A method and system for measuring bridge structural deformation based on intelligent sensors

By performing time alignment and low-frequency filtering on the initial deformation and state sequences of bridge structure measuring points, inflection points are identified and evolution feature vectors are constructed. This solves the problem of deformation trend deviation caused by the overlap between sensor zero drift and the actual deflection of the main beam, and achieves more accurate monitoring of bridge structural performance.

CN123111182APending Publication Date: 2026-11-17JIANGSU WUTONG ENG TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611627697.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-21
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In long-term continuous monitoring of long-span bridges across seasons, the inherent zero drift of the sensors and the actual slow deflection of the main beam overlap in the long-term trend, leading to a deviation in the attribution of the long-term deformation trend and affecting the accuracy of the analysis of the structural performance degradation trend at mid-span of the main beam.

Method used

By acquiring the initial deformation and state sequence of the bridge structure measuring points, time alignment and low-frequency filtering are performed to identify the inflection points of temperature and sensor operation status. The deformation time series is segmented and an evolution feature vector is constructed. Combined with preset screening conditions, the deformation type is determined, including sensor zero drift-dominated sections, structural deformation-dominated sections, and dual-source superposition sections, and the actual deformation increment of the structure is accurately extracted.

Benefits of technology

It effectively distinguishes between sensor zero drift and actual deflection of the main beam, improving the accuracy and reliability of bridge structural performance degradation trend analysis, avoiding false alarms or missed alarms, and providing more reliable monitoring data.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123111182A_ABST
    Figure CN123111182A_ABST
Patent Text Reader

Abstract

This application relates to a method and system for measuring bridge structural deformation based on intelligent sensors, specifically in the field of bridge structural deformation measurement technology. The method includes acquiring the initial deformation time series and initial state series of bridge structural measuring points; preprocessing the initial deformation time series and initial state series to generate a low-frequency deformation time series and a low-frequency state series; dividing the low-frequency deformation time series into multiple subsequences based on temperature and operational inflection points in the low-frequency state series; constructing an evolutionary feature vector for each subsequence; determining the deformation type of each subsequence based on the evolutionary feature vector and preset screening conditions; determining the actual structural deformation increment corresponding to each subsequence based on the deformation type; and obtaining a continuous trend sequence of actual structural deformation based on the actual structural deformation increments of each subsequence in chronological order. This application can improve the accuracy and reliability of monitoring data.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of bridge structural deformation measurement technology, and more specifically, to a bridge structural deformation measurement method and system based on intelligent sensors. Background Technology

[0002] In long-term, continuous monitoring of long-span bridges across seasons, equipment inevitably undergoes prolonged temperature cycling and component aging. At this point, the overlap between the inherent zero drift of the sensors and the actual, slow deflection of the main beam in the long-term trend becomes apparent. In actual operation, after a considerable period, the zero point of the sensors will slowly accumulate and shift due to aging and environmental changes. Simultaneously, the main beam will also experience a similar magnitude of actual, slow deflection due to material creep, prestress loss, or long-term load accumulation. Both of these exhibit monotonic or near-monotonic offset components in the long-term deformation trend, and their directions are often similar or consistent.

[0003] Because there is no direct physical distinction between the sensor's inherent zero drift and the actual slow deflection of the main beam in a single-point time series, the long-term offset of the mid-span measuring point cannot be naturally correlated with equipment drift or structural deformation. In scenarios involving long-term, continuous monitoring of the micron-level deflection of the main beam at mid-span across seasons, when the sensor's inherent zero drift and the actual slow deflection of the main beam accumulate at similar rates and directions in the long-term trend, the lack of distinguishable conditions between the two in a single-point time series leads to a deviation in the attribution of the long-term trend, which in turn seriously affects the accuracy of the analysis of the structural performance degradation trend at mid-span of the main beam. Summary of the Invention

[0004] This application aims to address at least one of the technical problems existing in the prior art. To this end, this application discloses a method and system for measuring bridge structural deformation based on intelligent sensors, which aims to solve the problem that in long-term continuous monitoring of long-span bridges across seasons, the inherent zero drift of the sensors overlaps with the actual slow deflection of the main beam in the long-term trend, leading to a deviation in the attribution of the long-term deformation trend, and thus seriously affecting the accuracy of the analysis of the structural performance degradation trend at mid-span of the main beam.

[0005] In a first aspect, this application discloses a method for measuring bridge structural deformation based on intelligent sensors, comprising the following steps: The initial deformation time series of the bridge structure measuring points and the initial state series used to characterize the zero-point stability of the measurement sensor are obtained, wherein the initial state series includes at least the temperature series, the sensor power supply voltage series and the sensor cumulative running time series; Time alignment and low-frequency filtering are performed on the initial deformation time series and the initial state series to generate a low-frequency deformation time series and a low-frequency state series. Based on the temperature change inflection point and the sensor operation inflection point in the low-frequency state sequence, the low-frequency deformation time sequence is divided into multiple sub-low-frequency deformation time sequences. The temperature change inflection point is the time point when the temperature change gradient in the low-frequency state sequence changes abruptly, and the sensor operation inflection point is the time point when the sensor power supply voltage and operation continuity change abruptly. For each sub-low-frequency deformation time series, construct the evolution feature vector of each sub-low-frequency deformation time series; Based on the evolutionary feature vector and combined with the preset screening conditions, the deformation type of each sub-low-frequency deformation time series is determined. The deformation type includes sensor zero drift-dominated segment, structural deformation-dominated segment and dual-source superposition segment. The dual-source superposition segment is the segment where sensor zero drift and bridge structural deformation are superimposed. Based on the deformation type, determine the actual structural deformation increment corresponding to each sub-low-frequency deformation time series; By accumulating the actual structural deformation increments of each sub-low-frequency deformation time series in chronological order, a continuous actual structural deformation trend series is obtained.

[0006] Further, the steps of performing time alignment and low-frequency filtering on the initial deformation time series and the initial state series to generate low-frequency deformation time series and low-frequency state series include: Within the defined sliding window, time alignment is performed on the initial deformation time series and the initial state series; The initial deformation time series and the initial state series, which are time-aligned, are used to remove transient spikes and impact interferences using a threshold determination method to obtain a denoised deformation time series and a denoised state series. Weighted moving averages are applied to the denoised deformation time series and the denoised state series respectively to generate low-frequency deformation time series and low-frequency state series.

[0007] Based on this, according to the inflection points of temperature change and sensor operation in the low-frequency state sequence, the steps of dividing the low-frequency deformation time sequence into multiple sub-low-frequency deformation time sequences include: Within a set sliding window, the variance of the temperature change gradient within the set sliding window is continuously calculated. The set sliding window contains N sampling points, where N is a natural number from 10 to 1000. When the variance of the temperature change gradient is greater than the preset variance threshold, the sampling point where the sign of the first time derivative of the temperature changes within the sliding window is determined as the inflection point of the temperature change. When the change in the sensor power supply voltage exceeds the preset voltage change threshold, the sampling point where the sign of the first time derivative of the voltage changes within the sliding window is determined as the inflection point of the sensor's operating state. The low-frequency deformation time series is cut off at the time points corresponding to the inflection point of the temperature change and the inflection point of the sensor's operating state, and the low-frequency deformation time series is divided into multiple sub-low-frequency deformation time series.

[0008] Furthermore, the steps for constructing the evolutionary feature vectors of each sub-low-frequency deformation time series include: Multiple linear deformation time curves are obtained by linearly fitting each sub-low-frequency deformation time series. Based on multiple linear deformation time curves, generate the predicted value sequence at the corresponding time point to obtain multiple sub-linear trend deformation sequences; For each of these low-frequency deformation time series, remove the corresponding sub-linear trend deformation series to obtain the deformation residual series; The residual deformation rate is calculated based on the deformation residual sequence. Calculate the Pearson correlation coefficient between the deformation residual sequence and the temperature sequence in the same region; The residual deformation rate and the Pearson correlation coefficient between the deformation residual sequence and the temperature sequence in the same region are summarized to construct the evolution feature vector of each sub-low-frequency deformation time series.

[0009] In some preferred embodiments, the step of determining the deformation type of each sub-low-frequency deformation time series based on the evolutionary feature vector and in conjunction with preset screening conditions includes: If the absolute value of the Pearson correlation coefficient in the evolution feature vector of the sub-low-frequency deformation time series is higher than the preset coefficient threshold, and the residual deformation rate matches the preset sensor calibration aging drift model, then the sub-low-frequency deformation time series is marked as the sensor zero drift dominant segment. If the absolute value of the Pearson correlation coefficient in the evolution feature vector of the sub-low-frequency deformation time series is lower than the preset coefficient threshold, and the residual deformation rate matches the theoretical creep prediction rate of the measured structure, then the sub-low-frequency deformation time series is marked as the structural deformation-dominant segment. If the sub-low-frequency deformation time series does not meet the conditions of the sensor zero-drift dominant section and the structural deformation dominant section, then the sub-low-frequency deformation time series is marked as a dual-source superposition section.

[0010] Based on this, the steps to determine the actual structural deformation increment corresponding to each sub-low-frequency deformation time series according to the deformation type include: The actual deformation increment of the sub-low-frequency deformation time series of the deformation type, which is the sensor zero-drift dominant segment, is determined to be 0; The actual structural deformation increment of the sub-low-frequency deformation time series with the deformation type being the dominant structural deformation segment is determined as the deformation in the corresponding low-frequency deformation time series. The actual structural deformation increment of the sub-low-frequency deformation time series of the deformation type of the dual-source superposition section is determined as the deformation amount obtained by subtracting the sensor zero drift increment from the deformation in the low-frequency deformation time series corresponding to the dual-source superposition section.

[0011] Furthermore, the step of determining the actual structural deformation increment of the sub-low-frequency deformation time series of the deformation type being a dual-source superimposed segment as the deformation amount obtained by subtracting the sensor zero drift increment from the deformation in the low-frequency deformation time series corresponding to the dual-source superimposed segment includes: Obtain the drift parameter benchmark of the sensor zero-drift dominant section that corresponds to the dual-source superposition section of the deformation type and is in a stable state; The theoretical increment of sensor zero drift is calculated based on the drift parameter benchmark and the sub-low frequency state sequence corresponding to the dual-source superposition section. The actual structural deformation increment of the sub-low-frequency deformation time series of the deformation type of the dual-source superposition section is determined by subtracting the theoretical zero-drift increment of the sensor from the deformation in the low-frequency deformation time series corresponding to the dual-source superposition section.

[0012] Preferably, the step of obtaining the drift parameter reference of the sensor zero-drift dominant section corresponding to the deformation type of the dual-source superposition section and in a stable state includes: Extract the most recent sensor zero-drift dominant segment from the historical zero-drift parameter evolution queue, which is a collection of sub-low-frequency deformation time series that have been confirmed in the continuously recorded history as sensor zero-drift dominant segments; The drift rate of the most recent sensor zero-drift dominant segment is extracted as the benchmark for the drift parameter.

[0013] Based on the above, the steps for calculating the theoretical increment of sensor zero drift according to the drift parameter benchmark and the low-frequency state sequence within the dual-source superposition section include: Calculate the first time derivative of the temperature sequence in the low-frequency state sequence corresponding to the dual-source superposition section, and divide the dual-source superposition section into a temperature rising branch period and a temperature falling branch period according to the sign of the first time derivative. The positive unit temperature drift rate corresponding to the heating process and the negative unit temperature drift rate corresponding to the cooling process are extracted from the corresponding sensor zero drift dominant section to construct an asymmetric thermal hysteresis loop model. This asymmetric thermal hysteresis loop model is used to calculate the transient thermal hysteresis temperature drift. According to the asymmetric thermal hysteresis loop model, if the sampling point corresponding to the dual-source superposition section is in the temperature rise branch period, the positive unit temperature drift rate is multiplied by the temperature change at the current moment relative to the starting point of the current temperature rise branch to obtain the transient thermal hysteresis temperature drift at the current moment. According to the asymmetric thermal hysteresis loop model, if the sampling point corresponding to the dual-source superposition section is in the temperature drop branch period, the reverse unit temperature drift rate is multiplied by the temperature change at the current moment relative to the starting point of the current cooling branch to obtain the transient thermal hysteresis temperature drift at the current moment. The cumulative aging drift is calculated based on the sensor's cumulative operating time and the drift parameter benchmark. The cumulative aging drift is then summed with the transient thermal hysteresis temperature drift to obtain the theoretical increment of the sensor's zero drift at the current moment within the dual-source superposition section.

[0014] Secondly, this application also discloses a bridge structural deformation measurement system based on intelligent sensors, comprising: The acquisition module is used to acquire the initial deformation time series of the bridge structure measuring points and the initial state sequence used to characterize the zero-point stability of the measurement sensor, wherein the initial state sequence includes at least the temperature sequence, the sensor power supply voltage sequence, and the sensor cumulative running time sequence. The preprocessing module is used to perform time alignment and low-frequency filtering on the initial deformation time series and the initial state series to generate a low-frequency deformation time series and a low-frequency state series. The subsequence division module is used to divide the low-frequency deformation time series into multiple sub-low-frequency deformation time series based on the temperature change inflection point and the sensor operation state inflection point in the low-frequency state series. The temperature change inflection point is the time point when the temperature change gradient in the low-frequency state series changes abruptly, and the sensor operation state inflection point is the time point when the sensor power supply voltage and operation continuity change abruptly. The evolution feature vector construction module is used to construct the evolution feature vectors of each sub-low-frequency deformation time series for each sub-low-frequency deformation time series. The deformation type determination module is used to determine the deformation type of each sub-low-frequency deformation time series based on the evolution feature vector and in combination with preset screening conditions. The deformation type includes sensor zero drift-dominated segment, structural deformation-dominated segment and dual-source superposition segment, wherein the dual-source superposition segment is the segment where sensor zero drift and bridge structural deformation are superimposed. The actual deformation increment determination module is used to determine the actual structural deformation increment corresponding to each sub-low-frequency deformation time series based on the deformation type. The deformation determination module is used to accumulate the actual structural deformation increments of each sub-low-frequency deformation time series in chronological order to obtain a continuous actual structural deformation trend series.

[0015] This application discloses a bridge structure deformation measurement method based on intelligent sensors. It acquires an initial deformation time series and an initial state series, performs time alignment and low-frequency filtering to generate a low-frequency deformation time series and a low-frequency state series. Based on this, according to the temperature change inflection point and sensor operating state inflection point in the low-frequency state series, the low-frequency deformation time series is divided into multiple sub-low-frequency deformation time series. For each sub-low-frequency deformation time series, an evolutionary feature vector is constructed, and combined with preset screening conditions, the deformation type is determined, including sensor zero-drift dominant segments, structural deformation dominant segments, and dual-source superposition segments. Finally, the actual structural deformation increment is determined according to the deformation type and accumulated over time to obtain a continuous trend sequence of actual structural deformation.

[0016] This method effectively solves the problem in existing technologies where the inherent zero drift of sensors overlaps with the actual slow deflection of the main beam in the long-term trend, leading to attribution bias in the long-term deformation trend. By introducing an initial state sequence (including temperature, power supply voltage, and cumulative operating time) and segmenting the data by combining temperature change inflection points and sensor operating state inflection points, deformation characteristics under different influencing factors can be identified more precisely. The construction of evolutionary feature vectors and the determination of deformation types enable effective differentiation between sensor zero drift and structural deformation, especially in dual-source superposition sections, where the actual structural deformation can be accurately obtained by subtracting the sensor zero drift increment. Therefore, this application can provide more reliable bridge structural performance degradation trend analysis data, avoiding false alarms or missed alarms, and significantly improving the accuracy and reliability of monitoring data.

[0017] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0018] The accompanying drawings are used to provide a further understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application.

[0019] Figure 1 A flowchart illustrating a bridge structure deformation measurement method based on smart sensors provided in one embodiment of this application; Figure 2 This is a schematic diagram of a process for generating low-frequency deformation time series and low-frequency state series according to an embodiment of this application; Figure 3 This is a flowchart illustrating the process of dividing multiple sub-low-frequency deformation time series according to one embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical methods, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0021] See Figure 1 , Figure 1 This is a flowchart illustrating a bridge structure deformation measurement method based on intelligent sensors, provided in one embodiment of this application. This application proposes a bridge structure deformation measurement method based on intelligent sensors, including but not limited to steps S110 to S170, which will be described in detail below.

[0022] S110. Obtain the initial deformation time series of the bridge structure measuring points and the initial state sequence used to characterize the zero-point stability of the measurement sensor, wherein the initial state sequence includes at least the temperature sequence, the sensor power supply voltage sequence, and the sensor cumulative running time sequence. S120. Perform time alignment and low-frequency filtering on the initial deformation time series and initial state series to generate a low-frequency deformation time series and a low-frequency state series. S130. Based on the temperature change inflection point and the sensor operation state inflection point in the low-frequency state sequence, the low-frequency deformation time sequence is divided into multiple sub-low-frequency deformation time sequences. The temperature change inflection point is the time point when the temperature change gradient in the low-frequency state sequence changes abruptly, and the sensor operation state inflection point is the time point when the sensor power supply voltage and operation continuity change abruptly. S140. For each sub-low-frequency deformation time series, construct the evolution feature vector of each sub-low-frequency deformation time series; S150. Based on the evolution feature vector and the preset screening conditions, determine the deformation type of each sub-low-frequency deformation time series. The deformation types include sensor zero drift-dominated segment, structural deformation-dominated segment and dual-source superposition segment. The dual-source superposition segment is the segment where sensor zero drift and bridge structural deformation are superimposed. S160. Based on the deformation type, determine the actual structural deformation increment corresponding to each sub-low-frequency deformation time series; S170. According to the chronological order, the actual structural deformation increments of each sub-low-frequency deformation time series are accumulated over time to obtain a continuous actual structural deformation trend series.

[0023] To better understand the method proposed in this application, it is necessary to explain some of the key terms involved.

[0024] The initial deformation time series refers to the sequence of raw deformation data collected directly by the sensor without any processing, which changes over time.

[0025] The initial state sequence refers to the raw data sequence acquired synchronously with the deformation data, reflecting the sensor's operating state and environmental conditions. It includes at least the temperature sequence, the sensor power supply voltage sequence, and the sensor's cumulative runtime sequence. The temperature sequence reflects the temperature changes in the sensor's environment; the sensor power supply voltage sequence reflects the stability of the sensor's power supply; and the sensor's cumulative runtime sequence records the total operating time of the sensor from startup, which is crucial for assessing the sensor's aging degree.

[0026] Low-frequency filtering aims to remove high-frequency noise and transient interference from data while preserving the long-term trends of deformation and state changes.

[0027] Temperature change inflection points and sensor operating status inflection points are indicators for identifying key change points in a data sequence. They help to divide a continuous deformation time series into subsequences with different characteristics.

[0028] Evolutionary eigenvectors are mathematical descriptions used to characterize the properties of low-frequency deformation time series. They contain multiple feature parameters that reflect deformation trends and correlations.

[0029] Deformation type is the result of classifying the low-frequency deformation time series, including sensor zero-drift dominant segment, structural deformation dominant segment, and dual-source superposition segment. These three types correspond to different deformation sources.

[0030] The actual deformation increment of the structure refers to the actual increment caused by the deformation of the bridge structure itself after deducting the influence of sensor zero drift.

[0031] The structural true deformation trend sequence is the final data sequence that can accurately reflect the long-term deformation trend of the bridge structure.

[0032] Initial deformation time series can be obtained in various ways. For example, the deformation of the bridge structure can be directly measured using resistance strain gauges, fiber optic grating sensors, or MEMS sensors, and these measurements can be recorded in chronological order. The initial state series can be obtained similarly to the deformation time series. For instance, temperature series can be obtained using temperature sensors such as thermistors or thermocouples; sensor power supply voltage series can be obtained by monitoring the power supply voltage in real time using voltage sensors; and sensor cumulative runtime series can be recorded using internal system timers or external counters. The acquisition of these series forms the basis for subsequent data processing and analysis.

[0033] Time alignment can be achieved through interpolation algorithms. For example, when two sequences have different sampling frequencies, one sequence can be interpolated to the timestamp of the other to ensure temporal consistency of data points. Low-frequency filtering can be implemented using various digital filters, such as moving average filters, Butterworth filters, or Chebyshev filters. By setting an appropriate cutoff frequency, high-frequency noise can be removed while preserving the long-term trends of deformation and state changes.

[0034] Inflection points in temperature changes can be identified by analyzing the first or second derivatives of the temperature sequence. For example, an inflection point can be considered to exist when the temperature gradient changes significantly. Inflection points in sensor operation can be determined by monitoring sudden changes in the sensor's power supply voltage or interruptions in operation continuity. For example, an inflection point can be considered to exist when the power supply voltage suddenly drops or when the sensor restarts after a period of inactivity. These inflection points allow the entire low-frequency deformation time series to be divided into several subsequences with relatively stable characteristics, facilitating subsequent independent analysis.

[0035] The construction of an evolutionary feature vector can include various feature parameters. For example, statistical features such as the mean, variance, and slope of each subsequence can be calculated. Furthermore, the correlation coefficients between the subsequences and environmental factors such as temperature and humidity sequences can be calculated to characterize the relationship between deformation and environmental factors. The combination of these feature parameters constitutes the evolutionary feature vector, which is used for subsequent deformation type identification.

[0036] Screening criteria can be set based on empirical knowledge or machine learning models. For example, when a certain feature parameter in the evolutionary feature vector (such as the correlation coefficient with temperature) exceeds a certain threshold, it may indicate that the subsequence is mainly affected by temperature zero drift. When another feature parameter (such as deformation rate) matches the structural theory creep model, it may indicate that the subsequence is mainly affected by structural deformation. When neither of these conditions is met, it may be a dual-source superposition segment.

[0037] For subsequences identified as sensor zero-drift dominant segments, their true structural deformation increment can be set to zero, as the deformation in this segment is primarily caused by sensor zero-drift. For subsequences identified as structural deformation dominant segments, their true structural deformation increment can be directly taken as the deformation value of that subsequence, as the deformation in this segment is primarily caused by the deformation of the structure itself. For subsequences identified as dual-source superposition segments, the sensor zero-drift increment needs to be subtracted from their total deformation to obtain the true structural deformation increment.

[0038] The time-series accumulation can be achieved through a simple summation operation, which adds up the actual deformation increments of each subsequence in order of their position in the original time series, thus forming a continuous sequence that reflects the actual deformation trend of the bridge structure.

[0039] This application effectively distinguishes between inherent zero drift from sensors and the actual slow deflection of the main beam, solving the problem of attribution bias in long-term trend quantities. Traditional methods often struggle to accurately separate these two different sources of offset, easily misattributing zero drift contributions to the actual deflection increment of the main beam, or misclassifying the actual deflection portion as zero drift. This application forms a complete solution through a series of steps, including introducing initial state sequences, time alignment and low-frequency filtering, inflection point-based subsequence partitioning, evolutionary feature vector construction, and deformation type determination. In particular, by identifying inflection points in temperature changes and sensor operating states, and through detailed analysis of evolutionary feature vectors, this application can accurately divide deformation data into sensor zero drift-dominated segments, structural deformation-dominated segments, and dual-source superposition segments, and adopt different processing strategies for different segments, thereby accurately extracting the actual structural deformation increment. This multi-dimensional, phased approach significantly improves the accuracy and reliability of bridge structural deformation monitoring, enabling the micron-level deformation trend at mid-span of the main girder to be reliably used to determine the true performance degradation of the main girder, avoiding false alarms or missed alarms, and providing a more scientific basis for the safety assessment and maintenance of bridge structures.

[0040] In some embodiments, such as Figure 2 As shown, the steps of performing time alignment and low-frequency filtering on the initial deformation time series and initial state series to generate low-frequency deformation time series and low-frequency state series are not limited to S210 to S230, and will be described one by one below.

[0041] S210. Within the set sliding window, perform time alignment on the initial deformation time sequence and the initial state sequence; S220. The initial deformation time sequence and the initial state sequence that are time-aligned are used to remove transient spikes and impact interference using a threshold determination method to obtain a denoised deformation time sequence and a denoised state sequence. S230. Perform weighted moving averages on the denoised deformation time series and the denoised state series respectively to generate a low-frequency deformation time series and a low-frequency state series.

[0042] Specifically, performing time alignment within a defined sliding window refers to synchronizing data collected by different sensors within a limited time frame to eliminate time deviations caused by differences in sampling frequencies or transmission delays. For example, interpolation, resampling, or nearest neighbor matching can be used to ensure precise correspondence between deformation data and state data on the time axis, aiming to provide a consistent time reference for subsequent data processing. Threshold-based removal of transient spikes and impact interference can be understood as identifying and removing abnormal instantaneous high or low values ​​in the data by setting reasonable numerical thresholds. These outliers are usually caused by external instantaneous impacts or sensor malfunctions. Specifically, when a data point's value exceeds preset upper and lower thresholds, it is identified as an outlier and removed or replaced. This aims to purify the data and prevent outliers from misleading subsequent filtering and analysis. In practical applications, weighted moving averages are applied to both the denoised deformation time series and the denoised state series. Specifically, a smoothing technique is used, assigning different weights to the data points within the sliding window for averaging to generate a smoother low-frequency sequence. For example, linear weighting, exponential weighting, or other custom weighting methods can be used to effectively filter out high-frequency noise, retain the long-term trend and low-frequency components of the data, and thus more accurately reflect the true deformation of the bridge structure and the stability of the sensors.

[0043] Through the above technical solutions, this application effectively solves the problems of transient spikes, impact interference, and time asynchrony in raw sensor data existing in traditional methods. Time alignment processing ensures the accurate temporal correspondence between deformation data and state data, providing a unified benchmark for subsequent analysis. The threshold determination method can efficiently remove outliers from the data, significantly improving data purity. Weighted moving average further smooths the data, effectively filtering out high-frequency noise, enabling the generated low-frequency deformation time series and low-frequency state series to more accurately reflect the long-term deformation trend of the bridge structure and the zero-point stability of the sensors. Compared with the basic solution, this application significantly improves data quality and the accuracy of subsequent deformation analysis through refined preprocessing, providing a more reliable data foundation for bridge structural health monitoring.

[0044] In some embodiments, such as Figure 3 As shown, the steps for dividing the low-frequency deformation time series into multiple sub-low-frequency deformation time series include, but are not limited to, S310 to S340, which will be described one by one below.

[0045] S310. Within a set sliding window, continuously calculate the variance of the temperature change gradient within the set sliding window. The set sliding window contains N sampling points, where N is a natural number from 10 to 1000. S320. When the variance of the temperature change gradient is greater than the preset variance threshold, the sampling point where the sign of the first time derivative of the temperature changes within the sliding window is determined as the inflection point of the temperature change. S330. When the change in the sensor power supply voltage exceeds the preset voltage change threshold, the sampling point where the sign of the first-order time derivative of the voltage changes within the sliding window is determined as the inflection point of the sensor's operating state. S340. Cut the low-frequency deformation time series at the time points corresponding to the inflection points of temperature change and sensor operation status, and divide the low-frequency deformation time series into multiple sub-low-frequency deformation time series.

[0046] Specifically, within a defined sliding window, the variance of the temperature change gradient is continuously calculated. The sliding window is a commonly used time-series data processing technique that performs local analysis of the data within a fixed-size window by moving the data across the time series. Here, the sliding window is used to monitor the dynamic characteristics of the temperature change gradient in real time. The defined sliding window contains N sampling points, where N is a natural number between 10 and 1000. This value of N is chosen to balance computational efficiency with the ability to capture the details of temperature changes, ensuring effective identification of trends and abrupt changes in temperature.

[0047] When the variance of the temperature change gradient exceeds a preset variance threshold, it indicates a significant change in the rate or direction of temperature change within the current sliding window, meaning the temperature change gradient is no longer stationary. In this case, the sampling points where the sign of the first-order time derivative of temperature changes within the sliding window are further identified as inflection points of temperature change. The first-order time derivative of temperature reflects the rate of temperature change over time, and its sign change typically indicates the turning point of the temperature's upward or downward trend, i.e., the moment when the rate of temperature change reaches a local maximum or minimum. These points are considered inflection points of temperature change.

[0048] Similarly, when the sensor's power supply voltage changes beyond a preset voltage change threshold, this typically indicates a significant change in the sensor's operating state, such as power instability, restart, or malfunction. In this case, the sampling points within the sliding window where the sign of the first-order time derivative of the voltage changes are identified as inflection points in the sensor's operating state. The change in the sign of the first-order time derivative of the voltage indicates the turning point in the trend of power supply voltage changes; these points are considered inflection points in the sensor's operating state and can effectively capture abrupt changes in the continuity of sensor operation or power supply status.

[0049] Therefore, the low-frequency deformation time series is segmented at the time points corresponding to the inflection points of the temperature change and the sensor operating state, thereby dividing the low-frequency deformation time series into multiple sub-low-frequency deformation time series. In this way, the original low-frequency deformation time series can be decomposed into multiple segments that are relatively stable in terms of temperature or sensor operating state, providing a more homogeneous data foundation for subsequent deformation type identification and calculation of the actual deformation increment.

[0050] Through the above technical solution, this application can more precisely and accurately identify the sections affected by temperature changes and sensor operating conditions during bridge structural deformation measurement. This inflection point-based division method avoids the problems of coarse division or ignoring key change points that may exist in traditional methods, making each sub-low-frequency deformation time series more homogeneous internally. Therefore, in subsequent deformation type identification and calculation of the actual structural deformation increment, the sensor zero drift and the actual structural deformation can be separated more effectively, significantly improving the accuracy and reliability of deformation measurement results and providing more accurate data support for bridge structural health monitoring.

[0051] In some embodiments, this application further proposes a step of constructing the evolution feature vector of each sub-low-frequency deformation time series, including: Multiple linear deformation time curves are obtained by linearly fitting each sub-low-frequency deformation time series. Based on the multiple linear deformation time curves, a sequence of predicted values ​​at corresponding time points is generated to obtain multiple sub-linear trend deformation sequences; For each of the aforementioned sub-low-frequency deformation time series, the corresponding sub-linear trend deformation series is removed to obtain the deformation residual series; The residual deformation rate is calculated based on the deformation residual sequence. Calculate the Pearson correlation coefficient between the deformation residual sequence and the temperature sequence in the same region; The residual deformation rate and the Pearson correlation coefficient between the deformation residual sequence and the temperature sequence in the same segment are summarized to construct the evolution feature vector of each sub-low-frequency deformation time series.

[0052] Specifically, linear fitting is performed on each sub-low-frequency deformation time series to extract the overall linear trend of the deformation time series. Through linear fitting, a linear deformation time curve is obtained, representing the average rate and direction of deformation change over time within the current sub-low-frequency deformation time series. Based on this linear deformation time curve, a predicted value sequence at the corresponding time point can be generated, thus obtaining the sub-linear trend deformation sequence. This sub-linear trend deformation sequence reflects the expected value of deformation at each time point under ideal linear change conditions. In practical applications, removing the corresponding sub-linear trend deformation sequence from each sub-low-frequency deformation time series yields the deformation residual sequence. The deformation residual sequence is the remaining part after subtracting the linear trend from the original deformation time series. It mainly contains information such as nonlinear changes, periodic fluctuations, and random noise, which helps to more clearly observe and analyze nonlinear deformation characteristics. Furthermore, the residual deformation rate can be calculated based on the deformation residual sequence. The residual deformation rate characterizes the rate of change of the deformation residual sequence after removing the linear trend, reflecting the dynamic characteristics of nonlinear deformation or zero drift. Furthermore, calculating the Pearson correlation coefficient between the deformation residual sequence and the temperature sequence in the same region is a crucial step. The Pearson correlation coefficient quantifies the degree of linear correlation between the deformation residual and temperature changes. Since sensor zero drift is usually closely related to temperature changes, calculating this correlation coefficient can effectively identify zero drift components that are significantly affected by temperature. Finally, by summing the residual deformation rate and the Pearson correlation coefficient between the deformation residual sequence and the temperature sequence in the same region, the evolutionary feature vector of each sub-low-frequency deformation time series can be constructed. This evolutionary feature vector is a multi-dimensional numerical set that comprehensively reflects the key characteristics of the sub-low-frequency deformation time series, such as linear trend, nonlinear fluctuations, rate of change, and sensitivity to temperature.

[0053] Through the above technical solution, this application can construct an evolutionary feature vector with higher discriminative power and robustness. Compared with schemes using only single or simple features, this application extracts multi-dimensional features such as linear trends, residual deformation rates, and correlations with temperature, enabling the evolutionary feature vector to more comprehensively capture the complex behavior of the sub-low-frequency deformation time series. This multi-feature fusion method significantly enhances the ability to identify the differences between sensor zero drift and structural deformation, effectively avoiding misjudgments caused by insufficient features. As a result, the subsequent deformation type determination module can more reliably divide the sub-low-frequency deformation time series into sensor zero drift-dominated segments, structural deformation-dominated segments, or dual-source superposition segments based on more accurate feature vectors, thus laying a solid foundation for finally obtaining a continuous and true structural deformation trend sequence and improving the overall accuracy and reliability of bridge structural deformation measurement.

[0054] Specifically, the steps for determining the deformation type of each sub-low-frequency deformation time series based on the evolutionary feature vector and preset screening conditions include: If the absolute value of the Pearson correlation coefficient in the evolution feature vector of the sub-low-frequency deformation time series is higher than the preset coefficient threshold, and the residual deformation rate matches the preset sensor calibration aging drift model, then the sub-low-frequency deformation time series is marked as the sensor zero drift dominant segment. If the absolute value of the Pearson correlation coefficient in the evolution feature vector of the sub-low-frequency deformation time series is lower than the preset coefficient threshold, and the residual deformation rate matches the theoretical creep prediction rate of the measured structure, then the sub-low-frequency deformation time series is marked as the structural deformation-dominant segment. If the sub-low-frequency deformation time series does not meet the conditions of the sensor zero-drift dominant section and the structural deformation dominant section, then the sub-low-frequency deformation time series is marked as a dual-source superposition section.

[0055] The absolute value of the Pearson correlation coefficient is a quantitative indicator of the linear correlation between the deformation residual sequence and the temperature sequence in the same region. A larger absolute value indicates a stronger correlation between the deformation residual and temperature changes, and is more likely to be caused by temperature-induced sensor zero drift. The preset coefficient threshold is an empirical or theoretical value used to distinguish between strong and weak correlations. The residual deformation rate refers to the rate of change of the deformation residual sequence over time after removing the linear trend, reflecting the evolution characteristics of the nonlinear deformation component. The preset sensor calibration aging drift model is established based on historical sensor data or laboratory calibration results and is used to describe the aging drift of the sensor zero point with time, temperature, and other factors. The theoretical creep prediction rate of the measured structure is the long-term deformation rate of the structure calculated using structural mechanics or materials mechanics theories based on the material properties, load conditions, and environmental factors of the bridge structure.

[0056] The above technical solution enables precise classification of different deformation sources in bridge structural deformation measurement data. This classification mechanism allows for differentiated processing strategies in subsequent deformation increment calculations based on different data types. For example, in sensor-dominated zero-drift sections, the actual structural deformation increment can be directly determined as zero, while in deformation-dominated sections, the measured deformation can be directly used as the actual deformation. Especially in dual-source superposition sections, it provides a clear identification basis for subsequent zero-drift separation. Therefore, this application significantly improves the accuracy and reliability of bridge structural deformation monitoring, providing more precise data support for bridge health monitoring and safety assessment.

[0057] Specifically, in the above method, in order to accurately extract the true deformation of the bridge structure from the measurement data, it is necessary to precisely quantify the actual deformation increment of the structure corresponding to each sub-low-frequency deformation time series according to the determined deformation type.

[0058] The steps described above for determining the actual structural deformation increments corresponding to each sub-low-frequency deformation time series based on the deformation type include: The actual deformation increment of the sub-low-frequency deformation time series with deformation type of sensor zero drift dominant segment is determined to be 0; The actual structural deformation increment of the sub-low-frequency deformation time series whose deformation type is the dominant structural deformation segment is determined as the deformation in the corresponding low-frequency deformation time series. The actual structural deformation increment of the sub-low-frequency deformation time series of the deformation type of the dual-source superposition section is determined as the deformation amount obtained by subtracting the sensor zero drift increment from the deformation in the low-frequency deformation time series corresponding to the dual-source superposition section.

[0059] Specifically, when a sub-low-frequency deformation time series is identified as a sensor zero-drift-dominated segment, it means that the deformation observed in this segment is mainly caused by the sensor's own zero-point drift, rather than the actual deformation of the bridge structure. Therefore, to accurately reflect the structural behavior, the actual structural deformation increment of this sub-low-frequency deformation time series is determined to be 0.

[0060] Furthermore, when a sub-low-frequency deformation time series is identified as the dominant segment of structural deformation, it indicates that the deformation within that segment primarily reflects the actual deformation of the bridge structure, and the influence of sensor zero drift can be ignored. In this case, the actual structural deformation increment of the sub-low-frequency deformation time series is directly determined as the deformation observed in the corresponding low-frequency deformation time series.

[0061] Furthermore, when a sub-low-frequency deformation time series is identified as a dual-source superposition segment, it indicates that the deformation within that segment is the result of the combined effects of sensor zero drift and bridge structural deformation. To obtain the true structural deformation increment, the estimated sensor zero drift increment needs to be subtracted from the total deformation in the low-frequency deformation time series corresponding to this dual-source superposition segment. In this way, the deformation component of the structure itself can be effectively separated.

[0062] Through the above technical solutions, this application can accurately and specifically quantify the true deformation increment of a bridge structure according to different deformation types. Specifically, for sections dominated by sensor zero drift, by setting the true deformation increment of the structure to 0, the interference of sensor zero drift on the structural deformation assessment is effectively avoided, improving the purity of the measurement results. For sections dominated by structural deformation, the deformation in the low-frequency deformation time series is directly used as the true increment, ensuring measurement efficiency and data directness. Especially in dual-source superposition sections, by subtracting the sensor zero drift increment, this application can accurately separate the deformation of the structure itself from the complex superimposed signals, significantly improving the accuracy and reliability of structural deformation measurement under the influence of multiple factors, and providing more accurate data support for bridge structural health monitoring.

[0063] Specifically, the step of determining the actual structural deformation increment of the sub-low-frequency deformation time series of the deformation type of the dual-source superimposed segment as the deformation amount obtained by subtracting the sensor zero drift increment from the deformation in the low-frequency deformation time series corresponding to the dual-source superimposed segment can be further refined into the following process.

[0064] The steps for determining the true structural deformation increment of a sub-low-frequency deformation time series with a deformation type of dual-source superposition segment as the deformation amount obtained by subtracting the sensor zero drift increment from the deformation in the low-frequency deformation time series corresponding to the dual-source superposition segment include: Obtain the drift parameter benchmark of the sensor zero-drift dominant section that corresponds to the deformation type of the dual-source superposition section and is in a stable state; The theoretical increment of sensor zero drift is calculated based on the drift parameter benchmark and the sub-low frequency state sequence corresponding to the dual-source superposition section. The actual structural deformation increment of the sub-low-frequency deformation time series of the deformation type of the dual-source superposition section is determined by subtracting the theoretical increment of sensor zero drift from the deformation in the low-frequency deformation time series corresponding to the dual-source superposition section.

[0065] The acquisition of drift parameter benchmarks for the sensor's zero-drift dominant segment, corresponding to the dual-source superposition deformation type and in a stable state, refers to obtaining a set of parameters characterizing the sensor's drift characteristics in a stable zero-drift state through historical data analysis or pre-calibration. This drift parameter benchmark can be understood as the sensor's inherent drift behavior under specific environmental conditions, such as its drift rate or model parameters varying with factors like temperature and time. Its purpose is to provide a reliable reference for subsequent zero-drift quantification.

[0066] Furthermore, calculating the theoretical zero-drift increment of the sensor based on the drift parameter benchmark and the corresponding low-frequency state sequence of the dual-source superposition section refers to using the acquired drift parameter benchmark, combined with the actual environmental conditions within the dual-source superposition section (such as temperature sequence, sensor power supply voltage sequence, sensor cumulative operating time sequence, etc.), and through an established mathematical model or empirical formula, to predict the deformation increment of the sensor due to its own zero-drift effect during this time period. This theoretical zero-drift increment of the sensor is a key value used to quantify the sensor's own error.

[0067] Therefore, the actual deformation increment of the sub-low-frequency deformation time series of the deformation type of the dual-source superimposed section is determined by subtracting the theoretical increment of sensor zero drift from the deformation in the low-frequency deformation time series corresponding to the dual-source superimposed section. This means that in the dual-source superimposed section, the total deformation in the low-frequency deformation time series actually measured by the sensor is subtracted from the theoretical increment of sensor zero drift calculated by the above method, thereby separating the actual increment caused only by the deformation of the bridge structure itself.

[0068] Through the above technical solution, this application can effectively solve the problem of accurately separating sensor zero drift from structural deformation in dual-source superposition sections. Specifically, by introducing a drift parameter benchmark and calculating the theoretical increment of sensor zero drift, the quantification of the sensor's own error is made more accurate, thereby improving the ability to extract the true structural deformation from the total deformation. Therefore, this application can provide more accurate and reliable bridge structural deformation data, significantly improving the accuracy and effectiveness of bridge health monitoring and avoiding misjudgments or omissions caused by sensor zero drift.

[0069] In some embodiments, this application further proposes a step of obtaining a drift parameter reference for a sensor zero-drift dominant segment that corresponds to a segment with a deformation type of dual-source superposition and is in a stable state, including: Extract the most recent sensor zero-drift dominant segment from the historical zero-drift parameter evolution queue, which is a collection of sub-low-frequency deformation time series that have been confirmed in the continuous history as sensor zero-drift dominant segments. The drift rate of the most recent sensor zero-drift dominant segment is extracted as the drift parameter benchmark.

[0070] Specifically, the historical zero-drift parameter evolution queue can be understood as a dynamically updated database or storage structure that continuously records and stores the sub-low-frequency deformation time series and their associated drift parameters that are identified and confirmed by the system as the sensor's zero-drift dominant segment within different time periods. The purpose of this queue is to accumulate historical zero-drift behavior data of the sensor for querying and utilization when needed. The sub-low-frequency deformation time series identified as the sensor's zero-drift dominant segment refers to those deformation data segments that, through analysis of their evolutionary characteristic vectors (e.g., Pearson correlation coefficient and residual deformation rate), are clearly determined to be deformation data segments primarily caused by the sensor's own zero drift.

[0071] Furthermore, extracting the most recent dominant segment of sensor zero drift from the historical zero drift parameter evolution queue means that when a drift parameter benchmark needs to be determined, the system will retrieve and select the most recent confirmed dominant segment of sensor zero drift from the aforementioned historical queue. The purpose of selecting the most recent segment is to ensure that the selected benchmark can reflect the current physical state and aging characteristics of the sensor to the greatest extent, because the zero drift characteristics of the sensor may change slowly over time, due to environmental factors, or usage conditions.

[0072] Therefore, the drift rate of the most recent sensor zero-drift dominant section is extracted as the drift parameter benchmark. The drift rate refers to the rate at which the sensor output signal changes over time within the sensor zero-drift dominant section; it quantifies the degree of sensor zero-point drift over time. This drift rate, as a benchmark, will be used to subsequently calculate the theoretical increment of sensor zero-drift within the dual-source superposition section, thereby achieving the goal of accurately separating the actual structural deformation from the total deformation.

[0073] Through the above technical solution, this application ensures that the obtained drift parameter benchmark has higher timeliness and accuracy, thereby significantly improving the accuracy of the theoretical increment calculation of sensor zero drift. Since the drift parameter benchmark is key to separating sensor zero drift and structural deformation in a dual-source superposition section, its improved accuracy directly leads to enhanced reliability in determining the actual deformation increment of the structure. Furthermore, by continuously maintaining the historical zero drift parameter evolution queue, the solution of this application also possesses adaptive and learning capabilities, better able to cope with changes in zero drift characteristics that may occur during long-term sensor operation, making the entire bridge structural deformation measurement method exhibit stronger robustness and stability in long-term monitoring applications.

[0074] In some embodiments, this application further proposes a step for calculating the theoretical increment of sensor zero drift based on a drift parameter benchmark and a low-frequency state sequence within the dual-source superposition segment, including: Calculate the first time derivative of the temperature sequence in the low-frequency state sequence corresponding to the dual-source superposition section, and divide the dual-source superposition section into a temperature rising branch period and a temperature falling branch period according to the sign of the first time derivative. The positive unit temperature drift rate corresponding to the heating process and the negative unit temperature drift rate corresponding to the cooling process are extracted from the corresponding sensor zero drift dominant section to construct an asymmetric thermal hysteresis loop model. The asymmetric thermal hysteresis loop model is used to calculate the transient thermal hysteresis temperature drift. According to the asymmetric thermal hysteresis loop model, if the sampling point corresponding to the dual-source superposition section is in the temperature rise branch period, the positive unit temperature drift rate is multiplied by the temperature change at the current moment relative to the starting point of the current temperature rise branch to obtain the transient thermal hysteresis temperature drift at the current moment. According to the asymmetric thermal hysteresis loop model, if the sampling point corresponding to the dual-source superposition section is in the temperature drop branch period, the reverse unit temperature drift rate is multiplied by the temperature change at the current moment relative to the starting point of the current cooling branch to obtain the transient thermal hysteresis temperature drift at the current moment. The cumulative aging drift is calculated based on the sensor's cumulative runtime and the drift parameter benchmark. The cumulative aging drift is then summed with the transient thermal hysteresis temperature drift to obtain the theoretical increment of sensor zero drift at the current moment within the dual-source superposition section.

[0075] Specifically, when calculating the theoretical increment of sensor zero drift, it is first necessary to analyze the temperature sequence in the low-frequency state sequence corresponding to the dual-source superposition section. By calculating the first-order time derivative of the temperature sequence, the trend of temperature change can be identified. When the first-order time derivative is positive, it indicates that the temperature is on an upward trend, forming a temperature rising branch period; when the first-order time derivative is negative, it indicates that the temperature is on a downward trend, forming a temperature falling branch period. This division aims to accurately capture the impact of temperature changes on sensor zero drift, because the sensor's response may differ during heating and cooling processes.

[0076] To more accurately quantify zero drift caused by temperature changes, this application constructs an asymmetric thermal hysteresis hysteresis loop model. This model is established by extracting the forward unit temperature drift rate corresponding to the heating process and the reverse unit temperature drift rate corresponding to the cooling process from the historical zero drift-dominant region of the sensor. The forward unit temperature drift rate characterizes the amount of drift of the sensor during a unit temperature increase, while the reverse unit temperature drift rate characterizes the amount of drift of the sensor during a unit temperature decrease. Due to the nonlinearity of sensor material properties and internal stress release mechanisms, the drift rates during heating and cooling processes are often asymmetric; therefore, using an asymmetric model can more realistically reflect the thermal hysteresis characteristics of the sensor. This asymmetric thermal hysteresis hysteresis loop model is specifically used to calculate the transient thermal hysteresis temperature drift.

[0077] In practical applications, based on the constructed asymmetric thermal hysteresis hysteresis model, the calculation of transient thermal hysteresis temperature drift is performed according to the current temperature change branch. Specifically, if the sampling point of the dual-source superposition section is in the temperature rising branch period, the pre-extracted positive unit temperature drift rate is multiplied by the temperature change at the current moment relative to the starting point of the current temperature rising branch to obtain the transient thermal hysteresis temperature drift at the current moment. Conversely, if the sampling point is in the temperature falling branch period, the reverse unit temperature drift rate is multiplied by the temperature change at the current moment relative to the starting point of the current temperature falling branch to obtain the transient thermal hysteresis temperature drift at the current moment. This differentiated processing ensures the accuracy of temperature drift calculation and avoids errors introduced by simple linear models.

[0078] In addition to the transient thermal hysteresis temperature drift, long-term sensor operation also leads to the accumulation of zero drift, i.e., cumulative aging drift. This cumulative aging drift is calculated based on the sensor's cumulative operating time and a preset drift parameter benchmark. The drift parameter benchmark can be obtained from a historical zero drift parameter evolution queue, such as the drift rate of the most recent sensor zero drift-dominant segment. Finally, by summing the calculated cumulative aging drift with the transient thermal hysteresis temperature drift, the theoretical increment of sensor zero drift at the current moment within the dual-source superposition segment can be obtained.

[0079] Through the above technical solution, this application can more accurately calculate the theoretical increment of sensor zero drift, significantly improving the identification and quantification accuracy of sensor zero drift in dual-source superposition sections. Compared with basic schemes that only use a single or simplified model to estimate zero drift, this application comprehensively captures the complex dynamic characteristics of zero drift by introducing the division of temperature rise / fall branches, constructing an asymmetric thermal hysteresis loop model, and considering the cumulative aging drift of the sensor. This refined zero drift estimation method effectively avoids the zero drift estimation errors caused by the asymmetry of temperature changes, thermal hysteresis effect, and long-term aging effect, thus making the zero drift component subtracted from the low-frequency deformation time series closer to the actual situation. As a result, the final obtained true structural deformation increment has higher accuracy and reliability, providing more accurate data support for bridge structural health monitoring and improving the overall performance of the monitoring system and the scientific basis of decision-making.

[0080] Secondly, this application proposes a bridge structural deformation measurement system based on intelligent sensors, comprising: The acquisition module is used to acquire the initial deformation time series of the bridge structure measuring points and the initial state sequence used to characterize the zero-point stability of the measurement sensor, wherein the initial state sequence includes at least the temperature sequence, the sensor power supply voltage sequence and the sensor cumulative running time sequence. The preprocessing module is used to perform time alignment and low-frequency filtering on the initial deformation time series and the initial state series to generate a low-frequency deformation time series and a low-frequency state series. The subsequence division module is used to divide the low-frequency deformation time series into multiple sub-low-frequency deformation time series based on the temperature change inflection point and the sensor operation state inflection point in the low-frequency state series. The temperature change inflection point is the time point when the temperature change gradient in the low-frequency state series changes abruptly, and the sensor operation state inflection point is the time point when the sensor power supply voltage and operation continuity change abruptly. The evolution feature vector construction module is used to construct the evolution feature vectors of each sub-low-frequency deformation time series for each sub-low-frequency deformation time series. The deformation type determination module is used to determine the deformation type of each sub-low-frequency deformation time series based on the evolution feature vector and in combination with preset screening conditions. The deformation type includes sensor zero drift-dominated segment, structural deformation-dominated segment and dual-source superposition segment, wherein the dual-source superposition segment is the segment where sensor zero drift and bridge structural deformation are superimposed. The actual deformation increment determination module is used to determine the actual structural deformation increment corresponding to each sub-low-frequency deformation time series based on the deformation type. The deformation determination module is used to accumulate the actual structural deformation increments of each sub-low-frequency deformation time series in chronological order to obtain a continuous actual structural deformation trend series.

[0081] The acquisition module can be configured in various forms. For example, it can be a data acquisition unit containing an analog-to-digital converter and a data storage device, used to receive analog signals from various sensors and convert them into digital signals for storage; or, the acquisition module can be a communication interface used to receive pre-collected deformation and state data from external data sources (such as remote servers or local storage devices). As a preferred implementation, the acquisition module can be integrated inside the smart sensor to directly perform data acquisition and preliminary processing.

[0082] The preprocessing module can be implemented as a software program running on a processor, which includes time alignment algorithms (such as linear interpolation and spline interpolation) and low-frequency filtering algorithms (such as moving average and Butterworth filters). Alternatively, the preprocessing module can be implemented as a dedicated digital signal processor or field-programmable gate array (FPGA) to efficiently perform data alignment and filtering operations in a hardware-accelerated manner.

[0083] The subsequence partitioning module can be implemented as a software algorithm that continuously monitors the temperature gradient and sensor power supply voltage in the low-frequency state sequence, and identifies inflection points based on preset thresholds and derivative changes. This module can run on the main controller and is responsible for logically dividing the continuous data stream into discrete subsequences for independent processing by subsequent modules.

[0084] The evolutionary feature vector construction module can be implemented as a set of functions in a data analysis software library to calculate the statistical features (such as mean, variance, and slope) and correlation features (such as Pearson correlation coefficient) of subsequences. This module can utilize multithreading or parallel computing techniques to extract features from multiple subsequences simultaneously, thereby improving processing efficiency.

[0085] The deformation type determination module can be implemented as a rule-based expert system or a machine learning classifier. For example, a series of logical rules can be preset to determine the deformation type based on a specific value range in the evolutionary feature vector; alternatively, a support vector machine or neural network model can be trained to automatically classify the deformation type based on the input evolutionary feature vector. The specific details of determining the actual structural deformation increment corresponding to each sub-low-frequency deformation time series based on the deformation type have already been described in the above embodiments and will not be repeated here. It is important to emphasize that the actual deformation increment determination module can be implemented as a set of computational logic that calls different computational subroutines based on the deformation type. For example, for sensor zero-drift dominant sections, zero increment is directly output; for structural deformation dominant sections, the original deformation value is directly output; and for dual-source superposition sections, a zero-drift subtraction algorithm is executed. This module ensures accurate separation of different deformation sources.

[0086] The deformation determination module can be implemented as a data accumulator or a cumulative summation algorithm. This module is responsible for accumulating the discrete structural real deformation increments output by the previous module according to their original time sequence, and finally generating a smooth and continuous sequence of structural real deformation trends. This sequence can be directly used for long-term health monitoring and assessment of bridge structures.

[0087] The bridge structure deformation measurement system based on intelligent sensors proposed in this application achieves efficient differentiation and accurate separation between sensor zero drift and actual structural deformation. In particular, the introduction of subsequence segmentation and deformation type determination modules enables the system to intelligently identify and classify deformation sources based on multi-dimensional state information and evolutionary characteristics, and to adopt differentiated processing strategies for different types. This system-level integration and intelligent analysis capability significantly improves the accuracy and reliability of bridge structure deformation monitoring, providing more scientific and reliable data support for the safety assessment and maintenance of bridge structures, and effectively avoiding false alarms or missed alarms that may occur in traditional methods.

[0088] The above provides a detailed description of the preferred embodiments of this application. However, this application is not limited to the above-described embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application. All such equivalent modifications or substitutions are included within the scope defined in this application.

Claims

1. A method for measuring bridge structural deformation based on intelligent sensors, characterized in that, Includes the following steps: The initial deformation time series of the bridge structure measuring points and the initial state series used to characterize the zero-point stability of the measurement sensor are obtained, wherein the initial state series includes at least the temperature series, the sensor power supply voltage series and the sensor cumulative running time series. Time alignment and low-frequency filtering are performed on the initial deformation time series and the initial state series to generate a low-frequency deformation time series and a low-frequency state series. Based on the temperature change inflection point and the sensor operating state inflection point in the low-frequency state sequence, the low-frequency deformation time sequence is divided into multiple sub-low-frequency deformation time sequences. The temperature change inflection point is the time point when the temperature change gradient in the low-frequency state sequence changes abruptly, and the sensor operating state inflection point is the time point when the sensor power supply voltage and operating continuity change abruptly. For each sub-low-frequency deformation time series, construct the evolution feature vector of each sub-low-frequency deformation time series; Based on the evolution feature vector and combined with the preset screening conditions, the deformation type of each sub-low-frequency deformation time series is determined. The deformation type includes sensor zero drift-dominated segment, structural deformation-dominated segment and dual-source superposition segment, wherein the dual-source superposition segment is the segment where sensor zero drift and bridge structural deformation are superimposed. Based on the deformation type, determine the actual structural deformation increment corresponding to each sub-low-frequency deformation time series; By accumulating the actual structural deformation increments of each sub-low-frequency deformation time series in chronological order, a continuous actual structural deformation trend series is obtained.

2. The method according to claim 1, characterized in that, The step of performing time alignment and low-frequency filtering on the initial deformation time series and the initial state series to generate a low-frequency deformation time series and a low-frequency state series includes: Within a defined sliding window, time alignment is performed on the initial deformation time sequence and the initial state sequence; The initial deformation time series and the initial state series that are time-aligned are used to remove transient spikes and impact interference using a threshold determination method to obtain a denoised deformation time series and a denoised state series. Weighted moving averages are applied to the denoised deformation time series and the denoised state series respectively to generate a low-frequency deformation time series and a low-frequency state series.

3. The method according to claim 1, characterized in that, The step of dividing the low-frequency deformation time series into multiple sub-low-frequency deformation time series based on the temperature change inflection point and the sensor operating state inflection point in the low-frequency state sequence includes: Within a set sliding window, the variance of the temperature change gradient within the set sliding window is continuously calculated. The set sliding window contains N sampling points, where N is a natural number from 10 to 1000. When the variance of the temperature change gradient is greater than a preset variance threshold, the sampling point where the sign of the first time derivative of the temperature changes within the sliding window is determined as the temperature change inflection point. When the change in the sensor power supply voltage exceeds the preset voltage change threshold, the sampling point where the sign of the first time derivative of the voltage changes within the sliding window is determined as the inflection point of the sensor's operating state. The low-frequency deformation time series is cut off at the time points corresponding to the inflection point of the temperature change and the inflection point of the sensor's operating state, and the low-frequency deformation time series is divided into multiple sub-low-frequency deformation time series.

4. The method according to claim 1, characterized in that, The step of constructing the evolution feature vector of each sub-low-frequency deformation time series includes: Multiple linear deformation time curves are obtained by linearly fitting each sub-low-frequency deformation time series. Based on the multiple linear deformation time curves, a sequence of predicted values ​​at corresponding time points is generated to obtain multiple sub-linear trend deformation sequences; For each of the aforementioned sub-low-frequency deformation time series, the corresponding sub-linear trend deformation series is removed to obtain the deformation residual series; The residual deformation rate is calculated based on the deformation residual sequence. Calculate the Pearson correlation coefficient between the deformation residual sequence and the temperature sequence in the same region; The residual deformation rate and the Pearson correlation coefficient between the deformation residual sequence and the temperature sequence in the same segment are summarized to construct the evolution feature vector of each sub-low-frequency deformation time series.

5. The method according to claim 1, characterized in that, The step of determining the deformation type of each sub-low-frequency deformation time series based on the evolutionary feature vector and in combination with preset screening conditions includes: If the absolute value of the Pearson correlation coefficient in the evolution feature vector of the sub-low-frequency deformation time series is higher than the preset coefficient threshold, and the residual deformation rate matches the preset sensor calibration aging drift model, then the sub-low-frequency deformation time series is marked as the sensor zero drift dominant segment. If the absolute value of the Pearson correlation coefficient in the evolution feature vector of the sub-low-frequency deformation time series is lower than the preset coefficient threshold, and the residual deformation rate matches the theoretical creep prediction rate of the measured structure, then the sub-low-frequency deformation time series is marked as the structural deformation-dominant segment. If the sub-low-frequency deformation time series does not meet the conditions of the sensor zero-drift dominant section and the structural deformation dominant section, then the sub-low-frequency deformation time series is marked as a dual-source superposition section.

6. The method according to claim 1, characterized in that, The step of determining the actual structural deformation increment corresponding to each sub-low-frequency deformation time series based on the deformation type includes: The actual structural deformation increment of the sub-low-frequency deformation time series whose deformation type is sensor zero-drift dominant segment is determined to be 0; The actual structural deformation increment of the sub-low-frequency deformation time series whose deformation type is the structural deformation-dominant segment is determined as the deformation in the corresponding low-frequency deformation time series. The actual structural deformation increment of the sub-low-frequency deformation time series of the deformation type of the dual-source superposition section is determined as the deformation amount obtained by subtracting the sensor zero drift increment from the deformation in the low-frequency deformation time series corresponding to the dual-source superposition section.

7. The method according to claim 6, characterized in that, The step of determining the actual structural deformation increment of the sub-low-frequency deformation time series of the deformation type being a dual-source superimposed segment as the deformation amount obtained by subtracting the sensor zero drift increment from the deformation in the low-frequency deformation time series corresponding to the dual-source superimposed segment includes: Obtain the drift parameter benchmark of the sensor zero-drift dominant section that corresponds to the dual-source superposition section of the deformation type and is in a stable state; The theoretical increment of sensor zero drift is calculated based on the drift parameter benchmark and the sub-low frequency state sequence corresponding to the dual-source superposition section. The actual structural deformation increment of the sub-low-frequency deformation time series of the deformation type of the dual-source superposition section is determined by subtracting the theoretical increment of sensor zero drift from the deformation in the low-frequency deformation time series corresponding to the dual-source superposition section.

8. The method according to claim 7, characterized in that, The step of obtaining the drift parameter reference of the sensor zero-drift dominant section corresponding to the deformation type of the dual-source superposition section and in a stable state includes: Extract the most recent sensor zero-drift dominant segment from the historical zero-drift parameter evolution queue, which is a collection of sub-low-frequency deformation time series that have been confirmed in the continuous history as sensor zero-drift dominant segments. The drift rate of the most recent sensor zero-drift dominant segment is extracted as the drift parameter benchmark.

9. The method according to claim 7, characterized in that, The step of calculating the theoretical increment of sensor zero drift based on the drift parameter benchmark and the sub-low-frequency state sequence corresponding to the dual-source superposition segment includes: Calculate the first time derivative of the temperature sequence in the low-frequency state sequence corresponding to the dual-source superposition section, and divide the dual-source superposition section into a temperature rising branch period and a temperature falling branch period according to the sign of the first time derivative. The positive unit temperature drift rate corresponding to the heating process and the negative unit temperature drift rate corresponding to the cooling process are extracted from the corresponding sensor zero drift dominant section to construct an asymmetric thermal hysteresis loop model. The asymmetric thermal hysteresis loop model is used to calculate the transient thermal hysteresis temperature drift. According to the asymmetric thermal hysteresis loop model, if the sampling point corresponding to the dual-source superposition section is in the temperature rise branch period, the positive unit temperature drift rate is multiplied by the temperature change at the current moment relative to the starting point of the current temperature rise branch to obtain the transient thermal hysteresis temperature drift at the current moment. According to the asymmetric thermal hysteresis loop model, if the sampling point corresponding to the dual-source superposition section is in the temperature drop branch period, the reverse unit temperature drift rate is multiplied by the temperature change at the current moment relative to the starting point of the current cooling branch to obtain the transient thermal hysteresis temperature drift at the current moment. The cumulative aging drift is calculated based on the sensor's cumulative runtime and the drift parameter benchmark. The cumulative aging drift is then summed with the transient thermal hysteresis temperature drift to obtain the theoretical increment of sensor zero drift at the current moment within the dual-source superposition section.

10. A bridge structural deformation measurement system based on intelligent sensors, characterized in that, include: The acquisition module is used to acquire the initial deformation time series of the bridge structure measuring points and the initial state sequence used to characterize the zero-point stability of the measurement sensor, wherein the initial state sequence includes at least the temperature sequence, the sensor power supply voltage sequence and the sensor cumulative running time sequence. The preprocessing module is used to perform time alignment and low-frequency filtering on the initial deformation time series and the initial state series to generate a low-frequency deformation time series and a low-frequency state series. The subsequence division module is used to divide the low-frequency deformation time series into multiple sub-low-frequency deformation time series based on the temperature change inflection point and the sensor operation state inflection point in the low-frequency state series. The temperature change inflection point is the time point when the temperature change gradient in the low-frequency state series changes abruptly, and the sensor operation state inflection point is the time point when the sensor power supply voltage and operation continuity change abruptly. The evolution feature vector construction module is used to construct the evolution feature vectors of each sub-low-frequency deformation time series for each sub-low-frequency deformation time series. The deformation type determination module is used to determine the deformation type of each sub-low-frequency deformation time series based on the evolution feature vector and in combination with preset screening conditions. The deformation type includes sensor zero drift-dominated segment, structural deformation-dominated segment and dual-source superposition segment, wherein the dual-source superposition segment is the segment where sensor zero drift and bridge structural deformation are superimposed. The actual deformation increment determination module is used to determine the actual structural deformation increment corresponding to each sub-low-frequency deformation time series based on the deformation type. The deformation determination module is used to accumulate the actual structural deformation increments of each sub-low-frequency deformation time series in chronological order to obtain a continuous actual structural deformation trend series.