Damage evaluation and early warning method for ultra-high performance concrete panel based on multi-source sensing data

Abstract

This invention discloses a method for damage assessment and early warning of ultra-high performance concrete masonry slabs based on multi-source sensor data, belonging to the field of concrete masonry slab damage assessment technology. The method includes the following steps: calculating the deviation value of the change amplitude, the distribution index of the change rate, and the change duration index of identified high-fluctuation data; determining whether it constitutes a severe fluctuation and marking it as an abnormal signal source using multi-dimensional thresholds; calculating the consistency of change direction, similarity of fitting slope, and range of fitting residuals for other sensor data within the time period corresponding to the marked abnormal signal source, used to generate a trend stability score. Other sensors refer to sensors not marked as abnormal signal sources. This invention solves the problem of misjudgment of abnormal signals in multi-source sensor data fusion, realizes dynamic identification and weight suppression of abnormal signals, and significantly improves the accuracy and reliability of structural damage early warning.

Description

Technical Field

[0001] This invention relates to the field of concrete masonry slab damage assessment technology, specifically to a method for damage assessment and early warning of ultra-high performance concrete masonry slabs based on multi-source sensor data. Background Technology

[0002] Damage assessment and early warning for ultra-high performance concrete (UHPC) slabs based on multi-source sensor data is a method that acquires multi-dimensional and multi-temporal physical parameters of the structure during operation by deploying multiple types of sensors (such as strain gauges, accelerometers, temperature and humidity sensors, vibration sensors, etc.) on or inside the slab structure. The collected data is then fused and processed to achieve real-time assessment of the structure's health status and early warning of potential damage trends. Existing technologies typically achieve this goal through the following steps: First, the scientific deployment and calibration of multi-source sensors ensures coverage of key stress locations and guarantees data reliability; second, the synchronous acquisition and preprocessing of multi-source data includes noise reduction, normalization, and data alignment; next, feature extraction and fusion modeling employ data fusion algorithms or machine learning models to comprehensively analyze multiple dimensions such as stress response, vibration frequency changes, and temperature effects; then, condition identification is performed by setting damage identification indicators or training models to identify potential cracks, fatigue, or other forms of damage; finally, risk assessment and early warning output are based on the identification results, including early warning level determination, alarm information dissemination, and structural maintenance recommendations. This complete process enables a comprehensive understanding and early warning of the operational status of UHPC masonry structures, thereby improving the safety and service life of the infrastructure.

[0003] The existing technology has the following shortcomings:

[0004] In the process of damage assessment and early warning of ultra-high performance concrete masonry based on multi-source sensor data, if a strain sensor generates a non-structural abrupt signal within a specific acquisition period due to poor installation contact, while the data collected by other types of sensors (such as crack, acoustic emission, temperature, etc.) remain stable, the amplitude of this abnormal signal will be given an unreasonable evaluation weight during the data fusion stage because it is significantly higher than that of other data sources. This leads to an abnormally high overall damage assessment result and triggers an erroneous high-level early warning. Because the current fusion assessment mechanism mainly relies on a fixed sensor weight allocation method, it lacks the ability to dynamically compare and identify the trend stability of other data sources under drastic fluctuations in a single data source. This results in the system being unable to determine whether the abrupt signal is non-structural interference, thus failing to suppress the triggering of erroneous early warnings. Therefore, existing high-performance concrete slab damage assessment and early warning technologies based on multi-source sensor data cannot reasonably judge the reliability of a signal and suppress false early warnings caused by abnormal data based on the trend stability of other data sources when a single signal source suddenly fluctuates drastically during the fusion assessment process. Ultimately, this may lead to the structural state being misjudged as severely damaged, resulting in unnecessary maintenance scheduling, traffic control, or operational interruptions. At the same time, it weakens the system's sensitivity to actual damage, reduces early warning reliability, and decreases engineering response efficiency.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a method for damage assessment and early warning of ultra-high performance concrete masonry slabs based on multi-source sensor data, so as to solve the problems in the background art mentioned above.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for damage assessment and early warning of ultra-high performance concrete masonry slabs based on multi-source sensor data, specifically including the following steps:

[0008] S1. Construct a time-series response trajectory set from the continuous data acquired by multiple sensors within a unified acquisition period, and extract the rate of change index of each sensor to identify high-fluctuation data.

[0009] S2. Calculate the deviation of the change amplitude, the distribution index of the change rate, and the change duration index for the identified high-fluctuation data. Use multi-dimensional thresholds to determine whether it constitutes a violent fluctuation and mark it as an abnormal signal source.

[0010] S3. Calculate the consistency of change direction, similarity of fitting slope, and range of fitting residuals for other sensor data within the time period corresponding to the marked abnormal signal source, and use them to generate a trend stability score. Other sensors refer to sensors that have not been marked as abnormal signal sources.

[0011] S4. The results of the judgment of violent fluctuations are jointly mapped with the trend stability score to a credibility value, which serves as the basis for the reliability of participation of abnormal signal sources.

[0012] S5. Adjust the participation weight of abnormal signal sources in the fusion evaluation calculation based on the credibility value, generate structural damage evaluation results by weighted fusion of the evaluation parameters of each sensor, and complete the early warning output.

[0013] Preferably, S2 specifically includes the following steps:

[0014] S201. Calculate the difference between the maximum and minimum values ​​for each segment of highly fluctuating data, and compare it with the average difference of the corresponding data of the same type of sensor in the historical acquisition period to obtain the deviation value of the change range.

[0015] S202. For each segment of highly fluctuating data, the numerical change interval between adjacent data points is statistically analyzed, a change rate distribution map is constructed, and the frequency concentration interval and the width of the change interval are extracted as change rate distribution indicators.

[0016] S203. Calculate the length of time that each segment of identified high-fluctuation data remains above the rate of change judgment benchmark in a continuous sampling period, and use it as an indicator of the duration of change.

[0017] S204. Compare the deviation value of the change amplitude with the change amplitude deviation threshold, compare the change rate distribution index with the change rate distribution threshold, and compare the change duration with the change duration threshold. When all three comparison results are greater than the threshold, the high fluctuation data is identified as a violent fluctuation, and the corresponding sensor is marked as an abnormal signal source.

[0018] Preferably, S201 specifically refers to:

[0019] For each segment of highly volatile data, extract the maximum and minimum values ​​within the current acquisition period, and calculate the difference as the current change amplitude value;

[0020] In the historical data of the same type of sensor, select multiple time periods with the same acquisition cycle length, calculate the difference between the maximum and minimum values ​​in each time period, and take the arithmetic mean of these differences as the average difference.

[0021] Subtract the average difference from the current change value to obtain the change deviation value, which is used to determine subsequent drastic fluctuations.

[0022] Preferably, S202 specifically refers to:

[0023] For each segment of highly volatile data, the numerical difference between every two adjacent data points is calculated sequentially, and the change interval sequence composed of all differences is statistically analyzed.

[0024] The frequency statistics of the changing interval sequence are statistically analyzed and a histogram is plotted to generate a rate of change distribution map.

[0025] In the rate of change distribution spectrum, the interval of continuous values ​​with the highest frequency is identified as the frequency concentration interval, and the difference between the minimum and maximum change values ​​in the distribution spectrum is calculated as the width of the change interval.

[0026] Preferably, S203 is as follows:

[0027] Each segment of high-fluctuation data is compared point by point with the rate of change judgment benchmark, and the continuous sampling point sequence with values ​​greater than the rate of change judgment benchmark is marked.

[0028] In each sequence of marked sampling points, the number of consecutive adjacent sampling points is counted, and this number is multiplied by the sampling interval to obtain the length of time in each sequence where the rate of change is greater than the threshold for determining the rate of change.

[0029] The maximum value among all time periods is selected as the indicator of the duration of change in the currently identified high-volatility data.

[0030] Preferably, S3 specifically includes the following steps:

[0031] S301. Select the sensor data that is not marked as an abnormal signal source within the time period corresponding to the sensor marked as an abnormal signal source, construct the time-series response trajectory of each sensor within the time period, and extract the change direction of each response trajectory.

[0032] S302. Compare the direction of change of each sensor that is not marked as an abnormal signal source with the direction of change of the sensor that is marked as an abnormal signal source point by point, and count the percentage of data points with the same direction as the consistency of the direction of change.

[0033] S303. Perform linear least squares regression on the response trajectories of all sensors that are not marked as abnormal signal sources, obtain the fitting slope and fitting residual, and then compare the fitting results with those of sensors marked as abnormal signal sources. Calculate the similarity of fitting slope and the range of fitting residual respectively.

[0034] S304. Based on the consistency of the direction of change, the similarity of the fitting slope, and the range of the fitting residual, weight coefficients are assigned respectively, and a trend stability score is generated by weighted averaging to determine whether to perform weighted suppression processing on abnormal signal sources.

[0035] Preferably, S302 is as follows:

[0036] Extract the sequence of changes in the direction of adjacent data points of the sensor marked as an abnormal signal source within the corresponding time period. The direction of change is determined by the sign of the difference between adjacent data points.

[0037] Extract the change direction sequence of sensors that are not marked as abnormal signal sources within the same time period, and maintain a one-to-one correspondence with the time points of the direction sequence of abnormal signal sources;

[0038] Compare the directions of each pair of corresponding data points in the two change direction sequences, count the number of data points with the same direction, divide the number by the total number of data points, calculate the direction consistency ratio, and use it as the change direction consistency index.

[0039] Preferably, S303 is as follows:

[0040] For each sensor that was not marked as an anomalous signal source, the response trajectory within the target time period was fitted using the linear least squares method to obtain the corresponding fitting slope and residual mean square value.

[0041] The same fitting process was performed on the sensor response trajectories marked as abnormal signal sources, and the corresponding fitting slope and residual mean square value were obtained as a comparison benchmark.

[0042] Calculate the mean difference between the fitted slope of all unlabeled signal sources and the baseline slope, as the fitted slope similarity; extract the maximum difference between the fitted residual of all unlabeled signal sources and the baseline residual, as the fitted residual range.

[0043] Preferably, S4 is as follows:

[0044] The result of the violent fluctuation judgment is mapped to a numerical order of magnitude. When the violent fluctuation judgment is true, it is set to the first preset value, and when the violent fluctuation judgment is false, it is set to the second preset value, thus forming a violent fluctuation numerical factor.

[0045] The trend stability score is normalized by interval, and the score is converted into a trend stability factor with a unified standard to ensure its additive integration with the drastic fluctuation numerical factor on the same numerical scale.

[0046] Using the drastic fluctuation numerical factor and the trend stability factor as joint input variables, a weighted linear combination function is executed to output a confidence value. The confidence value serves as the reliability basis for the participation of the corresponding abnormal signal source in subsequent information fusion calculations.

[0047] Preferably, S5 is as follows:

[0048] The participation weight of sensors marked as anomalous signal sources in the fusion evaluation calculation is adjusted based on the confidence level value. The participation weight of sensors not marked as anomalous signal sources is set to the full weight value, and the participation weight of sensors marked as anomalous signal sources is equal to the corresponding confidence level value.

[0049] Extract the evaluation parameters of all sensors in the current evaluation period, multiply the evaluation parameters of each sensor with their corresponding participation weights to obtain the weighted evaluation values ​​of each sensor, and then normalize and aggregate all weighted evaluation values ​​to generate structural damage evaluation results.

[0050] The structural damage assessment result is numerically compared with the damage judgment threshold. When the structural damage assessment result is greater than the damage judgment threshold, a structural damage early warning signal is output. At the same time, the structural location information, damage level value and participation weight information within the current assessment period are output to complete the early warning output.

[0051] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0052] 1. This invention effectively solves the problem of misjudgment caused by sudden changes in individual sensor signals in structural damage assessment by constructing an anomaly identification and dynamic fusion mechanism based on multi-source sensor data. This method not only integrates data from multiple sensors such as strain, cracks, acoustic emission, and temperature, but also introduces multi-dimensional indicators such as deviation of change amplitude, distribution of change rate, and duration of change to accurately identify highly fluctuating data. Furthermore, it establishes a logic for judging drastic fluctuations by comparing with historical data, achieving preliminary screening of anomaly signal sources. Based on this, a trend stability score is formed by calculating the consistency of change direction, similarity of fitting slope, and range of fitting residuals between the anomaly signal and other sensors. This score is then jointly mapped with the fluctuation judgment result to a credibility score, quantifying the credibility of the anomaly signal and providing a scientific basis for subsequent weighted fusion.

[0053] 2. This invention adjusts the participation weight of abnormal signal sources in real time by adjusting the confidence level value, and performs weighted aggregation in combination with other sensor evaluation parameters. This effectively suppresses the interference of non-structural, violent fluctuation signals on the final damage assessment results, avoiding false high-level warnings. Simultaneously, while outputting the structural damage assessment results, this method can also provide multi-dimensional information such as structural location, damage level, and participation weight, improving the interpretability and decision-making value of the warning, and greatly enhancing the system's sensitivity to actual structural damage and overall warning reliability. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0055] Figure 1 This is a flowchart illustrating the method for damage assessment and early warning of ultra-high performance concrete masonry slabs based on multi-source sensor data according to the present invention. Detailed Implementation

[0056] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0057] This invention provides, for example Figure 1 The damage assessment and early warning method for ultra-high performance concrete masonry slabs based on multi-source sensor data shown includes the following steps:

[0058] S1. Construct a time-series response trajectory set from the continuous data acquired by multiple sensors within a unified acquisition period, and extract the rate of change index of each sensor to identify high-fluctuation data.

[0059] In this embodiment, S1 specifically refers to:

[0060] Continuous data acquired by multiple sensors within a unified acquisition period are bidirectionally sorted according to sensor type and acquisition time to construct the time series corresponding to each type of sensor, and combined to form a set of time-series response trajectories.

[0061] In the process of bidirectionally sorting continuous data acquired by multiple sensors within a unified acquisition period according to sensor category and acquisition time, and constructing the time series corresponding to each type of sensor and combining them to form a time-series response trajectory set, it is first necessary to set a unified acquisition period boundary and initially classify the raw data returned by each sensor according to the sensor number. Each category represents a data source of a certain type of physical quantity, such as strain, vibration, temperature, or acoustic emission. Subsequently, within each type of sensor data, the timestamp field of each data record is read and sorted in ascending order according to time sequence to ensure that the data points of each sensor are completely consistent in order on the time axis. Next, the time series of multiple sensors under the same physical quantity category are horizontally spliced ​​so that the responses of the same type at each moment form a set of response vectors, ultimately forming a complete time series matrix for that type of sensor. For data of multiple categories, the above time series matrix is ​​arranged one by one according to the physical quantity category, vertically integrated into a multi-dimensional time-sensor response structure, and organized and stored in the form of a two-dimensional or three-dimensional array, forming a complete time-series response trajectory set across categories and time axes for subsequent rate of change analysis and high fluctuation identification. Throughout the process, efficient data processing and accurate data construction can be achieved through timestamp alignment algorithms, multidimensional array concatenation operations, and data structure indexing mechanisms.

[0062] Extract the time series corresponding to each type of sensor from the time series response trajectory set, and construct the rate of change index sequence by the numerical difference between adjacent acquisition time points;

[0063] In the process of extracting the time series corresponding to each type of sensor from the time-series response trajectory set and constructing the rate of change index sequence by the numerical difference between adjacent acquisition time points, the time-series response trajectory set first needs to be traversed by sensor category to extract the time series matrix corresponding to each type of sensor. Each row in this matrix represents continuous data of a sensor within a unified acquisition period, and each column corresponds to a standardized time sampling point. Subsequently, for each row of time series data, a difference algorithm is used to perform a difference operation on the values ​​of adjacent acquisition time points, that is, subtracting the nth sampling value from the (n+1)th sampling value to obtain a continuous change value sequence. This sequence is the rate of change index sequence, used to quantify the rate of change of the sensor response in the time dimension. This operation can be implemented through a sliding window mechanism with a window length of 2 and a step size of 1 to ensure that the difference sequence and the original sequence length are consistent. The resulting sequence is then standardized to uniformly adjust the units, making the rate of change indices between different sensors comparable. Throughout the process, the rate of change index sequence not only retains the dynamic trend characteristics of the original signal but also effectively eliminates the information redundancy caused by low-frequency slow changes, facilitating the subsequent high-fluctuation identification module to accurately identify abrupt changes.

[0064] For each type of sensor, the mean and standard deviation of the rate of change index sequence are statistically analyzed. Continuous segments with a rate of change greater than the sum of the mean and standard deviation are selected as high-fluctuation data, and the identification of high-fluctuation data is completed.

[0065] In the process of performing mean and standard deviation statistics on the rate of change index sequences of each type of sensor and selecting continuous segments with a rate of change greater than the sum of the mean and standard deviation as high-fluctuation data, the rate of change index sequences first need to undergo mathematical statistical processing. The arithmetic mean of the entire sequence is calculated to reflect the central trend of its overall change, while the standard deviation is calculated to measure the degree of fluctuation distribution within the sequence. Next, a fluctuation identification threshold is set as the sum of the mean and standard deviation, serving as the boundary for judging abnormal fluctuations. By traversing the rate of change index sequence point by point, each rate of change value is compared to the threshold. If the rate of change of multiple consecutive sampling points is greater than the threshold, the continuous segment is extracted as high-fluctuation data. This operation can be accomplished through a combination of Boolean flags and a sliding window; when the number of consecutive points meeting the conditions in the window reaches a set lower limit, it is identified as a high-fluctuation segment. The design logic of this method is that the combined threshold of mean and standard deviation takes into account both the overall trend and local fluctuations, can adapt to changes in the response characteristics of different sensors, and avoids misidentifying normal micro-fluctuations as abnormalities. This helps improve identification accuracy and accurately extract atypical drastic fluctuation behaviors, providing a reliable foundation for subsequent reliability assessment and early warning control.

[0066] S2. Calculate the deviation of the change amplitude, the distribution index of the change rate, and the change duration index for the identified high-fluctuation data. Use multi-dimensional thresholds to determine whether it constitutes a violent fluctuation and mark it as an abnormal signal source.

[0067] In this embodiment, S2 specifically includes the following steps:

[0068] S201. Calculate the difference between the maximum and minimum values ​​for each segment of highly fluctuating data, and compare it with the average difference of the corresponding data of the same type of sensor in the historical acquisition period to obtain the deviation value of the change range.

[0069] S202. For each segment of highly fluctuating data, the numerical change interval between adjacent data points is statistically analyzed, a change rate distribution map is constructed, and the frequency concentration interval and the width of the change interval are extracted as change rate distribution indicators.

[0070] S203. Calculate the length of time that each segment of identified high-fluctuation data remains above the rate of change judgment benchmark in a continuous sampling period, and use it as an indicator of the duration of change.

[0071] S204. Compare the deviation value of the change amplitude with the change amplitude deviation threshold, compare the change rate distribution index with the change rate distribution threshold, and compare the change duration with the change duration threshold. When all three comparison results are greater than the threshold, the high fluctuation data is identified as a violent fluctuation, and the corresponding sensor is marked as an abnormal signal source.

[0072] To ensure the robustness of the judgment results during the determination of severe fluctuations, three key indicators need to be incorporated into the judgment logic simultaneously. Specifically, the deviation of the change amplitude of each high-fluctuation data segment is compared with a preset change amplitude deviation threshold. Secondly, the width of the frequency concentration interval extracted from the corresponding rate of change distribution indicator is compared with the rate of change distribution threshold. Thirdly, the calculated duration of the change is compared with the duration threshold. When all three comparison results are greater than a certain value, it indicates that the data segment has reached a significantly abnormal level in terms of amplitude deviation, rate fluctuation amplitude, and time dimension, comprehensively demonstrating strong abnormal fluctuation characteristics. At this point, the high-fluctuation data segment can be identified as constituting severe fluctuation, and its corresponding sensor is marked as an abnormal signal source within the current acquisition period for subsequent reliability judgment and early warning weight adjustment. By simultaneously introducing three types of indicators and setting a strict multi-threshold joint judgment mechanism, misjudgment due to a single indicator anomaly can be effectively avoided, improving the accuracy and anti-interference capability of severe fluctuation identification.

[0073] The deviation threshold for amplitude, the rate of change distribution threshold, and the duration threshold are used to determine whether high-fluctuation data meets the criteria for severe fluctuation in three dimensions: fluctuation intensity, rate dispersion, and duration. The amplitude deviation threshold is calculated by statistically analyzing the difference between the maximum and minimum values ​​of similar sensors in each acquisition cycle under normal conditions, and using this deviation from the mean as a criterion. The rate of change distribution threshold is based on the variation interval sequence of historical fluctuation data. By analyzing the distribution characteristics of the variation interval width in its frequency distribution spectrum, and combining engineering experience, an acceptable upper limit of fluctuation range is selected as the standard. The duration threshold is calculated by simulating high-fluctuation samples under various known normal and abnormal conditions, extracting the duration of continuous high-rate-change segments, and setting it as a lower limit that stably reflects the persistence of abnormalities, combined with false trigger tolerance. All three thresholds are derived from the collaborative calculation of historical data statistics, experimental verification, and early warning false judgment tolerance, ensuring the engineering adaptability and accuracy of the judgment criteria.

[0074] To accurately determine whether highly volatile data constitutes severe fluctuations, a quantitative assessment is needed using three complementary dimensions. The deviation of the amplitude of change measures the degree to which the current data fluctuation intensity deviates from historical levels, reflecting the suddenness of abnormal fluctuations. The rate of change distribution index reflects the severity of data fluctuations in a short period through the frequency concentration interval and the width of the change interval, assessing the discreteness of its dynamic changes. The duration of change measures whether high-intensity fluctuations have continuity in the time dimension, revealing the stable abnormal characteristics of the fluctuations. Comparing these three indicators with set thresholds for amplitude deviation, rate of change distribution, and duration of change allows for a comprehensive assessment of the intensity, rate, and duration of a single fluctuation segment. Only when all three indicators exceed their respective thresholds can a truly disruptive severe fluctuation be identified, thus marking the corresponding sensor as an abnormal signal source and avoiding misjudgments due to localized random fluctuations. This multi-dimensional cross-validation method enhances the accuracy of the judgment and the system's noise resistance, making it particularly suitable for damage assessment and early warning scenarios with complex interference factors in multi-source sensor data.

[0075] In this embodiment, S201 specifically refers to:

[0076] For each segment of highly volatile data, extract the maximum and minimum values ​​within the current acquisition period, and calculate the difference as the current change amplitude value;

[0077] After acquiring each segment of identified high-fluctuation data, the start and end times of the current acquisition period must first be determined. Within this time period, all measured values ​​in the data segment are extracted. The maximum and minimum values ​​are then extracted by iterating through the data, and the maximum value is subtracted from the minimum value to calculate the current amplitude of change. The purpose of this operation is to quantify the response fluctuation range of the data segment within the time interval. For example, if a sensor records data [3.2, 4.1, 3.8, 5.6, 4.9] in a certain acquisition period, the maximum value is 5.6, the minimum value is 3.2, and the difference is 2.4, meaning the current amplitude of change for this data segment is 2.4. This amplitude value can be used for subsequent deviation analysis from the historical average fluctuation level to determine whether the segment belongs to an abnormal fluctuation range. In this technical feature, the "current acquisition period" limits the analysis time range, the "difference between the maximum and minimum values" constitutes a quantitative discrimination index, and the "amplitude of change" serves as an evaluation benchmark in subsequent reliability derivation. This process can be implemented through sliding window traversal, efficient numerical extraction algorithms, or matrix filtering methods, and is applicable to various sensor types and data structures.

[0078] In the historical data of the same type of sensor, select multiple time periods with the same acquisition cycle length, calculate the difference between the maximum and minimum values ​​in each time period, and take the arithmetic mean of these differences as the average difference.

[0079] When analyzing the magnitude deviation of highly volatile data, it is necessary to establish a reference standard based on historical data. First, the historical records of sensors of the same category are divided into multiple non-overlapping historical periods according to time intervals of equal length to the current acquisition cycle. Within each period, the maximum and minimum values ​​of the sensors are extracted, and their difference is calculated as the historical fluctuation amplitude for that period. Then, the fluctuation amplitude values ​​of all historical periods are summed and divided by the total number of periods to obtain the arithmetic mean, which is the average difference, used for subsequent comparison with the current fluctuation amplitude value. For example, if the current acquisition cycle is 5 minutes, and three consecutive 5-minute periods are selected from the historical data, with maximum and minimum value differences of 1.5, 2.0, and 1.8 respectively, the average difference is (1.5 + 2.0 + 1.8) / 3 = 1.77. In the technical characteristics, "same category of sensors" ensures consistent data sources, "same acquisition cycle length" ensures time scale alignment, "difference between maximum and minimum values" serves as a measure of local fluctuation, and the "arithmetic mean" provides a stability reference benchmark. This calculation process can be achieved through window sliding capture, extreme value function extraction, and mean value calculation modules, ensuring both calculation accuracy and processing efficiency.

[0080] Subtract the average difference from the current change value to obtain the change deviation value, which is used to determine subsequent drastic fluctuations.

[0081] To quantify the deviation of current data fluctuations from historical normal fluctuation levels, it is necessary to perform a difference calculation between the current fluctuation amplitude and the historical average difference. Specifically, the following steps are taken: First, obtain the difference between the maximum and minimum values ​​of the high-fluctuation data identified within the current acquisition period, as the current fluctuation amplitude value; then, obtain the average fluctuation amplitude of the same type of sensor over multiple equal-length time periods from historical data, as the average difference; subtract the average difference from the current fluctuation amplitude value, and the resulting value is the fluctuation amplitude deviation value, used to characterize the degree of abnormality of the current fluctuation. For example, if the current fluctuation amplitude value is 2.5 and the average difference is 1.8, then the fluctuation amplitude deviation value is 2.5 − 1.8 = 0.7. The larger this value, the more drastic the current fluctuation is compared to historical levels, and the more likely it is an abnormal signal. In the technical characteristics, the "current fluctuation amplitude value" comes from the target acquisition period, and the "average difference" comes from the statistical characteristics of historical data; the difference between the two constitutes an important parameter for determining the significance of fluctuations. This calculation process can be completed through basic arithmetic subtraction operations and serves as the core input basis for subsequent multi-dimensional threshold judgments.

[0082] In this embodiment, S202 specifically refers to:

[0083] For each segment of highly volatile data, the numerical difference between every two adjacent data points is calculated sequentially, and the change interval sequence composed of all differences is statistically analyzed.

[0084] To extract the variation characteristics of highly volatile data in a time series, it is necessary to calculate the difference between adjacent points for each identified highly volatile data segment to construct a variation interval sequence. This is achieved by arranging the data segment in chronological order, selecting each pair of adjacent data points sequentially, calculating their numerical difference, and retaining the positive and negative directions of the difference. This process is repeated until a complete variation interval sequence is formed. For example, if the continuous measurements of a highly volatile data segment are [2.1, 2.4, 2.0, 1.9, 2.3], then the difference between adjacent data points is [+0.3, -0.4, -0.1, +0.4]. This difference sequence is the variation interval sequence for that highly volatile data segment. This sequence not only reflects the amplitude of data fluctuations but also the direction of change, and can be used for subsequent frequency statistics and distribution analysis. Among the technical features, the "difference between adjacent data points" reflects the micro-level change trend, while the "variation interval sequence" is the basis for constructing a rate of change distribution map, possessing the ability to quantitatively describe high-frequency fluctuation behavior. This calculation process can be implemented in the data processing program using a cyclic difference function or a sliding window subtraction.

[0085] The frequency statistics of the changing interval sequence are statistically analyzed and a histogram is plotted to generate a rate of change distribution map.

[0086] To comprehensively describe the rate of change characteristics of highly volatile data, it is necessary to perform frequency statistics on the change interval sequence and generate a rate of change distribution map. The specific implementation method is as follows: First, a fixed width of the numerical interval is set. Each difference in the change interval sequence is assigned to a corresponding interval based on its absolute value, and the number of differences appearing in each interval is counted to obtain the frequency value of each interval. Then, a histogram is plotted with the interval as the horizontal axis and the frequency as the vertical axis, forming a frequency distribution graph of the rate of change. For example, if the difference range is from -0.6 to +0.6, and the interval width is set to 0.2, then the statistical intervals are [-0.6~-0.4], [-0.4~-0.2], ..., [+0.4~+0.6], and the number of differences within each of these intervals is counted. This method can intuitively display the central tendency and fluctuation amplitude of the differences, thereby generating a complete rate of change distribution map. Among the technical features, "frequency statistics" are used to quantify the frequency of occurrence of different rates of change, "histogram" provides an intuitive representation of the rate distribution, and "rate of change distribution map" provides a structured basis for subsequent extraction of concentration intervals and change widths. This process can be accomplished using a combination of histogram statistics algorithms and visualization plotting functions.

[0087] In the rate of change distribution spectrum, the interval of continuous values ​​with the highest frequency is identified as the frequency concentration interval, and the difference between the minimum and maximum change values ​​in the distribution spectrum is calculated as the width of the change interval.

[0088] In the rate of change distribution map, to extract the statistical region that best reflects the concentrated characteristics of data fluctuations, it is necessary to identify the continuous numerical intervals with the highest frequency and calculate the width of the variation interval. The specific implementation is as follows: First, based on the frequency values ​​of each numerical interval in the histogram, traverse all adjacent interval combinations to find the continuous interval with the largest cumulative frequency value, and use its corresponding numerical range as the frequency concentration interval. Then, identify the minimum and maximum boundary values ​​in this interval group and calculate their difference as the variation interval width. For example, if the frequency concentration interval is [-0.2, 0.2], then the variation interval width is 0.4. In the technical features, the "continuous numerical interval with the highest frequency" reflects the main distribution range of data fluctuations, while the "numerical difference between the minimum and maximum change values" characterizes the extent of the fluctuation. This process can be implemented using a sliding window plus frequency accumulation discrimination strategy to accurately define the core area of ​​rate fluctuations.

[0089] In this embodiment, S203 specifically refers to:

[0090] Each segment of high-fluctuation data is compared point by point with the rate of change judgment benchmark, and the continuous sampling point sequence with values ​​greater than the rate of change judgment benchmark is marked.

[0091] To determine the duration characteristics of highly volatile data, the rate of change sequence corresponding to each identified highly volatile data segment needs to be compared point-by-point with a preset rate of change criterion. Specifically, each value in the rate of change sequence is read sequentially and compared with the set rate of change criterion. If the value is greater than the criterion, the corresponding time point is marked as a high-change point. If multiple consecutive sampling points meet this condition, this set of consecutive time points is marked as a high-change sampling point sequence. In the technical characteristics, the "rate of change sequence" reflects the temporal rate of change of sensor data, the "rate of change criterion" is a quantitative standard for high-volatile behavior, "point-by-point comparison" ensures that each sampling moment is accurately evaluated, and the marking of the "continuous sampling point sequence" provides a basis for subsequent calculation of the duration of the volatility. This process can be implemented by setting logical conditions for iterative comparison and using Boolean arrays for marking.

[0092] In each sequence of marked sampling points, the number of consecutive adjacent sampling points is counted, and this number is multiplied by the sampling interval to obtain the length of time in each sequence where the rate of change is greater than the threshold for determining the rate of change.

[0093] To calculate the duration for which the rate of change in each marked sampling point sequence remains above the threshold for determining the rate of change, it is necessary to count the number of adjacent points for each group of consecutively marked "high-change" sampling points. Specifically, this involves scanning the Boolean marker array; when encountering a continuous segment of "true" values, recording its start and end positions, calculating the number of sampling points contained in that segment, and multiplying this number by the sampling interval to obtain the duration of that segment. For example, if a high-change segment consists of 5 adjacent sampling points with a sampling interval of 0.1 seconds, then the duration of that segment is 0.5 seconds. In the technical features, "the number of consecutive adjacent sampling points" represents the number of consecutive samples in a high-fluctuation state, and "the sampling interval" reflects the data acquisition frequency. Multiplying the two together converts them to a physical time scale, thus achieving precise quantification of the duration of drastic fluctuations. This statistical process can be implemented using a loop accumulation method combined with state transition judgment logic.

[0094] The maximum value among all time periods is selected as the indicator of the duration of change in the currently identified high-volatility data.

[0095] After calculating the duration of changes for multiple consecutive high-variance sampling point sequences identified in each segment of high-fluctuation data, a representative value needs to be selected from these durations to reflect the persistence of the drastic fluctuations. Specifically, all calculated durations are iterated through, their values ​​are compared, and the largest value is selected as the duration index for that segment of high-fluctuation data. This index reflects the longest consecutive drastic change interval during the entire high-fluctuation process and is a crucial basis for determining whether a drastic fluctuation has occurred. In the technical features, the "duration" is derived from the comparison of the preceding rate of change with a threshold, while the "maximum value" serves as an upper bound on the persistence of change, helping to avoid underestimating the intensity of fluctuations due to short-term anomalies. This can be quickly extracted using array iteration or a built-in maximum value function, ensuring the accuracy and timeliness of subsequent judgments.

[0096] S3. Calculate the consistency of change direction, similarity of fitting slope, and range of fitting residuals for other sensor data within the time period corresponding to the marked abnormal signal source, and use them to generate a trend stability score. Other sensors refer to sensors that have not been marked as abnormal signal sources.

[0097] In this embodiment, S3 specifically includes the following steps:

[0098] S301. Select the sensor data that is not marked as an abnormal signal source within the time period corresponding to the sensor marked as an abnormal signal source, construct the time-series response trajectory of each sensor within the time period, and extract the change direction of each response trajectory.

[0099] Within a unified acquisition period, the first step is to determine the time periods corresponding to sensors already marked as anomalous signal sources. Then, within these time periods, all sensor data not marked as anomalous signal sources are filtered out. For each sensor not marked as an anomalous signal source, continuous data points within that time period are extracted in chronological order of sampling time, constructing a time-series response trajectory with time as the horizontal axis and sensor data as the vertical axis. This trajectory describes the sensor's response change process within that time period. After constructing all trajectories, the numerical differences between adjacent sampling points in each trajectory are extracted one by one. The sign of the difference determines the direction of change between each pair of sampling points, where a positive value indicates an upward direction, a negative value indicates a downward direction, and zero indicates that the direction remains unchanged. Finally, the sequence of change directions for each sensor is preserved for subsequent comparative analysis with the change directions of anomalous signal sources. This entire process provides a directional feature basis for trend consistency calculation, while ensuring that the constructed time-series response trajectories are comparable and consistent in both the time and data dimensions.

[0100] S302. Compare the direction of change of each sensor that is not marked as an abnormal signal source with the direction of change of the sensor that is marked as an abnormal signal source point by point, and count the percentage of data points with the same direction as the consistency of the direction of change.

[0101] S303. Perform linear least squares regression on the response trajectories of all sensors that are not marked as abnormal signal sources, obtain the fitting slope and fitting residual, and then compare the fitting results with those of sensors marked as abnormal signal sources. Calculate the similarity of fitting slope and the range of fitting residual respectively.

[0102] S304. Based on the consistency of the direction of change, the similarity of the fitting slope, and the range of the fitting residual, weight coefficients are assigned respectively, and a trend stability score is generated by weighted averaging to determine whether to perform weighted suppression processing on abnormal signal sources.

[0103] When generating a trend stability score, the three indicators—consistency of change direction, similarity of fitting slope, and range of fitting residuals—first need to be standardized in terms of dimensions to ensure their values ​​fall within the same range, for example, by normalizing them to the interval between 0 and 1. Then, a weight coefficient is assigned to each indicator. This weight coefficient can be set based on the relative importance of each indicator to trend judgment in practical applications; for example, consistency of change direction could have a weight of 0.4, similarity of fitting slopes 0.3, and range of fitting residuals 0.3. The weighted average is then calculated as the trend stability score by multiplying each normalized indicator value by its corresponding weight coefficient and summing the results. For example, if the normalized values ​​of the three indicators are 0.9, 0.7, and 0.6, the score would be 0.9 × 0.4 + 0.7 × 0.3 + 0.6 × 0.3 = 0.78. The trend stability score is used to quantify the consistency of the abnormal signal source with other sensors in trend behavior. The higher the score, the stronger the overall consistency and the higher the reliability. If the score is lower than the preset threshold, it indicates that the abnormal signal may deviate significantly from other data sources. It is recommended to suppress its evaluation weight in subsequent fusion calculations to reduce the risk of misjudgment caused by local anomalies and improve the stability and accuracy of the early warning system.

[0104] In this embodiment, S302 specifically refers to:

[0105] Extract the sequence of changes in the direction of adjacent data points of the sensor marked as an abnormal signal source within the corresponding time period. The direction of change is determined by the sign of the difference between adjacent data points.

[0106] Within a defined time period, continuous response data collected by sensors identified as anomalous signal sources are extracted and arranged chronologically to form a time series. For each pair of adjacent data points in this time series, the numerical difference between the current data point and the previous data point is calculated. The sign of this difference indicates the direction of change: a positive difference indicates an upward trend in the sensor value; a negative difference indicates a downward trend; and a zero difference indicates that the sensor value remains unchanged during that period. The direction of change between each pair of adjacent sampling points is recorded sequentially to form a complete sequence of change directions. This sequence of change directions describes the dynamic trend of the anomalous signal source within the target time period and serves as the basis for subsequent consistency comparisons with other sensors. Its accuracy and the precision of the direction determination directly affect the judgment of trend stability. This method enables a refined expression of the response change pattern of anomalous signal sources in the time dimension.

[0107] Extract the change direction sequence of sensors that are not marked as abnormal signal sources within the same time period, and maintain a one-to-one correspondence with the time points of the direction sequence of abnormal signal sources;

[0108] Within a defined time period, response data are extracted from all sensors not marked as anomalous signal sources. The data from each sensor are then arranged chronologically according to their acquisition time to construct a corresponding time series. For each time series, the difference between adjacent data points is calculated, and the direction of change is determined based on the sign of the difference, resulting in a corresponding change direction sequence. The direction points in each change direction sequence must be strictly aligned on the time axis with the change direction sequences of sensors marked as anomalous signal sources, ensuring a one-to-one correspondence between the data directions of different sensors at the same time point. This one-to-one correspondence can be established through a unified sampling index or timestamp matching mechanism, ensuring the comparability of the change trends of each sensor within the same time window in subsequent direction consistency comparisons. This method guarantees the accuracy and logical integrity of the direction consistency index statistics, which is a prerequisite for constructing a trend stability score.

[0109] Compare the directions of each pair of corresponding data points in the two change direction sequences, count the number of data points with the same direction, divide the number by the total number of data points, calculate the direction consistency ratio, and use it as the change direction consistency index.

[0110] For each pair of change direction sequences, the change directions of corresponding data points from the two sensors at the same time point are compared to determine if the directions are consistent. The condition for consistency is that the difference between the two data points has the same sign, i.e., both are positive or both are negative. The number of data points satisfying the consistency condition across all time points is counted and recorded as the consistency count. Then, the consistency count is divided by the total number of data points within the time period to calculate the direction consistency percentage. For example, in a time period with 10 sampling points, if the change directions of the two sensors are completely consistent at 8 of the time points, the direction consistency percentage is 8 divided by 10, resulting in 0.8. This percentage serves as a quantitative indicator of the synchronization of change trends between different sensors. A higher value indicates that the abnormal signal is closer to the trend direction of other sensors, which helps in subsequent trend stability assessment. The comparison method can be implemented through sign array comparison, which has the advantages of clear calculation logic and simple operation.

[0111] In this embodiment, S303 specifically refers to:

[0112] For each sensor that was not marked as an anomalous signal source, the response trajectory within the target time period was fitted using the linear least squares method to obtain the corresponding fitting slope and residual mean square value.

[0113] When performing linear least squares fitting on the response trajectory collected by a sensor not marked as an anomalous signal source within a target time period, a set of time-value pairs is first constructed using sampling time as the independent variable and sensor data values ​​as the dependent variable. In linear least squares, the goal is to determine a straight line y = ax + b such that the sum of the squares of the perpendicular distances from all points to this line is minimized. The slope 'a' is obtained by calculating the covariance and variance of this sequence, and the intercept 'b' is determined by combining the mean. After fitting, the squares of all differences between each actual sampled value and the estimated value at the corresponding time point on the fitted line are calculated, and their average is obtained to obtain the residual mean square value. For example, in a sensor with a time series of [1,2,3,4,5] and response data of [2.1,2.5,2.9,3.2,3.8], the slope obtained through linear fitting is approximately 0.42. The residual mean square value reflects the degree of deviation between the data and the linear fit; the smaller the value, the closer the data is to a linear trend. This method has a clear mathematical basis and is repeatable, making it suitable for trend extraction from continuous time series data.

[0114] The same fitting process was performed on the sensor response trajectories marked as abnormal signal sources, and the corresponding fitting slope and residual mean square value were obtained as a comparison benchmark.

[0115] When fitting the response trajectory of a sensor marked as an anomalous signal source, the continuous sampling data within the target time period is first extracted. The sampling time is used as the independent variable, and the response value as the dependent variable, constructing a time-value data point set. A fitting model is established using the linear least squares method. By calculating the covariance and variance of the time series and the data series, the fitting slope is obtained, which is the rate of change of the dependent variable caused by a unit change in the independent variable. The fitting intercept is then determined by combining this with the mean relationship, forming a complete univariate linear expression. Subsequently, based on the difference between the actual sampling point values ​​and the fitted values, the squared differences are calculated point by point and averaged to obtain the residual mean square value, which reflects the degree of deviation from the fit. For example, if the time series of an anomalous sensor is [1,2,3,4,5], and the response value is [5.0,7.0,6.5,9.0,8.2], the fitted slope might be 0.8, and the residual mean square value might be 0.36, indicating that the data fluctuates relatively little in the overall linear trend but still exhibits slight instability. The fitting slope and residual mean square value serve as the benchmark for subsequent multi-source sensing trend comparison, reflecting the change pattern of abnormal signal sources in the time series.

[0116] Calculate the mean difference between the fitted slope of all unlabeled signal sources and the baseline slope, as the fitted slope similarity; extract the maximum difference between the fitted residual of all unlabeled signal sources and the baseline residual, as the fitted residual range.

[0117] When calculating the mean difference between the fitted slopes of all sensors not labeled as anomalous signal sources and the baseline slope, firstly, linear least squares fitting is performed on each sensor not labeled as anomalous signal source to obtain the corresponding set of fitted slopes. Then, the difference between these fitted slopes and the fitted slopes of the sensors labeled as anomalous signal sources is calculated one by one. The absolute values ​​of these differences are taken and their arithmetic mean is calculated as the fitted slope similarity, used to measure the degree of similarity in the overall trend between the two slope sets. For example, if the baseline slope is 0.8, and the fitted slopes of the other three sensors are 0.7, 0.6, and 0.9, the absolute values ​​of the differences are 0.1, 0.2, and 0.1, respectively, and the average is 0.133, which is used as the fitted slope similarity. When calculating the fitted residual range, firstly, the mean square value of the residuals of each sensor not labeled as anomalous signal source is obtained. Then, the absolute value of the difference between these values ​​and the mean square value of the baseline residuals is calculated, and the maximum value is extracted as the fitted residual range. For example, if the baseline residual is 0.36, and the other residuals are 0.40, 0.55, and 0.33, then the absolute values ​​of the differences are 0.04, 0.19, and 0.03, with a maximum value of 0.19, which is the range of the fitting residuals. This range measures the maximum deviation of the fluctuation level of each sensor. These two indicators are used together to determine the similarity of the abnormal signal source to other sensors in terms of trend and fitting error.

[0118] S4. The results of the judgment of violent fluctuations are jointly mapped with the trend stability score to a credibility value, which serves as the basis for the reliability of participation of abnormal signal sources.

[0119] In this embodiment, S4 specifically refers to:

[0120] The result of the violent fluctuation judgment is mapped to a numerical order of magnitude. When the violent fluctuation judgment is true, it is set to the first preset value, and when the violent fluctuation judgment is false, it is set to the second preset value, thus forming a violent fluctuation numerical factor.

[0121] The judgment result of severe fluctuation is usually represented in Boolean logic form, that is, true when it is judged as severe fluctuation, and false otherwise. To use this result for subsequent confidence level calculations, the Boolean result needs to be mapped to a numerical form. Specifically, two different preset values ​​are set; for example, a true judgment result is mapped to the value 1, and a false judgment result is mapped to the value 0. In this way, the logical judgment result can be converted into a quantifiable numerical input. Taking a sensor data point as an example, if its deviation in amplitude, rate of change distribution, and duration of change all exceed the corresponding thresholds, it is judged as a severe fluctuation, and the corresponding mapped value is 1; if the judgment conditions are not met, the mapped value is 0. This numerical result is defined as a severe fluctuation numerical factor, which can be used as one of the input parameters for joint confidence level calculations, allowing the identification result of severe fluctuation to participate in the subsequent weighted calculation logic in numerical form. This numerical factor has clear discriminative power, which is beneficial for reflecting the severity of signal anomalies in subsequent scoring models.

[0122] The trend stability score is normalized by interval, and the score is converted into a trend stability factor with a unified standard to ensure its additive integration with the drastic fluctuation numerical factor on the same numerical scale.

[0123] Trend stability scores are typically derived from a weighted average of multiple feature indicators (consistency of change direction, similarity of fitting slope, and range of fitting residuals). The original values ​​may fall within a non-uniform numerical range, such as between 0 and 10. To enable the fusion calculation of trend stability scores and volatile numerical factors (e.g., set to 0 or 1) on the same numerical scale, interval normalization is required. Interval normalization can be achieved using a linear normalization formula: subtract the minimum score from the original score and divide by the difference between the maximum and minimum values, thus compressing the result to the standard interval of 0 to 1. For example, if the historical minimum score is 2, the maximum score is 8, and the current score is 5, the normalization result is (5-2) / (8-2) = 0.5, ultimately converting the trend stability score into a standardized trend stability factor. This factor possesses the same scaling characteristics as the volatile numerical factor, ensuring additivity and relative weighting in subsequent reliability fusion calculations. A unified numerical scale helps to build numerical consistency in reliability assessment models and improves the interpretability and controllability of credibility values.

[0124] Using the drastic fluctuation numerical factor and the trend stability factor as joint input variables, a weighted linear combination function is executed to output a confidence value. The confidence value serves as the reliability basis for the participation of the corresponding abnormal signal source in subsequent information fusion calculations.

[0125] The numerical factor of drastic fluctuation and the trend stability factor reflect the reliability evaluation basis of anomalous signal sources under abrupt change characteristics and trend backgrounds, respectively. Using these two factors as joint input variables, a reliability numerical model can be constructed through a weighted linear combination function, where each factor corresponds to a weight parameter to adjust its influence on the final reliability result. The weighted linear combination function can be expressed as: Reliability value = α × drastic fluctuation factor + β × trend stability factor, where α and β are preset weight coefficients, satisfying α + β = 1, used to control the relative contribution ratio of the two types of information. For example, when the drastic fluctuation factor is 1 (indicating strong fluctuations), the trend stability factor is 0.3 (indicating unstable trends), and α = 0.7 and β = 0.3 are set, then the reliability value = 0.7 × 1 + 0.3 × 0.3 = 0.79. This value can serve as the weighting basis for the reliability of the anomalous signal source data in subsequent multi-source information fusion calculations, realizing quantitative measurement and dynamic control of the signal source reliability. By fusing evaluation factors from two different sources through linear combination, the accuracy and robustness of anomaly identification decisions can be improved.

[0126] S5. Adjust the participation weight of abnormal signal sources in the fusion evaluation calculation based on the credibility value, generate structural damage evaluation results by weighted fusion of the evaluation parameters of each sensor, and complete the early warning output.

[0127] In this embodiment, S5 specifically refers to:

[0128] The participation weight of sensors marked as anomalous signal sources in the fusion evaluation calculation is adjusted based on the confidence level value. The participation weight of sensors not marked as anomalous signal sources is set to the full weight value, and the participation weight of sensors marked as anomalous signal sources is equal to the corresponding confidence level value.

[0129] This process controls the impact of anomalous signal sources on structural damage assessment results by dynamically adjusting the weights of sensors involved, aiming to improve the overall reliability of the assessment. In implementation, each sensor is first classified within the current assessment period to identify which sensors have been marked as anomalous signal sources. Sensors not marked as anomalous are assigned a fixed full weight value, such as 1.0, indicating that their data is fully accepted in the fusion calculation. For sensors marked as anomalous, the confidence value generated in the preceding steps is used as their weight in the fusion process. For example, if the confidence value of an anomalous sensor is 0.72, its assessment parameters will only participate in the weighted fusion of structural damage indicators with a proportion of 0.72. This mechanism ensures that the reference value of anomalous signals is preserved during information fusion while controlling the interference caused by their uncertainty, thus improving the accuracy and stability of structural damage identification.

[0130] In practice, all sensors are first numbered and a corresponding weight vector is established. Then, the list of anomaly markers and their corresponding confidence results for the current period are read. The sensor set is iterated sequentially, and unmarked sensors are directly assigned a value of 1.0. For marked sensors, the corresponding values ​​are extracted from the confidence mapping table and filled into the weight vector. After the weight vector is updated, it is multiplied by the current evaluation parameters of each sensor to generate a weighted evaluation value vector, which serves as the input parameter for subsequent structural damage index fusion calculation. Throughout the process, the confidence value and sensor status markers constitute the core basis for weight adjustment, supporting the system in maintaining the stability and reliability of the overall evaluation results when facing interference from local abnormal signal sources.

[0131] Extract the evaluation parameters of all sensors in the current evaluation period, multiply the evaluation parameters of each sensor with their corresponding participation weights to obtain the weighted evaluation values ​​of each sensor, and then normalize and aggregate all weighted evaluation values ​​to generate structural damage evaluation results.

[0132] In structural damage assessment, to ensure that the fusion result fully reflects the numerical contribution and confidence level of each sensor, a weighted fusion process is required for the assessment parameters and corresponding participation weights of all sensors. Specifically, the assessment parameter values ​​calculated for each sensor within the current assessment period are first extracted, and a set of assessment parameter vectors is constructed. Simultaneously, a participation weight vector of equal length is prepared, containing fixed values ​​for sensors with full weights and confidence values ​​for sensors with abnormal signal sources. By multiplying each assessment parameter by its corresponding weight, the weighted assessment value for each sensor is obtained. To ensure the final fusion result is consistent in numerical scale and comparable, weighted assessment values ​​need to be weighted and normalized. This is achieved by dividing the weighted sum of all weighted assessment values ​​by the total weight sum, thereby generating a standardized structural damage assessment result. This result can be used to determine the health status level of the target structure and drive subsequent early warning decisions.

[0133] In practice, two one-dimensional arrays of the same length are constructed to store the original evaluation parameters and their weights for each sensor in the current period. For example, if the parameter array is P1, P2, ..., Pn and the weight array is W1, W2, ..., Wn, then the corresponding weighted evaluation value array is P1×W1, P2×W2, ..., Pn×Wn. Then, the weighted evaluation value arrays are summed to obtain the total weighted evaluation value, and the weight arrays are summed again to obtain the total weight value. The final structural damage assessment result is equal to the total weighted evaluation value divided by the total weight value. For example, if the evaluation values ​​of the three sensors are [0.8, 0.6, 0.4], and the corresponding weights are [1.0, 0.85, 0.72], then the weighted value is [0.8, 0.51, 0.288], the weighted sum is 1.598, the total weight is 2.57, and the structural damage assessment result is 1.598 divided by 2.57, which is approximately 0.622. This mechanism ensures that the fusion results not only consider the evaluation values ​​themselves, but also incorporate the credibility control logic of the data sources.

[0134] The structural damage assessment result is numerically compared with the damage judgment threshold. When the structural damage assessment result is greater than the damage judgment threshold, a structural damage early warning signal is output. At the same time, the structural location information, damage level value and participation weight information within the current assessment period are output to complete the early warning output.

[0135] In structural health monitoring, to respond promptly to potential risks, it is necessary to compare the structural damage assessment results for the current period with a pre-set damage threshold. This threshold, determined based on historical data, structural material properties, environmental factors, and empirical models, represents the maximum acceptable damage tolerance of the structure. Once the structural damage assessment result exceeds this threshold, the structure is considered to be in a potentially hazardous state, and an early warning mechanism must be triggered immediately. At this time, key data related to the current assessment results must be output simultaneously, including the structural spatial location information corresponding to the assessment time point, indicating the specific damage area; the assessed damage level value, used to quantify the degree of risk; and the participation weight information of each sensor in this fusion, used to further assist in reliability tracing and subsequent analysis. This mechanism constitutes a complete closed loop for early warning output, providing both the judgment result and support for response decisions.

[0136] In this specific implementation, the calculated structural damage assessment result is first denoted as D, and the preset damage judgment threshold is T. When D is greater than T, the system calls the early warning submodule to trigger the generation of an alarm signal. Location information is traced back using the sensor deployment area index and timestamp, combined with the geographic identifier of the identified abnormal signal source, to output a precise structural damage area number. The damage level value can be mapped to a qualitative level (e.g., mild, moderate, severe) according to the segmentation rules of the assessment result within the 0-1 range. The weighting information consists of the weight values ​​of each sensor participating in the fusion within the current period, output in a structured array format, and can be used to assist in interface display or subsequent feedback learning and optimization. This mechanism ensures the accuracy, interpretability, and operability of the early warning response.

[0137] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions according to the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means (e.g., infrared, wireless, microwave, etc.). A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0138] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0139] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0140] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0141] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0142] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0143] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for damage assessment and early warning of ultra-high performance concrete masonry slabs based on multi-source sensor data, characterized in that, Specifically, the following steps are included: S1. Construct a time-series response trajectory set from the continuous data acquired by multiple sensors within a unified acquisition period, and extract the rate of change index of each sensor to identify high-fluctuation data. S2. Calculate the deviation value of change amplitude, the distribution index of change rate, and the duration index of change for the identified high-fluctuation data. Compare the deviation value of change amplitude with the deviation threshold of change amplitude, the distribution index of change rate with the distribution threshold of change rate, and the duration of change with the duration threshold of change. When all three comparison results are greater than the threshold, the high-fluctuation data is identified as a violent fluctuation, and the corresponding sensor is marked as an abnormal signal source. S3. Select the sensor data that is not marked as an abnormal signal source within the time period corresponding to the sensor marked as an abnormal signal source, construct the time-series response trajectory of each sensor within the time period and extract the direction of change. Compare the direction of change of each sensor not marked as an abnormal signal source with the direction of change of the abnormal signal source point by point to obtain the consistency of the direction of change. Perform linear least squares regression on the response trajectory of each sensor not marked as an abnormal signal source to obtain the fitting slope and fitting residual, and compare the fitting results with the fitting results of the abnormal signal source to obtain the similarity of the fitting slope and the range of the fitting residual. Assign weight coefficients according to the consistency of the direction of change, the similarity of the fitting slope, and the range of the fitting residual, and generate a trend stability score value according to the weighted average method to determine whether to perform weight suppression processing on the abnormal signal source. S4. Map the results of the violent fluctuation judgment to numerical magnitude to form a violent fluctuation numerical factor. Perform interval normalization on the trend stability score to form a trend stability factor. Use the violent fluctuation numerical factor and the trend stability factor as joint input variables to execute a weighted linear combination function to output a confidence value. The confidence value serves as the reliability basis for the participation of the corresponding abnormal signal source in subsequent information fusion calculation. S5. Adjust the participation weight of sensors marked as abnormal signal sources in the fusion evaluation calculation based on the confidence value. The participation weight of sensors not marked as abnormal signal sources is set to the full weight value. The participation weight of sensors marked as abnormal signal sources is equal to the corresponding confidence value. Extract the evaluation parameters of all sensors in the current evaluation period and multiply the evaluation parameters of each sensor with their corresponding participation weights to obtain the weighted evaluation value of each sensor. Normalize all weighted evaluation values ​​and aggregate them to generate structural damage evaluation results. Compare the structural damage evaluation results with the damage judgment threshold. When the structural damage evaluation result is greater than the damage judgment threshold, output a structural damage warning signal and complete the warning output.

2. The method for damage assessment and early warning of ultra-high performance concrete slabs based on multi-source sensor data according to claim 1, characterized in that, S2 specifically includes the following steps: S201. Calculate the difference between the maximum and minimum values ​​for each segment of highly fluctuating data, and compare it with the average difference of the corresponding data of the same type of sensor in the historical acquisition period to obtain the deviation value of the change range. S202. For each segment of highly fluctuating data, the numerical change interval between adjacent data points is statistically analyzed, a change rate distribution map is constructed, and the frequency concentration interval and the width of the change interval are extracted as change rate distribution indicators. S203. Calculate the length of time that each segment of identified high-fluctuation data remains above the rate of change judgment benchmark in a continuous sampling period, and use it as an indicator of the duration of change. S204. Compare the deviation value of the change amplitude with the change amplitude deviation threshold, compare the change rate distribution index with the change rate distribution threshold, and compare the change duration with the change duration threshold. When all three comparison results are greater than the threshold, the high fluctuation data is identified as a violent fluctuation, and the corresponding sensor is marked as an abnormal signal source.

3. The method for damage assessment and early warning of ultra-high performance concrete slabs based on multi-source sensor data according to claim 2, characterized in that, S201 specifically refers to: For each segment of highly volatile data, extract the maximum and minimum values ​​within the current acquisition period, and calculate the difference as the current change amplitude value; In the historical data of the same type of sensor, select multiple time periods with the same acquisition cycle length, calculate the difference between the maximum and minimum values ​​in each time period, and take the arithmetic mean of these differences as the average difference. Subtract the average difference from the current change value to obtain the change deviation value, which is used to determine subsequent drastic fluctuations.

4. The method for damage assessment and early warning of ultra-high performance concrete slabs based on multi-source sensor data according to claim 2, characterized in that, S202 specifically refers to: For each segment of highly volatile data, the numerical difference between every two adjacent data points is calculated sequentially, and the change interval sequence composed of all differences is statistically analyzed. The frequency statistics of the changing interval sequence are statistically analyzed and a histogram is plotted to generate a rate of change distribution map. In the rate of change distribution spectrum, the interval of continuous values ​​with the highest frequency is identified as the frequency concentration interval, and the difference between the minimum and maximum change values ​​in the distribution spectrum is calculated as the width of the change interval.

5. The method for damage assessment and early warning of ultra-high performance concrete slabs based on multi-source sensor data according to claim 2, characterized in that, S203 specifically refers to: Each segment of high-fluctuation data is compared point by point with the rate of change judgment benchmark, and the continuous sampling point sequence with values ​​greater than the rate of change judgment benchmark is marked. In each sequence of marked sampling points, the number of consecutive adjacent sampling points is counted, and this number is multiplied by the sampling interval to obtain the length of time in each sequence where the rate of change is greater than the threshold for determining the rate of change. The maximum value among all time periods is selected as the indicator of the duration of change in the currently identified high-volatility data.

6. The method for damage assessment and early warning of ultra-high performance concrete slabs based on multi-source sensor data according to claim 1, characterized in that, S3 specifically includes the following steps: S301. Select the sensor data that is not marked as an abnormal signal source within the time period corresponding to the sensor marked as an abnormal signal source, construct the time-series response trajectory of each sensor within the time period, and extract the change direction of each response trajectory. S302. Compare the direction of change of each sensor that is not marked as an abnormal signal source with the direction of change of the sensor that is marked as an abnormal signal source point by point, and count the percentage of data points with the same direction as the consistency of the direction of change. S303. Perform linear least squares regression on the response trajectories of all sensors that are not marked as abnormal signal sources, obtain the fitting slope and fitting residual, and then compare the fitting results with those of sensors marked as abnormal signal sources. Calculate the similarity of fitting slope and the range of fitting residual respectively. S304. Based on the consistency of the direction of change, the similarity of the fitting slope, and the range of the fitting residual, weight coefficients are assigned respectively, and a trend stability score is generated by weighted averaging to determine whether to perform weighted suppression processing on abnormal signal sources.

7. The method for damage assessment and early warning of ultra-high performance concrete slabs based on multi-source sensor data according to claim 6, characterized in that, S302 specifically refers to: Extract the sequence of changes in the direction of adjacent data points of the sensor marked as an abnormal signal source within the corresponding time period. The direction of change is determined by the sign of the difference between adjacent data points. Extract the change direction sequence of sensors that are not marked as abnormal signal sources within the same time period, and maintain a one-to-one correspondence with the time points of the direction sequence of abnormal signal sources; Compare the directions of each pair of corresponding data points in the two change direction sequences, count the number of data points with the same direction, divide the number by the total number of data points, calculate the direction consistency ratio, and use it as the change direction consistency index.

8. The method for damage assessment and early warning of ultra-high performance concrete slabs based on multi-source sensor data according to claim 6, characterized in that, S303 specifically refers to: For each sensor that was not marked as an anomalous signal source, the response trajectory within the target time period was fitted using the linear least squares method to obtain the corresponding fitting slope and residual mean square value. The same fitting process was performed on the sensor response trajectories marked as abnormal signal sources, and the corresponding fitting slope and residual mean square value were obtained as a comparison benchmark. The mean difference between the fitted slopes of all unmarked anomalous signal sources and the baseline slope is calculated as the fitted slope similarity. Extract the maximum difference between the fitting residuals of all unmarked signal sources and the baseline residuals, and use it as the fitting residual range.

9. The method for damage assessment and early warning of ultra-high performance concrete slabs based on multi-source sensor data according to claim 1, characterized in that, S4 specifically refers to: The result of the violent fluctuation judgment is mapped to a numerical order of magnitude. When the violent fluctuation judgment is true, it is set to the first preset value, and when the violent fluctuation judgment is false, it is set to the second preset value, thus forming a violent fluctuation numerical factor. The trend stability score is normalized by interval, and the score is converted into a trend stability factor with a unified standard to ensure its additive integration with the drastic fluctuation numerical factor on the same numerical scale. Using the drastic fluctuation numerical factor and the trend stability factor as joint input variables, a weighted linear combination function is executed to output a confidence value. The confidence value serves as the reliability basis for the participation of the corresponding abnormal signal source in subsequent information fusion calculations.

10. The method for damage assessment and early warning of ultra-high performance concrete slabs based on multi-source sensor data according to claim 1, characterized in that, S5 specifically refers to: The participation weight of sensors marked as anomalous signal sources in the fusion evaluation calculation is adjusted based on the confidence level value. The participation weight of sensors not marked as anomalous signal sources is set to the full weight value, and the participation weight of sensors marked as anomalous signal sources is equal to the corresponding confidence level value. Extract the evaluation parameters of all sensors in the current evaluation period, multiply the evaluation parameters of each sensor with their corresponding participation weights to obtain the weighted evaluation values ​​of each sensor, and then normalize and aggregate all weighted evaluation values ​​to generate structural damage evaluation results. The structural damage assessment result is numerically compared with the damage judgment threshold. When the structural damage assessment result is greater than the damage judgment threshold, a structural damage early warning signal is output. At the same time, the structural location information, damage level value and participation weight information within the current assessment period are output to complete the early warning output.

Images