Bridge anti-collision sensor data anomaly detection method and system

Abstract

The present application belongs to the technical field of bridge health monitoring data processing, and particularly relates to a bridge anti-collision sensor data anomaly detection method and system, which comprises the following steps: collecting the vibration acceleration sequence and the absolute displacement sequence of the bridge pier; using the second integral of the acceleration data to calculate the theoretical displacement change, comparing it with the measured displacement, and calculating the kinematic consistency residual; calculating the vibration energy density of the vibration acceleration sequence and the displacement fluctuation power of the absolute displacement sequence, constructing the energy response coupling confidence model, and calculating the logic coupling anomaly index; comparing the kinematic consistency residual and the logic coupling anomaly index with the preset threshold, respectively, to realize the anomaly detection of the bridge anti-collision sensor data. The present application has the authenticity verification ability under the constraint of the physical field, can solve the sensor false alarm problem, and has extremely high numerical robustness.

Description

Technical Field

[0001] This invention relates to the field of bridge health monitoring data processing technology. More specifically, this invention relates to a method and system for detecting anomalies in bridge collision avoidance sensor data. Background Technology

[0002] Bridge collision avoidance monitoring systems are an important means of ensuring the safety of transportation infrastructure. They typically integrate multiple sensors such as GNSS displacement gauges, accelerometers, and hydrostatic levels. These sensors are used to detect collision events of ships or vehicles in real time. However, since these sensors operate in the humid, vibrating, and electromagnetically complex environment of outdoor bridges for extended periods, the stability and reliability of their data face severe challenges.

[0003] Current technologies for detecting anomalies in sensor data primarily rely on simple threshold methods: such as setting a fixed displacement reading threshold, triggering an alarm once the sensor data exceeds the threshold. However, simple threshold methods have significant limitations: firstly, sensors are susceptible to temperature fluctuations causing zero-point drift, resulting in a slow increase in readings without actual displacement; secondly, lightning strikes or electromagnetic interference can induce electrical spikes, leading to instantaneous maxima. These data fluctuations cause frequent false alarms, severely reducing the reliability of the monitoring system.

[0004] More importantly, a single data source cannot achieve self-verification. When sensor data shows that the bridge pier has undergone significant displacement, the system has difficulty determining whether this is a real structural deformation or an abnormal jump caused by satellite signal obstruction. Without external reference, the authenticity of data from a single sensor cannot be verified. Summary of the Invention

[0005] To address the problem that single sensors are susceptible to interference and false alarms and cannot verify the authenticity of their data, this invention provides a method and system for detecting abnormal data from bridge collision avoidance sensors. This method utilizes physical laws to perform dual verification of sensor data, effectively eliminating abnormal data.

[0006] Solutions are provided in the following areas.

[0007] In a first aspect, the present invention provides a method for detecting anomalies in bridge collision avoidance sensor data, comprising: acquiring vibration acceleration sequences and absolute displacement sequences of bridge piers, and interpolating and resampling the absolute displacement sequence using the time axis of the vibration acceleration sequence as a reference to achieve spatiotemporal alignment of multidimensional heterogeneous data; based on the aligned data, using the quadratic integral of the acceleration data to estimate the theoretical displacement change, and comparing it with the measured displacement to calculate the kinematic consistency residual characterizing the degree of deviation between the measured displacement and the physically estimated displacement; calculating the vibration energy density of the vibration acceleration sequence and the displacement fluctuation power of the absolute displacement sequence, constructing an energy response coupling confidence model based on the logarithmic linear correlation between the two, and calculating the logical coupling anomaly index; comparing the kinematic consistency residual and the logical coupling anomaly index with preset thresholds respectively, determining whether the sensor data has drift, crash, or noise faults based on the comparison results, and outputting cleaned monitoring data to achieve anomaly detection of bridge collision avoidance sensor data.

[0008] By adopting the above technical solution, a data cleaning mechanism based on multi-physics spatiotemporal causal constraints is proposed. This mechanism constructs a kinematic residual model and an energy response coupling model based on the calculus-integral physical relationship between acceleration and displacement, and the energy correlation between high-frequency vibration and low-frequency deformation, respectively. This invention possesses the ability to verify the authenticity of data under physical field constraints and does not rely on large-sample training. Any data that violates physical laws can be identified in real time, thus fundamentally solving the problem of sensor false alarms caused by data distortion. Simultaneously, by introducing a zero-reduction protection term and adopting a logarithmic ratio calculation form, the numerical stability of the algorithm is ensured under conditions such as zero sensor output or extreme values, resulting in high system robustness. Furthermore, its key parameters can be acquired through self-learning, eliminating the need for manual calibration and achieving adaptive startup under zero-sample conditions.

[0009] Preferably, the step of interpolating and resampling the absolute displacement sequence specifically includes: obtaining the high-frequency sampling rate of the vibration acceleration sequence and the low-frequency sampling rate of the absolute displacement sequence; using Lagrange interpolation to upsample the low-frequency absolute displacement sequence so that its time resolution is consistent with that of the vibration acceleration sequence, thereby constructing a unified time series matrix.

[0010] By adopting the above technical solution, the low-frequency absolute displacement sequence is sampled using the Lagrange interpolation method, which solves the problem that the sampling rates of different sensors are inconsistent and cannot be directly used for collaborative calculation. This provides a unified time reference for subsequent multiphysics joint analysis and ensures strict synchronization and correspondence of different data sources in the time domain.

[0011] Preferably, the kinematic consistency residuals satisfy the following relationship: ; In the formula, for kinematic consistency residuals at any given time; These are measured displacement data; The displacement reference value is calculated at the beginning of the calculation window; The acceleration is the result of high-pass filtering; For unit conversion and sensitivity correction coefficients; This represents the numerical second integral operation; This is the start time of the integration process; , It is the integral variable.

[0012] By adopting the above technical solution and utilizing the inherent calculus physical relationship between acceleration and displacement, it is possible to accurately identify sensor drift or abnormal jumps that violate basic kinematic laws, thus realizing data verification and cleaning based on physical essence.

[0013] Preferably, the high-pass filtered acceleration is obtained by applying a Butterworth high-pass filter to the collected raw acceleration data to filter out the gravity component and the DC bias of the sensor.

[0014] By adopting the above technical solution, a Butterworth high-pass filter is applied to the raw acceleration data, effectively filtering out the gravity component and the DC bias of the sensor. This avoids the problem of integral divergence caused by infinitely amplifying the DC component in the integral calculation, thus ensuring that only the dynamic acceleration component caused by the impact event is retained for subsequent calculations.

[0015] Preferably, the logical coupling anomaly index satisfies the following relationship: ; In the formula, for The logical coupling anomaly index at any given moment; Vibrational energy density; This refers to the displacement fluctuation power; To prevent the protection constant from being reduced to zero; The reference coupling constant represents the inherent vibration-displacement conversion ratio of the bridge pier structure.

[0016] Preferably, the calculation method for the vibration energy density and the displacement fluctuation power includes: calculating the root mean square value of the vibration acceleration sequence within a sliding window as the vibration energy density; and calculating the standard deviation of the absolute displacement sequence within the sliding window as the displacement fluctuation power.

[0017] Preferably, the reference coupling constant is obtained by taking the average value of the logarithmic ratio of vibration energy density to displacement fluctuation power during a period of system fault-free operation.

[0018] Preferably, the step of determining whether the sensor data has drift, crash, or noise fault based on the comparison result includes: when the kinematic consistency residual is greater than a first preset threshold and the logical coupling anomaly index is less than a second preset threshold, it is determined that the displacement sensor is drifting, and the displacement data is baseline corrected using the trend line obtained by acceleration integration; when the logical coupling anomaly index is greater than the second preset threshold, it is determined that the sensor hardware has failed, the corresponding data source weight is cut off, and an alarm message is sent.

[0019] By adopting the above technical solution, a multi-level anomaly decision matrix is ​​constructed, which can take targeted processing strategies according to different fault types. For example, baseline correction is performed on drift data, and alarms are triggered and the impact is cut off for hardware failure data. This mechanism ultimately outputs clean and reliable monitoring data, which significantly reduces the system's false alarm rate.

[0020] Preferably, the determination of sensor hardware failure further includes: if the displacement fluctuation power is close to zero and the vibration energy density is greater than zero, it is determined to be a system crash; if the displacement fluctuation power is very large and the vibration energy density is close to zero, it is determined to be a noise state.

[0021] Secondly, the present invention provides a bridge collision avoidance sensor data anomaly detection system, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned bridge collision avoidance sensor data anomaly detection method is implemented.

[0022] By adopting the above technical solution, a computer program is generated from the above-mentioned method for detecting abnormal data of bridge collision avoidance sensors and stored in a memory so that it can be loaded and executed by a processor. A terminal device can then be made based on the memory and the processor for convenient use.

[0023] The beneficial effects of this invention are as follows: This invention proposes a data cleaning method based on multi-physics spatiotemporal causal constraints. Based on the inherent calculus-integral physical relationship between acceleration and displacement, and the energy correlation between high-frequency vibration and low-frequency deformation, this method constructs a kinematic residual model and an energy response coupling model, respectively, thereby achieving physical consistency verification of sensor data. This method does not rely on large-sample training and can verify data authenticity in real time under physical constraints. Once data anomalies violating physical laws are detected, they can be identified immediately, thus fundamentally suppressing false alarms caused by data distortion.

[0024] Furthermore, to improve the stability of the algorithm under conditions such as zero sensor output or extreme values, this invention introduces a zero-reduction protection mechanism and adopts a logarithmic ratio-based computational architecture, significantly enhancing the numerical robustness of the system. Simultaneously, key parameters can be acquired through self-learning, eliminating the need for manual calibration and enabling zero-sample adaptive initialization and operation of the system. Attached Figure Description

[0025] Figure 1 This is a flowchart of a method for detecting abnormal data from a bridge collision avoidance sensor according to the present invention; Figure 2 This is a comparison diagram of sensor drift fault identification and correction in an embodiment of the present invention; Figure 3 This is a scatter plot for energy-response coupling anomaly diagnosis in an embodiment of the present invention; Figure 4 This is a timing diagram of real-time monitoring and alarm of abnormal index in an embodiment of the present invention. Detailed Implementation

[0026] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0027] This invention discloses a method for detecting abnormal data from bridge collision avoidance sensors, referring to... Figure 1 This includes steps S1-S4: S1. Collect the vibration acceleration sequence and absolute displacement sequence of the bridge pier, and use the time axis of the vibration acceleration sequence as a reference to interpolate and resample the absolute displacement sequence to achieve spatiotemporal alignment of multidimensional heterogeneous data.

[0028] In an optional embodiment, it is first necessary to collect the vibration acceleration sequence of the bridge piers. and absolute displacement sequence When collecting vibration acceleration sequences of bridge piers At the same time, the vibration acceleration sequence of the bridge piers is first collected using sensors capable of acquiring high-frequency dynamic information, such as piezoelectric or MEMS accelerometers installed on the piers. It is important to note that the vibration acceleration sequence of the bridge piers was collected. The sampling frequency of the sensor needs to be set to a high level, as it is used to collect the vibration acceleration sequence of the bridge pier in this invention. One implementation of the sensor's setting frequency is to set its sampling frequency to 100 Hz.

[0029] When collecting the absolute displacement sequence of bridge piers At that time, the absolute displacement sequence of the bridge piers is collected by sensors capable of acquiring low-frequency static information, such as GNSS receivers or hydrostatic levels installed around the bridge piers. It is important to note that the absolute displacement sequence of the bridge piers should be collected. The sampling frequency of the sensor does not need to be set to a very high level, as it is used in this invention to collect the absolute displacement sequence of the bridge piers. One implementation of the sensor's setting frequency is to set its sampling frequency to 1 Hz.

[0030] Because of the acquisition of vibration acceleration sequences of bridge piers Sensors and acquisition of absolute displacement sequences of bridge piers The sensors used in this invention have different sampling frequencies, making it impossible to directly process these two types of data. Therefore, in this embodiment of the invention, a high-frequency vibration acceleration sequence is used. Using the time axis as a reference, Lagrange interpolation is employed for low frequencies. Upsampling is performed; specifically, assuming at time point... To calculate the displacement value, a Lagrange polynomial is constructed using low-frequency displacement sampling points before and after the given time point, thereby estimating the displacement. The displacement value at each moment is used to improve its time resolution to be consistent with the acceleration data, and finally a unified time series matrix is ​​constructed.

[0031] Thus, by resampling and spatiotemporally aligning heterogeneous data using Lagrange interpolation, the frequency differences between different sensors are eliminated, laying a data foundation for subsequent joint calculations of physical fields based on a unified time reference.

[0032] S2. Based on the aligned data, the theoretical displacement change is estimated by the second integral of the acceleration data, and compared with the measured displacement to calculate the kinematic consistency residual, which characterizes the degree of deviation between the measured displacement and the physically estimated displacement.

[0033] In an optional embodiment, since displacement sensors such as GNSS receivers or hydrostatic levels are easily affected by satellite signal quality, resulting in instantaneous jumps or slow drifts, and any real displacement change is inevitably accompanied by acceleration, if the displacement changes drastically but the acceleration display is calm, it is clearly an abnormal situation. Therefore, this invention uses acceleration data for double integration to calculate the theoretical displacement change, and compares it with the measured displacement to determine whether the sensor has produced instantaneous jumps or slow drifts. The specific operation steps are as follows: Due to direct exposure to vibration acceleration sequences Integrating the components would amplify the gravitational component and the DC bias of the sensor infinitely, causing the integral to diverge. Therefore, this invention first requires processing the vibration acceleration sequence. By applying a Butterworth high-pass filter with a cutoff frequency of 0.1 Hz, the acceleration after high-pass filtering is obtained. This filters out the DC component, retaining only the dynamic acceleration caused by the impact.

[0034] Subsequently, the kinematic consistency residuals were calculated. Kinematic consistency residuals The specific calculation method is as follows: ; In the formula, The unit is mm; These are measured displacement data; This is the displacement reference value at the start of the current calculation window, which is used to eliminate static deviations in the absolute coordinates; This is a numerical quadratic integral term, used to integrate acceleration (…). Convert the displacement change (m); The unit conversion and sensitivity correction coefficient is a constant consisting of two parts: one is 1000 for converting meters to millimeters, and the other is the lever arm compensation coefficient caused by the different installation positions of the accelerometer and displacement meter.

[0035] In one specific implementation, assume a measured absolute displacement sequence at a certain moment. 50mm; reference value It is 0 mm; however, its vibration acceleration sequence The theoretical displacement change after double integration and coefficient transformation is 10 mm. Therefore, the calculated kinematic consistency residual is... If the kinematic consistency residual were to be close to 0 for a real impact, the 40mm residual here indicates that the measured displacement deviates significantly from the physical facts.

[0036] Thus, by constructing a kinematically consistent residual model and introducing the iron law of calculus in physics as a criterion, it is possible to effectively identify spurious displacement signals that lack acceleration support, thereby determining whether the sensor has non-physical drift or flypoint faults.

[0037] S3. Calculate the vibration energy density of the vibration acceleration sequence and the displacement fluctuation power of the absolute displacement sequence. Based on the logarithmic linear correlation between the two, construct an energy response coupling confidence model and calculate the logical coupling anomaly index.

[0038] In an optional embodiment, since some fault manifestations may be caused by sensor malfunctions, in practical applications, the specific form of sensor malfunction may be sensor jamming, in which case there is energy input but no response behavior; or it may be that the sensor generates false signals due to interference, in which case there is no energy input but there is a violent response. In the embodiments of the present invention, energy input represents the vibration energy density of the vibration acceleration sequence, and response behavior represents the displacement fluctuation power of the absolute displacement sequence.

[0039] As one implementation of this invention, the invention constructs a logical coupling anomaly index. To identify sensor faults and construct a logic coupling anomaly index The specific steps are as follows: First, calculate the acceleration sequence. The root mean square value within the sliding window is used as the vibrational energy density. It is used to characterize the vibration intensity of bridge piers; Subsequently, the absolute displacement sequence was calculated. The standard deviation within the sliding window is used as the displacement fluctuation power. It is used to characterize the sway intensity of bridge piers.

[0040] Finally, the logical coupling anomaly index is calculated using the following method. : ; In the formula, It is a dimensionless exponent; To prevent the removal of the zero protection constant, it is set to [value] in the embodiments of the present invention. It is used to prevent overflow of logarithmic function calculation when the sensor malfunctions and outputs all 0s or the environment is absolutely still, and to ensure the mathematical stability of the formula under extreme conditions. The baseline coupling constant is obtained by statistically analyzing data from historical periods during which the system has been confirmed to be fault-free through self-testing. The average value represents the inherent vibration displacement conversion ratio of the bridge pier structure.

[0041] To more clearly illustrate the logical coupling anomaly index The function and calculation process will be illustrated with examples below: In an optional embodiment, it is assumed that The displacement fluctuation power is 2. The energy density approaches zero, but there are micro-vibrations in the environment; the vibration energy density... If it is 0.1; then ; In another alternative embodiment, it is assumed that The value remains at 2; the displacement fluctuation power is relatively large, at this time... It is 10; however, the vibrational energy density If it is 10; then .

[0042] It can be seen that, under normal circumstances, the logical coupling anomaly index The value should be close to 0, while the logic coupling anomaly indexes in the two cases above are both high, thus enabling the sensor to be sensitively identified as having a logic-exclusive fault.

[0043] Thus, by constructing an energy response coupled confidence model and utilizing the linear relationship in the logarithmic field and the design to prevent division by zero, we can accurately identify logically exclusive faults such as system crashes and open circuits, ensuring the robustness of the algorithm under extreme conditions.

[0044] S4. Compare the kinematic consistency residual and the logical coupling anomaly index with preset thresholds respectively. Based on the comparison results, determine whether the sensor data has drift, crash or noise fault, and output the cleaned monitoring data to realize the abnormal detection of bridge collision avoidance sensor data.

[0045] In an optional embodiment, the present invention uses a preset drift threshold. and abnormal thresholds kinematic consistency residuals and logical coupling anomaly index The system performs a judgment, establishes a multi-level anomaly decision matrix to determine the type of data fault, and takes targeted handling measures. The judgment logic is as follows: like and If the displacement sensor drifts, the system determines that the physical laws still exist, but there is a deviation. The system uses the trend line obtained by acceleration integration to perform baseline correction on the displacement data.

[0046] like If the sensor malfunctions or is disconnected, the data is deemed unreliable. The system automatically cuts off the weight of the data source and sends an offline alarm to the management terminal.

[0047] Finally, the system outputs clean monitoring data after the above steps to remove faults, which is then used by subsequent collision avoidance algorithms.

[0048] In this way, through multi-level anomaly decision-making and data cleaning, the system achieves automated classification and processing of different types of faults. It can correct recoverable drift data and promptly alarm hardware failures, ensuring the accuracy and reliability of the input data of the collision avoidance system.

[0049] To further illustrate the effects of the present invention, the following description is provided in conjunction with the accompanying drawings: Reference Figure 2The dashed line representing the original GNSS data shows a slow, linear upward trend over time, but it deviates significantly from the zero axis, exhibiting false characteristics. In contrast, the solid line representing the data after physical correction according to this invention successfully eliminated the slope trend of the gray curve, with the data stabilizing around the zero mark with slight fluctuations, verifying the algorithm's ability to remove zero-point drift.

[0050] Reference Figure 3 Points located in the strip area represent normal coupling and normal data; points located in the upper left of the strip area represent fault data with displacement but no vibration; and points located in the lower right of the strip area represent fault data with vibration but no displacement. This intuitively demonstrates the formula's ability to accurately identify logically mutually exclusive fault data using multi-physics coupling relationships.

[0051] Reference Figure 4 The figure shows the curve of the coupling anomaly index changing with monitoring time. As can be seen from the figure, the curve fluctuates smoothly in the first half of the time axis, while it suddenly spikes vertically in the latter half of the time axis and exceeds the fault judgment threshold line. This clearly shows the sensitive response and alarm triggering process of the anomaly index when the sensor fails.

[0052] This invention also discloses a bridge collision avoidance sensor data anomaly detection system, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a bridge collision avoidance sensor data anomaly detection method according to the present invention.

[0053] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0054] In the description of this specification, "multiple" or "several" means at least two, such as two, three or more, unless otherwise expressly and specifically defined.

Claims

1. A method for detecting abnormal data from bridge collision avoidance sensors, characterized in that, include: The vibration acceleration sequence and absolute displacement sequence of the bridge pier are collected, and the absolute displacement sequence is interpolated and resampled based on the time axis of the vibration acceleration sequence to achieve spatiotemporal alignment of multidimensional heterogeneous data. Based on the aligned data, the theoretical displacement change is estimated by the second integral of the acceleration data, and compared with the measured displacement to calculate the kinematic consistency residual, which characterizes the degree of deviation between the measured displacement and the physically estimated displacement. Calculate the vibration energy density of the vibration acceleration sequence and the displacement fluctuation power of the absolute displacement sequence, construct an energy response coupling confidence model based on the logarithmic linear correlation between the two, and calculate the logical coupling anomaly index. The kinematic consistency residual and the logical coupling anomaly index are compared with preset thresholds respectively. Based on the comparison results, it is determined whether the sensor data has drift, crash or noise fault, and the cleaned monitoring data is output to realize the abnormal detection of bridge anti-collision sensor data. The logical coupling anomaly index satisfies the following relationship: In the formula, for The logical coupling anomaly index at any given moment; Vibrational energy density; This refers to the displacement fluctuation power; To prevent the protection constant from being reduced to zero; The reference coupling constant represents the inherent vibration-displacement conversion ratio of the bridge pier structure; The calculation methods for the vibration energy density and the displacement fluctuation power include: Calculate the root mean square value of the vibration acceleration sequence within the sliding window, and use it as the vibration energy density; Calculate the standard deviation of the absolute displacement sequence within the sliding window, and use it as the displacement fluctuation power; The reference coupling constant is obtained as follows: During the historical period when the system was fault-free, the average value of the logarithmic ratio of vibration energy density to displacement fluctuation power was used as the reference coupling constant.

2. The method for detecting abnormal data from bridge collision avoidance sensors according to claim 1, characterized in that, The interpolation and resampling of the absolute displacement sequence specifically includes: Obtain the high-frequency sampling rate of the vibration acceleration sequence and the low-frequency sampling rate of the absolute displacement sequence; The low-frequency absolute displacement sequence is upsampled using the Lagrange interpolation method to make its time resolution consistent with the vibration acceleration sequence, thus constructing a unified time series matrix.

3. The method for detecting abnormal data from bridge collision avoidance sensors according to claim 1, characterized in that, The kinematic consistency residuals satisfy the following relationship: In the formula, for kinematic consistency residuals at any given time; These are measured displacement data; The displacement reference value is calculated at the beginning of the calculation window; The acceleration is the result of high-pass filtering; For unit conversion and sensitivity correction coefficients; This represents the numerical second integral operation; This is the start time of the integration process; , It is the integral variable.

4. The method for detecting abnormal data from bridge collision avoidance sensors according to claim 3, characterized in that, The high-pass filtered acceleration is obtained as follows: A Butterworth high-pass filter is applied to the raw acceleration data to filter out the gravity component and the DC bias of the sensor.

5. The method for detecting abnormal data from bridge collision avoidance sensors according to claim 1, characterized in that, The step of determining whether the sensor data has drift, crashes, or noise faults based on the comparison results includes: When the kinematic consistency residual is greater than the first preset threshold and the logical coupling anomaly index is less than the second preset threshold, it is determined that the displacement sensor is drifting, and the displacement data is baseline corrected using the trend line obtained by acceleration integration. When the logical coupling anomaly index is greater than the second preset threshold, it is determined that the sensor hardware has failed, the corresponding data source weight is cut off, and an alarm message is sent.

6. The method for detecting abnormal data from bridge collision avoidance sensors according to claim 5, characterized in that, The conditions under which the sensor hardware is determined to be faulty further include: If the displacement fluctuation power is less than the third preset threshold and the vibration energy density is greater than zero, it is determined to be in a frozen state. If the displacement fluctuation power is very large and the vibration energy density approaches zero, it is determined to be a noise state.

7. A bridge collision avoidance sensor data anomaly detection system, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement a method for detecting abnormal data from a bridge collision avoidance sensor according to any one of claims 1-6.

Images