Intelligent analysis and test method for seismic performance of curtain wall structure based on vibration monitoring

Abstract

This invention belongs to the field of building structure testing technology, and relates to an intelligent analysis and testing method for the seismic performance of curtain wall structures based on vibration monitoring. The method includes: collecting and preprocessing multi-source synchronous excitation and response data to obtain dynamic acceleration sequences and standard wind pressure sequences; extracting structural dynamic features from the dynamic acceleration sequences to obtain high-frequency and low-frequency vibration energy; processing the standard wind pressure sequences to obtain average wind pressure and wind pressure abrupt change range; constructing a physical response index that integrates nonlinear hysteresis effects and using this index to reconstruct the multi-dimensional spatial feature data through clustering distance to obtain a physical state weighted distance; performing adaptive benchmark calibration based on the physical state weighted distance and outputting seismic performance degradation assessment results. This invention solves the problem of false alarms in strong winds and missed alarms in light winds that easily occur in anomaly detection algorithms under complex wind pressure and meteorological conditions.

Description

Technical Field

[0001] This invention belongs to the field of building structure testing technology, specifically relating to an intelligent analysis and testing method for the seismic performance of curtain wall structures based on vibration monitoring. Background Technology

[0002] As the core external envelope of modern high-rise buildings, the overall seismic and wind pressure resistance of curtain walls highly depends on the pre-embedded flexible metal connection nodes between the main curtain wall keel and the main building structure. Over a decades-long service life, the repeated coupling effects of high-frequency temperature fluctuations, continuous directional wind loads, and even geological micro-seismic events in the external environment can easily induce material fatigue relaxation and fastener degradation at concealed connection nodes, leading to physical hazards. Since these metal components are often concealed within narrow cavities behind the external decorative panels, traditional surface inspection methods relying on manual suspended platforms are not only risky but also completely unable to detect subtle mechanical variations in the internal structure. Therefore, continuous vibration monitoring using sensing technology has become an essential technical means to ensure the safe operation of curtain walls.

[0003] The current common technical approach in the engineering field is to directly attach high-frequency triaxial accelerometers to key stress nodes of the curtain wall to obtain structural vibration response data. After the front-end hardware completes data acquisition, the back-end server typically uses signal processing algorithms to extract characteristic frequency band data. The extracted features are then input into an unsupervised machine learning model for anomaly clustering and detection based on spatial distance distribution. While this type of purely data-driven anomaly detection algorithm demonstrates good data generalization ability in the laboratory, it has certain technical limitations in real outdoor meteorological environments.

[0004] Specifically, existing anomaly detection algorithms based on data clustering face the problem of benchmark drift and decoupling failure of physical features in practical engineering. In real-world environments, curtain wall structures are constantly exposed to highly unstable random wind pressure excitation. When external wind pressure increases sharply, even perfectly healthy and stable curtain wall nodes will experience huge vibration amplitudes and high-frequency energy surges due to the overall building dynamics response. Conversely, when external wind pressure is extremely low, the abnormal vibration characteristics of defective nodes with latent fatigue relaxation are severely masked and extremely inconspicuous.

[0005] Existing algorithms such as the local outlier factor rely solely on the absolute geometric distance between data points in a multidimensional feature space to calculate density and identify anomalies. This approach severs the physical mapping relationship between external wind pressure excitation energy and internal structural vibration response. This failure of mechanical decoupling at the feature level inevitably leads to the algorithm misclassifying normal macroscopic severe responses as node relaxation and degradation under strong gusts, while easily missing implicit structural damage that has already occurred under light winds. This makes it impossible to meet the rigid engineering requirements of continuous, high-precision testing under variable weather conditions. Summary of the Invention

[0006] The purpose of this invention is to propose an intelligent analysis and testing method for the seismic performance of curtain wall structures based on vibration monitoring, in order to solve the technical problem that existing anomaly detection algorithms fail to detect strong winds and miss light winds due to the failure of physical feature decoupling under complex and variable wind load environments.

[0007] Therefore, the technical solution of the intelligent analysis and testing method for the seismic performance of curtain wall structures based on vibration monitoring provided by this invention is as follows: Intelligent analysis and testing methods for the seismic performance of curtain wall structures based on vibration monitoring include: Collect and preprocess multi-source synchronous excitation and response data to obtain dynamic acceleration sequences and standard wind pressure sequences; Structural dynamics features are extracted from the dynamic acceleration sequence to obtain high-frequency and low-frequency vibration energy; the standard wind pressure sequence is processed to obtain the average wind pressure and the wind pressure abrupt change range. A physical response index incorporating nonlinear hysteresis effects is constructed based on the high-frequency vibration energy, low-frequency vibration energy, average wind pressure, and wind pressure abrupt change range. By combining the physical response index with the multidimensional spatial feature data, clustering distance reconstruction is performed to obtain the physical state weighted distance; Adaptive benchmark calibration is performed based on the weighted distance of the physical state, and the seismic performance degradation assessment result is output.

[0008] The beneficial effects are as follows: This invention conducts a full-process analysis by integrating multi-source external excitation and internal response data, establishing a physical mapping relationship between external wind pressure excitation and internal structural vibration response. This effectively avoids analytical biases caused by physical feature decoupling failures and fundamentally overcomes the shortcomings of traditional algorithms that are susceptible to misjudgments due to complex meteorological environments. Furthermore, this invention constructs a physical response index that integrates nonlinear hysteresis effects based on structural dynamics and wind pressure characteristics. This allows for deep integration of the structure's own vibration characteristics with external wind pressure environmental characteristics, enabling the analysis of the curtain wall structure's state to closely match its actual mechanical performance and accurately capture subtle mechanical changes in the curtain wall structure. Simultaneously, by combining the physical response index with clustering distance reconstruction of multi-dimensional spatial feature data, the analysis logic of multi-dimensional spatial feature data is matched with the actual physical state of the curtain wall structure, making subsequent state analysis more scientific. Based on this, adaptive benchmark calibration is performed, and seismic performance degradation assessment results are output, achieving accurate determination of the seismic performance degradation state of the curtain wall structure and significantly improving the accuracy of seismic performance degradation assessment of concealed nodes under complex meteorological conditions.

[0009] Furthermore, the process of acquiring and preprocessing multi-source synchronous excitation and response data to obtain dynamic acceleration sequences and standard wind pressure sequences includes: Triaxial acceleration data were obtained at typical connection nodes of the curtain wall, and surface wind pressure data were obtained at the same elevation position on the outer surface of the curtain wall. A fixed time window length is set, and the triaxial acceleration data and the surface wind pressure data are synchronously captured with the set time window as the period to obtain acceleration time series data and wind pressure time series data. A high-pass digital filter is used to filter the acceleration time series data to remove the DC bias signal caused by gravity components and temperature drift, thereby obtaining the dynamic acceleration sequence. The wind pressure time series data is smoothed by moving average filtering to obtain the standard wind pressure sequence.

[0010] The beneficial effects are: by synchronously collecting acceleration and wind pressure data at typical locations on the curtain wall, the time consistency between the excitation source and the response end data is ensured. Then, by selectively filtering the DC bias signal in the acceleration data and the transient electromagnetic noise in the wind pressure data, a low-level dynamic dataset with no bias deviation and high fidelity can be obtained.

[0011] Furthermore, the dynamic acceleration sequence is processed using a variational mode decomposition algorithm to analyze the original non-stationary vibration signal into multiple intrinsic mode function components with specific center frequencies. The extracted intrinsic mode function components are divided into low-frequency principal mode components and high-frequency principal mode components according to their center frequencies. The root mean square energy value of the low-frequency principal mode component within a set time window is calculated and denoted as the low-frequency vibration energy. The root mean square energy value of the high-frequency principal mode component within the time window is calculated and denoted as the high-frequency vibration energy. The arithmetic mean of the absolute values ​​of all wind pressure sampling points within the time window is calculated and denoted as the average wind pressure. The absolute value of the difference between the maximum and minimum values ​​of the standard wind pressure sequence within the time window is calculated and denoted as the wind pressure abrupt change range.

[0012] The beneficial effects are: by analyzing non-stationary vibration signals into modal components in different frequency domains and completing the high-low frequency division through variational mode decomposition, it is possible to separate the characteristic measurements of macroscopic forced displacement and microstructural attenuation of buildings, and at the same time accurately calculate the external excitation characteristics related to wind pressure, and effectively extract the dynamic characteristics that reflect the overall and local vibration state of the curtain wall structure.

[0013] Furthermore, the formula for calculating the physical response index is as follows:

[0014] In the formula, This represents the physical response index; This represents the high-frequency vibration energy; This represents the low-frequency vibration energy; Indicates the energy margin of the sensor's noise floor; Represents the base constant of the natural logarithm; This indicates a very large variation in wind pressure. This represents the average wind pressure; This represents the standard atmospheric pressure reference fluctuation constant.

[0015] The beneficial effects are as follows: by establishing a comprehensive energy evaluation system that integrates nonlinear hysteresis effects, the ratio of high-frequency to low-frequency vibration energy can be used to accurately track the internal micro-collision characteristics caused by fastener loosening. By using logarithmic calculations to buffer the extreme impact of transient gusts, the interference phenomenon of synchronous amplification of the overall structural energy caused by the rise of absolute wind load can be eliminated. The reference drift interference caused by environmental and meteorological changes can be offset from a mathematical perspective. The physical mapping relationship between external wind pressure excitation and internal structural vibration response can be established, and a single monitoring parameter can be used to capture the subtle mechanical defects of the curtain wall with high sensitivity.

[0016] Furthermore, before obtaining the weighted distance of the physical state, the process also includes: Extract the frequency domain peak features within a set time window, and combine the frequency domain peak features with the high-frequency vibration energy and the low-frequency vibration energy to form an initial feature vector; The initial feature vector is dimensionless using a standardization process to generate standard feature sample data. The standard Euclidean distance between corresponding samples is calculated based on the standard feature sample data.

[0017] Furthermore, the formula for calculating the weighted distance of the physical state is:

[0018] In the formula, This represents the data samples corresponding to the historical time window. Data samples corresponding to the current time window The physical state-weighted distance between them; The data sample represents With the data sample The standard Euclidean distance between them; The data sample represents The corresponding physical response indicators; The data sample represents The corresponding physical response indicators; This represents an exponential function with the natural constant as its base.

[0019] The beneficial effects are: the formula can push out the cluster center of normal samples away from the potential hidden dangers of abnormal data samples, amplify the spatial segmentation sensitivity of local outlier detection in strong background noise environment, remove the abnormal hidden danger samples from the cluster group of normal data to the far end, significantly optimize the classification isolation of feature space, and make the distance measurement more in line with the actual physical state of the curtain wall structure.

[0020] Furthermore, the adaptive benchmark calibration based on the weighted distance of the physical state and the output of the seismic performance degradation assessment result include: Based on the physical state weighted distance, calculate the k-th nearest neighbor distance between the feature sample and its neighboring samples; Compare the physical state weighted distance between the feature sample and other samples in the neighborhood with the k-th nearest neighbor distance, and determine the maximum value of the two as the reachable distance; The local reachability density is calculated based on the reachability distance, and a local outlier factor score sequence is generated based on the local reachability density.

[0021] Furthermore, the adaptive benchmark calibration based on the weighted distance of the physical state and the output of the seismic performance degradation assessment result also includes: In the baseline self-learning mode of the first month cycle, assuming that all nodes are in good connection status, the arithmetic mean of the local outlier factor score sequence of all time windows during the self-learning phase is calculated and denoted as the health mean. The standard deviation of the local outlier score sequence for all time windows within the self-learning phase is calculated and denoted as the health standard deviation.

[0022] Furthermore, the adaptive benchmark calibration based on the weighted distance of the physical state and the output of the seismic performance degradation assessment result also includes: During the routine engineering monitoring phase, the local outlier score sequence for the current time window is acquired in real time. When a monitoring node at a certain location has outlier scores exceeding the sum of the healthy mean and three times the healthy standard deviation for three consecutive time windows, the corresponding monitoring node is deemed to have experienced physical fatigue relaxation.

[0023] Furthermore, after determining that the corresponding monitoring node has experienced physical fatigue relaxation, the method also includes: sending a structural seismic performance deterioration warning command to the engineering operation and maintenance terminal, and outputting the three-dimensional physical coordinates of the corresponding monitoring node that has deteriorated.

[0024] The beneficial effects of this invention are: This invention effectively solves the problem of false alarms in strong winds and missed alarms in light winds caused by benchmark drift in pure data anomaly detection algorithms under complex wind pressure and meteorological conditions. At the same time, it overcomes the defect of traditional pure data clustering algorithms being easily affected by the environment and causing misjudgments. It improves the accuracy of seismic performance degradation assessment of concealed nodes of curtain walls under complex meteorological conditions, realizes efficient monitoring and operation and maintenance of the seismic performance of curtain wall structures, and ensures the safe operation of curtain wall structures. Attached Figure Description

[0025] Figure 1 This is a flowchart of the intelligent analysis and testing method for the seismic performance of curtain wall structures based on vibration monitoring, provided in an embodiment of the present invention. Figure 2 It is a feature space data point distribution map using existing technology algorithms; Figure 3 This is a feature space data point distribution map after adopting the technical solution of this invention; Figure 4 This is a two-dimensional mapping diagram of the distribution of seismic resistance nodes on the curtain wall facade provided in an embodiment of the present invention. Detailed Implementation

[0026] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0027] like Figure 1 As shown in this embodiment, the intelligent analysis and testing method for the seismic performance of curtain wall structures based on vibration monitoring includes the following steps: S1. Collect multi-source synchronous excitation and response data and perform preprocessing to obtain dynamic acceleration sequence and standard wind pressure sequence.

[0028] Specifically, high-frequency triaxial accelerometers are installed at the metal connection nodes of the typical load-bearing main keel of the curtain wall using rigid bonding or bolt fastening to acquire triaxial acceleration data reflecting the microscopic vibrations of the structure. To measure the external kinetic energy input that induces structural vibrations, high-precision micro-differential pressure sensors are deployed at corresponding elevations on the outer surface of the curtain wall to continuously acquire transient environmental wind pressure data, which is then used as the surface wind pressure data. The system sets a fixed high-frequency sampling rate at the hardware control layer to ensure that the surface wind pressure data (the excitation source) and the triaxial acceleration data (the response end) are strictly aligned in timestamps.

[0029] The system's main control unit further sets a fixed time window length as the data stream segmentation standard. Using this set time window as the loop period, it synchronously truncates and reads the triaxial acceleration data and surface wind pressure data from the hardware cache, thereby obtaining corresponding and synchronous acceleration and wind pressure time series data. This looping truncation mechanism based on a set time window can transform infinitely long continuous environmental monitoring data into finite-length discrete computational data blocks, providing a standardized large-sample input sequence for subsequent structural dynamics analysis and frequency domain conversion.

[0030] For the extracted acceleration time-series data, a high-pass digital filter configured with a specific cutoff frequency is invoked to perform filtering. In real outdoor meteorological environments, sensors attached to curtain wall nodes are inevitably affected by the constant component of Earth's gravity, and the temperature-sensing elements inside the sensors are prone to temperature drift due to high-frequency temperature alternation between day and night. These objective physical phenomena directly superimpose a very strong low-frequency DC bias signal onto the original acceleration signal. This step, by using a high-pass digital filter to filter the data, can block and filter out the DC bias signal caused by the gravity component and temperature drift, ultimately restoring the true dynamic acceleration sequence containing only high-frequency alternating information, thus eliminating baseline offset errors introduced by non-structural dynamic factors.

[0031] For the extracted wind pressure time series data, a moving average filter is used for smoothing to suppress abrupt disturbances. The wind field environment on the exterior surface of the curtain wall is extremely complex. Local air eddies around the building facade and transient electromagnetic noise caused by the external environment can lead to irregular high-frequency spikes in the original wind pressure data. Iterative calculation using a moving average filter algorithm can effectively suppress transient electromagnetic noise caused by air eddies while preserving the macroscopic trend of wind load changes and effective gust extremes, thus generating a stable standard wind pressure sequence.

[0032] Through high-precision time synchronization acquisition from multiple hardware sources and the aforementioned targeted frequency band filtering and smoothing preprocessing, the system successfully acquired a high-fidelity underlying dynamic dataset with no DC bias deviation.

[0033] S2. Extract structural dynamic features from the dynamic acceleration sequence to obtain high-frequency vibration energy and low-frequency vibration energy; process the standard wind pressure sequence to obtain the average wind pressure and the wind pressure abrupt change range.

[0034] This step aims to perform frequency domain stripping of the hybrid structural dynamic response and evaluate the excitation intensity of external wind loads. Specifically, the preprocessed dynamic acceleration sequence is imported into the variational mode decomposition (VMD) algorithm module for deep signal analysis. The VMD algorithm, through iterative solutions within a variational framework, adaptively deconstructs and separates the original highly non-stationary time-domain vibration signal into multiple eigenmode function components with non-overlapping center frequencies. This adaptive decomposition mechanism effectively avoids the mode aliasing problem easily caused by traditional fixed-band filtering methods and can accurately reconstruct the implicit frequency domain characteristics in the signal.

[0035] Based on the building dynamics response mechanism, the system directly divides all extracted intrinsic mode function components into low-frequency principal mode components and high-frequency principal mode components according to the numerical value of the center frequency. The reason for this frequency domain physical mapping is that the low-frequency principal mode components physically represent the macroscopic overall elastic displacement phenomenon of the curtain wall following the main building under wind load, and its energy fluctuation is strongly positively correlated with the magnitude of external wind pressure; while the high-frequency principal mode components directly point to the microscopic local high-frequency vibrations induced by material fatigue relaxation, bolt tightening force degradation, and structural friction and collision within the metal connection nodes, and are direct reflection parameters of the substantial decline in the structural health status.

[0036] Based on the principle of time-domain energy equivalence, the root-mean-square (RMS) energy value of the low-frequency dominant mode component within a set time window is calculated and recorded as low-frequency vibration energy. Using the same numerical calculation steps, the system calculates the RMS energy value of the high-frequency dominant mode component within the same time window and records it as high-frequency vibration energy. Through RMS energy statistical operations, the system can convert the rapidly fluctuating transient vibration amplitude within the time window into a stable energy scale with a cumulative effect, providing a stable basic physical quantity for subsequently establishing an energy hysteresis dissipation index.

[0037] In terms of external excitation parameter extraction, the system calculates the arithmetic mean of the absolute values ​​of all wind pressure sampling points within the current time window and records it as the average wind pressure. Using absolute values ​​in the calculation aims to prevent anomalies caused by alternating positive and negative wind pressure due to air vortices on the building facade surface, which could lead to the cancellation of mean values. This ensures that the calculation results accurately reflect the overall macroscopic environmental wind pressure intensity within that time period. Simultaneously, the system retrieves the maximum and minimum values ​​of the standard wind pressure sequence within the aforementioned time window, calculates the absolute value of the difference between them, and records it as the wind pressure abrupt change range. This wind pressure abrupt change range parameter is used to evaluate the rate of change of the impact load gradient exerted on the curtain wall facade by extreme gusts or transient strong winds.

[0038] By decomposing complex physical vibration signals into different frequency bands and performing root mean square energy statistics, and by accurately calculating the average intensity and extreme values ​​of external meteorological excitation, the system successfully separates the characteristic measurements of macroscopic forced displacement and microstructural attenuation of buildings, solving the problem of high coupling between environmental interference and structural response in traditional time-domain analysis.

[0039] S3. Construct a physical response index that incorporates the nonlinear hysteresis effect based on the high-frequency vibration energy, low-frequency vibration energy, average wind pressure, and wind pressure abrupt change range.

[0040] For metal connection nodes in a healthy state with adequate fastening force, their structure exhibits typical linear stiffness characteristics. When external wind pressure increases drastically, the macroscopic low-frequency vibration energy and microscopic high-frequency vibration energy of the node itself will be amplified synchronously and linearly, and the transmission ratio between the two will always remain near a stable safety baseline. Once the connection node experiences fatigue degradation and mechanical gaps due to long-term loads, when it encounters a sudden, abrupt impact from a gust of wind, the stressed components will undergo nonlinear secondary collisions within the gaps. This microscopic mechanical variation will cause a nonlinear surge in high-frequency vibration energy relative to low-frequency vibration energy, exhibiting a significant energy hysteresis and dissipation effect.

[0041] Based on the above physical and dynamic characteristics, this step constructs a physical response index that incorporates nonlinear hysteresis effects, and its calculation formula is as follows:

[0042] In the formula, Indicates physical response index; Represents high-frequency vibration energy; This represents low-frequency vibration energy; This parameter represents the sensor's noise floor energy margin and is a calibration constant. It represents the base constant of the natural logarithm, and its value is a natural constant; This indicates a very poor sudden change in wind pressure; Indicates average wind pressure; This represents the standard atmospheric pressure reference fluctuation constant.

[0043] The formula comprises two core components: the energy transfer ratio component and the excitation abrupt response component. The first term, the energy transfer ratio component, is calculated by dividing the high-frequency vibration energy by the sum of the low-frequency vibration energy and the sensor's intrinsic noise energy margin. This term aims to eliminate the interference of energy synchronous amplification caused by the increase in the absolute value of the overall wind load. Simultaneously, the introduction of the sensor's intrinsic noise energy margin, a fixed constant parameter, ensures that the denominator is not zero even in extremely windless and vibration-free environments, thus preventing program anomalies such as computational system overflow and crashes from the underlying algorithm architecture.

[0044] The second term of the formula represents the excitation abrupt change response component, calculated by dividing the wind pressure abrupt change range by the sum of the average wind pressure and the standard atmospheric pressure reference fluctuation constant, and then combining this with the natural constant to obtain the natural logarithm. Introducing the natural logarithmic function effectively reflects the damping buffering effect of gust impacts on the nonlinear excitation of the structure in aerodynamics. Due to the smoothing characteristics of the logarithmic function, even if the absolute value of the wind pressure is extremely large, resulting in both the average wind pressure and the wind pressure abrupt change range being extremely high, extreme changes in external meteorological conditions will not cause divergence in the calculation results. Analysis of the formula shows that when the node is in a healthy state, regardless of the changes in the gust abrupt change range, the energy transfer ratio component remains in a very low baseline range, and the calculated physical response index remains in a low steady state. When physical gaps exist within the node, small wind pressure abrupt changes will excite internal collisions, causing the energy transfer ratio component to instantly amplify by more than a factor of two, thus resulting in a significant peak mapping in the physical response index.

[0045] By integrating physical collision mechanisms and aerodynamic characteristics, the algorithm mathematically counteracts the benchmark drift interference caused by environmental and meteorological changes, achieving high-sensitivity capture of subtle mechanical defects in curtain walls by a single monitoring parameter and accurate restoration of data dimensions.

[0046] S4. Combine the physical response index to reconstruct the multidimensional spatial feature data by clustering distance to obtain the physical state weighted distance.

[0047] Before using the distance metric algorithm, this step performs dimensionless processing on various heterogeneous data. Specifically, frequency domain peak features within a defined time window are extracted and sequentially concatenated with previously extracted high-frequency and low-frequency vibration energy to form an initial feature vector containing multi-dimensional dynamic information. Subsequently, the initial feature vector is dimensionless using a standardization criterion. Through mathematical transformation, the features of each dimension within the initial feature vector are uniformly converted into standard feature sample data with a mean of 0 and a variance of 1. This data alignment operation avoids weight biases caused by different physical dimensions in spatial metric calculations. Based on this, the standard Euclidean distance between corresponding samples is calculated using the standard feature sample data.

[0048] In real-world outdoor weather conditions, when faced with complex and variable wind loads, even perfectly healthy data samples may exhibit severe divergence in the feature space due to significant differences in the amplitude of external forced vibrations. To overcome the baseline drift defect caused by this purely data-driven algorithm, this step introduces the calculated physical response index with an exponential penalty to comprehensively reconstruct the underlying multidimensional spatial geometric distance measurement logic. Combining the physical response index with the multidimensional spatial feature data for clustering distance reconstruction, the formula for calculating the weighted distance of the physical state is as follows:

[0049] In the formula, This represents the data samples corresponding to the historical time window. Data samples corresponding to the current time window The physical state-weighted distance between them. This represents a data sample calculated based on standard feature sample data. With data samples The standard Euclidean distance between the data samples represents the basic geometric interval of the data samples in the original multidimensional feature space. Represents data samples The corresponding physical response indicators, at the same time Represents data samples The corresponding physical response indicators reflect the true physical health status of the building nodes. This represents an exponential function with the natural constant as its base.

[0050] The formula utilizes the nonlinear amplification property of the exponential function to implement spatial isolation. Analysis reveals that when all data samples are in a healthy state, the difference in their physical response indicators is minimal, the exponential penalty term approaches the natural value of 1, and the safe distance relationship between the relatively clustered samples remains intact in space. Conversely, if the data samples corresponding to the current time window suffer from severe fatigue relaxation defects, their physical response indicators will undergo abnormal abrupt changes, causing a surge in the absolute value of the difference between the two indicators. In this case, the huge indicator difference, amplified by the exponential function, generates an extremely high penalty weight, which is then directly multiplied by the basic standard Euclidean distance.

[0051] This step transforms the response index, which reflects the actual mechanical damage within the structure, into an exponential amplification factor of the bottom-level distance. This allows the algorithm to forcibly remove data samples with abnormal potential from the cluster of normal data and push them to the far end of the feature space, thereby optimizing the classification isolation of the multidimensional feature space and eliminating spatial interference from environmental excitation fluctuations.

[0052] S5. Perform adaptive benchmark calibration based on the weighted distance of the physical state, and output the seismic performance degradation assessment results.

[0053] This step embeds the reconstructed weighted physical state distance into the kernel of the local outlier factor algorithm. Once the algorithm starts, it calculates the k-th nearest neighbor distance between the feature sample and its neighboring samples based on the weighted physical state distance obtained from the reconstruction. Subsequently, the system executes a rigorous spatial comparison mechanism, comparing the weighted physical state distance and the k-th nearest neighbor distance between the feature sample and other samples in its neighborhood, and determining the maximum value as the reachable distance. The logic behind using the maximum value as the reachable distance is to smooth out statistical fluctuations in the underlying distances, resulting in a more stable spatial density benchmark for normal samples within the same neighborhood. Based on the acquired large batch of reachable distances, the system further reverse-engineers the local reachability density and finally generates a local outlier factor score sequence corresponding to the current time window based on the differences in the distribution gradient of the local reachability density. The larger the value of this score sequence, the deeper the physical deviation of the tested object from the healthy baseline.

[0054] To address the issue of traditional models' heavy reliance on engineers manually setting experience thresholds, the system is configured to automatically enter a baseline self-learning mode for an initial one-month period after its first on-site deployment. During this specific initial setup phase, the engineering logic assumes that all node connections are intact and meet factory safety standards. During this period, the control center continuously accumulates and rigorously statistically analyzes the local outlier score sequences across all time windows within the self-learning phase. Through statistical calculations, the arithmetic mean of all score sequences within the self-learning phase is extracted and used as the building's specific health mean. Simultaneously, data dispersion analysis is performed, calculating and statistically analyzing the standard deviation of the local outlier score sequences across all time windows within the self-learning phase, which is recorded as the health standard deviation.

[0055] After entering the routine engineering monitoring phase, the processing cabinet receives the circulating data in real time and obtains the local outlier score sequence for the current time window. The system background continuously runs a triggered out-of-bounds judgment program. When the outlier score data of a monitoring node at a critical location exceeds the sum of the healthy mean and three times the healthy standard deviation for three consecutive time windows, the program determines that the corresponding monitoring node has experienced physical fatigue relaxation. Setting up continuous judgment across multiple time windows and combining it with the statistical criterion of three times the standard deviation not only adapts to the individual background response differences of different buildings but also effectively eliminates interference from single calculation jumps caused by occasional strong environmental noise, ensuring high confidence in anomaly judgment. After determining that the corresponding monitoring node has experienced physical fatigue relaxation, the system immediately generates a hazard warning and sends a structural seismic performance degradation warning command to the engineering operation and maintenance terminal via the local area network. Simultaneously, the system automatically highlights the corresponding deteriorated monitoring node in the building information model interface on the interactive screen and outputs the node's three-dimensional physical coordinates on the macroscopic building facade. High-precision spatial coordinate mapping directly transforms abstract underlying data anomalies into visualized engineering location information, thereby guiding the on-site maintenance team to carry out targeted reinforcement work.

[0056] The following combination Figures 2 to 4 The effects of the present invention will be further explained.

[0057] like Figure 2 As shown in the figure, this diagram illustrates the distribution of data points in the feature space of existing technologies. It can be seen that, due to the lack of physical response indicators to correct the underlying spatial measurement benchmark, the data point set representing hidden dangerous defects is largely overlapped and intrudes into the data point cluster area representing healthy and safe states. The two distinct physical state samples exhibit a highly coupled diagonal distribution trend in the feature space, with no buffer zone between them that can be defined and isolated by the algorithm. This chaotic clustering result demonstrates that traditional pure data geometric distance suffers from severe physical mapping failure and judgment interference problems when facing fluctuating external environmental stimuli.

[0058] like Figure 3 As shown in the figure, the distribution of feature space data points after adopting the technical solution of this invention is illustrated. It can be seen from the figure that all nodes in a healthy state spontaneously cluster at the bottom of the horizontal axis, forming an extremely smooth and low safety baseline band. This safety baseline band does not experience any baseline drift regardless of extreme external environmental wind pressure gradient changes. Simultaneously, all defect nodes with potential local relaxation defects are forcefully stripped away by the exponential penalty term in the formula and forcibly pushed up to a very high position on the vertical axis. This reshaping mechanism creates a very clear and wide blank safety isolation band separating the normal healthy group from the abnormal defective group, improving the classification isolation of the abnormal feature space.

[0059] like Figure 4 As shown in the diagram, the macroscopic facade of the entire building is divided into coordinate grid matrices corresponding to each actual physical connection node. The diagram demonstrates that after the background algorithm completes adaptive benchmark calibration and strictly executes the boundary judgment procedure, the system can accurately filter and identify the coordinates of abnormal node grids exhibiting physical fatigue decay from thousands of conventional grid matrices representing healthy and secure foundation states. This intuitive physical coordinate mapping transformation provides the on-site maintenance team with engineering spatial location guidance for conducting targeted maintenance and high-altitude reinforcement operations.

[0060] In summary, by deeply integrating a local proximity spatial distance determination mechanism with a dynamic health parameter self-learning mode and finally providing an intuitive and clear visual early warning terminal output, this invention not only effectively eliminates the delayed early warning and missed reporting blind spots caused by traditional single static threshold determination, but also delivers a stable and reliable closed-loop safety control system for modern curtain wall maintenance projects facing highly complex and variable weather conditions.

[0061] While various embodiments of the invention have been shown and described in this specification, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention.

Claims

1. A method for intelligent analysis and testing of the seismic performance of a curtain wall structure based on vibration monitoring, characterized in that, include: Collect and preprocess multi-source synchronous excitation and response data to obtain dynamic acceleration sequences and standard wind pressure sequences; Structural dynamics features are extracted from dynamic acceleration sequences to obtain high-frequency and low-frequency vibration energy; standard wind pressure sequences are processed to obtain average wind pressure and wind pressure abrupt change range. A physical response index incorporating nonlinear hysteresis effects is constructed based on the high-frequency vibration energy, low-frequency vibration energy, average wind pressure, and wind pressure abrupt change range. By combining the physical response index with the multidimensional spatial feature data, clustering distance reconstruction is performed to obtain the physical state weighted distance; Adaptive benchmark calibration is performed based on the weighted distance of the physical state, and the seismic performance degradation assessment result is output.

2. The intelligent analysis and testing method for the seismic performance of curtain wall structures based on vibration monitoring according to claim 1, characterized in that, The process of collecting and preprocessing multi-source synchronous excitation and response data to obtain dynamic acceleration sequences and standard wind pressure sequences includes: Triaxial acceleration data were obtained at typical connection nodes of the curtain wall, and surface wind pressure data were obtained at the same elevation position on the outer surface of the curtain wall. A fixed time window length is set, and the triaxial acceleration data and the surface wind pressure data are synchronously captured with the set time window as the period to obtain acceleration time series data and wind pressure time series data. A high-pass digital filter is used to filter the acceleration time series data to remove the DC bias signal caused by gravity components and temperature drift, thereby obtaining the dynamic acceleration sequence. The wind pressure time series data is smoothed by moving average filtering to obtain the standard wind pressure sequence.

3. The intelligent analysis and testing method for the seismic performance of curtain wall structures based on vibration monitoring according to claim 1, characterized in that, The dynamic acceleration sequence is processed using a variational mode decomposition algorithm to analyze the original non-stationary vibration signal into multiple intrinsic mode function (IMF) components with specific center frequencies. The extracted IMF components are then divided into low-frequency principal mode components and high-frequency principal mode components based on their center frequencies. The root mean square (RMS) energy value of the low-frequency principal mode component within a set time window is calculated and denoted as the low-frequency vibration energy. Similarly, the RMS energy value of the high-frequency principal mode component within the same time window is also calculated and denoted as the high-frequency vibration energy. Calculate the arithmetic mean of the absolute values ​​of all wind pressure sampling points within the time window, and denot it as the average wind pressure; The absolute value of the difference between the maximum and minimum values ​​of the standard wind pressure sequence within the time window is calculated and denoted as the wind pressure abrupt change range.

4. The intelligent analysis and testing method for the seismic performance of curtain wall structures based on vibration monitoring according to claim 1, characterized in that, The formula for calculating the physical response index is as follows: In the formula, This represents the physical response index; This represents the high-frequency vibration energy; This represents the low-frequency vibration energy; Indicates the energy margin of the sensor's noise floor; Represents the base constant of the natural logarithm; This indicates a very large variation in wind pressure. This represents the average wind pressure; This represents the standard atmospheric pressure reference fluctuation constant.

5. The intelligent analysis and testing method for the seismic performance of curtain wall structures based on vibration monitoring according to claim 1, characterized in that, Before obtaining the weighted distance of the physical state, the process also includes: Extract the frequency domain peak features within a set time window, and combine the frequency domain peak features with the high-frequency vibration energy and the low-frequency vibration energy to form an initial feature vector; The initial feature vector is dimensionless using a standardization process to generate standard feature sample data. The standard Euclidean distance between corresponding samples is calculated based on the standard feature sample data.

6. The intelligent analysis and testing method for the seismic performance of curtain wall structures based on vibration monitoring according to claim 5, characterized in that, The formula for calculating the weighted distance of the physical state is: In the formula, This represents the data samples corresponding to the historical time window. Data samples corresponding to the current time window The physical state-weighted distance between them; The data sample represents With the data sample The standard Euclidean distance between them; The data sample represents The corresponding physical response indicators; The data sample represents The corresponding physical response indicators; This represents an exponential function with the natural constant as its base.

7. The intelligent analysis and testing method for the seismic performance of curtain wall structures based on vibration monitoring according to claim 6, characterized in that, The adaptive benchmark calibration based on the weighted distance of the physical state, and the output of the seismic performance degradation assessment results, include: Based on the physical state weighted distance, calculate the k-th nearest neighbor distance between the feature sample and its neighboring samples; Compare the physical state weighted distance between the feature sample and other samples in the neighborhood with the k-th nearest neighbor distance, and determine the maximum value of the two as the reachable distance; The local reachability density is calculated based on the reachability distance, and a local outlier factor score sequence is generated based on the local reachability density.

8. The intelligent analysis and testing method for the seismic performance of curtain wall structures based on vibration monitoring according to claim 7, characterized in that, The adaptive benchmark calibration based on the weighted distance of the physical state, and the output of the seismic performance degradation assessment result, also includes: In the baseline self-learning mode of the first month cycle, assuming that all nodes are in good connection status, the arithmetic mean of the local outlier factor score sequence of all time windows during the self-learning phase is calculated and denoted as the health mean. The standard deviation of the local outlier score sequence for all time windows within the self-learning phase is calculated and denoted as the health standard deviation.

9. The intelligent analysis and testing method for the seismic performance of curtain wall structures based on vibration monitoring according to claim 8, characterized in that, The adaptive benchmark calibration based on the weighted distance of the physical state, and the output of the seismic performance degradation assessment result, also includes: During the routine engineering monitoring phase, the local outlier score sequence for the current time window is acquired in real time. When a monitoring node at a certain location has outlier scores exceeding the sum of the healthy mean and three times the healthy standard deviation for three consecutive time windows, the corresponding monitoring node is deemed to have experienced physical fatigue relaxation.

10. The intelligent analysis and testing method for the seismic performance of curtain wall structures based on vibration monitoring according to claim 9, characterized in that, After determining that the corresponding monitoring node has experienced physical fatigue relaxation, the method further includes: sending a structural seismic performance deterioration warning command to the engineering operation and maintenance terminal, and outputting the three-dimensional physical coordinates of the corresponding monitoring node that has deteriorated.

Images