, 3, ..., N j N j Let j be the number of strain data points in the j-th iteration. The k-th smoothed mean of the strain data in the j-th iteration is obtained after smoothing and averaging.

[0043] Step 2: Calculate strain data With smoothed mean Difference between Mean of difference Standard deviation of sum and difference And according to and σ j Set the difference discrimination threshold

[0044] e j,k This represents the k-th difference in the j-th iteration; σ is the mean difference in the j-th iteration; j T is the standard deviation of the difference in the j-th iteration; e,h The threshold for difference discrimination in the j-th iteration;

[0045] Step 3: If the mean of the difference is e j,k Greater than the difference discrimination threshold T e,j Then Marked as pseudo-vehicle-induced strain Conversely, it is marked as pseudo-temperature strain.

[0046] Step 4: Apply pseudo-temperature strain Perform the (j+1)th smoothing process to obtain the smoothed mean. Calculating pseudo-temperature strain With smoothed mean The difference e j+1,k Then calculate the rate of change of the mean difference.

[0047] Step 5: If r is greater than 0.01, repeat steps 2-5; otherwise, end the iteration. Due to noise interference in the total strain data, the pseudo-temperature strain contains a small amount of vehicle-induced strain data, and the pseudo-vehicle-induced strain is interspersed with temperature strain data.

[0048] Example 3, based on Example 2, reclassifies the vehicle-induced strain data segments in the pseudo-temperature strain data segments. Since the initial iterative extraction of vehicle-induced strain is performed using a hard threshold, during the iterative segmentation of the total strain, data exceeding the difference threshold is considered vehicle-induced strain, while data below the difference threshold is classified as temperature strain. Therefore, any data below the set threshold will also be classified as temperature strain data. Because vehicle-induced strain in the total strain fluctuates around the temperature strain, there will inevitably be local vehicle-induced strain segments below the difference threshold during iterative segmentation, leading to misclassification of vehicle-induced strain as temperature strain. This is why the data after the initial iterative segmentation is called pseudo-vehicle-induced strain and pseudo-temperature strain. Given that this phenomenon is unavoidable, it is necessary to reclassify the vehicle-induced strain in the pseudo-temperature strain data segments. Since the vehicle-induced strain data segments interspersed in the pseudo-temperature strain data segments have varying lengths but are relatively short, setting an appropriate data segment length threshold can reclassify these vehicle-induced strain data segments to maximize the extraction of vehicle-induced strain. The specific steps are as follows:

[0049] Step 1: Transfer the pseudo-temperature strain data segment The length of each sub-segment in the data is denoted as l. k Statistical length l k The frequency of occurrence of identical sub-data segments;

[0050] Step 2: Given the pseudo-temperature strain data segment Characteristics of length distribution, statistics k The frequency distribution of sub-data segments of equal length is analyzed. Then, based on the magnitude of frequency change, the length of the sub-data segment corresponding to the largest frequency change is determined, and this length is used as the threshold T for judging the sub-data segment length. l ;

[0051] Step 3: Transfer the pseudo-temperature data segment Neutron data segment length l k Less than the length discrimination threshold T l The sub-data segments are considered as vehicle-induced strain data segments and reclassified as pseudo-vehicle-induced strain data segments. The remaining data segments are labeled as temperature strain data.

[0052] Example 4, based on Example 3, removes temperature strain data segments from the pseudo vehicle-induced strain data segments. Because vehicle-induced strain exhibits large short-term fluctuations, the short-term range of the vehicle-induced strain data segments in the pseudo vehicle-induced strain data segments is significantly greater than that of the temperature strain data segments. By setting a range discrimination threshold, temperature strain data segments can be effectively removed from the pseudo vehicle-induced strain data segments. The specific steps are as follows:

[0053] Step 1: Calculate the strain data segment caused by the pseudo-vehicle The range R of each sub-data segment k And arrange the range values ​​in ascending order;

[0054] Step 2: Calculate the difference between two adjacent range values ​​after sorting, select the point with the smallest difference as the jump point, and use one of the corresponding two adjacent range values ​​as the range discrimination threshold T. R ;

[0055] Step 3: Transfer the pseudo-vehicle strain data segment The range R of each sub-data segment in k Greater than the range discrimination threshold T R The data segment is considered as vehicle-induced strain The remaining data is labeled as temperature strain.

[0056] Example 5: Based on Example 2, the size of the smoothing mean window is selected. Since Allan variance can effectively assess signal volatility, it has been used to determine the window size for simple moving averages, especially in the presence of noise and abrupt changes. Therefore, this method also utilizes Allan variance to set the size of the smoothing mean window. Allan variance is defined as the mean square of the second difference between the means of adjacent windows within a non-overlapping sliding window of length m, as shown in formula (1):

[0057]

[0058] In the formula, m is the size of the smoothing mean window; It is the cumulative value of all data before the l-th strain data; N i It is a time period [t] i ,t i+1The number of strain data points received within a given window. When the smoothing window is small, the Allan variance increases rapidly as the window size increases; when the smoothing window increases to a certain extent, the Allan variance tends to level off. The ideal window size should be located in the stable region after a period of significant fluctuation in Allan variance, so that the data can still retain sufficient signal characteristics after smoothing and averaging.

[0059] To adaptively determine the smoothed mean window size using Allan variance, this method optimizes the window size by calculating the rate of change (CR) of the Allan variance. The window size is determined when CR stabilizes. Specifically, if the smoothed mean window size changes continuously 50 times and CR remains essentially constant, the calculation is stopped, and the window size at the time of the first change is selected as the final smoothed mean window size, as shown in formula (2).

[0060]

[0061] σ 2 (m) represents the Allan variance when the smoothed mean window size is m; σ 2 (m+1) represents the Allan variance when the smoothed mean window size is m+1.

[0062] Example 6: Based on Examples 1 to 5 above, an evaluation standard for this method is designed, which is divided into:

[0063] (1) Cutting accuracy, which is expressed by cutting rate and misjudgment rate. Cutting rate is the ratio of the amount of target strain data in the cut strain data segment to the amount of target strain data in the total strain data, as shown in formula (3):

[0064]

[0065] P vs N is the cut rate of the vehicle-induced strain. v N represents the amount of vehicle-induced strain data in the extracted strain data segment. V The vehicle-induced strain data is the amount of total strain data. The misclassification rate is the proportion of non-target strain data in the extracted strain data segments, as shown in formula (4):

[0066]

[0067] FPR is the misjudgment rate of vehicle-induced strain, N t N represents the amount of temperature-strain data in the extracted strain data segment, where N is the amount of strain data extracted.

[0068] (2) Noise resistance: In practical applications, the data collected by health monitoring systems are often affected by various noises such as environmental interference, equipment errors, and signal attenuation. Therefore, it is necessary to discuss the impact of noise on segmentation quality. Measurement noise is added to the original signal in the form of signal-to-noise ratio (SNR). By systematically adjusting the SNR, monitoring conditions under different noise levels can be simulated. The calculation of the SNR is shown in formula (5):

[0069]

[0070] A signal A represents the amplitude of the signal. noise The amplitude of the noise is represented. The noise immunity of the cutting method is quantified using a coefficient called noise immunity strength, as shown in Equation (6):

[0071]

[0072] D N and D S These represent the strain cut-off rates obtained under noise-affected and noise-free conditions, respectively.

[0073] Example 7: Based on the above examples, the proposed cutting method is verified. First, a finite element model is established. To verify the effectiveness of the proposed method, as follows... Figure 2 As shown, this method uses a T-beam bridge as a reference object and establishes a finite element model of a single T-beam. The main beam is 2.3m high, 2.5m wide, and 50m long, and is a simply supported structure. To reduce the computational load and improve the model's computational speed, beam elements are used for modeling, and the selected element type is BEAM188. The verification of the proposed method requires three types of strain: pure temperature strain, pure vehicle-induced strain, and the total strain caused by both temperature and vehicle load. Therefore, this study conducted simulations under pure temperature, pure vehicle load, and simultaneous temperature and vehicle load conditions. The simulation of the simultaneous temperature and vehicle load condition was achieved through data combination.

[0074] The measured annual temperature variations at 6 measuring points on a concrete bridge are as follows: Figure 3 As shown. In this study, the pure temperature condition was achieved by applying only a uniform temperature to the T-beam cross-section. The beam cross-section temperature is the annual temperature variation corresponding to the T6 measuring point (Note: Due to sensor failure, some annual temperature data was lost, but the selected temperature range still includes the maximum and minimum values ​​of the annual temperature variation and is representative). The reference temperature in the finite element model was set to 20℃, and the coefficient of linear expansion was set to 1.2×10⁻⁶. -6 / ℃.

[0075] For the pure vehicle-borne condition, only the scenario of a vehicle moving at a constant speed in one direction from the center of the beam section was considered, with concentrated forces of varying magnitudes simulating different vehicle weights. Based on the fitting results of measured vehicle weight data, vehicle weights of 15kN, 25kN, 35kN, 45kN, and 55kN were selected. Given that rough road surfaces cause fluctuations in axle load on the bridge, the concentrated forces simulated in this study varied randomly with an amplitude of 5%. Since the focus of this condition is on obtaining vehicle-induced strain, and vehicle speed does not affect the segmentation method, 10m / s was selected as the simulated vehicle speed. To simulate high-frequency data sampling for small-to-medium span bridges, the sampling frequency was set to 50Hz. The magnitudes of the concentrated forces corresponding to different positions of the vehicle on the main beam are shown below. Figure 4 As shown in (a), the beam bottom strain generated by each vehicle passing through the main beam is as follows: Figure 4 As shown in (b).

[0076] Then, the effect of vehicle-induced strain extraction was verified. The strain under pure temperature conditions was as follows: Figure 5 As shown, the strain exhibits a significant positive correlation with the applied temperature, while no temperature strain is generated when there is no temperature change. The strain under pure vehicle-load conditions is generated by randomly selecting different types of vehicles and superimposing them sequentially over time. The annual vehicle-induced strain is shown below. Figure 6 As shown. The total strain is obtained by adding pure mechanical strain to the pure temperature strain. To more realistically simulate actual strain data, 60dB of Gaussian white noise is applied, and the total strain with noise is as follows. Figure 7 As shown.

[0077] To demonstrate the real-time segmentation effect, this embodiment segments the annual strain data into 15-minute intervals, comprising 45,000 data points, to simulate a real-time segmentation scenario. The Allan variance and rate of change of a given 15-minute data point under different smoothing window sizes are shown below. Figure 8 As shown, when the smoothing window m increases from 0 to 30, the Allan variance increases to a maximum of 2.213. With further increases in the smoothing window size, the Allan variance gradually decreases to 0.5 and tends to stabilize. When the smoothing window m is 256, the rate of change of the Allan variance has remained essentially constant for 50 consecutive cycles, and the calculation automatically stops. Since a larger m results in slower computation, to ensure the algorithm's real-time performance as much as possible, the smoothing window size was ultimately chosen to be the m value corresponding to the initial stable rate of change of variance, i.e., 206, as the optimal smoothing window. The strain data were then smoothed and mean-valued based on the adaptively obtained smoothing window size.

[0078] The iterative process of initial segmentation of vehicle-induced strain is as follows: Figure 9 As shown. The initial smoothed mean and difference discrimination threshold are as follows. Figure 9As shown in (a), as expected, the data exceeding the difference discrimination threshold represents the vehicle-induced strain portion. The initial segmentation results based on the initial difference discrimination threshold are as follows: Figure 9 As shown in (b), with the increase in the number of iterations, the extracted vehicle-induced strain database may contain temperature strain data segments introduced due to excessive iterations. Therefore, the strain extracted in this stage is called pseudo-vehicle-induced strain. Subsequently, based on the analysis of the segmentation results, the difference discrimination threshold is dynamically adjusted, and multiple iterations are performed. Each time, the threshold is updated and the pseudo-temperature strain data is segmented, thereby gradually extracting the pseudo-vehicle-induced strain portion. The segmentation result of the second iteration is shown in [image / description]. Figure 9 As shown in (c), the segmentation results of the vehicle-induced strain and temperature strain data gradually stabilize as the iteration progresses. The iteration stops when the convergence condition is met; a total of 6 iterations have been performed. The initial segmentation results are shown in (c). Figure 9 As shown in (d).

[0079] To more intuitively analyze the pseudo-temperature strain data segments obtained after the initial segmentation, a 5-minute pseudo-temperature strain data segment is selected from the initial segmentation results for display, such as... Figure 10 As shown in (a), the pseudo-temperature strain data obtained from the initial segmentation consists of a large number of short-time data segments, which also include some fragmented vehicle-induced strain data segments. The frequency distribution of data segment lengths is as follows: Figure 10 As shown in (b), the number of data segments with a length less than 10 is significantly greater than that of data segments with other lengths. Based on this, the locations where the frequency of data segment length changes significantly are adaptively identified, and the data segment length at these locations is used as a discrimination threshold. According to this threshold, the portion of data segments with a length less than 10 is reclassified into pseudo-vehicle-induced strain data segments. The reclassified pseudo-temperature strain data segments are shown below. Figure 10 As shown in (c). However, in the process of reclassifying fragmented vehicle-induced strain data segments into pseudo-vehicle-induced strain data segments, some temperature strain data segments of similar length may also be misclassified. Therefore, to improve the accuracy of data classification, further processing of pseudo-vehicle-induced strain data segments is required.

[0080] The pseudo-vehicle-induced strain data obtained after reclassification are as follows: Figure 11 As shown in (a), this data contains many temperature strain data segments with very small ranges. The frequency distribution of the ranges of the data segments is as follows: Figure 11 As shown in (b), the range can intuitively reflect the short-term fluctuation characteristics of the data segment. In contrast, the range of the vehicle strain data segment is significantly larger than that of the temperature strain data segment. The system adaptively identifies locations where the range changes significantly and uses the range value corresponding to that location as the range discrimination threshold. Subsequently, data segments with ranges smaller than this threshold are marked as temperature strain data segments and discarded. The final vehicle strain data segment obtained after removing the temperature strain data segments is shown below. Figure 11As shown in (c). After the above processing, the vehicle-induced strain data segment was successfully extracted, and the final segmentation result of the 15-minute time period data is as follows. Figure 12 As shown.

[0081] Next, the total strain data for the entire year is segmented sequentially according to time sequence to simulate the real-time segmentation effect in actual engineering. It is worth noting that the data length can be adjusted according to actual needs, such as selecting the duration or data volume corresponding to a single data storage file in a small-span bridge health monitoring system. This embodiment selects 15 minutes as the time window unit for segmentation, and the segmentation process is as follows... Figure 13 As shown in (a), the annual results obtained by segmenting the data are as follows: Figure 13 As shown in (b), the final vehicle-induced strain cutoff rate was 98.01%, with a misclassification rate of 0.89%. The data misclassified by the proposed vehicle-induced strain cutoff method mainly consisted of the strain of the bridge during free vibration after the vehicle crossed the bridge, such as... Figure 13 As shown in (b), in some cases, the strain data and noise data during bridge free vibration are nearly equal in amplitude. Since the proposed method automatically extracts vehicle-induced strain through an amplitude threshold, this leads to misjudgment. Because this method focuses more on the main beam strain under vehicle load, while bridge free vibration occurs after vehicles cross the bridge, the dynamic strain during free vibration can be considered either vehicle-induced or non-vehicle-induced strain, exhibiting significant subjectivity. Considering that the main purpose of this study is to extract dynamic strain data under vehicle load, this type of dynamic strain data was not included in the misjudgment rate statistics.

[0082] Finally, the impact of noise was verified. The results of strain data segmentation under different levels of noise are shown below. Figure 14 As shown in Table 1, with increasing noise amplitude, the noise gradually approaches the vibration strain generated by the vehicle passing over the bridge. Therefore, the noise mainly affects the vibration strain component. With increasing noise amplitude, the cutoff rate and misjudgment rate of vehicle-induced strain gradually increase. The cutoff accuracy under different noise levels is shown in Table 1.

[0083] Table 1 Noise Resistance

[0084]

[0085] Example 8: Application on a real bridge to verify the bridge and lightweight monitoring overview. The method was applied to a T-shaped continuous beam bridge structure. The bridge has several spans, each with a beam length of 50 meters and a beam height of 2.5 meters, and the bridge deck is a one-way four-lane road. Since heavy vehicles travel in the second and third lanes, and the mid-span section of the bridge structure is a typical stress area for the maximum bending moment of the main beam, it can directly reflect vehicle-induced strain characteristics. The transverse diaphragms, rigidly connected to adjacent main beams, can simultaneously characterize lateral coordinated deformation and temperature effects. For these reasons, a total of 12 sensors were arranged at the mid-span position of the third main beam and its transverse diaphragms, such as... Figure 15 As shown, the system includes strain sensors and temperature sensors. To ensure the acquisition of complete vehicle-induced strain data, the sampling frequency of the monitoring system should be at least eight times the natural frequency of the bridge. The natural frequency of the bridge is approximately 2.5 Hz, and the sampling frequency of the monitoring system is set to 50 Hz. The strain measured at measuring point MR-2-S from January 4th to 6th, 2025 is shown below. Figure 16 As shown in the figure, the strain data reveals that the strain amplitude generated by some of the vehicles is relatively small, not significantly different from the temperature strain fluctuations caused by sensor accuracy limitations.

[0086] The MR-2-S strain sensor, located at the bottom of the mid-span beam, captures the largest dynamic strain changes, which is beneficial for vehicle-induced strain segmentation. Therefore, this study uses it as a benchmark to determine the time points corresponding to vehicle-induced strain in order to segment the data from all measuring points. First, the Allan variance under different smoothing mean window sizes is calculated, and an adaptive window size of 468 is selected. This window size is used to perform initial smoothing mean processing on the strain data. Subsequently, the difference discrimination threshold is automatically set and iterative segmentation is performed. The final iterative result is as follows: Figure 17 As shown. The pseudo-temperature strain data segments and pseudo-vehicle-induced strain data segments generated during the iteration process are as follows: Figure 18 As shown, the pseudo-temperature strain data segment contains fragmented vehicle-induced strain data segments, and the pseudo-vehicle-induced strain data segment contains temperature strain data segments with small ranges. Next, adaptively calculating the fragmented data segment length discrimination threshold and the range discrimination threshold, and performing data segmentation, the final segmentation result is shown below. Figure 19 As shown. From Figure 19 As shown in (a), some smaller vehicle-induced strains were misclassified as temperature strains. This phenomenon is mainly due to the smaller strains generated by lighter vehicles, and the slight difference between these strains and noise signals, making them difficult to identify accurately. Nevertheless, bridge load-bearing capacity assessment mainly utilizes the vehicle-induced strains of heavier vehicles. The unidentified strains caused by lighter vehicles have a negligible impact on the bridge, so the misclassified vehicle-induced strains of lighter vehicles are acceptable.

[0087] The strain data was segmented in real time according to the time period sequence, and the final segmentation effect is as follows: Figure 20 As shown, the vehicle-induced strain generated by the heavy vehicle has been successfully captured. To achieve zero baseline for the vehicle-induced strain signal, the strain values ​​of adjacent intervals at both ends of the captured strain segment are first averaged to obtain an estimate of the temperature strain for that period. This estimate is then subtracted from the entire vehicle-induced strain data to eliminate the influence of temperature strain on the vehicle-induced strain signal, thus achieving zero baseline. In terms of real-time performance, this method captures bridge vehicle-induced strain data online in time-sequential blocks, enabling real-time identification of vehicle-induced strain characteristics. This efficient data capture method not only achieves rapid response in actual bridge monitoring but also provides reliable technical support for real-time health monitoring.

[0088] All dynamic data of the bridge were extracted based on the timestamps obtained from the MR-2-S measuring point data segmentation. The storage size of the strain data collected over 20 days was 7,055,568 KB. After segmentation, only the vehicle-induced strain data segment was retained, with a storage size of 3,282,584 KB, a reduction of approximately 54%. This demonstrates that this method can effectively compress the data storage size of the monitoring system in real time.

[0089] For lightweight monitoring of small and medium-span bridges, the proposed method has three advantages: First, the proposed algorithm has high real-time performance, a simple process, and low computational requirements, making it easy to embed into monitoring systems; second, it can significantly compress the data size in high-frequency health monitoring systems, reducing system storage and data transmission costs; finally, the proposed algorithm completely preserves the original vehicle-induced strain data while compressing data storage size, ensuring the integrity of vehicle-induced strain and providing a reliable data source for subsequent structural analysis and evaluation based on vehicle-induced strain.

[0090] The above description is only for illustrating the present invention and should be understood as not being limited to the above embodiments. Various modifications that conform to the spirit of the present invention are within the protection scope of the present invention.

Claims

1. An adaptive online truncation method for the high-frequency vehicle actuation response of a bridge, characterized in that, Includes the following steps: S1: First, based on the characteristic of large short-term fluctuations in vehicle-induced strain, the total strain data is initially divided into pseudo-temperature strain data segment and pseudo-vehicle-induced strain data segment using the idea of ​​smooth iteration. S2: Subsequently, based on the length characteristics of the fragmented vehicle-induced strain data segments, the vehicle-induced strain data segments in the pseudo-temperature strain data segments in S1 are correctly classified by the threshold-based length segmentation method. S3: Finally, based on the short-time range characteristics of the temperature strain data segment in the pseudo-vehicle strain data segment, the temperature strain data segment is removed from the pseudo-vehicle strain data segment in S1 by a range-based screening method. Step S2 specifically includes: S2-1: Transform the pseudo-temperature strain data segment The length of each sub-segment in the data is denoted as . Statistical length The frequency of occurrence of identical sub-data segments; S2-2: Given the pseudo-temperature strain data segment Characteristics of length distribution, statistics The frequency distribution of sub-data segments of equal length is analyzed. Then, based on the magnitude of frequency change, the length of the sub-data segment corresponding to the largest frequency change is determined and used as the threshold for judging the length of the sub-data segment. ; S2-3: Transfer pseudo-temperature data segment Neutron data segment length Less than the length threshold The sub-data segments are considered as vehicle-induced strain data segments and reclassified as pseudo-vehicle-induced strain data segments. The remaining data segments are labeled as temperature strain data. ; Step S3 specifically includes: S3-1: Calculation of strain data segment caused by pseudo-vehicle Range of each sub-data segment And arrange the range values ​​in ascending order; S3-2: Calculate the difference between two adjacent range values ​​after sorting, select the point with the smallest difference as the jump point, and use one of the corresponding two adjacent range values ​​as the range discrimination threshold. ; S3-3: The pseudo-vehicle-induced strain data segment The range of each sub-data segment in Greater than the range discrimination threshold The data segment is considered as vehicle-induced strain The remaining data is labeled as temperature strain. .

2. The adaptive online truncation method for high-frequency vehicle actuation response of bridges according to claim 1, characterized in that, Step S1 specifically includes: S1-1: Based on the short-term fluctuation characteristics of vehicle-induced strain and temperature strain, Allan variance is used to balance the smoothing effect of the total strain data signal with the ability to retain short-term fluctuations, and the window size of the smoothing mean is adaptively determined. and for the time period [ Strain data within ] Perform smoothing and mean processing to obtain the smoothed mean. ; in, For time period [ [Middle] In the nth iteration One strain data, , , , For the first Number of strain data in each iteration For the first In the nth iteration, the strain data is processed by smoothing and averaging. A smoothed mean; S1-2: Calculation of strain data With smoothed mean Difference between Mean of difference Standard deviation of sum and difference and according to and Set the difference discrimination threshold ; S1-3: If the difference Greater than the difference discrimination threshold Then Marked as pseudo-vehicle-induced strain Conversely, it is marked as pseudo-temperature strain. ; S1-4: Strain on pseudo-temperature Perform the (j+1)th smoothing process to obtain the smoothed mean. Calculate pseudo-temperature strain With smoothed mean Difference between Then calculate the rate of change of the mean difference. ; S1-5: If If the value is greater than 0.01, repeat steps S1-2 to S1-5; otherwise, end the iteration. Due to noise interference in the total strain data, the pseudo-temperature strain contains a small amount of vehicle-induced strain data, and the pseudo-vehicle-induced strain also contains temperature strain data.

3. The adaptive online truncation method for high-frequency vehicle actuation response of a bridge according to claim 2, characterized in that, The process of balancing the signal by incorporating Allan variance includes: defining the Allan variance as a length... Within a non-overlapping sliding window, the mean square value of the second difference between the means of adjacent windows is shown in formula (1): （1） In the formula, It is the size of the smoothed mean window. It is the first The cumulative value of all data prior to the strain data. It is a time period [ The number of strain data received within the [input / output].

4. The adaptive online truncation method for high-frequency vehicle actuation response of a bridge according to claim 2, characterized in that, The adaptive window size for determining the smoothed mean The process includes: calculating the rate of change of Allan variance To optimize the window size, as shown in formula (2): （2） The smoothed mean window size is... Allan variance at time The smoothed mean window size is... The Allan variance at time CR is determined, and the window size can be determined when CR tends to stabilize.

Images