Building structure health monitoring and early warning system based on internet of things

Abstract

This invention relates to the field of building health management technology and discloses an IoT-based building structural health monitoring and early warning system, comprising: collecting environmental and acceleration signals and uploading them to the blockchain for normalization; extracting parameters representing the damp heat accumulation in beam sockets to characterize the hazards of moisture retention; extracting the proportion of local slip energy consumption by restoring displacement and filtering out overall translational interference; correlating the temporal evolution of the above two to find the response time difference and constructing a slip hysteresis index; fusing this hysteresis index with interface stiffness attenuation parameters to deduce the continuous degradation risk of deviating from historical benchmarks; finally, deduce the global propagation risk and adaptively divide the early warning levels, output resource scheduling instructions, and update the benchmark based on on-site verification. This invention avoids false alarms from empirical thresholds and discovers early degradation of hidden beam socket connections.

Description

Technical Field

[0001] This invention relates to the field of building health management technology, and more specifically, to an Internet of Things-based building structural health monitoring and early warning system. Background Technology

[0002] In historical masonry and old brick-and-wood houses, the ends of wooden floor beams are typically embedded into the beam recesses of the thick masonry walls using an inlay technique, relying mainly on end bearing, contact friction, and local interlocking to transfer loads. In recent years, IoT continuous sensing technology has gradually replaced manual inspections, using multi-source monitoring parameters (such as temperature, humidity, crack displacement, acceleration, etc.) to assess the real-time safety status of such buildings.

[0003] However, existing IoT monitoring solutions have significant cognitive and analytical limitations when dealing with such embedded nodes. On the one hand, traditional data processing paradigms tend to treat fluctuations in ambient temperature and humidity as external disturbances, attempting to eliminate them through conventional filtering methods to obtain a pure mechanical response. However, for the aforementioned embedded wooden beam ends, deep masonry enclosure, ventilation obstruction, and repeated internal and external insulation modifications make it extremely easy for a humid and hot microenvironment to form inside the beam socket. Long-term moisture retention not only exacerbates the decline in the transverse grain bearing capacity of the wood but also directly drives the loss of frictional interlocking ability. In this specific scenario, temperature and humidity are not simply environmental noise but rather the core driving variables that induce weakening of node connections.

[0004] On the other hand, when early degradation occurs at such hidden nodes, visible cracks in the external walls often haven't yet expanded significantly, and the overall macroscopic modal frequency shift of the building is easily masked by seasonal temperature effects. Existing early warning systems mostly rely on global frequency shifts or local vibration extremes to trigger alarms, easily mistaking local relative slippage caused by beam-end friction degradation for normal elastic expansion and contraction due to environmental disturbances. Furthermore, the use of a uniform static threshold logic fails to consider individual differences in material degradation timing and load transfer paths among different old buildings, leading to frequent large-scale false alarms during rainy and humid seasons, making it difficult to achieve early identification and closed-loop management of truly hidden risks. Summary of the Invention

[0005] This invention provides an Internet of Things-based building structure health monitoring and early warning system, which solves the technical problems mentioned in the background art.

[0006] This invention provides an IoT-based building structure health monitoring and early warning system, comprising multi-source sensors, a trusted data base, and an asset management platform, configured to execute:

[0007] The wall temperature and humidity, air temperature and humidity, and wall-floor acceleration signals collected by the multi-source sensors are acquired. After outlier removal by dividing the data window according to the structural environment period, the data is written into the trusted data base. The relative water vapor gradient between the wall and indoor air at the beam-socket joint is extracted. The water vapor sealing effect of the beam-socket geometry is combined with the historical time-domain attenuation superposition calculation to extract the characteristic quantity of beam-socket humid heat accumulation that characterizes the harm of moisture retention. The displacement is recovered by frequency domain integration of the acceleration signal of the wall and floor slab, the rigid body rotation component of the wall is stripped to filter out the overall motion interference, the relative slip displacement of the actual connection misalignment is extracted, and then the proportion of local slip energy consumption that characterizes the depth of connection loosening is calculated. Based on the evolution time sequence of the characteristic quantity of wet heat accumulation in the beam socket and the proportion of local slip energy consumption, the response time difference under the maximum correlation matching between the two is found, and the slip hysteresis evolution index characterizing the friction degradation of the wooden beam end is obtained. By integrating the relative slip displacement and the acceleration signal, a dynamic equilibrium equation is constructed to obtain the stiffness degradation parameter characterizing the bearing capacity of the node. Combined with the slip hysteresis evolution index, the continuous degradation risk amount deviating from the historical normal benchmark is calculated. Based on the continuous degradation risk of each node, the global propagation risk is inferred. The risk level interval is divided by dynamic clustering to obtain the early warning level. Based on this, resource scheduling instructions for targeted inspection and reinforcement are generated and output to the asset management platform.

[0008] The beneficial effects of this invention are as follows: by introducing environmental temperature and humidity data into the degradation driving mechanism and extracting relative slip energy consumption parameters, a hysteresis evolution fingerprint is constructed, enabling the early warning system to effectively identify the difference between seasonal natural disturbances and actual connection loosening; at the same time, by combining an adaptive baseline model and a full life cycle economic cost evaluation, the false alarms caused by fixed threshold alarms are avoided, and early warning of potential hazards and reasonable scheduling of maintenance resources are realized. Attached Figure Description

[0009] Figure 1 This is a flowchart of the working process of a building structure health monitoring and early warning system based on the Internet of Things according to the present invention. Detailed Implementation

[0010] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0011] like Figure 1 As shown, an IoT-based building structure health monitoring and early warning system includes multi-source sensors, a trusted data base, and an asset management platform, configured to execute: The wall temperature and humidity, air temperature and humidity, and wall-floor acceleration signals collected by the multi-source sensors are acquired. After outlier removal by dividing the data window according to the structural environment period, the data is written into the trusted data base. The relative water vapor gradient between the wall and indoor air at the beam-socket joint is extracted. The water vapor sealing effect of the beam-socket geometry is combined with the historical time-domain attenuation superposition calculation to extract the characteristic quantity of beam-socket humid heat accumulation that characterizes the harm of moisture retention. The displacement is recovered by frequency domain integration of the acceleration signal of the wall and floor slab, the rigid body rotation component of the wall is stripped to filter out the overall motion interference, the relative slip displacement of the actual connection misalignment is extracted, and then the proportion of local slip energy consumption that characterizes the depth of connection loosening is calculated. Based on the evolution time sequence of the characteristic quantity of wet heat accumulation in the beam socket and the proportion of local slip energy consumption, the response time difference under the maximum correlation matching between the two is found, and the slip hysteresis evolution index characterizing the friction degradation of the wooden beam end is obtained. By integrating the relative slip displacement and the acceleration signal, a dynamic equilibrium equation is constructed to obtain the stiffness degradation parameter characterizing the bearing capacity of the node. Combined with the slip hysteresis evolution index, the continuous degradation risk amount deviating from the historical normal benchmark is calculated. Based on the continuous degradation risk of each node, the global propagation risk is inferred. The risk level interval is divided by dynamic clustering to obtain the early warning level. Based on this, resource scheduling instructions for targeted inspection and reinforcement are generated and output to the asset management platform.

[0012] The system is deployed in a three-tiered collaborative architecture consisting of a cloud computing cluster, an edge IoT gateway, and on-site multi-source sensing terminals. The on-site multi-source sensing terminals include temperature and humidity sensors deployed at beam-socket nodes in the target building, triaxial accelerometers deployed at floor slabs and corresponding wall locations, and tilt sensors and crack gauges deployed at key wall locations. The sampling period for the temperature and humidity sensors is set to 60 seconds, the sampling frequency for the accelerometers is set to 100Hz, and the sampling period for the tilt sensors and crack gauges is set to 300 seconds. All sensing terminals achieve clock synchronization via the NTP network time protocol, with a clock synchronization error not exceeding 10% of the corresponding sampling period. The edge IoT gateway is deployed on-site and is responsible for local caching, preprocessing, and retransmission of sensor data in the event of a network outage. During network outages, the local cache capacity is no less than 30 days of full sensor data, which is then retransmitted to the cloud computing cluster in timestamp order after network recovery. The cloud computing cluster is responsible for storing all data, calculating core algorithms, generating early warnings, and optimizing decisions. All algorithm steps are executed in a fixed sequence, with the core calculation cycle set to 24 hours to match the structural environment cycle.

[0013] The timing logic and synchronization rules for algorithm execution are as follows: After each computation cycle starts, the edge gateway first completes clock alignment and upload of all sensor data from the previous cycle. After the cloud completes data reception, it first performs data preprocessing and trusted on-chain operations, and then performs parallel calculations of beam-socket damp heat accumulation characteristics and local slip energy consumption ratios. The two parallel computation steps must use data from the same source within the same time window. After the calculation is completed, the subsequent slip lag evolution index calculation is triggered by the thread synchronization lock. Then, the continuous degradation risk calculation, global propagation risk and early warning level classification, and resource scheduling instruction generation are performed sequentially. Finally, the full lifecycle decision optimization and model update are performed. If data is missing or computational anomalies occur in a single step, three retries are performed. If the retry fails, the node is marked as an abnormal node, and subsequent calculations for the current cycle of that node are skipped. At the same time, a sensor anomaly alarm is generated.

[0014] Data on wall temperature and humidity, air temperature and humidity, triaxial acceleration of the wall and floor slabs, wall tilt angle, and crack width are acquired from multi-source sensors. Data windows are divided according to the structural environmental cycle. The structural environmental cycle adopts a 24-hour calendar day cycle. The data window length is consistent with the structural environmental cycle, the window sliding step is set to 12 hours, the overlap rate of adjacent windows is 50%, and the number of effective sampling points in each data window is not less than 95% of the number of sampling points of the corresponding sensor in a single cycle. When the number of effective sampling points is insufficient, linear interpolation of the same source data in adjacent cycles is used to complete the data. The number of sampling points completed by interpolation shall not exceed 5% of the total number of points in the window.

[0015] Extract the statistical central value of each monitoring data point within the data window. The statistical central value is the median of the monitoring data within the data window. For monitoring data sequences of even length, the median is the arithmetic mean of the two middle values ​​of the sequence. The median has the characteristic of resisting interference from extreme outliers and can avoid the impact of single sampling jumps on the statistical benchmark.

[0016] The absolute dispersion of each monitoring data point from the statistical central value is calculated to eliminate background meteorological noise interference caused by environmental baseline drift. The absolute dispersion is expressed as the median absolute deviation, which is the median of the absolute values ​​of the differences between each monitoring data point and the statistical central value within the data window. This median absolute deviation characterizes the inherent dispersion level of the monitoring data and eliminates baseline shifts caused by background noise such as periodic drift in environmental temperature and humidity and normal building vibrations. The formula for calculating the median absolute deviation is:

[0017] In the formula, The data collected by sensor s at time t is of type p. The data includes six types: wall temperature and humidity, air temperature and humidity, wall and floor acceleration, wall tilt angle, and crack width. Let t be the monitoring data window corresponding to time t. For time iteration variables within the data window; The median extraction operator is used to calculate the statistical central value of a sequence; Let be the median absolute deviation of the p-th type of monitoring data from sensor s at time t, with the same dimensions as the corresponding monitoring data.

[0018] The monitoring data is normalized using absolute dispersion, and a zero-limit constant with the same dimensions as the monitoring data is added to maintain the convergence of the underlying calculation, resulting in a standardized state quantity that is stripped of environmental trends. The calculation formula is as follows:

[0019] In the formula, 1.4826 is an empirical constant that makes the median absolute deviation asymptotically consistent with the standard deviation of the normal distribution. It is applicable to monitoring data that conforms to the normal distribution. When the monitoring data does not conform to the normal distribution, the constant can be adjusted through the normality test results in the data window. The adjustment range is 1.0 to 2.0. To prevent zeroing constants with the same dimensions as the p-th type of monitoring data, the value is taken as 1e-6 times the full scale of the corresponding monitoring data, in order to avoid calculation divergence caused by the denominator being 0; The standardized state quantity corresponding to the monitoring data at time t is dimensionless. This state quantity is separated from environmental trends and baseline drift, and has dimensional consistency across sensors and monitoring types.

[0020] Perform outlier removal. For sampling points where the absolute value of the standardized state variable is greater than 3, they are identified as outliers. The outliers are replaced with the median of the monitoring data within the data window to complete the outlier removal.

[0021] The underlying information fusion process integrates historical state hash chains, sensor hardware identifiers, acquisition timestamps, standardized state variables, and device operating status parameters. The historical state hash chain, generated from the previous calculation cycle, is used to construct a chain-like data structure. The sensor hardware identifier is a unique device identification code for each sensor, using a 16-byte globally unique identifier, ensuring global uniqueness. The acquisition timestamp is the Unix timestamp corresponding to the data sampling time, synchronized with the sensor sampling clock. Device operating status parameters include sensor power supply voltage, sampling success rate, and communication link quality. The sampling success rate is the ratio of the number of valid sampling points to the theoretical number of sampling points within the data window, dimensionless, ranging from 0 to 1. The communication link quality is the normalized value of the gateway's received signal strength indication, dimensionless, ranging from 0 to 1. The underlying information fusion uses a fixed byte order binary concatenation method, sequentially concatenating the historical state hash chain, sensor hardware identifier, acquisition timestamp, standardized state variables, and device operating status parameters. The byte length of each parameter is fixed; any length less than the fixed length is padded with 0x00 to the target length.

[0022] A current-cycle storage block feature with tamper-proof properties is generated through cryptographic one-way hash transformation and written into a trusted data base to achieve multi-source data chain traceability. The calculation formula is as follows:

[0023] In the formula, A hash chain of historical states generated in the previous calculation cycle. To calculate the period length, it should be consistent with the sliding step size of the data window; For sensor hardware identification; To collect time tags; This is a column vector formed by concatenating the standardized state variables of all monitoring types corresponding to sensor s; The column vector is formed by concatenating the device operating status parameters of sensor s at time t. It is a cryptographic one-way hash transformation operator, implemented using the SHA-256 algorithm, and has the characteristics of one-way irreversibility and collision resistance; This is a data concatenation identifier used to concatenate binary streams with multiple parameters. The current period stores block features, which correspond one-to-one with the spliced ​​original data. Any tampering with the original data will cause irreversible changes to the block features. By synchronously writing the block features and the original standardized state quantities into the trusted data base, the full-cycle anti-tampering and chain traceability of monitoring data can be achieved.

[0024] Based on the wall temperature and humidity and the air temperature and humidity, the internal pore water vapor pressure and the indoor ambient water vapor pressure are calculated separately. The difference between these two values ​​is used to eliminate substrate interference under identical internal and external temperature and humidity conditions, thus deriving the relative water vapor gradient driving moisture migration. The calculation formula is as follows:

[0025]

[0026] In the formula, This is a function for calculating actual water vapor pressure, used to calculate the actual water vapor pressure under corresponding environments based on temperature and relative humidity, with the unit being Pascals. This is a temperature parameter, in degrees Celsius, applicable from -45℃ to 60℃. When the temperature is below 0℃, the modified ice-phase saturated vapor pressure formula is used. The modified formula is: ; , where is the relative humidity parameter, in percentage; 610.94, 17.625 and 243.04 are empirical constants used in the Tetens formula to fit the saturated vapor pressure of the aqueous phase; Let t be the temperature value in the wall temperature and humidity monitoring data corresponding to node b in the beam socket at time t. The temperature and humidity sensor is deployed at the contact surface between the end of the wooden beam and the masonry wall inside the beam socket. The sampling point is not less than 1 / 2 of the length of the wall embedded at the end of the beam socket. Let be the relative humidity value in the wall temperature and humidity monitoring data corresponding to beam socket node b at time t; Let t be the air temperature value of the indoor environment corresponding to beam socket node b. The air temperature and humidity sensor is deployed on the same floor as the beam socket node, at a horizontal distance of no more than 2 meters from the beam socket, avoiding direct sunlight and direct airflow from the ventilation opening. Let be the relative humidity value of the indoor environment corresponding to beam socket node b at time t; Let be the relative water vapor gradient at beam-socket node b at time t, in Pascals. This value is the difference between the water vapor pressure inside the wall pores and the water vapor pressure in the indoor environment. When this value is positive, it means that the water vapor pressure inside the wall is greater than that in the indoor environment, and water vapor diffuses outward. When this value is negative, it means that the water vapor pressure in the indoor environment is greater than that inside the wall, and water vapor penetrates into the wall. This difference eliminates substrate interference under the same temperature and humidity conditions inside and outside, and can accurately characterize the thermodynamic driving force for the cross-interface migration of moisture.

[0027] Autocorrelation time-domain evolution analysis was performed on the relative water vapor gradient to extract the hygrothermal memory period, which characterizes the lag in natural water evaporation. The autocorrelation function was calculated using an unbiased estimation method, and the formula is as follows:

[0028]

[0029] In the formula, This is the time-shifted variable of the autocorrelation function, with units consistent with the sampling period; This represents the total number of sampling points within the data window. This represents the variance of the relative water vapor gradient sequence within the data window; This is the arithmetic mean of the relative water vapor gradient sequence within the data window; Let be the autocorrelation function of the relative water vapor gradient sequence of node b in the beam socket within the data window corresponding to time t. It is dimensionless and is used to characterize the degree of self-similarity of the sequence under different time shifts. The maximum shift for autocorrelation analysis is set to half the length of the data window; This is a time-division-zero constant, with a value of 1e-3 hours, used to avoid calculation divergence caused by a denominator of 0; denoted as the humid heat memory period corresponding to node b in the beam-socket joint at time t, in hours. This value is the first moment of the autocorrelation function, which characterizes the temporal persistence of the relative water vapor gradient sequence. It corresponds to the lag time length of natural evaporation after water intrusion at the beam-socket joint. The larger this value is, the longer the water stays inside the beam-socket joint, and the more significant the lag characteristic of evaporation.

[0030] By selecting positive gradient segments of water vapor infiltration, a decay-forgetting mechanism based on the humid-thermal memory cycle is introduced for time-domain accumulation calculation to obtain the initial humid-thermal accumulation. For discrete sampled data, the trapezoidal integral method is used to perform time-domain integration, and the calculation formula is as follows:

[0031]

[0032] In the formula, For historical time-domain integral variables, the units are consistent with the sampling period; To characterize the non-negative value function of the positive gradient segment of water vapor infiltration into the wall, only the gradient segment of water vapor infiltration into the wall is retained; The decay and forgetting factor is based on the damp heat memory cycle. The further away the historical data is from the current time t, the smaller the decay weight, which corresponds to the decay characteristics of the moisture inside the beam cavity naturally evaporating over time. The lower limit of integration is set to the time 5 times the damp heat memory cycle before the current time to cut off historical data beyond 5 times the damp heat memory cycle and avoid numerical divergence caused by infinite integration. Let be the initial moisture and heat accumulation at node b in the beam cavity at time t, expressed in Pascal-hours. This value is the time-domain weighted cumulative result of the inward water vapor infiltration gradient, characterizing the moisture and heat accumulation effect formed by historical water vapor intrusion within the beam cavity. When the relative water vapor gradient remains positive and water vapor diffuses outward, the initial moisture and heat accumulation is expressed as a time constant. It decays exponentially, and the decay formula is: This is to reflect the attenuation of the cumulative effect of moisture and heat caused by water evaporation.

[0033] The system ledger is used to obtain the beam end wall embedment constraint ratio, wall recess heat dissipation space ratio, and mortar joint airflow exchange ratio. These are then nonlinearly fused to obtain the local structural water vapor sealing coefficient, which reflects the local structural water retention capacity. The calculation formula is as follows:

[0034] In the formula, The length of the wooden beam end embedded in the wall corresponding to beam socket node b is in millimeters. The data is from the building structure system ledger. The width of the wooden beam end corresponding to beam socket node b is in millimeters. The data is from the building structure system ledger. The ratio of the wooden beam end embedded in the wall is dimensionless. The larger the ratio, the greater the depth of the wooden beam end embedded in the wall, the longer the ventilation path between the inside of the beam socket and the outside, and the more difficult it is for moisture to escape. The wall thickness corresponding to beam socket node b is in millimeters, and the data comes from the building structure system ledger. The effective height of the beam socket corresponding to beam socket node b is in millimeters. The data is from the building structure system ledger. This is a division-zero constant with the dimension of length, with a value of 1e-3 mm, used to avoid calculation divergence caused by a denominator of 0; The ratio of heat dissipation space in the wall recess is dimensionless. The larger the ratio, the smaller the effective heat dissipation and ventilation space inside the beam recess, and the stronger the water vapor sealing effect. The number of connected mortar joint channels corresponding to beam socket node b is dimensionless and the data comes from the building structure system ledger. This represents the ventilation openness of the beam socket corresponding to node b, a dimensionless value ranging from 0 to 1, where 0 represents a completely closed beam socket and 1 represents a completely open beam socket. This value is calculated using the formula... Calculation, where The effective ventilation area of ​​the beam recess opening is expressed in square millimeters. This represents the total surface area of ​​the beam recess in contact with the interior space, expressed in square millimeters. The mortar joint airflow exchange ratio is dimensionless. The larger the ratio, the more mortar joint channels there are around the beam socket, the lower the ventilation openness, and the easier it is for water vapor to stay inside the beam socket. is the local structural water vapor sealing coefficient corresponding to beam-socket node b. It is dimensionless and is a nonlinear product of the ratios of three structures. It comprehensively characterizes the local structural sealing and retention capacity of the beam-socket node for water vapor. The larger the value, the stronger the water retention capacity inside the beam-socket and the higher the risk of decay under the same water vapor intrusion.

[0035] By weighting the initial moist heat accumulation amount with a risk level using the local structural moisture sealing coefficient, the characteristic quantity of moist heat accumulation in the beam cavity is obtained. The calculation formula is as follows:

[0036] In the formula, The value is the characteristic quantity of moisture and heat accumulation in the beam socket corresponding to node b at time t, in Pascal-hours. This value is the result of the initial moisture and heat accumulation after weighting by the local structural water vapor sealing coefficient. It comprehensively characterizes the actual degree of harm caused by water vapor retention at the beam socket node and can be directly used for subsequent degradation time series correlation analysis.

[0037] The triaxial acceleration signals of the wall and floor slab were preprocessed, including mean removal, windowing, and frequency domain transformation, to eliminate DC components and spectral leakage. Hanning windows were used for acceleration signal preprocessing, with the window length matching the data window length and an overlap rate of 50% between adjacent windows. The windowed acceleration signals were then converted to the frequency domain using a Fast Fourier Transform (FFT). The number of FFT points was set to be an integer power of 2, not less than the minimum number of sampling points in the data window. Sample sequences shorter than the target length were padded with zeros to the end.

[0038] During the frequency domain integration of the acceleration signal, an adaptive cross-validation method is used to introduce a frequency offset suppression constraint term to eliminate integration drift distortion caused by low-frequency noise accumulation, thereby obtaining the global displacement of the floor slab and the global displacement of the walls. The frequency offset suppression constraint term is automatically optimized using a generalized cross-validation criterion. The optimization process employs a grid search algorithm with a fixed random seed. The random seed is set to a fixed value of 20240506 to ensure stable reproducibility of the optimization results. The calculation formula is as follows:

[0039]

[0040] In the formula, The frequency offset suppression regularization parameter is dimensionless, and the optimization range is set to 1e-8 to 1e2. The grid search step size is set to 1e-8. The generalized cross-validation regularization matrix is ​​constructed using the Tikhonov regularization matrix, with dimensions matching the length of the frequency domain acceleration vector, and diagonal elements of length 1. Off-diagonal elements are 0; Let be the frequency domain acceleration vector of a single measurement point in the frequency domain of the acceleration signal collected by sensor s, with units of meters per second squared. The square operation of the L2 norm of a vector is used to quantify the total energy of the fitting bias; The trace operator for matrices, used to calculate the sum of the diagonal elements of a matrix; For the identity matrix, dimension and regularization matrix Consistent; It is a dimensionless divide-by-zero constant with a value of 1e-8, used to avoid calculation divergence caused by a denominator of 0; For sensor s at time t, the frequency offset suppression constraint term is automatically optimized. Its dimension is 1 per square second. This value is optimized through the generalized cross-validation criterion and achieves the optimal balance between fitting bias and regularization strength. It can adaptively suppress integral drift caused by the accumulation of low-frequency noise. For acceleration signals at angular frequency The frequency domain acceleration component at that point, in meters per second squared; The angular frequency component for frequency domain analysis is expressed in radians per second. The effective passband range is set to the angular frequency corresponding to 0.1Hz to 20Hz. The amplitude of frequency components outside the passband range is set to 0 to eliminate low-frequency drift and high-frequency noise. Angular frequency The frequency domain displacement recovery, in meters, is obtained through quadratic integration in the frequency domain. By introducing a frequency offset suppression constraint, the integral drift distortion caused by low-frequency acceleration noise can be eliminated. By performing an inverse fast Fourier transform on the frequency domain displacement recovery, the global displacement of the floor and the global displacement of the wall in the time domain can be obtained, which correspond to the displacement recovery results of the accelerometer at the floor and the corresponding accelerometer at the wall, respectively. The dimensions of the displacement results are unified in millimeters.

[0041] The projection of the global displacement of the wall onto the contact normal direction is extracted, and the spatial geometric lever arm is derived using acceleration signals to calculate the pure boundary translational quantity, eliminating local overturning errors of the wall. The contact normal direction is the normal direction of the contact surface between the wooden beam and the wall at the beam-socket joint, perpendicular to the wall surface in the horizontal direction, used to project the three-dimensional spatial displacement onto the dominant direction of the beam end slippage. The rigid body motion of the wall includes three translational degrees of freedom and three rotational degrees of freedom. Using the displacement recovery results from at least two triaxial accelerometers at different height positions of the wall, the three rotational components of the wall are calculated, including the overturning angle perpendicular to the wall surface, the in-plane rotation angle along the wall surface, and the torsional angle along the wall height direction. The calculation formula is as follows:

[0042]

[0043]

[0044] In the formula, Let be the overturning angle of the wall at beam-socket node b around the x-axis at time t, in radians, with the x-axis parallel to the wall surface in the horizontal direction; Let t be the in-plane rotation angle of the wall at beam socket node b around the y-axis at time t, in radians. The y-axis is perpendicular to the wall surface in the horizontal direction, i.e., the direction of the contact normal. Let be the angle of twist of the wall about the z-axis at beam socket node b at time t, in radians, with the z-axis along the vertical direction; , These represent the global displacements of the high and low measuring points along the z-axis, respectively, in millimeters. , The installation heights of the high and low measuring points are respectively, in millimeters, and the height difference is not less than 1 meter; , These represent the global displacements of the high and low measuring points along the x-axis, respectively, in millimeters. , These represent the global displacements along the y-axis of the left and right measuring points at the same height on the wall, respectively, in millimeters. , The horizontal installation positions of the right and left measuring points are shown in millimeters, with a horizontal spacing of not less than 2 meters.

[0045] Subtracting the pure boundary translational moment from the contact surface projection of the global displacement of the floor slab, and separating the overall coordinated deformation of the structure from the six-degree-of-freedom rigid body motion, yields the relative sliding displacement. The calculation formula is as follows:

[0046] In the formula, is the unit vector of the contact surface normal direction corresponding to beam socket node b, which is a three-dimensional column vector, dimensionless, along the y-axis direction; Let be the global displacement vector of the floor slab corresponding to the beam socket node b at time t. It is a three-dimensional column vector with units of millimeters. Let be the global displacement vector of the wall position corresponding to beam socket node b at time t. It is a three-dimensional column vector with units of millimeters. The projection of the global displacement difference between the floor and the wall onto the normal direction of the contact surface represents the initial relative displacement before eliminating rigid body rotation errors. , where b is the geometric arm of the spatial rotation angle of the beam socket node, in millimeters, and is the vertical distance of the beam socket node relative to the rotation center of the wall. The data comes from the building structure system ledger. , where b is the horizontal geometric lever arm in millimeters, and , is the horizontal distance between the beam socket node and the torsional center of the wall. The data is from the building structure system ledger. This refers to the displacement component caused by the rigid body rotation of the wall, which is the rigid body motion error that needs to be eliminated. The relative slip displacement of beam socket node b at time t is expressed in millimeters. This value is stripped of the interference of overall structural deformation and six-degree-of-freedom rigid body motion of the wall, and only retains the actual connection displacement between the beam end and the wall contact surface, which can directly characterize the degree of looseness of the node connection.

[0047] The kinetic energy time integral corresponding to the velocity sequence of relative slip displacement is calculated to obtain the local slip dissipation energy characterizing the friction interface. The velocity sequence of relative slip displacement is obtained by performing a first-order time-domain difference calculation on the relative slip displacement. Before the difference, a 5th-order Butterworth low-pass filter is used to filter the relative slip displacement sequence, with the filter cutoff frequency set to 20Hz to eliminate high-frequency noise amplified by the difference operation. The velocity sequence calculation formula is as follows:

[0048] In the formula, The sampling interval for the acceleration signal is in seconds. The time-domain velocity of the relative slip displacement is expressed in millimeters per second.

[0049] The local slip energy dissipation is divided by the overall structural response kinetic energy to correct for fluctuations in the input energy, yielding the proportion of local slip energy dissipation. For discrete sampled data, the trapezoidal integral method is used to perform time-domain integration, and the calculation formula is as follows:

[0050] In the formula, For time iteration variables within the data window; The square-time integral of the relative sliding velocity within the data window represents the local sliding dissipation energy generated by the sliding motion of the friction interface, and is positively correlated with the total energy of the work done by the sliding friction. The time-domain velocity vector of the global displacement of the floor slab, in millimeters per second, is obtained by performing a first-order time-domain difference on the global displacement of the floor slab. The time-domain velocity vector of the global displacement of the wall, in millimeters per second, is obtained by performing a first-order time-domain difference on the global displacement of the wall. It represents the total kinetic energy of the global motion velocity of the floor slab and walls, and the input energy characterizing the overall structural response; This is a kinematic constant to prevent division by zero, with a value of 1e-3 square millimeters per square second, used to avoid calculation divergence caused by a denominator of 0; The value is the proportion of local slip energy dissipation at beam socket node b at time t. It is dimensionless and ranges from 0 to 1. This value is the ratio of local slip energy dissipation to the overall structural response kinetic energy. It corrects for the input energy disturbance caused by fluctuations in the intensity of external excitation and can stably characterize the proportion of friction energy dissipation caused by loose node connection. The larger the value, the higher the degree of looseness of the node connection and the more significant the slip energy dissipation at the friction interface.

[0051] The static DC components representing the characteristic quantities of heat accumulation in the beam cavity and the proportion of energy consumption due to local slip were filtered out separately, extracting the AC sequences of heat and energy consumption changes that characterize dynamic fluctuations. The static DC component is the arithmetic mean of the corresponding sequences within the data window. The DC component was filtered out by subtracting the mean from the value at each time step of the sequence within the data window, resulting in an AC sequence that retains only the dynamic fluctuations, thus eliminating the interference of the static baseline on the time-series correlation analysis. The heat and energy consumption AC sequence is... ,in The mean value of the characteristic quantity of humid heat accumulation in the beam socket within the data window; the AC sequence of energy consumption change is... ,in This represents the average percentage of energy consumed during local sliding within the data window.

[0052] A series of time-shifting operations are applied to an alternating current sequence with varying energy consumption to construct a candidate time delay set characterizing the hysteresis properties of the physical response. The values ​​of the candidate time delay set range from 0 to min( , The translation step size is consistent with the sampling interval of the acceleration signal. Only the positive time delay is considered, i.e., the change in humid heat accumulation precedes the change in energy consumption, which conforms to the physical causal sequence of water vapor intrusion leading to frictional performance degradation. The candidate time delay set is denoted as... The elements in the set are all the time shift steps to be verified. .

[0053] The cross-correlation coupling integral between the hydrothermal change AC sequence and the shifted energy consumption change AC sequence is calculated iteratively. The shift step size that maximizes the absolute value of this integral is determined as the response time difference from water vapor intrusion to frictional loss under the physical disaster mechanism. For the portion of the sequence that exceeds the data window after shifting, truncation is performed, retaining only the overlapping portion within the data window for cross-correlation calculation. The calculation formula is as follows:

[0054] In the formula, The search step size variable for constructing the time delay set has the same unit as the sampling period; This is the positive causal delay verification domain, i.e., the candidate time delay set; The time shift operation applied to the alternating current sequence of energy consumption changes characterizes the time delay of energy consumption changes relative to changes in heat and moisture accumulation. This represents the overlapping time window of the two sequences after translation; The cross-correlation coupling integral of the heat and humidity change AC sequence and the shifted energy consumption change AC sequence is used to quantify the degree of linear correlation between the two sequences under the corresponding time shift. Let t be the response time difference corresponding to node b in the beam socket at time t, with the unit consistent with the sampling period. This value is the translation step size that maximizes the absolute value of the cross-correlation coupling integral. It characterizes the time delay from the formation of humid heat accumulation due to water vapor intrusion into the beam socket to the increase in slip energy consumption caused by the degradation of the friction performance at the beam end. This is consistent with the causal time sequence relationship between the disaster-causing factor and the degradation response.

[0055] Peak significance screening is performed. When the normalized cross-correlation coefficient corresponding to the maximum cross-correlation coupling integral is less than 0.3, it is determined that there is no significant causal relationship, the response time difference is set to 0, and the glide lag evolution index is set to 0, in order to filter out spurious causal relationships. The formula for calculating the normalized cross-correlation coefficient is:

[0056] In the formula, It is a dimensionless divide-by-zero constant with a value of 1e-8; is the normalized cross-correlation coefficient corresponding to the maximum cross-correlation coupling integral, which is dimensionless and ranges from -1 to 1.

[0057] The maximum cross-correlation coupling integral is extracted, and numerical normalization is performed using the variance of the autocorrelation energy of each of the two sequences to eliminate absolute amplitude scale interference. Then, hysteresis penalty amplification is applied using the response time difference to derive a glide hysteresis evolution index characterizing the hidden degradation depth. The calculation formula is as follows:

[0058] In the formula, The lag penalty factor is dimensionless. The larger the ratio of the response time difference to the humid heat memory period, the more significant the degradation lag effect caused by water vapor intrusion, and the deeper the implicit degradation of the node, thus amplifying the penalty on the index. The square root term in the denominator is the geometric mean of the variance of the autocorrelation energy of the two AC sequences. It is used to normalize the cross-correlation coupling integral and eliminate the interference caused by the difference in the absolute amplitude scale of the sequences. The normalized value ranges from -1 to 1. is the slip hysteresis evolution index corresponding to beam socket node b at time t. It is dimensionless and integrates the structural hazard of the beam socket, the hysteresis characteristics of the degradation response, and the correlation with the humid heat-energy dissipation sequence. It can accurately characterize the hidden depth of frictional degradation of the timber at the beam end caused by water vapor intrusion. The larger the value, the more severe the early degradation of the node.

[0059] The equivalent mass of the floor slab supported by the beam socket is obtained using the system configuration ledger. The inertial force is then calculated by combining the relative difference in acceleration between the floor slab and the wall, thus restoring the equivalent horizontal shear force at the interface. The calculation formula is as follows:

[0060] In the formula, The equivalent mass of the floor slab borne by beam socket node b is expressed in kilograms. The data is derived from the building structure system ledger and is calculated based on the floor slab area, surface load, and material density. Let be the time-domain acceleration vector of the floor slab at time t, in meters per second squared. Let be the time-domain acceleration vector of the wall at time t, in meters per second squared. The projection of the relative acceleration difference between the floor slab and the wall onto the normal direction of the contact surface, expressed in meters per second squared; This is a dimension conversion factor used to convert millimeters (in terms of displacement) to meters (in terms of displacement) to ensure consistent dimensions in calculations. Let be the equivalent horizontal shear force at the interface corresponding to beam-socket node b at time t, in Newtons. It is obtained by solving for relative acceleration and equivalent mass according to d'Alembert's principle, and represents the horizontal shear force transmitted between the beam end and the wall contact surface.

[0061] Based on the hysteresis work envelope area of ​​the equivalent horizontal shear force and relative slip displacement at the interface, the equivalent shear stiffness characterizing the node's resistance to slippage is derived. For discrete sampled data, the trapezoidal integral method is used to perform time-domain integration, and the calculation formula is as follows:

[0062] In the formula, The integral of the product of the equivalent horizontal shear force and the relative slip displacement within the data window corresponds to the envelope area of ​​the shear force-slip hysteresis loop, representing the total deformation work done by the node under cyclic load. The square integral of the relative slip displacement within the data window represents the total deformation of the node. This is a zero-prevention constant for the mechanics-kinematic coupling dimension, with a value of 1e-3 square millimeters per second, used to avoid calculation divergence caused by a denominator of 0; Let be the equivalent shear stiffness of beam socket node b at time t, expressed in Newtons per millimeter. This value is the ratio of the work done by deformation to the total deformation, characterizing the node's ability to resist horizontal slip deformation. The smaller this value, the weaker the node's shear bearing capacity.

[0063] By comparing the current equivalent shear stiffness with the healthy stiffness sequence of the structure during its undamaged historical period, stiffness degradation parameters characterizing structural resistance loss are extracted. The healthy stiffness sequence is the median sequence of equivalent shear stiffness during the first 3 months after structure completion, excluding extreme weather and load disturbances, and serves as the baseline for healthy stiffness in the undamaged state of the structure. The calculation formula is as follows:

[0064]

[0065] In the formula, The baseline for healthy stiffness corresponding to beam socket node b is expressed in Newtons per millimeter. This is a zero-division constant for rigid units, with a value of 1e-3 Newtons per millimeter, used to avoid calculation divergence caused by a denominator of 0; Let be the stiffness degradation parameter corresponding to beam socket node b at time t. It is dimensionless and ranges from 0 to 1. This value is 1 minus the ratio of the current stiffness to the healthy stiffness baseline, which characterizes the degree of loss of shear stiffness of the node. The larger the value, the more severe the resistance degradation of the node. , These are the time sampling points within the data window, and ; The rate of change of the stiffness degradation parameter between the two sampling points; The degradation rate of node b at time t is expressed in units of 1 second. This value is the median of all degradation rate samples within the data window, representing the rate of node stiffness degradation. The larger the value, the faster the node degradation process.

[0066] The slip hysteresis evolution index, stiffness degradation parameter, degradation evolution rate, crack propagation rate, nodal rotation rate, and beam socket hygrothermal accumulation characteristic quantity are all dimensionlessly processed and fused into a multi-dimensional set of nodal health status. All dimensional parameters are mapped to the 0-1 interval using a min-max normalization method. The upper and lower limits of the normalization are determined based on the extreme values ​​of historical full-cycle data, and the calculation formula is as follows:

[0067]

[0068] In the formula, For the i-th dimension of the node health status set, see the original parameters. , These are the historical full-cycle minimum and maximum values ​​of the i-th dimension parameter, respectively; It is a dimensionless divide-by-zero constant with a value of 1e-8; This is the dimensionless result of the i-th dimension parameter, with a value ranging from 0 to 1; It is a dimensionless glide hysteresis evolution index; The dimensionless stiffness degradation parameter; The degeneracy rate after dimensionless transformation; The crack propagation rate is a dimensionless value, calculated from the first difference of the crack width data collected by the crack gauge, and is expressed in millimeters per second. The nodal angle change rate is a dimensionless value, which is calculated from the first difference of the wall tilt angle data collected by the tilt sensor, and the unit is radians per second. The dimensionless characteristic quantity of heat accumulation in the beam socket; Let be the set of multi-dimensional node health states corresponding to node b in the beam socket at time t. It is a dimensionless column vector that integrates multi-dimensional information such as the degree of node degradation, stiffness loss, degradation rate, apparent damage and disaster-causing factors, and comprehensively characterizes the health state of the node.

[0069] The multidimensional spatial variation distance of this state set relative to the centroid of the historical normal service distribution is calculated to eliminate false alarms caused by single abnormal data jumps. The multidimensional spatial variation distance is calculated using Mahalanobis distance. For cases where the covariance matrix is ​​close to singular, a ridge regression method is used to add a regularization term to ensure the positive definiteness of the matrix. The calculation formula is as follows:

[0070]

[0071] In the formula, This is a sample of the health status of a dimensionless node under normal service conditions in history. The sample size is the number of historical samples in normal service status, with a minimum sample size of 100 groups. The centroid vector of the distribution of the dimensionless node health status set under historical normal service conditions is calculated from the arithmetic mean of historical normal samples; The coefficient of the ridge regression regularization term takes a value of 1e-6 and is used to ensure the positive definiteness of the covariance matrix and avoid the failure of inversion caused by matrix singularity. It is an identity matrix with the same dimensions as the covariance matrix; The multidimensional covariance matrix is ​​a set of dimensionless node health states under historical normal service conditions, used to characterize the correlation and dispersion between parameters of each dimension; The inverse of the covariance matrix is ​​calculated using the Moore-Penrose pseudo-inverse, which further improves the numerical stability of matrix inversion. The distance is the Mahalanobis distance between the current set of health statuses and the historical normal distribution, also known as the multidimensional spatial variation distance. It is dimensionless and eliminates false alarms caused by a single abnormal data jump. The distance value will only increase significantly when multiple parameters deviate from the normal distribution simultaneously.

[0072] Multiplying the multidimensional spatial variation distance by the component failure risk weight, which characterizes the load-bearing importance of the node, yields the continuous degradation risk. The component failure risk weight is quantitatively calculated based on the node's load-sharing ratio, structural safety level, and the severity of failure consequences. The calculation formula is as follows:

[0073]

[0074] In the formula, The standard value of the vertical load on the floor slab borne by beam socket node b is given in kilonewtons. The data is sourced from the building structure system ledger. This represents the standard value of the total vertical load of the building, in kilonewtons. The structural safety level coefficient is dimensionless, with 1.2 for Level 1, 1.0 for Level 2, and 0.9 for Level 3. The failure consequence coefficient is dimensionless. It is 1.5 for critical load-bearing nodes, 1.0 for general load-bearing nodes, and 0.5 for non-load-bearing nodes. The component failure risk weight corresponding to beam socket node b is dimensionless and ranges from 0 to 1. It is determined based on the load-bearing importance of the node in the structural system and the severity of the failure consequences. Let be the continuous degradation risk quantity corresponding to node b at time t. It is dimensionless and combines the deviation of the node's health status with the structural importance. It can continuously quantify the failure risk of the node. The larger the value, the higher the degradation risk of the node.

[0075] Based on the structural spatial topology, the sensitivity gradient of the stiffness variation of the source beam-socket joint to the displacement response of the associated supporting members is calculated, and the weights of the mechanical transmission influence are constructed. The sensitivity gradient is calculated using the finite difference method of the structural finite element model, and the calculation formula is as follows:

[0076]

[0077] In the formula, The displacement response parameter of the associated support member v, in millimeters, is calculated by the linear elastic finite element model of the building structure. The finite element model is constructed based on the geometric and material parameters of the building structure system ledger. The value is the small stiffness disturbance of beam socket node b, which is 1% of the healthy stiffness baseline, and the unit is Newtons per millimeter; Let V be the partial derivative of the displacement response of component v with respect to the stiffness variation of node b, i.e., the sensitivity gradient, in millimeters per Newton, characterizing the degree of influence of the stiffness variation of node b on the displacement response of component v. The pool of all associated source nodes that affect the target component v includes all beam socket nodes that have a mechanical transfer relationship with component v; This is a zero-prevention constant for the mechanical-displacement coupling dimension, with a value of 1e-3 mm per Newton, used to avoid calculation divergence caused by a denominator of 0; Let be the weight of the mechanical transmission influence of the source beam socket node b on the associated support component v at time t. It is dimensionless and ranges from 0 to 1. This value is the proportion of the sensitivity gradient of node b to the sum of the sensitivity gradients of all associated nodes, representing the weight of the degradation risk of node b being transmitted to component v.

[0078] By spatially weighting and summing the influence of mechanical transmission with the continuous degradation risk of the source beam-socket node, the global propagation risk of the overall domino failure tendency of the structure is quantified. The calculation formula is as follows:

[0079] In the formula, The summation of the continuous degradation risk of all associated source nodes after weighting by the influence of mechanical transmission is dimensionless and represents the total risk input converging to component v; the exponential function adopts a negative exponential form to achieve spatial weighted attenuation of risk, which conforms to the attenuation characteristics of structural mechanics transmission. Let be the global propagation risk of component v at time t. It is dimensionless and ranges from 0 to 1. This value quantifies the tendency of component v to fail as a result of the degradation of associated nodes. The larger the value, the higher the overall failure risk of component v due to the degradation of associated nodes.

[0080] Unsupervised dynamic clustering is performed on the global propagation risk values ​​of all building components to find the hierarchical interval partitioning boundary that maximizes intra-cluster convergence and inter-cluster discrepancy, thereby suppressing false alarms caused by seasonal drift of the building's synchronization parameters. The clustering algorithm employs a Gaussian mixture model optimized using the Bayesian information criterion, with upper and lower limits for the number of clusters ranging from 2 to 5 levels, corresponding to the number of warning levels. Model training uses an expectation-maximization algorithm with a fixed random seed, set to a fixed value of 20240506, to ensure stable reproducibility of the clustering results. The objective function for optimal cluster partitioning is:

[0081] In the formula, For all possible clustering partitioning schemes, each scheme contains multiple risky clustering subsets; For a single risk cluster subset in the clustering partitioning scheme; For a single component within a cluster subset; This is the arithmetic mean of the global propagation risk of all components within the cluster subset g; The total sum of squares within each cluster is used to quantify the convergence of the data within each cluster. The smaller this value is, the lower the dispersion of the data within each cluster. This represents the number of cluster subsets in the clustering partitioning scheme. The total number of all building components participating in the clustering; The variance of the global propagation risk value across all components of the building; To limit the structural risk information penalty term of fragmented grouping, in order to avoid overfitting caused by too many clusters, and to suppress false alarms caused by seasonal drift of the building-wide synchronization parameters; The optimal risk level interval corresponding to time t is defined by minimizing the objective function, which achieves optimal clustering partitioning that maximizes intra-class convergence and inter-class dispersion. This partitioning boundary is dynamically updated with the risk distribution of the entire building, without the need to set a fixed threshold.

[0082] The warning level is determined by comparing the relative severity of each component's risk level range. To avoid drastic fluctuations in warning levels caused by abrupt changes in cluster boundaries between adjacent periods, a sliding window smoothing process is used. The warning level is the mode of the warning levels for the current period and the previous two periods, calculated using the following formula:

[0083]

[0084] In the formula, The risk cluster subset to which component v belongs; The mean global propagation risk of component v within its cluster subset; To count the elements of a set that meet the Boolean criterion within the curly braces; Let v be the original warning level corresponding to component v at time t. It is a positive integer. This value is determined according to the risk mean of the cluster subset. The higher the risk mean of the cluster subset, the higher the warning level. The warning level is based on the relative distribution of risk in the whole building, rather than a fixed threshold, which can effectively avoid false alarms caused by seasonal parameter drift. Extracting operators for the mode; Let v be the smoothed warning level corresponding to component v at time t. It is a positive integer with a value range of 1 to 5, where level 1 is low risk and level 5 is extremely high risk.

[0085] Based on the proportion of early warning levels and the risk weight of component failure, differentiated material allocation rates are calculated, generating resource scheduling instructions to guide preventative temporary support and non-destructive testing of key areas. The calculation formula is as follows:

[0086]

[0087] In the formula, The weighting coefficient for the warning level is dimensionless, with 0.1 for Level 1, 0.3 for Level 2, 0.6 for Level 3, 1.0 for Level 4, and 1.5 for Level 5. It is a dimensionless divide-by-zero constant with a value of 1e-8; Let be the material allocation rate corresponding to component v at time t. It is dimensionless and ranges from 0 to 1. This value is the proportion of the risk-weight product of component v to the total of the entire building, representing the proportion of maintenance resources that should be allocated to component v. This refers to the total amount of maintenance resources currently managed by the system's pool, including the number of inspection personnel, the number of temporary support equipment, and the number of non-destructive testing equipment hours. The maintenance resource allocation for component v is issued at time t. Based on the resource allocation of each component and the warning level, the system generates resource scheduling instructions for targeted inspection, preventive temporary support, and non-destructive testing of key parts, and issues them to the asset management platform for execution. Among them, components with warning level 5 are subject to on-site verification and temporary support within 24 hours, components with warning level 4 are subject to non-destructive testing within 72 hours, components with warning level 3 are subject to targeted inspection within 7 days, and components with warning level 2 and below are subject to monthly routine inspection.

[0088] Based on the time-varying structural decay mechanics model and the state of the previous period, theoretical prior state prediction data are calculated by extrapolating to the next time series. The time-varying structural decay mechanics model adopts a nonlinear damage evolution model, which is constructed based on the coupling relationship between wood decay and stiffness degradation. The state vector is a dimensionless set of nodal healthy states, and the state transition equation is:

[0089] In the formula, This is the posterior state estimate of the beam socket node b obtained from the previous cycle update, i.e., the dimensionless set of node health states. This represents a dimensionless mapping of external interventions implemented within the current cycle, including quantitative parameters for maintenance measures such as reinforcement and repair. The value is 0 when there is no intervention and 10 when complete repair is achieved. ; The damage evolution rate coefficient is dimensionless and calibrated based on historical degradation data. This is a function for a time-varying decay mechanics model of the structure, used to characterize the temporal evolution of node states; The state prior prediction data for node b at time t is the theoretical prediction state obtained by extrapolation based on the state of the previous cycle and the decay model. The superscript - represents the prior prediction value.

[0090] The measured moisture content of timber, ultrasonic defect depth, and measured crack width, recorded during manual inspections, constitute a reliable feedback source for on-site verification, shielding against long-term sensor drift errors. The calculation formula for the observation vector is:

[0091] In the formula, The dimensionless result is the wood moisture content at beam socket node b measured on-site. The moisture content was obtained on-site by a needle moisture meter and is expressed as a percentage. The dimensionless result of the ultrasonic defect depth rate measured on site is calculated from the ratio of the wood decay depth obtained by ultrasonic non-destructive testing to the beam end section height, and is dimensionless. The dimensionless result of the crack width around the beam socket joint measured on site is obtained by crack caliper measurement in millimeters; The dimensionless result of the implemented maintenance recovery coefficient is a dimensionless value ranging from 0 to 1, which characterizes the degree of recovery of node performance by maintenance measures. Let be the field verification observation vector corresponding to beam socket node b at time t. It is a dimensionless column vector and serves as the true feedback source to shield the long-term drift error of the sensor, used to correct the model prediction deviation.

[0092] Based on the observation residuals of the actual feedback source and the prior state prediction data verified on-site, an unscented Kalman filter algorithm is used to generate a state correction gain. This gain is then used to perform closed-loop correction on the uncertainty bias of the decay model, resulting in a posterior state estimate to update the subsequent prediction baseline. The unscented Kalman filter is suitable for state estimation of nonlinear systems and avoids the Jacobian matrix calculation error of the extended Kalman filter. The calculation formula is as follows:

[0093]

[0094]

[0095] In the formula, The prior evolution error covariance matrix is ​​dimensionless and represents the uncertainty of the prior predicted state. It is obtained by updating the posterior error covariance matrix and the process noise covariance matrix of the previous period. This is a state-mapping observation matrix, dimensionless, used to map state space vectors to observation space, achieving dimensional alignment between prior prediction data and field verification observation data. The matrix dimension is the observation vector dimension × state vector dimension, with diagonal elements being 1 and off-diagonal elements being 0. This is the combined uncertainty matrix of the sensor and the manual verification, which is dimensionless and represents the measurement error of the observed data; Let be the state correction gain matrix corresponding to beam socket node b at time t, i.e., the unscented Kalman filter gain, used to minimize the mean square error of state estimation; The observation residual is the difference between the observed values ​​verified on-site and the observed values ​​predicted by the model. Let be the posterior estimate of the state of node b in the beam socket at time t. The superscript + represents the posterior estimate after observation correction, which serves as the update benchmark for subsequent periodic prediction. It is an identity matrix with the same dimensions as the error covariance matrix; The updated posterior error covariance matrix is ​​used for prior error prediction in the next period. Through the above unscented Kalman filtering steps, closed-loop correction of the uncertainty shift in the decay model is achieved, shielding long-term sensor drift errors and continuously optimizing the estimation accuracy of node states. When the filter diverges, i.e., when the observation residuals are greater than 3 times the observation standard deviation for 3 consecutive periods, the filter error covariance matrix and state vector are reset to restore the initial healthy state.

[0096] The expected risk evolution of each alternative maintenance scheme under the updated baseline is converted into the nodal failure proximity of the quantified ultimate failure probability. The alternative maintenance schemes include five categories: no intervention, surface anti-corrosion treatment, crack grouting reinforcement, beam end embedding reinforcement, and complete replacement of the timber beam, numbered from m=0 to m=4 respectively. The calculation formula is as follows:

[0097] In the formula, This is the number for the alternative maintenance plan; Future time points for full lifecycle accounting, in years; For alternative maintenance plan m, the future time nodes The corresponding node global propagation risk evolution sequence is obtained by extrapolation based on the updated state posterior estimate and the time-varying decay mechanics model. For future time nodes The corresponding median risk of all building components is used as the benchmark for normal risk. It is a dimensionless divide-by-zero constant with a value of 1e-8; As for alternative maintenance plan m, future time nodes The corresponding node failure proximity is dimensionless and ranges from 0 to 1. The degree of risk deviation is mapped to the probability of ultimate failure through the Sigmoid function. The closer the value is to 1, the closer the node is to the ultimate failure state.

[0098] The cost of pre-construction monitoring, current construction and repair costs, and anticipated secondary disaster losses driven by the approach of node failures are comprehensively calculated. A full life-cycle economic model is derived through discounting the time value of money, and the reinforcement method corresponding to the minimum value is selected as the final decision. The time span for the full life-cycle accounting is set as the remaining design service life of the building, and the discount rate is the industry benchmark rate of return. The calculation formula is as follows:

[0099] In the formula, The extreme time endpoints for full lifecycle accounting are expressed in years. As an alternative maintenance plan m, the timeline is as follows: The corresponding pre-monitoring cost, in yuan; As an alternative maintenance plan m, the timeline is as follows: The corresponding current construction and repair costs are expressed in yuan. As an alternative maintenance plan m, the timeline is as follows: The corresponding cost of secondary disasters caused by node failure, in yuan; The expected failure loss cost is the product of the failure probability and the failure loss. The discount rate parameter is dimensionless and adopts the benchmark rate of return for the construction industry of 4%. The time value of money discount factor discounts the costs of future periods to the present moment; The total lifecycle cost of alternative maintenance option m, expressed in yuan; The optimal maintenance solution is the one that minimizes the total cost over the entire lifecycle, which is the reinforcement and maintenance method that is the final decision output.

[0100] Based on the decision results of the optimal maintenance plan, the parameters of the structural time-varying decay mechanical model and the health status baseline are updated for the next cycle, realizing closed-loop management of the entire life cycle of monitoring, early warning, decision-making and maintenance.

[0101] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A building structure health monitoring and early warning system based on the Internet of Things, comprising multi-source sensors, a trusted data base, and an asset management platform, characterized in that, Configured for execution: The wall temperature and humidity, air temperature and humidity, and wall-floor acceleration signals collected by the multi-source sensors are acquired. After outlier removal by dividing the data window according to the structural environment period, the data is written into the trusted data base. The relative water vapor gradient between the wall and indoor air at the beam-socket joint is extracted. The water vapor sealing effect of the beam-socket geometry is combined with the historical time-domain attenuation superposition calculation to extract the characteristic quantity of beam-socket humid heat accumulation that characterizes the harm of moisture retention. The displacement is recovered by frequency domain integration of the acceleration signal of the wall and floor slab, the rigid body rotation component of the wall is stripped to filter out the overall motion interference, the relative slip displacement of the actual connection misalignment is extracted, and then the proportion of local slip energy consumption that characterizes the depth of connection loosening is calculated. Based on the evolution time sequence of the characteristic quantity of wet heat accumulation in the beam socket and the proportion of local slip energy consumption, the response time difference under the maximum correlation matching between the two is found, and the slip hysteresis evolution index characterizing the friction degradation of the wooden beam end is obtained. By integrating the relative slip displacement and the acceleration signal, a dynamic equilibrium equation is constructed to obtain the stiffness degradation parameter characterizing the bearing capacity of the node. Combined with the slip hysteresis evolution index, the continuous degradation risk amount deviating from the historical normal benchmark is calculated. Based on the continuous degradation risk of each node, the global propagation risk is inferred. The risk level interval is divided by dynamic clustering to obtain the early warning level. Based on this, resource scheduling instructions for targeted inspection and reinforcement are generated and output to the asset management platform.

2. The IoT-based building structure health monitoring and early warning system according to claim 1, characterized in that, The step of writing the outlier stripping data into the trusted data base includes: Extract the statistical central value of each monitoring data item within the data window; Calculate the absolute dispersion of each monitoring data point from the statistical central value to eliminate background meteorological noise interference caused by environmental baseline drift; The monitoring data is normalized and mapped using the absolute dispersion, and a zero-prevention constant is added to maintain the convergence of the underlying calculation, resulting in a standardized state quantity stripped of environmental trends. The historical state hash chain, sensor hardware identifier, acquisition time tag, the standardized state quantity and device operating status parameters are fused at the underlying level. The current period storage block features with tamper-proof properties are generated by cryptographic one-way hash transformation and written into the trusted data base to realize multi-source data chain traceability.

3. The IoT-based building structure health monitoring and early warning system according to claim 1, characterized in that, The steps for extracting the characteristic quantities of the humid heat accumulation in the beam socket include: Based on the wall temperature and humidity and the air temperature and humidity, the internal pore water vapor pressure and the indoor environment water vapor pressure are calculated respectively. The difference between the two is used to eliminate the substrate interference under the same temperature and humidity conditions inside and outside, and the relative water vapor gradient that drives moisture migration is obtained. Autocorrelation time-domain evolution analysis was performed on the relative water vapor gradient to extract the hygrothermal memory period, which characterizes the lag in natural evaporation of water. By screening positive gradient segments of water vapor infiltration, a decay and forgetting mechanism based on the aforementioned humid heat memory cycle is introduced for time-domain cumulative calculation to obtain the initial humid heat accumulation amount. The system ledger was used to obtain the wall-embedded constraint ratio of the wooden beam end, the heat dissipation space ratio of the wall recess, and the airflow exchange ratio of the mortar joint. These were then nonlinearly fused to obtain the local structural water vapor sealing coefficient, which reflects the local structural water retention capacity. By using the local structural moisture sealing coefficient to weight the initial amount of humid heat accumulation with a risk factor, the characteristic quantity of humid heat accumulation in the beam cavity is obtained.

4. The IoT-based building structure health monitoring and early warning system according to claim 1, characterized in that, The steps for extracting the proportion of energy consumed by the local slip include: During the frequency domain integration of the acceleration signal, a frequency offset suppression constraint term is introduced through adaptive cross-validation to eliminate the integration drift distortion caused by low-frequency noise accumulation, and to obtain the global displacement of the floor and the global displacement of the wall. The projection of the global displacement of the wall in the contact normal direction is extracted, and the spatial geometric lever arm is derived by combining the acceleration signal to calculate the pure boundary translational moment that eliminates the local overturning error of the wall. Subtract the pure boundary translational moment from the contact surface projection of the global displacement of the floor slab, and strip away the overall coordinated deformation of the structure to obtain the relative slip displacement; The kinetic energy time integral corresponding to the velocity sequence of the relative slip displacement is calculated to obtain the local slip dissipation energy characterizing the friction interface. The proportion of energy dissipated by local slip is obtained by dividing the total kinetic energy of the structure's response to correct for fluctuations in the input energy.

5. The IoT-based building structure health monitoring and early warning system according to claim 1, characterized in that, The steps for deriving the slip hysteresis evolution index include: The static DC components of the characteristic quantity of moisture heat accumulation in the beam socket and the proportion of energy consumption due to local slip are filtered out respectively, and the AC sequences of moisture heat change and energy consumption change that characterize dynamic fluctuations are extracted. A series of time-shifting operations are applied to the energy consumption change AC sequence to construct a candidate time delay set characterizing the physical response hysteresis characteristics; The cross-correlation coupling integral between the heat and humidity change sequence and the translated energy consumption change sequence is calculated. The translation step size that maximizes the absolute value of the integral is located as the response time difference from water vapor intrusion to loss of friction under the physical disaster mechanism. The maximum cross-correlation coupling integral is extracted, and numerical normalization is performed using the variance of the autocorrelation energy of the two sequences to eliminate absolute amplitude scale interference. Then, the hysteresis penalty amplification is performed in combination with the response time difference to obtain the slip hysteresis evolution index that characterizes the hidden degradation depth.

6. The IoT-based building structure health monitoring and early warning system according to claim 1, characterized in that, The steps for calculating the continuous degradation risk amount deviating from the historical normal baseline include: The equivalent mass of the floor slab bearing the beam socket is obtained by using the system configuration ledger. The inertial force is solved by combining the relative difference in acceleration between the floor slab and the wall, and the equivalent horizontal shear force at the interface is restored. Based on the hysteresis work envelope area of ​​the equivalent horizontal shear force at the interface and the relative slip displacement, the equivalent shear stiffness characterizing the node's resistance to slippage is derived. By comparing the current equivalent shear stiffness with the healthy stiffness sequence of the undamaged structure's historical periods, the stiffness degradation parameter characterizing the loss of structural resistance is extracted. The slip hysteresis evolution index, the stiffness degradation parameter, and the crack propagation rate aggregated from the system ledger are fused into a multi-dimensional set of node health status. Calculate the multidimensional spatial variation distance of the state set relative to the centroid of the historical normal service distribution to eliminate false alarms caused by a single abnormal data jump; The continuous degradation risk quantity is obtained by multiplying the multidimensional spatial variation distance by the component failure risk weight that characterizes the load-bearing importance of the node.

7. The IoT-based building structure health monitoring and early warning system according to claim 1, characterized in that, The steps for outputting the resource scheduling instruction include: Based on the structural spatial topology, the sensitivity gradient of the stiffness variation of the source beam-socket node to the displacement response of the associated support member is calculated, and the mechanical transmission influence weights are constructed. The weight of the mechanical transmission effect and the continuous degradation risk of the source beam socket node are spatially weighted and summed to obtain the global propagation risk of the overall domino failure tendency of the structure. Unsupervised dynamic clustering is performed on the global propagation risk values ​​of all building components to find the hierarchical interval partition boundary that maximizes intra-class convergence and inter-class dispersion, so as to suppress the false alarm phenomenon caused by seasonal building-wide synchronization parameter drift. The warning level is determined by comparing the relative severity of each component within its risk level range. Based on the proportion of early warning levels and the risk weight of component failure, a differentiated material allocation rate is calculated, and the resource scheduling instructions that guide preventive temporary support and non-destructive testing of key parts are generated.

8. The IoT-based building structure health monitoring and early warning system according to claim 1, characterized in that, The system is also configured to acquire on-site verification observation data to optimize lifecycle decision-making, including the following steps: Based on the structural time-varying decay mechanics model and the state of the previous cycle, theoretical state prior prediction data are calculated by extrapolating the time series. The measured moisture content of wood, ultrasonic defect depth and measured crack width recorded by manual inspection are used to form a real feedback source for on-site verification to shield the long-term drift error of the sensor. Based on the observation residuals of the actual feedback source of the on-site verification and the state prior prediction data, the state correction gain is generated by the dynamic optimal estimation filtering algorithm, the uncertainty offset of the decay model is corrected in a closed loop, and the state posterior estimate is obtained to update the subsequent prediction benchmark. The expected risks of each alternative maintenance scheme under the updated baseline are converted into the node failure approximation of the quantified limit of the probability of destruction. The costs of pre-monitoring, current construction and repair, and expected secondary disaster losses driven by the proximity of failure at the aforementioned nodes are comprehensively calculated. A full life-cycle economic model is derived by discounting the time value of money, and the reinforcement method corresponding to the minimum value is selected as the final decision.

Images