Method for evaluating health state of road and bridge structure based on multi-source data fusion

Abstract

The application relates to the technical field of structural health monitoring, and discloses a road-bridge structure health state evaluation method based on multi-source data fusion, which comprises the following steps: collecting environmental meteorological data, total strain data and vibration acceleration data; reconstructing and aligning the multi-source signals based on the acceleration data time as a benchmark, and extracting a quasi-static strain sequence; solving an equivalent structure temperature sequence, calculating the phase delay of the strain sequence for time delay compensation, and constructing an environmental quasi-static baseline; calculating the inherent frequency offset based on the compensated temperature sequence, dynamically adjusting the frequency domain integral boundary, and calculating the local modal energy focusing degree; calculating the environmental degradation index and the local stiffness abnormality index according to the quasi-static baseline and the energy focusing degree respectively, constructing a comprehensive feature vector for tolerance boundary comparison, and outputting the evaluation result. The application eliminates the data time base deviation, effectively decouples the environmental thermal effect interference, and improves the reliability of state evaluation and early warning caused by the feature distortion of adaptive frequency drift.

Description

Technical Field

[0001] This invention relates to the field of structural health monitoring technology, specifically to a method for assessing the health status of road and bridge structures based on multi-source data fusion. Background Technology

[0002] The road and bridge structural health monitoring system acquires operational status data of the structure during its service life by deploying various sensors such as strain, vibration, and meteorology sensors. In order to comprehensively assess the health status of the structure, it is usually necessary to fuse and analyze multi-source heterogeneous monitoring data. However, due to the differences in hardware sampling mechanisms and communication links of different types of sensors, the collected heterogeneous data often have time base deviations. Directly fusing data without strict physical time alignment will introduce fundamental timing errors and affect the accuracy of subsequent analysis.

[0003] Meanwhile, road and bridge structures are exposed to the natural environment for a long time, and their structural response is significantly affected by the fluctuation of environmental temperature. Existing condition assessment methods mostly use static fitting or conventional filtering to remove the environmental baseline when eliminating environmental temperature interference. They fail to effectively handle the physical time lag inherent in the transmission of environmental temperature changes to the interior of the structure and the strain response. If this dynamic phase delay is not compensated, it will be difficult to accurately decouple the environmental thermal effect from the stress state of the structure itself. In addition, environmental temperature fluctuations will also cause small changes in the local stiffness of the structure, which will lead to the drift of the natural frequency of the target order.

[0004] In current frequency domain analysis methods, fixed frequency integration boundaries are commonly used to extract modal energy features. When the natural frequency of a structure drifts due to temperature changes, the fixed integration boundary cannot adapt to this shift, resulting in numerical distortion of the extracted modal features. These errors, caused by data synchronization defects and insufficient environmental interference processing mechanisms, are superimposed on each other, making the final extracted structural anomaly indicators inaccurate and reducing the reliability of road and bridge structural health status assessment and early warning. Summary of the Invention

[0005] The technical problem solved by this invention is that existing road and bridge structure monitoring has multi-source data time base bias and does not process the dynamic phase delay between temperature and strain response, resulting in inaccurate decoupling of environmental effects; in addition, the fixed frequency integral boundary cannot adapt to the natural frequency drift caused by temperature, resulting in distorted modal feature extraction and ultimately reducing the reliability of condition assessment.

[0006] To address the above problems, the present invention provides the following technical solution: The first aspect of this invention provides a method for assessing the health status of road and bridge structures based on multi-source data fusion, comprising the following steps: Collect multi-source heterogeneous monitoring signals, including environmental meteorological data, total strain data, and vibration acceleration data of different spatial nodes; A global time axis is established based on the time of vibration acceleration data, multi-source heterogeneous monitoring signals are reconstructed and aligned, and the total strain data is filtered to extract the quasi-static strain sequence. Environmental meteorological data are coupled and solved to generate an equivalent structured temperature sequence; The cross-correlation between the equivalent structural temperature sequence and the quasi-static strain sequence is calculated within a sliding window. The time-varying phase delay coefficient is extracted and used to compensate for the equivalent structural temperature sequence. The compensated temperature sequence is then output and an environmental quasi-static baseline is constructed. The theoretical offset of the target order natural frequency is calculated based on the compensation temperature sequence to adjust the frequency domain integration boundary. The local modal energy focusing degree of each spatial node is calculated within the adjusted boundary. Based on the quasi-static baseline of the environment and the local modal energy focusing degree, the temporal environmental degradation index and the local stiffness anomaly index are calculated respectively. A comprehensive state feature vector is constructed for boundary comparison, and the evaluation results and early warning instructions are output.

[0007] Furthermore, the process of establishing a global time axis based on the time of vibration acceleration data, reconstructing and aligning multi-source heterogeneous monitoring signals, and filtering the total strain data to extract a quasi-static strain sequence includes: A third-order B-spline interpolation algorithm is used for time reconstruction, which maps the original sampling time nodes of environmental meteorological data and total strain data to the global time axis, and outputs surface temperature sequence, wind speed sequence, light intensity sequence and total strain sequence. An infinite impulse response low-pass filter is used to perform low-pass filtering on the total strain sequence after time reconstruction in order to extract the quasi-static strain sequence.

[0008] Furthermore, the coupled solution of environmental meteorological data to generate an equivalent structured temperature sequence includes: Based on the natural convection heat transfer constant under windless conditions, the forced convection wind speed sensitivity coefficient, and the wind speed sequence in environmental meteorological data, the wind resistance convection heat transfer coefficient of the structural surface is calculated through empirical linear relationships. By combining the solar radiation absorptivity of the bridge surface material, the ratio of the light intensity sequence in the environmental meteorological data to the wind resistance convective heat transfer coefficient is calculated to obtain the equivalent temperature increment. The equivalent temperature increment is then superimposed on the environmental surface temperature sequence to generate the equivalent structural temperature sequence.

[0009] Furthermore, the step of calculating the cross-correlation between the equivalent structural temperature sequence and the quasi-static strain sequence within a sliding window, and extracting the time-varying phase retardation coefficient, includes: The equivalent structural temperature sequence and quasi-static strain sequence within the sliding window are respectively subjected to mean removal and standard deviation normalization. Calculate the local discrete cross-correlation function of the two, search for the peak value of the local discrete cross-correlation function within the preset translation search interval, and convert the translation step number corresponding to the peak value into the time-varying phase delay coefficient at the current time point.

[0010] Furthermore, the compensation of the equivalent structural temperature sequence, the output of the compensated temperature sequence, and the construction of an environmental quasi-static baseline include: Based on the delay steps of the time-varying phase delay coefficient corresponding to the current time point, the equivalent structure temperature sequence is time-shifted using interpolation to obtain the compensated temperature sequence. Based on the compensated temperature sequence and the quasi-static strain sequence, a multinomial regression model with Huber loss function is used for fitting to construct an environmental quasi-static baseline; The environmental quasi-static baseline is separated from the total strain data, and the net strain sequence is output.

[0011] Further, the step of calculating the theoretical offset of the target order natural frequency based on the compensated temperature sequence to adjust the frequency domain integration boundary includes: The compensation temperature sequence is input into a preset polynomial nonlinear mapping function to calculate the theoretical offset of the target order natural frequency. The coefficients of the polynomial nonlinear mapping function are obtained by least squares fitting based on data within the healthy baseline period. Perform short-time Fourier transform on the vibration acceleration data at each spatial node to obtain the corresponding time-frequency amplitude matrix; Based on the natural frequency of the target order, the basic analysis bandwidth, and the theoretical offset under the healthy baseline state, the dynamic integration lower limit and dynamic integration upper limit of the frequency domain integration at the current time point are dynamically determined, and amplitude limiting is performed according to the preset 0 value and Nyquist cutoff frequency.

[0012] Further, the calculation of the local modal energy focusing degree of each spatial node within the adjusted boundary includes: The cumulative sum of the squared amplitudes at discrete frequency points is extracted from the time-frequency amplitude matrix based on the lower limit and upper limit of dynamic integration to generate the target modal energy sequence. The sum of the squared amplitudes of discrete frequency points is extracted in the full frequency band from 0 to the Nyquist cutoff frequency to generate the total energy sequence of the full frequency band. The ratio of the target modal energy sequence to the total energy sequence of the entire frequency band is calculated, and combined with the preset positive compensation term, the local modal energy focusing degree at the corresponding spatial node is generated.

[0013] Furthermore, the calculation of the time-domain environmental degradation index and the local stiffness anomaly index based on the quasi-static environmental baseline and the local modal energy focusing degree respectively includes: The time change rate is obtained by calculating the difference between the quasi-static baseline of the environment before and after a set number of discrete time intervals, and is used as an indicator of environmental degradation in the time domain. For any target node in each spatial node, select its neighboring nodes to form a spatial node set. Calculate the mean value of the local modal energy focusing degree of each node in the spatial node set, and calculate the spatial cross-sectional variance based on the local modal energy focusing degree and the mean value. Use the spatial cross-sectional variance as a local stiffness anomaly index.

[0014] Furthermore, the construction of the comprehensive state feature vector for boundary comparison, outputting evaluation results and early warning instructions, includes: Based on historical health benchmark data, statistical values ​​of indicators are extracted to determine the environmental degradation tolerance threshold and stiffness anomaly tolerance threshold. An elliptical two-dimensional tolerance boundary domain is constructed using environmental degradation tolerance threshold and stiffness anomaly tolerance threshold as parameters. The comprehensive state feature vector is mapped to the two-dimensional tolerance boundary domain as a two-dimensional coordinate point for judgment. When the comprehensive state feature vector of a consecutive preset number of time steps crosses the two-dimensional tolerance boundary domain, a structural health status assessment result with abnormal risk is generated and an early warning command is triggered.

[0015] Furthermore, in the step of acquiring multi-source heterogeneous monitoring signals: A global reference clock pulse is generated by a temperature-controlled crystal oscillator inside the synchronous triggering unit, and the global reference clock pulse is distributed in parallel to the analog-to-digital converters of each sensor that acquires multi-source heterogeneous monitoring signals. When the analog-to-digital converter receives the preset level edge of the global reference clock pulse, it latches and digitally quantizes the corresponding analog signal so that the multi-source heterogeneous monitoring signals are established on a unified physical time reference.

[0016] A second aspect of the present invention provides a road and bridge structural health status assessment system based on multi-source data fusion, comprising: The data acquisition module is used to collect multi-source heterogeneous monitoring signals, including environmental meteorological data, total strain data, and vibration acceleration data of different spatial nodes; The time-base reconstruction module is used to establish a global time axis based on the time of vibration acceleration data, reconstruct and align multi-source heterogeneous monitoring signals, and filter the total strain data to extract the quasi-static strain sequence. The environmental calculation module is used to perform coupled calculations on environmental meteorological data to generate equivalent structured temperature sequences. The baseline construction module is used to calculate the cross-correlation between the equivalent structural temperature sequence and the quasi-static strain sequence within a sliding window, extract the time-varying phase retardation coefficient and compensate for the equivalent structural temperature sequence, output the compensated temperature sequence and construct the environmental quasi-static baseline. The feature extraction module is used to calculate the theoretical offset of the target order natural frequency based on the compensation temperature sequence to adjust the frequency domain integration boundary, and to calculate the local modal energy focusing degree of each spatial node within the adjusted boundary. The state assessment module is used to calculate the time-domain environmental degradation index and the local stiffness anomaly index based on the quasi-static baseline of the environment and the local modal energy focusing degree, respectively, construct a comprehensive state feature vector for boundary comparison, and output the assessment results and early warning instructions.

[0017] This invention provides a method for assessing the structural health status of roads and bridges based on multi-source data fusion. It has the following beneficial effects: 1. This invention controls the underlying hardware to synchronously sample by a global reference clock pulse and combines it with a third-order B-spline interpolation algorithm to reconstruct time nodes, thereby achieving the alignment of environmental meteorological, total strain and vibration acceleration data on the same discrete time axis, providing an accurate time series basis for multi-source data fusion analysis in road and bridge condition assessment.

[0018] 2. This invention extracts the time-varying phase delay coefficient by calculating the local cross-correlation between equivalent structural temperature and quasi-static strain, performs feedforward time shift compensation on the temperature data, and combines a multinomial regression model with Huber loss function to strip away the environmental quasi-static baseline, thereby achieving effective decoupling between environmental thermal effects and the actual mechanical characteristics of the structure.

[0019] 3. This invention uses the compensated temperature sequence to calculate the frequency offset, dynamically adjusts the frequency domain integral boundary of modal energy extraction to prevent feature distortion, and constructs the calculated time-domain environmental degradation index and local stiffness anomaly index as feature vectors. The two-dimensional tolerance boundary domain is used for cross-comparison and judgment, which reduces the false alarm rate of assessment caused by a single index and improves the reliability of road and bridge structure health status assessment. Attached Figure Description

[0020] Figure 1 This is a system architecture diagram of the present invention; Figure 2 This is a flowchart of the method of the present invention; Figure 3 This is a schematic diagram illustrating the heterogeneous sensor arrangement and hardware synchronous triggering principle of the present invention. Figure 4 This is a flowchart of the multi-rate time-base reconstruction and preprocessing process of the present invention; Figure 5 This is a flowchart of the multi-environmental factor coupling solution of the present invention; Figure 6 This is a flowchart of the temporal phase delay tracking and baseline stripping process of the present invention; Figure 7 This is a flowchart of the frequency domain dynamic boundary tracking and feature extraction process of the present invention; Figure 8This is a flowchart of the multi-dimensional parameter linkage state evaluation and output process of the present invention.

[0021] The module includes: 10. Data acquisition module; 11. Heterogeneous sensor group; 12. Synchronization triggering unit; 20. Preprocessing module; 30. Solving module; 40. Baseline stripping module; 50. Frequency domain tracking module; and 60. State assessment module. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see the appendix Figure 1 This invention provides a road and bridge structure health status assessment system based on multi-source data fusion, including: a data acquisition module 10, a preprocessing module 20, a solution module 30, a baseline stripping module 40, a frequency domain tracking module 50, and a status assessment module 60.

[0024] The data acquisition module 10 is installed at the site of the road and bridge structure to be evaluated, and includes a heterogeneous sensor group 11 and a synchronous triggering unit 12. The heterogeneous sensor group 11 includes at least a surface temperature sensor, an anemometer, a light meter, a strain gauge, and multiple accelerometers. Among them, the surface temperature sensor is used to acquire surface temperature data of the structure, the anemometer is used to acquire on-site wind speed data, the light meter is used to acquire solar radiation intensity data, the strain gauge is used to acquire structural strain data, and the multiple accelerometers are used to acquire vibration acceleration data at different spatial nodes.

[0025] The synchronous triggering unit 12 is connected to each sensor acquisition channel in the heterogeneous sensor group 11, and is used to send hardware trigger clock signals to each acquisition channel so that all kinds of monitoring signals are established on a unified physical time reference.

[0026] The preprocessing module 20 is communicatively connected to the data acquisition module 10 and is used to receive the raw sampling signals output by the heterogeneous sensor group 11. The preprocessing module 20 reconstructs the time nodes of the data sequences with different sampling frequencies, and combines smooth interpolation and low-pass filtering to generate time-aligned discrete data sequences.

[0027] The calculation module 30 is connected to the preprocessing module 20 and is used to receive time-aligned environmental monitoring data. The calculation module 30 calculates the equivalent structural temperature sequence characterizing the thermal boundary state of the structure based on the surface temperature, solar radiation intensity, and wind resistance convective heat transfer coefficient determined by wind speed.

[0028] The baseline stripping module 40 receives the strain sequence output by the preprocessing module 20 and the equivalent structural temperature sequence output by the solution module 30, respectively. The baseline stripping module 40 is equipped with a sliding data window, and calculates the discrete cross-correlation relationship between the equivalent structural temperature sequence and the strain sequence within the sliding data window to extract the time-varying thermo-mechanical phase retardation coefficient.

[0029] The baseline stripping module 40 performs time delay compensation on the equivalent structural temperature sequence based on the time-varying thermo-mechanical phase delay coefficient, thereby constructing an environmental quasi-static baseline, and stripping the environmental quasi-static baseline from the strain sequence, outputting the compensated temperature sequence, the environmental quasi-static baseline, and the net strain sequence.

[0030] The frequency domain tracking module 50 receives the compensated temperature sequence output by the baseline stripping module 40, and the vibration acceleration sequence of each acceleration sensor node output by the preprocessing module 20, which is aligned with the global time axis.

[0031] The frequency domain tracking module 50 inputs the compensation temperature sequence into a preset polynomial nonlinear mapping function to obtain the theoretical offset of the target order natural frequency. The vibration acceleration sequence of each acceleration sensor node is subjected to a short-time Fourier transform to obtain the corresponding time-frequency amplitude data. The frequency domain tracking module 50 adjusts the upper and lower limits of the frequency domain integration in the short-time Fourier transform result according to the theoretical offset, and calculates the target modal energy within the adjusted frequency domain integration range, thereby obtaining the local modal energy focusing degree corresponding to each acceleration sensor node.

[0032] The state assessment module 60 receives the environmental quasi-static baseline output by the baseline stripping module 40 and the local modal energy focusing degree output by the frequency domain tracking module 50.

[0033] The state assessment module 60 calculates the time-domain environmental degradation index based on the quasi-static baseline of the environment, and calculates the local stiffness anomaly index based on the local modal energy focusing degree of adjacent acceleration sensor nodes. The time-domain environmental degradation index and the local stiffness anomaly index are then used to construct a comprehensive state feature vector.

[0034] The status assessment module 60 maps the comprehensive status feature vector to a preset two-dimensional tolerance domain for boundary comparison, and outputs the structural health status assessment result and corresponding early warning instructions based on the comparison result.

[0035] Please see the appendix Figure 2 This invention provides a method for assessing the health status of road and bridge structures based on multi-source data fusion, comprising the following steps: S10: Simultaneous acquisition of multi-source heterogeneous monitoring signals at the road and bridge structure site. These signals include environmental meteorological data, total strain data, and vibration acceleration data.

[0036] S20 uses the sampling frequency of the vibration acceleration signal or the acceleration sampling time axis formed by the synchronous triggering unit as the global time base to perform multi-rate time base reconstruction on the multi-source heterogeneous monitoring signal, so that the environmental meteorological data, total strain data and vibration acceleration data obtained at different sampling frequencies correspond to each other at the time nodes, and generate aligned discrete data sequences and quasi-static strain sequences.

[0037] S30 performs multi-environmental factor coupling calculation on the discretized environmental meteorological data, and integrates the thermal boundary effects reflected by surface temperature, wind speed and solar radiation intensity to generate an equivalent structural temperature sequence.

[0038] S40, calculate the local cross-correlation between the equivalent structural temperature sequence and the quasi-static strain sequence within the sliding data window, and extract the time-varying phase delay coefficient corresponding to the current time node; The equivalent structural temperature sequence is time-delayed by the time-varying phase delay coefficient, and the environmental quasi-static baseline is constructed from the compensated temperature sequence. The environmental quasi-static baseline is separated from the total strain data, and the compensated temperature sequence, environmental quasi-static baseline, and net strain sequence are output.

[0039] S50, calculate the target order natural frequency offset based on the temperature sequence after time delay compensation, and introduce the natural frequency offset as a dynamic boundary control quantity into the frequency domain integration range of the vibration acceleration signal. The target modal energy is extracted within the adjusted frequency domain integral range to generate a local modal energy focus sequence.

[0040] S60, calculate the temporal environmental degradation index based on the time change of the quasi-static baseline of the environment, and calculate the local stiffness anomaly index based on the dispersion of local modal energy focusing degree between adjacent spatial nodes. A comprehensive state feature vector is constructed from temporal environmental degradation index and local stiffness anomaly index. This comprehensive state feature vector is then mapped to a preset tolerance boundary domain for discrimination, generating structural health status assessment results and corresponding early warning instructions.

[0041] See attached document Figure 3 Step S10 specifically includes the following steps: S101, the heterogeneous sensor group 11 in the data acquisition module 10 is arranged on-site at the road and bridge structure.

[0042] Specifically, anemometers and irradiance meters are installed on the top of the main tower of the road and bridge structure, the outer side of the bridge deck, or other unobstructed locations to obtain on-site wind speed and solar radiation intensity data. Surface temperature sensors are installed at representative surface temperature measurement points on the road and bridge structure. These representative measurement points are preferably shaded points, points not directly exposed to sunlight, or points equipped with shading and radiation protection structures to obtain base surface temperature data that is not superimposed on the heating effect of direct solar radiation.

[0043] Strain gauges are fixed at the main reinforcing bars in the mid-span section of the main girder, at the bottom plate of the steel box girder, or at other strain-sensitive locations to obtain total structural strain data. Multiple accelerometers are respectively set at the alignment control points, modal vibration network points, or other vibration-sensitive nodes of the main girder of the road and bridge to obtain vibration acceleration data at different spatial nodes.

[0044] The aforementioned sensors are arranged in a distributed manner to adapt to the monitoring characteristics of slow environmental thermal conduction response and fast structural vibration response. Multiple accelerometers can be numbered according to their mileage location on the bridge structure, component connection relationship, or modal correlation; in subsequent condition assessment, any target acceleration node and its adjacent nodes together constitute a local spatial node set.

[0045] S102, start the heterogeneous sensor group 11 to collect multi-source heterogeneous monitoring signals of the road and bridge structure site.

[0046] Because the rates of change of environmental meteorological quantities, structural strain, and vibration acceleration measurements differ, different initial sampling frequencies are used for various sensors. Let the sampling frequency of the vibration acceleration data be... The sampling frequency of the total strain data is The sampling frequency of environmental meteorological data is ;in, , and satisfy > > .

[0047] In one specific implementation, The value range is 100–200Hz. The value range is 10–50 Hz. The value range is 0.1 to 1 Hz. The above sampling frequency configuration satisfies the Nyquist sampling requirements in structural vibration monitoring and can also accommodate the data storage load during long-term continuous monitoring.

[0048] Within a single continuous acquisition time window Within the same sensor, the number of discrete data points generated varies. The monitoring time can be set from 1 to 24 hours depending on the monitoring task; when When the time span required for subsequent sliding cross-correlation windows, time derivative calculation windows, or health status assessment windows is less than the time span required for these windows, the system calls upon monitoring data from the historical cache database that is continuously connected to the currently collected data to participate in the calculation, in order to ensure that phase delay tracking and degradation trend assessment have sufficient effective data length.

[0049] Define the length of the discrete vibration acceleration data sequence corresponding to any acceleration sensor node as . The total strain discrete data sequence length is The length of the discrete environmental meteorological data sequence is Then the lengths of each sequence satisfy: ; ; ; in, , and All units are in Hz. The unit is seconds.

[0050] S103, through the synchronization triggering unit 12 in the data acquisition module 10, hardware synchronization triggering is performed on each acquisition channel in the heterogeneous sensor group 11 to eliminate the start time deviation between independent acquisition channels.

[0051] In one specific implementation, the synchronization trigger unit 12 employs distributed synchronization hardware supporting the IEEE 1588 precision time protocol, or a central timing module based on an industry-standard hardware bus. The oven-controlled crystal oscillator inside the synchronization trigger unit 12 generates a global reference clock pulse, which is then distributed in parallel to the analog-to-digital conversion channels of each sensor in the heterogeneous sensor group 11 via coaxial cable, backplane bus, or other hardwired connections.

[0052] During the acquisition process, the analog-to-digital converters of each sensor channel do not use software instructions as the sampling time. Instead, they latch and quantize the corresponding analog signals when they receive the preset level edge of the global reference clock pulse. As a result, environmental meteorological data, total strain data, and vibration acceleration data can be aligned to the same physical time axis, reducing the impact of network transmission latency, operating system timestamp errors, and software task scheduling jitter on synchronization accuracy.

[0053] The specific circuit structure, trigger level settings, and pulse edge capture method of the analog-to-digital converter can be configured according to the interface specifications of the selected industrial data acquisition card or synchronous data acquisition device.

[0054] See attached document Figure 4 Step S20 specifically includes the following steps: S201, the preprocessing module 20 receives the discrete data sequence of multi-source heterogeneous monitoring signals output by the data acquisition module 10.

[0055] In multi-source heterogeneous monitoring signals, the sampling frequency of vibration acceleration data The sampling frequency is higher than that of the total strain data and environmental meteorological data. Since subsequent frequency domain analysis needs to retain the high-frequency modal information in the vibration acceleration signal, directly downsampling the vibration acceleration data can easily lead to the loss of high-frequency dynamic components. Therefore, the preprocessing module 20 uses the sampling time of the vibration acceleration signal as a reference to establish a global discrete time axis.

[0056] Define the global time step as The time points on the global discrete time axis are represented as ,in, Non-negative integer index, The vibration acceleration sequence is directly mapped to this global discrete time axis, denoted as... .

[0057] The global discrete time axis is mainly used for frequency domain analysis of vibration acceleration signals. For environmental meteorological data, quasi-static strain data, and thermo-mechanical phase delay tracking processes, the preprocessing module 20 can establish a separate quasi-static analysis time axis after synchronization alignment is completed, and perform low-frequency environmental effect calculations on this quasi-static analysis time axis to reduce the data scale of long-term window cross-correlation operations.

[0058] S202, for environmental meteorological data sequences and total strain data sequences with sampling frequencies lower than those of vibration acceleration signals, the preprocessing module 20 uses a third-order B-spline interpolation algorithm to reconstruct the time, mapping the original sampling time nodes to the global discrete time axis. .

[0059] Third-order B-spline interpolation has second-order continuous differentiability. While improving the time resolution of low-frequency data, it can maintain the smooth variation trend of slowly changing signals such as ambient temperature, wind speed, light intensity and structural strain, and reduce the numerical step that may be introduced by linear interpolation.

[0060] Suppose that the original discrete sequence of a certain low-frequency or mid-frequency frequency to be processed has the value at the original time node. Its position at the global time point is obtained through interpolation. Target value at Its mathematical expression is: ; In the formula, This refers to the control vertex vector obtained from the original discrete data points; For time points The third-order B-spline basis function at the location; To control the total number of vertices.

[0061] After the above interpolation process, the preprocessing module 20 outputs a surface temperature sequence aligned with the global time axis. Wind speed sequence Light intensity sequence and total strain sequence .

[0062] During the environmental quasi-static baseline stripping process, the preprocessing module 20 can also generate a quasi-static analysis sequence based on the alignment results described above. The sampling frequency of the quasi-static analysis sequence is lower than the vibration acceleration sampling frequency and meets the effective sampling requirements for environmental effects and quasi-static strain changes. If subsequent thermo-mechanical phase delay tracking, environmental quasi-static baseline stripping, and time-domain environmental degradation index calculations are performed on the quasi-static analysis time axis, then the time step of that quasi-static analysis time axis will be used; if the above calculations are performed directly on the global discrete time axis, then the corresponding time step will still be used. For ease of description, subsequent steps will still refer to... This indicates the current computation time point. When the computational object is located on the quasi-static analysis time axis, This represents the discrete time node on the time axis of the quasi-static analysis.

[0063] The recursive calculation of the third-order B-spline basis functions and the solution of the control vertices can be achieved using conventional numerical analysis methods.

[0064] S203, Preprocessing module 20 processes the total strain sequence after time reconstruction. Low-pass filtering was performed to extract the quasi-static strain sequence. .

[0065] Total strain sequence Typically, it contains two types of components: one is the low-frequency quasi-static strain component caused by changes in ambient temperature, and the other is the high-frequency dynamic strain component caused by transient excitations such as vehicle loads and wind-induced vibrations. To obtain the low-frequency strain response corresponding to the environmental thermal effects, the preprocessing module 20 uses an infinite impulse response low-pass filter. Process it.

[0066] In one specific implementation, the infinite impulse response low-pass filter is a Butterworth-type low-pass filter. This filter has a relatively smooth amplitude-frequency response within the passband, making it suitable for preserving the slowly varying strain components corresponding to environmental effects.

[0067] When the global time axis sampling frequency is high and the cutoff frequency is low, the Butterworth type infinite impulse response low-pass filter is preferably implemented in the form of a second-order cascade. Alternatively, filtering can be performed on the quasi-static analysis time axis to improve numerical stability at extremely low cutoff frequencies. Its time-domain difference form can be expressed as: ; In the formula, and These are the feedforward order and feedback order of the filter, respectively; Forward coefficients, The feedback coefficient is based on a preset cutoff frequency. Confirmed. In routine structural health monitoring scenarios, and A value of 4 can be chosen to balance filtering attenuation capability and computational delay.

[0068] For the situation where there is insufficient boundary data in the initial stage of filtering, i.e. when <0 or When <0, the preprocessing module 20 supplements the data required for calculation by mirroring and extending the initial edge data.

[0069] Since the cycles of ambient temperature and solar radiation intensity are typically measured in hours, and the first-order natural frequency of bridge structures and the dynamic load frequency of vehicles are usually above 1Hz, the cutoff frequency of the low-pass filter... It can be set to 0.01–0.05 Hz. This cutoff frequency can attenuate the high-frequency strain components corresponding to transient traffic dynamic loads, while retaining quasi-static strain data that changes slowly with environmental factors.

[0070] See attached document Figure 5 Step S30 specifically includes the following steps: S301, the solution module 30 receives the time-aligned discrete data sequence output by the preprocessing module 20. The discrete data sequence includes the surface temperature sequence. Wind speed sequence and light intensity sequence .

[0071] Among them, surface temperature sequence This is used to characterize the basic thermal state of the surface of road and bridge structures. To avoid the repeated accounting of radiation-induced temperature rise caused by direct sunlight, the surface temperature sensor is preferably installed at a shaded measuring point, a measuring point not exposed to direct sunlight, or a measuring point with a light-shielding and radiation-proof structure.

[0072] Bridge and road structural components typically have large cross-sectional dimensions and thermal inertia. Their overall thermal state is not entirely determined by surface temperature but is also affected by absorption of shortwave solar radiation and wind-induced convection heat dissipation. Therefore, the solution module 30 couples surface temperature, wind speed, and light intensity to obtain an equivalent structural temperature sequence for subsequent thermo-mechanical response analysis.

[0073] S302, Solving module 30 based on wind speed sequence Calculate the wind resistance convective heat transfer coefficient of the structural surface. .

[0074] In one specific implementation, the wind resistance convective heat transfer coefficient The empirical linear relationship is used to determine its calculation formula, which is as follows: ; In the formula, The natural convection heat transfer constant under windless conditions is expressed in W / (m³). 2 ⋅K); The forced convective wind speed sensitivity coefficient is expressed in J / (m). 3 ⋅K). For conventional concrete surfaces, asphalt pavement surfaces, or bridge steel structure coating surfaces, A value of 4.0 to 6.0 is acceptable. A value of 3.0 to 4.5 is acceptable. The above parameters can be pre-calibrated based on meteorological records of the bridge site area, the roughness of the bridge surface, and the surface materials of the structure.

[0075] because The value is positive, and the wind speed measured on-site is... Since it is a non-negative value, therefore Keeping it positive can prevent subsequent... When used as the denominator, it results in a division by zero situation.

[0076] S303, the calculation module 30 calculates based on the surface temperature sequence. Light intensity sequence and air resistance convective heat transfer coefficient Calculate the equivalent structural temperature series .

[0077] When using the temperature of a shaded measuring point, a measuring point not directly exposed to sunlight, or the surface temperature of a foundation after it has undergone shading and radiation protection treatment as input, the additional heat input resulting from the absorption of shortwave solar radiation by the structure can be converted into an equivalent temperature increment and superimposed on the foundation surface temperature. Correspondingly, the equivalent structural temperature sequence It can be calculated using the following formula: ; In the formula, The solar radiation absorptivity of the bridge surface material is a dimensionless parameter; its value varies depending on the surface coating and material type. A value of 0.4 to 0.8 is acceptable. For time points The corresponding light intensity is expressed in W / m². 2 .

[0078] If the measuring point where the surface temperature sensor is located already includes the direct solar heating effect, then the calculation module 30 will not repeatedly include the solar radiation term as a simple superposition term. In this case, the calculation module 30 can use surface temperature, light intensity, and wind resistance convective heat transfer coefficient as input variables to generate an equivalent structural temperature sequence through a pre-calibrated regression model or empirical fusion model.

[0079] Through the above processing, the discretized surface temperature, wind speed, and light intensity are fused into a one-dimensional equivalent structural temperature sequence. This equivalent structural temperature sequence simultaneously reflects the basic thermal state of the structural surface, solar radiation heat input, and wind-induced convection heat dissipation, and can be used as the temperature input for subsequent phase delay tracking and environmental quasi-static baseline stripping.

[0080] See attached document Figure 6 Step S40 specifically includes the following sub-steps: S401, the baseline stripping module 40 receives the quasi-static strain sequence output by the preprocessing module 20. And the equivalent structural temperature sequence output by the solution module 30 .

[0081] Because road and bridge structural materials possess a certain thermal inertia, there is typically a time lag when environmental heat is transferred from the structural surface to the interior, causing a mechanical strain response. This lag is not a fixed constant but varies with factors such as solar radiation, wind speed, and the thermal conductivity of the structural cross-section. Therefore, the equivalent structural temperature sequence... With quasi-static strain sequence There is a time-varying phase deviation between them on the time axis.

[0082] To obtain this phase deviation, the baseline stripping module 40 sets a sliding data window. Let the sliding data window contain... Each discrete data point has a window with a time span that can be configured to range from 12 to 24 hours to cover the main periodic segments of environmental solar radiation changes.

[0083] At any calculation time point The baseline stripping module 40 first performs mean-reduction and standard deviation normalization on the equivalent structural temperature data and quasi-static strain data within the sliding window, respectively, to reduce the influence of amplitude scale differences and long-term trend terms on the relevant peak positions; then it calculates the local discrete cross-correlation function of the two. Its expression is: ; In the formula, The number of discrete translation steps; Index of data points within the sliding window; Indicates the current calculation time point The point in time obtained by tracing back in historical direction; Indicates the number of translation steps The corresponding quasi-static strain time point under action.

[0084] S402, the baseline stripping module 40 searches for the peak value of the local cross-correlation function within the preset translation search interval, and determines the translation step number corresponding to the peak value as... When the strain gauge installation direction, stress state, or component constraint conditions cause the quasi-static strain sequence to be negatively correlated with the equivalent structural temperature sequence, the baseline stripping module 40 determines the number of translation steps based on the peak value of the absolute value of the normalized cross-correlation function. The corresponding search relationship can be represented as: ; Since temperature boundary changes physically precede the internal strain response of the structure, the translation search interval is set to 0 < 0. ≤ A positive finite interval. Among them, The maximum possible delay step is preset based on the structural cross-sectional dimensions, material thermal conductivity characteristics, and engineering experience. This is used to limit the search range and avoid mismatches across cycles.

[0085] In obtaining Subsequently, the baseline stripping module 40 converts it into the time-varying thermo-mechanical phase delay coefficient for the current calculation time point. The formula for its calculation is: ; In the formula, Global time step; time-varying thermo-mechanical phase delay coefficient It is used to characterize the actual hysteresis time of the structural temperature boundary change at the current time point in time, which is then transmitted to the strain response.

[0086] S403, Baseline stripping module 40 based on time-varying thermo-mechanical phase retardation coefficient Time delay compensation is performed on the equivalent structure temperature sequence.

[0087] Specifically, the baseline stripping module 40 performs a time shift on the equivalent structural temperature sequence according to the delay step corresponding to the current time point, thereby obtaining a compensated temperature sequence corresponding to the quasi-static strain response. In discrete implementation, when the delay corresponds to an integer step size, the equivalent structural temperature value at the corresponding historical time point is directly read; when the delay corresponds to a non-integer time point, linear interpolation or spline interpolation is used to obtain the compensated temperature value. Its expression can be represented as: ; Through the aforementioned time delay compensation, the temperature sequence is compensated. With quasi-static strain sequence Establishing a correspondence over time is beneficial for subsequently establishing a mapping between the effects of environmental temperature and the quasi-static strain of the structure.

[0088] In the early stages of calculation, if the length of data before the current time point is insufficient to form a complete sliding window, the baseline stripping module 40 uses linear extrapolation, historical health average filling, or pauses the output at the current time to avoid index out-of-bounds errors caused by insufficient window data.

[0089] S404, Baseline stripping module 40 based on compensated temperature sequence With quasi-static strain sequence Establish a robust multinomial regression model and construct a quasi-static environmental baseline. .

[0090] In one specific implementation, the robust polynomial regression model employs a second- or third-order polynomial to balance fitting accuracy and overfit suppression. The loss function of the regression model uses the Huber loss function, whose threshold parameter can be determined based on the local variance of the quasi-static strain sequence within a sliding window, thereby reducing the impact of outlier monitoring points on the baseline fitting results.

[0091] The quasi-static environmental baseline obtained through regression fitting This is used to characterize the gradually varying strain trend of the structure caused by changes in ambient temperature. Subsequently, the baseline stripping module 40 extracts the total strain sequence after time alignment. The quasi-static baseline of the environment was stripped to obtain the net strain sequence. The formula for its calculation is: ; Therefore, the baseline stripping module 40 outputs a compensation temperature sequence. 1. Environmental quasi-static baseline and net strain sequence Among them, the net strain sequence It retains the structural dynamic response components caused by external excitations such as traffic dynamic loads and wind-induced vibrations, which can be used as auxiliary data for subsequent identification of dynamic response anomalies and can be cross-validated with the local modal energy focusing index.

[0092] The solution method for the robust multinomial regression model, the specific expression of the Huber loss function, and the parameter optimization process can be configured according to conventional machine learning or numerical optimization methods.

[0093] See attached document Figure 7 Step S50 specifically includes the following sub-steps: S501, the frequency domain tracking module 50 receives the compensated temperature sequence output by the baseline stripping module 40. .

[0094] During the operation of road and bridge structures, changes in ambient temperature can cause variations in the elastic modulus, boundary constraints, and overall stiffness of the structural materials, resulting in a temperature-dependent drift in the target natural frequency. To quantify this drift, the frequency domain tracking module 50 uses a compensation temperature sequence. By inputting a preset nonlinear polynomial mapping function, the theoretical offset of the target order natural frequency is obtained. The formula for its calculation is: ; In the formula, , and The coefficients are second-order polynomials, which can be obtained by least squares fitting based on the compensated temperature data and the measured natural frequency data of the target order during the reference period of road and bridge structural health. This indicates the modal order of the target to be tracked.

[0095] The measured natural frequency data of the target order can be determined from the peak value of the acceleration spectrum and the modal identification results during the healthy reference period, or from the frequency domain peak tracking results during the low-disturbance period of operation. Therefore, the frequency domain tracking module 50 establishes the correspondence between the compensation temperature and the natural frequency offset of the target order.

[0096] S502, Frequency Domain Tracking Module 50 receives the vibration acceleration sequence of each accelerometer node after global time axis alignment. Short-time Fourier transforms were performed on the vibration acceleration sequences of each node to obtain the corresponding time-frequency amplitude matrices. .

[0097] During the short-time Fourier transform, each analysis window uses its center time or start time as the corresponding time node; the frequency domain tracking module 50 uses this time node to perform calculations on the compensated temperature sequence. The value is taken from the middle or interpolated to determine the theoretical offset of the natural frequency corresponding to the current analysis window.

[0098] The discrete form of the short-time Fourier transform can be expressed as: ; In the formula, These are discrete frequency components on the frequency axis; It is a time-domain truncation window function; This represents the number of discrete data points included in the truncated window function. It is the imaginary unit.

[0099] In one specific implementation, the time-domain truncated window function employs a Hamming window to reduce spectral sidelobe leakage; the window width... The corresponding time length can be set to 10-30 seconds to balance time resolution and frequency resolution; the overlap rate between adjacent analysis windows can be set to 50% to improve the continuity of the time-frequency matrix in the time direction.

[0100] S503, the frequency domain tracking module 50 determines the dynamic frequency domain integral boundary for target mode energy extraction based on the inherent frequency theoretical offset.

[0101] Specifically, let the natural frequency of the target order under the healthy baseline state be . The basic analysis bandwidth is Basic analysis of bandwidth The initial value can be set to 5-10%. To ensure sufficient frequency coverage for discrete frequency domain integration, Not less than the short-time Fourier transform frequency resolution Twice as; when according to 5-10% of the determined When the frequency resolution requirement is not met, the frequency domain tracking module 50 increases the length of the truncated window function. Or increase appropriately This ensures that the target analysis frequency band covers at least two discrete frequency points.

[0102] Frequency domain tracking module 50 offsets the target order's natural frequency theoretically. Calculate the lower limit of dynamic integration at the current time point. and dynamic integration limit The formula for its calculation is: ; ; In this way, the frequency domain integration interval can move synchronously with the temperature-induced natural frequency drift, thereby reducing the loss of modal energy caused by fixed frequency band extraction.

[0103] When the lower limit of dynamic integral When the value is less than 0, the frequency domain tracking module 50 will... The limit is 0; When the dynamic integration limit Greater than the Nyquist cutoff frequency At that time, Limiting This ensures that the integration interval is within the effective frequency range.

[0104] S504, after the dynamic integration boundary is determined, the frequency domain tracking module 50 tracks the time-frequency amplitude matrix. Calculate the target modal energy sequence and total energy sequence across the entire frequency band .

[0105] In the discrete frequency domain, integration can be converted into the summation of the squared amplitudes at frequency points. When the dynamic integration boundary does not fall on the discrete frequency points, the frequency domain tracking module 50 can select all discrete frequency points falling within the dynamic integration boundary to participate in the summation, or it can use linear interpolation to calculate the amplitude contribution of the boundary frequency points. The calculation formula is: ; ; In the formula, For discrete frequency resolution, determined by the sampling frequency Window function length Sure; The Nyquist cutoff frequency has a value of .

[0106] Subsequently, the frequency domain tracking module 50 calculates the local modal energy focusing sequence based on the target modal energy and the total energy of the entire frequency band. The formula for its calculation is: ; In the formula, A positive compensation term, used to avoid A division by zero error occurs when the value is zero. 10 can be taken -6 .

[0107] For multiple acceleration sensor nodes, the frequency domain tracking module 50 calculates and outputs the local modal energy focusing sequence for each node. ,in Assign spatial node numbers. Local modal energy focus is used to characterize the proportion of energy occupied by the target mode under broadband external excitation; When the local stiffness of a road and bridge structure becomes abnormal, the modal energy distribution between adjacent nodes will change. This focus sequence can be used as a frequency domain feature for subsequent identification of local stiffness anomalies.

[0108] See attached document Figure 8 Step S60 specifically includes the following steps: S601, the state assessment module 60 receives the environmental quasi-static baseline output by the baseline stripping module 40. .

[0109] For road and bridge structures, long-term changes such as material aging, loss of prestress in cables, and degradation of bearing performance typically cause the structure to exhibit different quasi-static responses under similar environmental boundary conditions. These changes are reflected in monitoring data as a continuous shift of the environmental quasi-static baseline over time.

[0110] To characterize this offset trend, the state assessment module 60 uses an environmental quasi-static baseline. Calculation of time-domain environmental degradation index based on rate of change In discrete-time series, the difference method can be used for calculation: ; In the formula, The number of steps for the set discrete time interval; This represents the global time step. In practical engineering configurations, the number of discrete time interval steps is used to filter out high-frequency numerical noise and extract long-term degradation trends. The corresponding time span is usually set to 24 hours or an integer multiple thereof.

[0111] In engineering applications, The corresponding time span can be set to 24 hours or an integer multiple of 24 hours to reduce the impact of short-term disturbances on the judgment of long-term degradation trends. If there are initial disturbances in the calculation... If the value is less than 0, the state assessment module 60 will pause the output of the current time-domain environmental degradation index, or fill it with the baseline mean of the initial healthy period.

[0112] The resulting temporal environmental degradation index Used to characterize the degree of slow degradation of a structure under environmental influences.

[0113] S602, the state assessment module 60 receives the local modal energy focusing sequence output by the frequency domain tracking module 50. .

[0114] To identify local stiffness anomalies, the state assessment module 60 selects neighboring nodes of the target acceleration sensor node to form a spatial node set. Spatial node set It can be determined based on the mileage location of multiple acceleration sensor nodes on the bridge structure, the connection relationship of components, or the correlation of modal vibration modes.

[0115] Let the set of spatial nodes be... Contains One sensor node, The density can be set according to the sensor arrangement, preferably 3 to 5 adjacent nodes. The state assessment module 60 first calculates the average value of the local modal energy focusing degree of each node in the set. Then, the variance of its spatial cross section is calculated to obtain the local stiffness anomaly index. Its expression is: ; In the formula, For a set of spatial nodes Inner Local modal energy focusing degree corresponding to each sensor node; This is the arithmetic mean of the local modal energy focusing degree of each node in the set.

[0116] When the local stiffness of a structure remains normal, the modal energy focusing at adjacent nodes typically exhibits a relatively flat spatial distribution. However, when local stiffness becomes abnormal, the energy distribution relationship between the target node and its neighboring nodes changes, manifesting as an increase in spatial variance. Therefore, local stiffness anomaly indices... It can be used to characterize the degree of anomaly in local dynamic properties.

[0117] S603, the state assessment module 60 is based on time-domain environmental degradation indicators. and local stiffness anomaly index Constructing a comprehensive state feature vector .

[0118] The comprehensive state feature vector is used to simultaneously describe the long-term slow-varying degradation state and the local stiffness anomaly state of the structure, and its expression is: ; in, It reflects the long-term changing trend of the quasi-static environmental baseline. It reflects the degree of anomaly in the modal energy distribution between adjacent spatial nodes. Through the above two-dimensional feature expression, the state assessment module 60 can comprehensively consider the structural durability degradation and local mechanical property changes within the same discrimination space.

[0119] S604, the state assessment module 60 establishes a two-dimensional tolerance boundary domain based on historical health benchmark data, and maps the comprehensive state feature vector into the two-dimensional tolerance boundary domain for discrimination.

[0120] In one specific implementation, the state assessment module 60 extracts statistical values ​​of indicators when the structure is in a healthy state, and determines the environmental degradation tolerance thresholds respectively. and stiffness anomaly tolerance threshold . and All are set to real numbers greater than zero to avoid division by zero during the subsequent normalization process.

[0121] Data corresponding to the health status can be obtained from the initial opening stage of the road and bridge structure, the periodic load test confirming no damage, or historical continuous monitoring periods confirmed by manual inspections that no structural damage has occurred. The tolerance threshold can be determined by the maximum absolute value of the healthy sample, the mean of the healthy sample plus a preset multiple of the standard deviation, or the confidence quantile of the healthy sample.

[0122] Based on the above threshold, the state evaluation module 60 constructs an elliptical two-dimensional tolerance boundary domain, and its determination condition is as follows: ; The state assessment module 60 will integrate the state feature vector. The coordinate point is mapped onto the tolerance boundary domain as a two-dimensional coordinate point. If the coordinate point meets the above judgment conditions, the road and bridge structure is determined to be in a normal state; if the coordinate point crosses the tolerance boundary domain, the structure is determined to be at risk of abnormality.

[0123] To reduce false alarms caused by single noise disturbances, the state assessment module 60 can introduce a continuous limit violation determination mechanism. Specifically, when continuous... An anomaly detection is triggered when the comprehensive state feature vectors at all time steps all cross the tolerance boundary region; among them, 3 to 5 are acceptable.

[0124] After triggering the anomaly determination, the status assessment module 60 generates structural health status assessment results and corresponding early warning instructions to prompt maintenance personnel to conduct targeted review or repair.

[0125] In this embodiment, for situations such as insufficient initial data, boundary index out-of-bounds errors, or incomplete historical windows that occur during filtering, interpolation, cross-correlation calculation, time delay compensation, short-time Fourier transform, and degradation index calculation, the system employs a preset boundary processing strategy. The preset boundary processing strategy includes at least one of mirror extension, linear extrapolation, historical health mean filling, or pausing the output of the current index. Within the same engineering instance, a fixed boundary processing strategy is preferentially used to maintain consistency in evaluation results across different time periods.

[0126] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for assessing the structural health status of roads and bridges based on multi-source data fusion, characterized in that, Includes the following steps: Collect multi-source heterogeneous monitoring signals, including environmental meteorological data, total strain data, and vibration acceleration data of different spatial nodes; A global time axis is established based on the time of the vibration acceleration data, the multi-source heterogeneous monitoring signals are reconstructed and aligned, and the total strain data is filtered to extract the quasi-static strain sequence. The environmental meteorological data are coupled and calculated to generate an equivalent structured temperature sequence; The cross-correlation between the equivalent structural temperature sequence and the quasi-static strain sequence is calculated within a sliding window. The time-varying phase delay coefficient is extracted and compensated for the equivalent structural temperature sequence. The compensated temperature sequence is output and an environmental quasi-static baseline is constructed. The theoretical offset of the target order natural frequency is calculated based on the compensation temperature sequence to adjust the frequency domain integration boundary, and the local modal energy focusing degree of each spatial node is calculated within the adjusted boundary. Based on the quasi-static baseline of the environment and the local modal energy focusing degree, the temporal environmental degradation index and the local stiffness anomaly index are calculated respectively. A comprehensive state feature vector is constructed for boundary comparison, and the evaluation results and early warning instructions are output.

2. The method for assessing the health status of road and bridge structures based on multi-source data fusion according to claim 1, characterized in that, The process of establishing a global time axis based on the time of the vibration acceleration data, reconstructing and aligning the multi-source heterogeneous monitoring signals, and filtering the total strain data to extract a quasi-static strain sequence specifically includes: A third-order B-spline interpolation algorithm is used for time reconstruction, which maps the original sampling time nodes of the environmental meteorological data and the total strain data to the global time axis, and outputs the surface temperature sequence, wind speed sequence, light intensity sequence and total strain sequence. The total strain sequence after time reconstruction is low-pass filtered using an infinite impulse response low-pass filter to extract the quasi-static strain sequence.

3. The method for assessing the health status of road and bridge structures based on multi-source data fusion according to claim 1, characterized in that, The process of coupling and solving the environmental meteorological data to generate an equivalent structured temperature sequence specifically includes: Based on the natural convection heat transfer constant under windless conditions, the forced convection wind speed sensitivity coefficient, and the wind speed sequence in the environmental meteorological data, the wind resistance convection heat transfer coefficient of the structural surface is calculated through empirical linear relationships. By combining the solar radiation absorptivity of the bridge surface material, the ratio of the light intensity sequence in the environmental meteorological data to the wind resistance convective heat transfer coefficient is calculated to obtain the equivalent temperature increment. The equivalent temperature increment is then superimposed on the environmental surface temperature sequence to generate the equivalent structural temperature sequence.

4. The method for assessing the structural health status of roads and bridges based on multi-source data fusion according to claim 1, characterized in that, The step of calculating the cross-correlation between the equivalent structural temperature sequence and the quasi-static strain sequence within a sliding window, and extracting the time-varying phase retardation coefficient, specifically includes: The equivalent structural temperature sequence and the quasi-static strain sequence within the sliding window are respectively subjected to mean removal and standard deviation normalization. Calculate the local discrete cross-correlation function of the two, search for the peak value of the local discrete cross-correlation function within the preset translation search interval, and convert the translation step number corresponding to the peak value into the time-varying phase delay coefficient at the current time point.

5. The method for assessing the structural health status of roads and bridges based on multi-source data fusion according to claim 1, characterized in that, The compensation of the equivalent structure temperature sequence, outputting the compensated temperature sequence, and constructing an environmental quasi-static baseline specifically includes: Based on the delay step of the time-varying phase delay coefficient corresponding to the current time point, the equivalent structure temperature sequence is time-shifted using interpolation to obtain the compensated temperature sequence. Based on the compensated temperature sequence and the quasi-static strain sequence, a multinomial regression model with Huber loss function is used for fitting to construct the environmental quasi-static baseline; The environmental quasi-static baseline is separated from the total strain data to output the net strain sequence.

6. The method for assessing the structural health status of roads and bridges based on multi-source data fusion according to claim 1, characterized in that, The step of calculating the theoretical offset of the target order natural frequency based on the compensated temperature sequence to adjust the frequency domain integration boundary specifically includes: The compensation temperature sequence is input into a preset polynomial nonlinear mapping function to calculate the theoretical offset of the target order natural frequency. The coefficients of the polynomial nonlinear mapping function are obtained by least squares fitting based on data within a healthy baseline period. Perform a short-time Fourier transform on the vibration acceleration data at each of the aforementioned spatial nodes to obtain the corresponding time-frequency amplitude matrix; Based on the natural frequency of the target order under healthy baseline conditions, the basic analysis bandwidth, and the theoretical offset, the dynamic integration lower limit and dynamic integration upper limit of the frequency domain integration at the current time point are dynamically determined, and amplitude limiting is performed according to the preset 0 value and Nyquist cutoff frequency.

7. The method for assessing the health status of road and bridge structures based on multi-source data fusion according to claim 6, characterized in that, The calculation of the local modal energy focusing degree of each spatial node within the adjusted boundary specifically includes: In the time-frequency amplitude matrix, the accumulated value of the squared amplitude of discrete frequency points is extracted according to the lower limit of dynamic integration and the upper limit of dynamic integration to generate the target mode energy sequence; The sum of the squared values ​​of discrete frequency points is extracted within the full frequency band from 0 to the Nyquist cutoff frequency to generate the total energy sequence of the full frequency band. The ratio of the target modal energy sequence to the total energy sequence of the entire frequency band is calculated, and combined with a preset positive compensation term, the local modal energy focusing degree at the corresponding spatial node is generated.

8. The method for assessing the health status of road and bridge structures based on multi-source data fusion according to claim 1, characterized in that, The calculation of the temporal environmental degradation index and the local stiffness anomaly index based on the quasi-static environmental baseline and the local modal energy focusing degree specifically includes: The time change rate is obtained by calculating the difference between the quasi-static baseline of the environment before and after a set number of discrete time intervals, and is used as the time-domain environmental degradation index. For any target node among the aforementioned spatial nodes, its adjacent nodes are selected to form a spatial node set. The mean value of the local modal energy focusing degree of each node in the spatial node set is calculated, and the spatial cross-sectional variance is calculated based on the local modal energy focusing degree and the mean value. The spatial cross-sectional variance is used as the local stiffness anomaly index.

9. The method for assessing the structural health status of roads and bridges based on multi-source data fusion according to claim 1, characterized in that, The process of constructing a comprehensive state feature vector, performing boundary comparison, and outputting evaluation results and early warning instructions specifically includes: Based on historical health benchmark data, statistical values ​​of indicators were extracted to determine the environmental degradation tolerance threshold and stiffness anomaly tolerance threshold. Using the environmental degradation tolerance threshold and the stiffness anomaly tolerance threshold as parameters, an elliptical two-dimensional tolerance boundary domain is constructed. The comprehensive state feature vector is mapped to the two-dimensional tolerance boundary domain as a two-dimensional coordinate point for judgment. When the comprehensive state feature vector crosses the two-dimensional tolerance boundary domain for a consecutive preset number of time steps, the structural health status assessment result with abnormal risk is generated and the warning instruction is triggered.

10. The method for assessing the health status of road and bridge structures based on multi-source data fusion according to claim 1, characterized in that, In the step of acquiring multi-source heterogeneous monitoring signals: A global reference clock pulse is generated by a temperature-controlled crystal oscillator inside the synchronous triggering unit, and the global reference clock pulse is distributed in parallel to the analog-to-digital converters of each sensor that acquires the multi-source heterogeneous monitoring signals. When the analog-to-digital converter receives the preset level edge of the global reference clock pulse, it latches and digitally quantizes the corresponding analog signal so that the multi-source heterogeneous monitoring signals are established on a unified physical time reference.

Images