A bim-based multi-bridge health state real-time monitoring and prediction method

Abstract

The application discloses a kind of based on BIM's multi-bridge health state real-time monitoring prediction method, it is related to intelligent monitoring technical field, the original data packet is written into distributed system in the application;Original data packet is carried out window synchronization, and missing value and space normalization processing are obtained Dimensionless modal response matrix;Modal response matrix is developed into one-dimensional column vector, whether there is abnormal change of bridge according to cosine similarity judgment;Decay factor is introduced to calculate window weight, and baseline is generated according to the modal response matrix of window weight to bridge normal window;Left singular matrix is obtained by SVD matrix singular value decomposition to baseline, and orthogonal projection operator is constructed, after real-time acquisition new sensing data of bridge sensor, residual column vector is calculated using orthogonal projection operator to the modal response matrix of new sensing data, drift entropy is calculated after purifying sequence monadic linear regression of pre-set time period, and bridge maintenance strategy is triggered using drift entropy.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology, specifically to a BIM-based method for real-time monitoring and prediction of the health status of multiple bridges. Background Technology

[0002] In recent years, there has been further promotion of the construction of structural health monitoring systems for long highway bridges. At the same time, the application of Building Information Modeling (BIM) technology in bridge engineering has become increasingly widespread. BIM technology can provide bridges with a three-dimensional visualization information management platform, digitally representing the bridge's geometric information, material properties, and component relationships. Combining BIM technology with structural health monitoring not only provides the interactive management platform required by the monitoring system but also significantly improves the management level of monitoring information.

[0003] Traditional technologies typically require data from a "healthy state" as a training benchmark, but scientifically defining this "healthy state" is itself a problem. For an old bridge that has been in service for many years, its "normal" state might simply be one with minor cracks and a predetermined deflection. Using an idealized "perfect model" as a benchmark could potentially misreport normal operating conditions as abnormalities. Summary of the Invention

[0004] The purpose of this invention is to provide a BIM-based method for real-time monitoring and prediction of the health status of multiple bridges, in order to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A BIM-based real-time monitoring and prediction method for the health status of multiple bridges, the method comprising the following steps: S100: Read sensor data from the data acquisition devices of each sensor on the bridge, construct raw data packets, and write the raw data packets into the distributed system; The specific steps for constructing the original data packet are as follows: S101. Sensors are installed on the bridge, including vibrating wire strain gauges, piezoelectric accelerometers, platinum resistance thermometers, etc. The sensor data is directly read from the data acquisition device of all sensors, including uncalibrated analog-to-digital conversion values ​​and raw voltage values. The read sensor data is encapsulated into a raw data packet in JSON format. The raw data packet includes a unique bridge code, sensor channel number, raw electrical signal value, nanosecond-level timestamp based on BeiDou module timing, ambient temperature of the sensor location, and self-test status bit of the data acquisition instrument. In the self-test status bit of the data acquisition instrument, 0 indicates normal, 1 indicates over-range, and 2 indicates communication abnormality. These bits are used to mark but not remove data. The raw data packets are partitioned by time and stored in a read-only directory of the distributed system.

[0006] It directly reads uncalibrated AD values ​​or raw voltage values ​​from the data acquisition instrument without any filtering or smoothing, avoiding the distortion or masking of the true structural response information caused by traditional preprocessing. The raw data packets are written to a read-only directory in a distributed file system, ensuring data immutability and high reliability, while also supporting horizontal scaling storage of massive amounts of data across multiple bridges.

[0007] S200. Set a fixed window to perform window synchronization on the original data packets, and perform missing value and spatial normalization processing to obtain a dimensionless modal response matrix. The specific steps for obtaining the dimensionless modal response matrix by performing missing value and spatial normalization processing are as follows: S201. Set a fixed window width L. The system background slides the window once according to a fixed period. Set the window number according to the window sequence. For the window number t, extract the raw data packets of all sensors in the window. Through the fixed window width and sliding mechanism, the asynchronously acquired multi-source sensor data are strictly aligned on the time axis, laying the foundation for subsequent cross-sensor collaborative analysis.

[0008] When the missing data in a single sensor within a window exceeds one-tenth of the total data of that sensor, all data of that sensor within this window is discarded; when the number of sensors retaining data in a window is less than n times N, the entire window is discarded; N represents the total number of sensors, and n represents the allowable multiple. S202, Construct an initial matrix M by arranging the retained sensor data in rows. t raw (N×L), where N×L represents a matrix with N rows and L columns; due to the significant differences in strain and temperature values, each row of the initial matrix is ​​zero-mean normalized, and the mean of the i-th row is calculated using the following formula: ; where μ i M represents the mean of the i-th row; t raw (i, j) represents the element in the i-th row and j-th column of the initial matrix; Calculate the standard deviation σ of the i-th row. i The initial matrix is ​​spatially normalized using the mean and standard deviation, as shown in the following formula: ; Among them, M t (i, j) represents the element in the normalized modal response matrix at row i and column j, which ultimately yields the modal response matrix M. t .

[0009] Normalize the initial matrix, and map data of different physical dimensions such as strain, temperature, and acceleration uniformly to a dimensionless space, so that all subsequent linear algebra operations are not dominated by differences in numerical magnitudes, and truly reflect the morphology rather than the amplitude of the structural response.

[0010] S300. Unfold the modal response matrix into a one-dimensional column vector, then calculate the cosine similarity of the one-dimensional column vectors of adjacent windows respectively, and judge whether there is an abnormal change of the bridge according to the cosine similarity; The specific steps for judging whether there is an abnormality in the bridge according to the cosine similarity are as follows: S301. Unfold the modal response matrix row by row to form a one-dimensional column vector v t , for two adjacent windows t and t+1, calculate the cosine similarity of the one-dimensional column vectors, the formula is: ; where the numerator is the dot product, and the denominator is the product of the Euclidean norms; v t+1 represents the one-dimensional column vector of window t+1, and S(t) represents the cosine similarity of the one-dimensional column vectors of two adjacent windows t and t+1; The cosine similarity measures the direction consistency between the high-dimensional vectors of adjacent windows, rather than the numerical difference of a single measuring point; it can capture the overall change of the joint response of all sensors of the bridge, and reflect the structural state more comprehensively than the single-point threshold method.

[0011] S302. Preset a difference threshold Sy. When S(t) < Sy, it is judged that there is a significant change in the bridge; when S(t) ≥ Sy, it is judged that the bridge is in a stable state; the difference threshold is preset by the staff according to experience, and is set to 0.85 through testing in this application; After it is judged that there is a significant change in the bridge, the damage warning will not be triggered immediately, and the daily drift entropy in S600 will be further queried in linkage, and the fallback period is set to TL, which is set by experts, and TL=3 days is selected in this application; when the drift entropy falls back to the level before this judgment within the fallback period after the end of this judgment, it is judged that the change of the bridge this time is a reversible environmental event; when the drift entropy does not fall back to the level before this judgment within the fallback period, it is judged as a real damage event. The level before this judgment refers to the drift entropy value before the significant change of the bridge is judged this time. Through the above steps, the industry pain point of the traditional system of "reporting every change, with a high false alarm rate" is effectively solved.

[0012] S400. Generate cosine similarity curves of all windows based on the time series, extract the normal window set of the bridge, introduce a attenuation factor to calculate the window weight, and generate a baseline by weighting the modal response matrix of the normal bridge windows according to the window weight; The specific steps of generating the baseline by weighting the modal response matrix of the normal bridge windows according to the window weight are as follows: S401. Generate cosine similarity curves for all windows based on the time series, find K consecutive window times that are all judged to be in a stable state, and record the set of window indices {t1, t2, t3, ..., tm} corresponding to the found window times; for the modal response matrix in the set of window indices, calculate the window weight based on the time difference between different windows and the current time and introduce a decay factor, the formula is: ; Among them, w r T represents the weight of the tr-th window, λ represents the decay factor, and T current T represents the current time. tr This represents the time of the tr-th window; r∈[1,m]; An attenuation factor is introduced so that the closer the stable window is to the current moment, the greater its contribution to the baseline; this allows the baseline to slowly track the natural aging process of the bridge and avoids the baseline from deviating from the actual state of the structure due to the use of outdated windows.

[0013] S402. The modal response matrix of the bridge's normal windows is weighted according to the window weights, using the following formula: ; The matrix operations described above form the baseline after element-wise operations. B represents the baseline of the bridge in its current state and is an N×L matrix.

[0014] Traditional methods require "clean data" from the initial stage of bridge construction as a health baseline, but existing old bridges often lack such data. This solution finds a continuous stable window and "self-emerges" from the current monitoring data to identify the baseline representing the current most stable state of the structure, thus completely solving the baseline definition problem for "bridges without maps" and "bridges in service with defects."

[0015] S500: Perform SVD matrix singular value decomposition on the baseline to obtain the left singular matrix, construct an orthogonal projection operator, and after real-time acquisition of new sensing data from bridge sensors, use the orthogonal projection operator to calculate the residual column vector of the modal response matrix of the new sensing data, and use the residual column vector to calculate the anomaly index of the real-time window for judgment and trigger modal change warning. The specific steps for using the residual column vector to calculate the anomaly index of the real-time window for judgment and triggering modal change warning are as follows: S501. Perform SVD (Singular Value Decomposition) on the baseline to obtain the left singular matrix U, with dimensions N×R, where R represents the matrix rank; construct the orthogonal projection operator using the left singular matrix, with the formula: P=U×U TP represents the orthogonal projection matrix as the orthogonal projection operator, with dimensions N×N. By performing SVD decomposition on the baseline and constructing the orthogonal projection operator, the new response vector is projected onto the subspace spanned by the baseline. The energy magnitude of the residual column vector directly reflects the new components in the current response that do not belong to the baseline mode. S502. After acquiring new sensing data from the bridge sensors in real time within the current window, take the j-th column of the modal response matrix of the new sensing data to form a column vector c of all sensor values ​​at the corresponding time. j The dimension is N×1; the residual column vector is calculated using the orthogonal projection operator and column vectors, with the formula: e j =c j -P×c j ;e j Represents the residual column vector; The anomaly index for the current window is calculated using the residual column vectors of all columns, using the following formula: ; Where Ω(t) represents the anomaly index of the current window, This represents the squared L2 norm of the residual column vector; When Ω(t) is greater than three times the average of the historical window anomaly index, a modal mutation warning is triggered, indicating that the bridge has developed a new abnormal mode. Since the baseline itself comes from "stable but defective" data, and the anomaly index is calculated based on relative residuals rather than absolute physical quantities, the slow drift of the sensor itself will not cause a significant change in Ω(t). Ω(t) will only spike when the structural response mode undergoes a fundamental change, greatly reducing the false alarm rate.

[0016] S600. Construct a daily average sequence, use ambient temperature to purify the daily average sequence to generate a purified sequence, perform univariate linear regression on the purified sequence for a preset time period to calculate the drift entropy, and use the drift entropy to trigger bridge maintenance strategies.

[0017] The specific steps for triggering bridge maintenance strategies using drift entropy are as follows: S601, The daily average series represents the median of the anomaly index of all retained windows at the end of each day, denoted as Ω. d The ambient temperature of the sensor sites in the original data packet was extracted, and the abnormal index values ​​with daily average temperature fluctuations within the range of ±Te were selected to obtain the purification sequence Ω. d filtered The term Te represents the purification threshold, which is determined by staff. In this application, the value can be 2℃. Extract the daily purification sequence from history, and perform a univariate linear regression with the number of days as the independent variable and the anomaly index as the dependent variable: ; in, This represents the average number of days. The value represents the average anomaly index; Q represents the total number of historical days extracted; β represents the cumulative drift slope. S602. Calculate the drift entropy using the cumulative drift slope. The formula is: ; Here, H represents drift entropy; drift entropy quantifies the irreversible migration rate of the baseline into a percentage indicator, making the changes in the "health" of the bridge numerically trackable, providing a clear quantitative basis for maintenance decisions, rather than relying solely on experience.

[0018] The preset migration threshold is Hy. When the drift entropy continuously exceeds the fallback period and is greater than the migration threshold, the system determines that the bridge baseline has permanently migrated, triggering advanced expert maintenance, which refers to a detailed manual inspection. When H continuously exceeds the migration threshold, the system determines that the bridge baseline has "permanently migrated," triggering a detailed manual inspection; this signifies the completion of the closed loop from "data monitoring" to "maintenance action," realizing the ultimate value of the monitoring system.

[0019] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention completely abandons the reliance of traditional methods on theoretical values ​​of bridge design or "pure data" at the completion period. Through the self-emergent principle of "stability as baseline", it is applicable to any bridge in service and has extremely strong engineering universality.

[0020] 2. By using the dual criteria of "short-term similarity drop + long-term drift entropy verification", it can accurately distinguish between reversible environmental events such as typhoons and temperature changes and permanent structural damage, thus solving the industry problems of high false alarm rate and waste of maintenance resources in traditional systems.

[0021] 3. The anomaly index and drift entropy output by this invention can be associated and mapped with the component codes in the BIM model, directly driving the color warning and status update in the BIM model, providing a high-quality data foundation for "digital twin". Attached Figure Description

[0022] Figure 1 This is a schematic diagram illustrating the steps of a BIM-based real-time monitoring and prediction method for the health status of multiple bridges according to the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example: Figure 1 As shown, the present invention provides a technical solution. This embodiment takes a cross-river highway bridge in a certain city, hereinafter referred to as "Bridge A", as an example. The bridge is a double-tower cable-stayed bridge with a main span of 500m. It has been in service for 15 years and no complete as-built design BIM model has been retained. A BIM-based real-time monitoring and prediction method for the health status of multiple bridges, the method comprising the following steps: S100: Read sensor data from the data acquisition devices of each sensor on the bridge, construct raw data packets, and write the raw data packets into the distributed system; The specific steps for constructing the original data packet are as follows: S101. Sensors are installed on the bridge, including vibrating wire strain gauges, piezoelectric accelerometers, platinum resistance thermometers, etc. Sensing data is directly read from the data acquisition device of all sensors. The sensing data includes uncalibrated analog-to-digital conversion values ​​and raw voltage values. A total of N=120 sensor channels are arranged at the mid-span, 1 / 4 span, 3 / 4 span of the main beam of Bridge A and the key sections of the two bridge towers. The read sensor data is encapsulated into a raw data packet in JSON format. The raw data packet includes a unique bridge code, sensor channel number, raw electrical signal value, nanosecond-level timestamp based on BeiDou module timing, ambient temperature of the sensor location, and self-test status bit of the data acquisition instrument. In the self-test status bit of the data acquisition instrument, 0 indicates normal, 1 indicates over-range, and 2 indicates communication abnormality. These bits are used to mark but not remove data. The raw data packets are partitioned by time and stored in a read-only directory of the distributed system.

[0025] It directly reads uncalibrated AD values ​​or raw voltage values ​​from the data acquisition instrument without any filtering or smoothing, avoiding the distortion or masking of the true structural response information caused by traditional preprocessing. The raw data packets are written to a read-only directory in a distributed file system, ensuring data immutability and high reliability, while also supporting horizontal scaling storage of massive amounts of data across multiple bridges.

[0026] The data acquisition instrument directly reads the uncalibrated raw voltage signal and encapsulates it into a raw data packet in JSON format. Each record contains: The bridge's unique code is: "Bridge A", sensor channel number "STRAIN_01"-"TEMP_40", and the original electrical signal value (for example, the strain gauge output is 1200μV). Based on the nanosecond-level timestamp 2026-08-13T10:00:00.123456789Z of Beidou module timing, the ambient temperature of the sensor location, and the self-test status bit 0 of the data acquisition instrument; S200. Set a fixed window to perform window synchronization on the original data packets, and perform missing value and spatial normalization processing to obtain a dimensionless modal response matrix. The specific steps for obtaining the dimensionless modal response matrix by performing missing value and spatial normalization processing are as follows: S201. Set a fixed window width L, which can be L=3600s in this application, and the sampling frequency is 1Hz. The system background slides the window once according to a fixed period and sets the window number according to the window sequence. For the window number t, the original data packets of all sensors in the window are extracted. Through the fixed window width and sliding mechanism, the asynchronously acquired multi-source sensor data are strictly aligned on the time axis, laying the foundation for subsequent cross-sensor collaborative analysis.

[0027] When the missing data in a single sensor within a window exceeds one-tenth of the total data of that sensor, all data of that sensor within this window is discarded; when the number of sensors retaining data within a window is less than n times N, the entire window is discarded; N represents the total number of sensors, and n represents the allowable multiple. In this application, to ensure the integrity of sensor data within the window, n can be 0.8. A window of t=100 is selected from 10:00 to 11:00 on August 13, 2026. The raw data from all 120 sensors within this window form the initial matrix M. t raw (N×L), with dimensions of 120×3600. If a sensor has more than 360 missing data points within this window, the entire row for that sensor is removed. All sensors within this window are valid, and all 120 rows are retained.

[0028] S202, Construct an initial matrix M by arranging the retained sensor data in rows. t raw (N×L), where N×L represents a matrix with N rows and L columns; due to the significant differences in strain and temperature values, each row of the initial matrix is ​​zero-mean normalized, and the mean of the i-th row is calculated using the following formula: ; where μ i M represents the mean of the i-th row; t raw (i, j) represents the element in the i-th row and j-th column of the initial matrix; For each row, zero mean is applied. Taking the first strain sensor as an example, its 3600 sampled values ​​have a mean of 102.5 μV and a standard deviation of 15.2 μV. Then, each element in that row is transformed according to the formula.

[0029] Calculate the standard deviation σ of the i-th row. i The initial matrix is ​​spatially normalized using the mean and standard deviation, as shown in the following formula: ; where M t (i, j) represents the element at the i-th row and j-th column of the normalized modal response matrix, and the modal response matrix M is finally obtained t .

[0030] The initial matrix is normalized, and data of different physical dimensions such as strain, temperature and acceleration are uniformly mapped to a dimensionless space, so that all subsequent linear algebra operations are not dominated by numerical magnitude differences, and the morphology of the structural response rather than the amplitude is truly reflected.

[0031] S300. Unfolding the modal response matrix into a one-dimensional column vector, then calculating the cosine similarity of the one-dimensional column vectors of adjacent windows respectively, and judging whether there is an abnormal change of the bridge according to the cosine similarity; The specific steps of judging whether there is an abnormality in the bridge according to the cosine similarity are as follows: S301. Unfolding the modal response matrix row by row to form a one-dimensional column vector v t , for two adjacent windows t and t+1, calculating the cosine similarity of the one-dimensional column vectors, the formula is: ; where the numerator is a dot product, and the denominator is the product of Euclidean norms; v t+1 represents the one-dimensional column vector of window t+1, and S(t) represents the cosine similarity of the one-dimensional column vectors of two adjacent windows t and t+1; The cosine similarity is calculated for adjacent windows t=100 and t=101, and it is assumed that the calculated S(100)=0.978. The preset difference threshold Sy=0.85, and since 0.978≥0.85, it is judged that the bridge is in a stable state in this period.

[0032] The cosine similarity measures the direction consistency between the high-dimensional vectors of adjacent windows, rather than the numerical difference of a single measurement point; it can capture the overall change of the joint response of all sensors of the bridge, and reflect the structural state more comprehensively than the single-point threshold method.

[0033] S302. Presetting a difference threshold Sy, when S(t) < Sy, judging that there is a significant change in the bridge; when S(t) ≥ Sy, judging that the bridge is in a stable state; the difference threshold is preset by the staff according to experience, and is set to 0.85 through testing in the present application; When a significant change in the bridge is detected, a damage warning is not immediately triggered. Instead, the system further queries the daily drift entropy in the S600 database, setting a fallback period of TL (3 days) to be determined by experts. If the drift entropy falls back to the pre-detection level within the fallback period after the current detection, the bridge change is considered a reversible environmental event. If the drift entropy does not fall back to the pre-detection level within the fallback period, it is considered a genuine damage event. The pre-detection level refers to the drift entropy value before the significant change in the bridge was detected. These steps effectively address the industry pain point of traditional systems, which often report changes immediately and have a high false alarm rate.

[0034] Suppose a typhoon hits a certain day, and the similarity between windows t=250 and t=251 drops sharply to S(250)=0.72, which is below the threshold, triggering a "significant change" flag. However, the system does not immediately issue an alarm, but instead links to the daily drift entropy in subsequent steps S600. If the daily drift entropy H=0.02% / day, and the drift entropy recovers to the pre-typhoon level within the following 3 days, then the change is determined to be a "reversible environmental event," and the system only records the event log without triggering a damage warning.

[0035] S400. Generate cosine similarity curves for all windows based on time series, extract the set of normal bridge windows, introduce a decay factor to calculate window weights, and generate a baseline by weighting the modal response matrix of the normal bridge windows according to the window weights. The specific steps for generating the baseline by weighting the modal response matrix of the normal bridge window according to the window weight are as follows: S401. Generate cosine similarity curves for all windows based on the time series, find K consecutive window times that are all judged to be in a stable state, and record the set of window indices {t1, t2, t3, ..., tm} corresponding to the found window times; for the modal response matrix in the set of window indices, calculate the window weight based on the time difference between different windows and the current time and introduce a decay factor, the formula is: ; Among them, w r T represents the weight of the tr-th window, λ represents the decay factor, and T current T represents the current time. tr This represents the time of the tr-th window; r∈[1,m]; An attenuation factor is introduced so that the closer the stable window is to the current moment, the greater its contribution to the baseline; this allows the baseline to slowly track the natural aging process of the bridge and avoids the baseline from deviating from the actual state of the structure due to the use of outdated windows.

[0036] S402. The modal response matrix of the bridge's normal windows is weighted according to the window weights, using the following formula: ; The matrix operations described above form the baseline after element-wise operations. B represents the baseline of the bridge in its current state and is an N×L matrix.

[0037] Traditional methods require "clean data" from the initial stage of bridge construction as a health baseline, but existing old bridges often lack such data. This solution finds a continuous stable window and "self-emerges" from the current monitoring data to identify the baseline representing the current most stable state of the structure, thus completely solving the baseline definition problem for "bridges without maps" and "bridges in service with defects."

[0038] After the system has been running for 7 days, a large amount of window similarity data has been accumulated. The S(t) curves of all windows are scanned to find time intervals with K=6 consecutive windows where each S(t) is ≥0.95. Assume the following stable window index set is found: {150, 151, 152, 153, 154, 155}.

[0039] Calculate the weights of these windows; assuming a decay factor λ = 0.01, and the current time T... current =200 hours, end time T of each window tr The timeframes are 150 to 155 hours respectively. For example, the weight of window 150 is approximately 0.6065, and the weight of window 155 is 0.6376. The weighted average yields the baseline matrix B (120×3600).

[0040] S500: Perform SVD matrix singular value decomposition on the baseline to obtain the left singular matrix, construct an orthogonal projection operator, and after real-time acquisition of new sensing data from bridge sensors, use the orthogonal projection operator to calculate the residual column vector of the modal response matrix of the new sensing data, and use the residual column vector to calculate the anomaly index of the real-time window for judgment and trigger modal change warning. The specific steps for using the residual column vector to calculate the anomaly index of the real-time window for judgment and triggering modal change warning are as follows: S501. Perform SVD (Singular Value Decomposition) on the baseline to obtain the left singular matrix U, with dimensions N×R, where R represents the matrix rank; construct the orthogonal projection operator using the left singular matrix, with the formula: P=U×U T P represents the orthogonal projection matrix as the orthogonal projection operator, with dimensions N×N. By performing SVD decomposition on the baseline and constructing the orthogonal projection operator, the new response vector is projected onto the subspace spanned by the baseline. The energy magnitude of the residual column vector directly reflects the new components in the current response that do not belong to the baseline mode. S502. After acquiring new sensing data from the bridge sensors in real time within the current window, take the j-th column of the modal response matrix of the new sensing data to form a column vector c of all sensor values ​​at the corresponding time. j The dimension is N×1; the residual column vector is calculated using the orthogonal projection operator and column vectors, with the formula: ej =c j -P×c j ;e j Represents the residual column vector; the system updates the U matrix offline once a day at 2:00 AM.

[0041] The anomaly index for the current window is calculated using the residual column vectors of all columns, using the following formula: ; Where Ω(t) represents the anomaly index of the current window, This represents the squared L2 norm of the residual column vector; When Ω(t) is greater than three times the average of the historical window anomaly index, a modal mutation warning is triggered, indicating that the bridge has developed a new abnormal mode. Since the baseline itself comes from "stable but defective" data, and the anomaly index is calculated based on relative residuals rather than absolute physical quantities, the slow drift of the sensor itself will not cause a significant change in Ω(t). Ω(t) will only spike when the structural response mode undergoes a fundamental change, greatly reducing the false alarm rate.

[0042] During real-time monitoring, for a new window, take the average column vector of its 3600 columns; Assuming the historical window anomaly index average is 0.12, and the current Ω(201) = 0.08, which is less than 3 times the average, no warning is triggered.

[0043] If structural damage occurs at a certain moment, Ω(t) suddenly increases to 0.55, exceeding 3 × 0.12 = 0.36, the system triggers a "modal mutation warning", indicating that the A-bridge may have a new abnormal mode, which requires close attention.

[0044] S600. Construct a daily average sequence, use ambient temperature to purify the daily average sequence to generate a purified sequence, perform univariate linear regression on the purified sequence for a preset time period to calculate the drift entropy, and use the drift entropy to trigger bridge maintenance strategies.

[0045] The specific steps for triggering bridge maintenance strategies using drift entropy are as follows: S601, The daily average series represents the median of the anomaly index of all retained windows at the end of each day, denoted as Ω. d The ambient temperature of the sensor sites in the original data packet was extracted, and the abnormal index values ​​with daily average temperature fluctuations within the range of ±Te were selected to obtain the purification sequence Ω. d filtered The term Te represents the purification threshold, which is determined by staff. In this application, the value can be 2℃. Extract the daily purification sequence from history, and perform a univariate linear regression with the number of days as the independent variable and the anomaly index as the dependent variable: ; in, This represents the average number of days. The value represents the average anomaly index; Q represents the total number of historical days extracted; β represents the cumulative drift slope. S602. Calculate the drift entropy using the cumulative drift slope. The formula is: ; Here, H represents drift entropy; drift entropy quantifies the irreversible migration rate of the baseline into a percentage indicator, making the changes in the "health" of the bridge numerically trackable, providing a clear quantitative basis for maintenance decisions, rather than relying solely on experience.

[0046] The preset migration threshold is Hy. When the drift entropy continuously exceeds the fallback period and is greater than the migration threshold, the system determines that the bridge baseline has permanently migrated, triggering advanced expert maintenance, which refers to a detailed manual inspection. When H continuously exceeds the migration threshold, the system determines that the bridge baseline has "permanently migrated," triggering a detailed manual inspection; this signifies the completion of the closed loop from "data monitoring" to "maintenance action," realizing the ultimate value of the monitoring system.

[0047] Assuming that the average daily temperature was within the range of 20℃±2℃ for 20 out of the past 30 days, the corresponding Ω d The values ​​form a purification sequence; assuming the calculated β=0.002 and the mean of this sequence is 0.12, the drift entropy is 1.67% / day; if the preset migration threshold Hy=0.5% / day, H is much larger than Hy, and exceeds the threshold for 5 consecutive days, exceeding the fallback period TL=3 days, the system determines that the A-bridge baseline has permanently migrated, triggering an "Expert Advanced Maintenance" work order, requiring a detailed manual investigation.

[0048] The anomaly index Ω(t), drift entropy H, and warning level output by this method are pushed to the BIM management platform in real time through a standardized data interface; the BIM model assigns a unique code to each component of Bridge A; the platform associates the Ω(t) value according to the component location, and at the same time, the BIM model can dynamically display the similarity curve of historical windows and the anomaly index trend chart.

[0049] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A BIM-based real-time monitoring and prediction method for the health status of multiple bridges, characterized in that: The method comprises the following steps: S100. Read sensing data from the data acquisition instrument of each sensor of the bridge, construct an original data packet, and write the original data packet into the distributed system; S200. Set a fixed window to perform window synchronization on the original data packet, and perform missing value processing and spatial normalization processing to obtain a dimensionless modal response matrix; S300. Expand the modal response matrix into a one-dimensional column vector, then calculate the cosine similarity of the one-dimensional column vectors of adjacent windows respectively, and judge whether there is an abnormal change of the bridge according to the cosine similarity; S400. Generate a cosine similarity curve of all windows based on the time series, extract a normal window set of the bridge, introduce an attenuation factor to calculate window weights, and generate a baseline by weighting the modal response matrices of the normal windows of the bridge according to the window weights; S500. Perform SVD matrix singular value decomposition on the baseline to obtain a left singular matrix, and construct an orthogonal projection operator. After new sensing data of the bridge sensors is collected in real time, calculate a residual column vector for the modal response matrix of the new sensing data by using the orthogonal projection operator, calculate the anomaly index of the real-time window by using the residual column vector for judgment and trigger a modal mutation early warning; S600. Construct a daily average sequence, purify the daily average sequence by using the ambient temperature to generate a purified sequence, calculate drift entropy after performing unary linear regression on the purified sequence in a preset time period, and trigger a bridge maintenance strategy by using the drift entropy.

2. The method for real-time monitoring and prediction of the health status of multiple bridges based on BIM according to claim 1, characterized in that: In S100, writing the original data packet into the distributed system is specifically: Sensors are arranged on the bridge, and sensing data is directly read from the data acquisition instruments of all the sensors, wherein the sensing data comprises uncalibrated analog-to-digital conversion values and original voltage values; The read sensing data is encapsulated into an original data packet in JSON format, and the original data packet includes a unique bridge code, a sensor channel number, an original electrical signal value, a nanosecond-level timestamp timed by a Beidou module, the ambient temperature of the sensor position, and a self-check status bit of the acquisition instrument; in the self-check status bit of the acquisition instrument, 0 represents normal, 1 represents over-range, and 2 represents communication abnormality, which is used for marking but not eliminating data; the original data packets are written into the read-only directory of the distributed system by time partition for storage.

3. The method for real-time monitoring and prediction of the health status of multiple bridges based on BIM according to claim 1, characterized in that: In S200, the window synchronization and missing value processing are specifically: A fixed window width L is set, the system background slides the window once according to a fixed period, window serial numbers are set according to the window sequence, and for the window serial number t, the original data packets of all sensors in the window are extracted; When the missing data of a single sensor in the window exceeds one tenth of the total data of the sensor, all data of the sensor in this window is discarded; when the number of sensors with reserved data in the window is less than n times of N, the entire window is discarded; N represents the total number of sensors, and n represents the allowable multiple.

4. The method for real-time monitoring and prediction of the health status of multiple bridges based on BIM according to claim 3, characterized in that: In S200, the spatial normalization processing to obtain the dimensionless modal response matrix is specifically: The retained sensor data will be arranged in rows to construct an initial matrix M. t raw (N×L), where N×L represents a matrix with N rows and L columns; zero-mean is applied to each row of the initial matrix, and the mean and standard deviation σ of the i-th row are calculated. i The initial matrix is ​​spatially normalized using the mean and standard deviation to obtain the modal response matrix M. t .

5. The method for real-time monitoring and prediction of the health status of multiple bridges based on BIM according to claim 1, characterized in that: In S300, judging whether there is an abnormal change of the bridge according to the cosine similarity is specifically: Expand the modal response matrix row-wise to form a one-dimensional column vector v. t For two adjacent windows t and t+1, calculate the cosine similarity S(t) of the one-dimensional column vectors. Preset a difference threshold Sy, when S(t) < Sy, it is judged that there is a significant change in the bridge; when S(t) ≥ Sy, it is judged that the bridge is in a stable state; the difference threshold is preset by the staff according to experience; When a significant change is detected in the bridge, a damage warning is not immediately triggered. Instead, the drift entropy of the day in the S600 is queried, and a fallback period of TL is set, which is determined by an expert. If the drift entropy falls back to the level before the current judgment within the fallback period after the current judgment ends, the bridge change is determined to be a reversible environmental event. If the drift entropy does not fall back to the level before the current judgment within the fallback period, it is determined to be a real damage event.

6. The method for real-time monitoring and prediction of the health status of multiple bridges based on BIM according to claim 4, characterized in that: The baseline generation in S400, which weights the modal response matrix of the bridge's normal window according to the window weight, is specifically as follows: Based on the time series, generate cosine similarity curves for all windows, find K consecutive window times that are all judged to be in a stable state, and record the set of window indices {t1, t2, t3, ..., tm} corresponding to the found window times; for the modal response matrix in the set of window indices, calculate the window weight based on the time difference between different windows and the current time and introduce a decay factor; the baseline B is obtained by weighting the modal response matrix of the normal window of the bridge according to the window weight, and B is an N×L matrix.

7. The method for real-time monitoring and prediction of the health status of multiple bridges based on BIM according to claim 6, characterized in that: The S500 mechanism utilizes residual column vectors to calculate the anomaly index of the real-time window for judgment and triggers modal change warnings as follows: Perform SVD matrix singular value decomposition on the baseline to obtain the left singular matrix U, with dimensions N×R, where R represents the matrix rank; construct the orthogonal projection operator P using the left singular matrix, where P represents the orthogonal projection matrix as the orthogonal projection operator, with dimensions N×N; After acquiring new sensing data from the bridge sensors in real time within the current window, the column vector c is constructed by taking all sensor values ​​at the corresponding time from the j-th column of the modal response matrix of the new sensing data. j The dimension is N×1; the residual column vector e is calculated using the orthogonal projection operator and column vectors. j The anomaly index of the current window is calculated using the residual column vectors of all columns. When Ω(t) is greater than 3 times the average anomaly index of the historical windows, a modal mutation warning is triggered, indicating that a new anomaly mode has appeared on the bridge.

8. The method for real-time monitoring and prediction of the health status of multiple bridges based on BIM according to claim 1, characterized in that: The bridge maintenance strategy triggered by drift entropy in S600 is specifically as follows: The daily average series represents the median of the anomaly index of all retained windows at the end of each day, denoted as Ω. d The ambient temperature of the sensor sites in the original data packet was extracted, and the abnormal index values ​​with daily average temperature fluctuations within the range of ±Te were selected to obtain the purification sequence Ω. d filtered The term Te represents the purification threshold, which is determined by staff. Extract the purification sequence of each day in history, use the number of days as the independent variable and the anomaly index as the dependent variable, and perform univariate linear regression to obtain the cumulative drift slope β; The drift entropy is calculated using the cumulative drift slope. The preset migration threshold is Hy. When the drift entropy continuously exceeds the fallback period and is greater than the migration threshold, the system determines that the bridge baseline has permanently migrated and triggers expert advanced maintenance, which means a detailed manual inspection.

Images