Building bearing capacity detection system based on Internet of Things

Abstract

The invention relates to the technical field of safety monitoring, in particular to a building bearing capacity detection system based on the Internet of Things, which comprises a strain load module, an offset gradient module, a rhythm risk module, a threshold judgment module and a map reconstruction module. According to the method, the maximum strain excursion and direction reversal characteristics of adjacent time windows are analyzed and captured through the dynamic frequency excursion gradient, the response sensitivity to microcosmic bearing capacity fluctuation is improved, a strain jump rhythm map is constructed, sudden change time point time difference, amplitude difference and direction continuity parameters are extracted, and an abnormal load section is accurately positioned. The method comprises the following steps: identifying sudden strain jump events, performing cross-level rhythm map multi-dimensional comparison by applying a dynamic time warping algorithm, screening synchronous mutation nodes to generate a risk set, performing density clustering sorting based on direction consistency parameters, reconstructing a security level topological map, and improving visual spatial relevance and dynamic evolution characteristics of a structural instability path.

Description

Technical Field

[0001] The present invention relates to the technical field of safety monitoring, and in particular to a building bearing capacity detection system based on the Internet of Things. Background Art

[0002] The field of safety monitoring technology includes a full-process technology system for sensing, transmitting and analyzing the safety of people and property and the status of key infrastructure. The core content of this technical field is to collect key data, such as stress changes, vibration frequencies, environmental parameters, etc., by deploying a variety of sensors, and transmit the data to the analysis platform in real time through the communication network to achieve continuous monitoring of the operating status of the target object. From a systematic perspective, safety monitoring technology covers the data perception layer, network transmission layer and data analysis and processing layer, and is often used in status monitoring, anomaly identification and information early warning in urban construction, power systems, bridges and tunnels and other fields. This field emphasizes the rationality of equipment deployment, the continuity of data collection and the intelligence of information processing to ensure that key systems are in a controlled and traceable monitoring state.

[0003] Among them, the building bearing capacity detection system refers to a bearing capacity monitoring device and its network integration system used in building structures. The subject of this patent is aimed at monitoring the changes in bearing capacity in building structures, and mainly covers key links such as real-time stress detection, data acquisition and transmission, and remote analysis and processing. It collects the strain values ​​and vibration frequencies of different parts under load by laying out resistive strain sensors and piezoelectric acceleration sensors in key load-bearing parts of the building, and uploads the collected data to a remote server using a low-power wireless communication module. The data is compared and analyzed in the server by setting thresholds and timing curves, thereby achieving accurate monitoring and long-term tracking of the overall and local bearing status of the building.

[0004] The existing technology uses a preset fixed threshold for static data comparison, which cannot adapt to the dynamic strain characteristics of building structures under complex loads. When the strain value shows a low-frequency and small offset, the static threshold masks the potential risk, resulting in the failure to identify the early bearing capacity attenuation. For example, in a bridge monitoring, the traditional method missed the bearing failure because it did not capture the 0.5% continuous offset gradient. The existing technology relies on single-dimensional time series curve analysis and lacks joint modeling of time difference, amplitude difference and directional continuity parameters, resulting in positioning deviations of abnormal load sections. For example, in the monitoring of a high-rise building, adjacent jump events were misjudged as a single signal. The existing technology aggregates data at a fixed period and ignores the correlation of synchronous mutations across nodes, resulting in a logical fault in the risk set screening. For example, in a tunnel monitoring, the synchronous settlement of the span vault and side wall was not associated and identified. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a building bearing capacity detection system based on the Internet of Things.

[0006] To achieve the above object, the present invention adopts the following technical solution: An Internet of Things-based building bearing capacity detection system includes: A strain load module, configured to collect real-time strain values of building nodes through distributed strain sensors, perform moving average filtering to generate reference strain load parameters, and transfer the real-time strain values and the reference strain load parameters to an offset gradient module; An offset gradient module, configured to call a gradient tracking algorithm, calculate a maximum strain offset based on the real-time strain values within adjacent time windows, generate a bearing capacity offset gradient sequence, identify nodes with continuously reverse directions, and transfer the bearing capacity offset gradient sequence and the reverse nodes to a rhythm risk module; A rhythm risk module, configured to detect strain mutation time points within a load period according to the bearing capacity offset gradient sequence and the reverse nodes, extract strain difference values, time intervals, and direction parameters between adjacent mutation time points, construct a strain jump rhythm map, mark abnormal load sections, and transfer the strain jump rhythm map to a threshold determination module.

[0007] As a further solution of the present invention, the reference strain load parameters are specifically the mean value, variance, and range after filtering. The bearing capacity offset gradient sequence includes the maximum offset, gradient direction, and time window number. The strain jump rhythm map includes the time difference between mutation time points, amplitude difference, and direction continuity parameters.

[0008] As a further solution of the present invention, the strain load module includes: A real-time data acquisition sub-module monitors the output signals of distributed strain sensors, intercepts the strain sampling values corresponding to building nodes at the corresponding time stamps, stores them in association with the sensor numbers and spatial coordinates, extracts the data of all sampling points within a fixed time period at the current moment, arranges them in chronological order as a continuous strain sequence, and generates an original strain sequence; A dynamic filtering processing sub-module, based on the original strain sequence, sets a sliding window with a fixed width, successively intercepts the data within the window at a preset time step, performs arithmetic mean operations on the strain values within each window, replaces the numerical value at the starting point position of the window with the window mean value, traverses all windows to complete the mean value replacement, and generates a smoothed strain sequence; A reference parameter generation sub-module calls the smoothed strain sequence, selects the numerical values within the initial time period, calculates its mean square deviation and fluctuation amplitude, compares the mean square deviation with a preset strain fluctuation threshold. If the mean square deviation is lower than the threshold, the mean value of the data within the initial time period is set as the reference value. If it is higher than the threshold, the time period is extended and the mean value is recalculated to generate reference strain load parameters.

[0009] As a further solution of the present invention, the offset gradient module includes: The strain window intercepting sub-module calls the real-time strain values within adjacent time windows, divides the sampling data according to the window time length, intercepts the difference between the maximum and minimum strain values within each window, arranges the differences in the order of window numbers, and generates a time window strain set; The gradient change calculating sub-module calculates the absolute value of the difference change rate between adjacent windows based on the time window strain set, extracts the top 10% of the candidate offsets after sorting the absolute values, counts the number of windows in which the candidate offsets continuously appear, compares the number with a preset offset persistence threshold, filters out the continuous segments that meet the conditions, and generates an offset gradient sequence; The reverse node identifying sub-module calls the offset gradient sequence and the reference strain load parameter, extracts the positive and negative direction signs of each offset segment in the sequence, compares the number of positive and negative alternations of the direction signs of adjacent segments, counts the node positions where the directions are continuously opposite three times, and performs reverse verification on the node positions with the load direction in the reference parameter to generate direction continuously reverse nodes.

[0010] As a further solution of the present invention, the rhythm risk module includes: The mutation parameter extracting sub-module calls the bearing capacity offset gradient sequence and the strain mutation time points within the reverse node detection load period, extracts the difference between the strain values at adjacent mutation time points, calculates the time interval between the mutation time points, records the positive and negative signs of the mutation direction parameter, associates and stores the difference, interval and direction parameter in the mutation order, and generates a mutation parameter set; The rhythm feature analyzing sub-module statistically analyzes the average time interval and the difference fluctuation range between adjacent mutation time points based on the mutation parameter set, calculates the proportion of the number of times when the direction parameters are continuously consistent, compares the average interval with a preset rhythm period threshold, filters out the mutation segments whose fluctuation range exceeds the threshold, and integrates the average interval, fluctuation range and direction consistency ratio to generate a rhythm feature sequence; The jump map generating sub-module calls the rhythm feature sequence, sorts the mutation segments from high to low according to the fluctuation range, maps the sorting result to the time axis, assigns node weights according to the direction consistency ratio, constructs a three-dimensional grid topological relationship of time-difference-direction, and converts the grid data into a format compatible with the map rendering interface to generate a strain jump rhythm map.

[0011] As a further solution of the present invention, the system further includes: A threshold determination module, which is used to adopt a dynamic time warping algorithm to compare the mutation time difference, strain difference threshold and direction synchronization rate of the current level and the previous level of the strain jump rhythm map, filter out cross-level synchronous mutation nodes to generate a bearing capacity risk node set, and transfer the bearing capacity risk node set to the map reconstruction module; The atlas reconstruction module is used to call the density clustering algorithm to sort the direction consistency parameters between nodes through the coordinates of the bearing capacity risk node set, generate a bearing capacity distribution topology atlas reflecting the safety level, and output a visual detection result; The bearing capacity risk node set includes a risk level, a synchronous timestamp, and a direction consistency coefficient; the bearing capacity distribution topology atlas is specifically a safety level label, node space coordinates, and a clustering path number.

[0012] As a further solution of the present invention, the threshold determination module includes: The hierarchical parameter alignment sub-module calls the mutation time difference parameter of the strain jump rhythm atlas of the current level and the previous level, aligns the mutation time points of the two levels along the time axis, calculates the absolute value of the time stamp difference at the corresponding time points, and adjusts the time window length. The adjusted time difference is associated and stored according to the hierarchical number to generate a cross-level time difference sequence; The synchronous threshold matching sub-module extracts the strain difference threshold and the direction synchronization rate parameter of the current level based on the cross-level time difference sequence, compares the absolute value of the strain difference at the corresponding time point of the previous level with the current threshold, calculates the direction synchronization rate as the percentage of the number of times in the same direction to the total number of times, and screens the time points where the strain difference is lower than the threshold and the synchronization rate is higher than the preset synchronization rate threshold to generate a threshold matching node set; The risk node screening sub-module calls the threshold matching node set, counts the number of nodes with a cross-level time difference continuously less than the preset alignment threshold, calculates the product of the time span and the direction synchronization rate of the continuous node segment as the risk weight, screens the node segments with the top 20% of the risk weight values, and integrates the node coordinates and weight parameters to generate a bearing capacity risk node set.

[0013] As a further solution of the present invention, the atlas reconstruction module includes: The node parameter analysis sub-module calls the coordinates of the bearing capacity risk node set, extracts the distance between adjacent nodes and the included angle of the direction vectors, screens the node pairs with an included angle difference less than the direction consistency threshold, associates and stores the direction vector parameters according to the node number, and counts the number of adjacent directions that meet the threshold for each node to generate a node direction parameter set; The clustering level division sub-module counts the density of adjacent directions of each node based on the node direction parameter set, compares the density with the preset core density threshold, marks the nodes with a density higher than the threshold as core nodes, and at the same time expands the adjacent nodes centered on the core nodes, merges the node groups with continuously consistent direction angles, sorts the groups according to the density from high to low to divide the safety levels, and generates a clustering group level; The security topology rendering sub-module calls the clustering group level, maps the security level numbers to node coordinates, constructs a three-dimensional grid topology relationship, performs coloring rendering by assigning a preset color gradient according to the level numbers, superimposes direction vector arrows for identification, converts the grid data into a format that can be parsed by the graphics interface, and generates a visual detection result.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In the present invention, by analyzing the dynamic frequency offset gradient, the maximum strain offset and its direction reversal characteristics in adjacent time windows are captured in real time, improving the response sensitivity to microscopic bearing capacity fluctuations. A strain jump rhythm map is constructed, and parameters such as the time difference, amplitude difference, and direction continuity at the mutation time point are extracted to accurately locate abnormal load sections and identify sudden strain jump events that are easily smoothed in traditional time series curve analysis. Through the dynamic time warping algorithm, multi-dimensional comparisons of time difference, strain difference threshold, and direction synchronization rate are performed on cross-level rhythm maps, screening cross-level synchronous mutation nodes to generate a risk set and reducing the misjudgment rate. Based on the direction consistency parameter between nodes, density clustering sorting is performed to reconstruct the security level topology map, enhancing the visual spatial correlation and dynamic evolution characteristics of the structural instability path. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is the system flow chart of the present invention; Figure 2 is the flow chart of the strain load module of the present invention; Figure 3 is the flow chart of the offset gradient module of the present invention; Figure 4 is the flow chart of the rhythm risk module of the present invention; Figure 5 is the flow chart of the threshold determination module of the present invention; Figure 6 is the flow chart of the map reconstruction module of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0016] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0017] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention. In addition, in the description of the present invention, the meaning of "a plurality of" is two or more, unless otherwise specifically defined.

[0018] Embodiment 1: Please refer to Figure 1 , an Internet of Things-based building bearing capacity detection system includes: A strain load module, configured to collect real-time strain values of building nodes through distributed strain sensors, perform moving average filtering to generate reference strain load parameters, and transfer the real-time strain values and the reference strain load parameters to the offset gradient module; An offset gradient module, configured to call a gradient tracking algorithm, calculate the maximum strain offset based on the real-time strain values within adjacent time windows, generate a bearing capacity offset gradient sequence, combine the reference strain load parameters, identify nodes with continuously reverse directions, and transfer the bearing capacity offset gradient sequence and the reverse nodes to the rhythm risk module; A rhythm risk module, configured to detect the strain mutation time points within the load period according to the bearing capacity offset gradient sequence and the reverse nodes, extract the strain difference, time interval, and direction parameters between adjacent mutation time points, construct a strain jump rhythm map, mark abnormal load sections, and transfer the strain jump rhythm map to the threshold determination module; A threshold determination module, configured to adopt a dynamic time warping algorithm, compare the mutation time difference, strain difference threshold, and direction synchronization rate of the strain jump rhythm maps of the current level and the previous level, screen cross-level synchronous mutation nodes to generate a bearing capacity risk node set, and transfer the bearing capacity risk node set to the map reconstruction module; A map reconstruction module, configured to sort the direction consistency parameters between nodes by calling a density clustering algorithm through the coordinates of the bearing capacity risk node set, generate a bearing capacity distribution topology map reflecting the safety level, and output a visual detection result.

[0019] The reference strain load parameters are specifically the filtered mean, variance, and range. The bearing capacity offset gradient sequence includes the maximum offset, gradient direction, and time window number. The strain jump rhythm map includes the time difference, amplitude difference, and direction continuity parameter of the mutation time point. The bearing capacity risk node set includes the risk level, synchronous timestamp, and direction consistency coefficient. The bearing capacity distribution topology map is specifically the safety level label, node spatial coordinates, and clustering path number.

[0020] Please refer toFigure 2 , the strain loading module includes: The real-time data acquisition submodule monitors the output signal of the distributed strain sensor, intercepts the strain sampling value at the corresponding timestamp of the building node, stores it according to the sensor number and spatial coordinate association, extracts the data of all sampling points within the fixed time period at the current moment, arranges them in chronological order as a continuous strain sequence, and generates the original strain sequence; The real-time data acquisition submodule continuously monitors five distributed fiber Bragg grating (FBG) strain sensors numbered FBG-A01 to FBG-A05 installed at specific monitoring points near the main cable saddle of the bridge, and continuously records their output signals from a specific date, such as 10:00:00 on April 15, 2025. This process is achieved by setting a sampling frequency of 10 Hz, that is, the central wavelength reading of each sensor is obtained every 0.1 second. The acquired wavelength data is converted into the strain value used in the project based on the sensitivity coefficient provided in the sensor calibration certificate (in this case, the coefficient is 1.2 pm / με, indicating that each microstrain με corresponds to a wavelength drift of 1.2 picometers). The specific conversion process is as follows: at 10:00:00.100, the wavelength monitored by sensor FBG-A03 is offset by 181.8 pm relative to the initial calibration wavelength. The strain value calculated by the conversion rule is Then, at the next sampling time 10:00:00.200, the wavelength shift was detected to be 183.0pm, and the corresponding converted strain value was , the strain value obtained by each sampling is bound to its timestamp accurate to milliseconds, the unique sensor number (FBG-A03) and the pre-set three-dimensional spatial coordinates of the sensor (here FBG-A03 is located at (150.5, 25.2, 80.0) meters) to form a structured data record, such as {Timestamp:'2025-04-1510:00:00.100', SensorID:'FBG-A03', Coordinate:(150.5, 25.2, 80.0), Strain:151.5με}. These records are stored in the central database in real time. Subsequently, the system extracts all the sampling data within the specified time period for subsequent analysis. The continuous 60 seconds of data before the current time 10:01:00.000 are selected, that is, the total data collected by all five sensors from 10:00:00.000 to 10:01:00.000. For sensor FBG-A03, the 600 strain sampling values ​​collected within 60 seconds are strictly based on the timestamp. Arrange in order to form a time series , the data at the beginning of the sequence is With me, after completing this step, the original strain sequence corresponding to each sensor is obtained.

[0021] The dynamic filtering processing sub-module sets a sliding window with a fixed width based on the original strain sequence, intercepts the data within the window successively at a preset time step, performs an arithmetic mean operation on the strain values within each window, replaces the value at the starting point of the window with the window mean, and traverses all windows to complete the mean replacement, generating a smoothed strain sequence; The dynamic filtering processing sub-module receives the original strain sequence generated by the real-time data acquisition sub-module, such as the sequence of sensor FBG-A03 , to eliminate the possible high-frequency noise interference in the sensor signal, a moving average filtering method is used for processing. A sliding window with a fixed width of 5 sampling points (time span of ) is set, and the time step is set to 1 sampling point (0.1 second) to move the window point by point backward. At the beginning of the processing, the data of the first window at the start of the sequence is intercepted, that is, the first 5 sampling values με, and the arithmetic mean of these 5 strain values is calculated: ,, the calculated window mean replaces the original value 151.5με at the starting point of this window as the first value of the filtered sequence. Subsequently, the window slides backward by one time step, and the data from the 2nd sampling point to the 6th sampling point is intercepted, that is με (where 151.9με is the original value of the 6th sampling point), and the arithmetic mean of the second window is calculated: , and this mean value 152.08με replaces the original value 152.5με at the starting point of the window as the second value of the filtered sequence. In this way, the window keeps sliding backward, repeating the operations of intercepting, calculating the mean, and replacing the value at the starting point until all possible window positions of the entire original strain sequence are traversed (the data at the end of the sequence that is less than the width of a complete window is not processed to ensure that the length of the filtering result is consistent with the effective processing section), and finally a new smoothed strain sequence with a reduced noise level is generated , and its starting part is με.

[0022] Table 1 Partial original and smoothed strain data of FBG-A03 sensor: ; As shown in Table 1, the original strain values of sensor FBG-A03 at the initial several sampling moments and the smoothed strain values obtained after 5-point moving average filtering are listed.

[0023] The reference parameter generation sub-module calls the smoothed strain sequence, selects the values within the initial time period, calculates its mean square deviation and fluctuation amplitude, and compares the mean square deviation with the preset strain fluctuation threshold. If the mean square deviation is lower than the threshold, the mean value of the data in the initial time period is set as the reference value. If it is higher than the threshold, the time period is extended and the mean value is recalculated to generate the reference strain load parameter.

[0024] The reference parameter generation sub-module calls the smoothed strain sequence output by the dynamic filtering processing sub-module, such as that of FBG-A03 for the purpose of determining a reference strain level representing the structural position where the sensor is located in a normal and stable state. Select the data in the initial time period of the smoothed sequence , specifically select the data within the first 10 seconds at the start of the sequence, that is, the first 100 data points, as the samples for establishing the reference, and calculate the mean square deviation of these 100 smoothed strain values and the fluctuation amplitude (defined as the difference between the maximum and minimum values within the sample ). Through calculation, the mean square deviation is , and the fluctuation amplitude is . Next, set a strain fluctuation threshold . The setting of this threshold needs to reflect the normal working state of the structure. Referring to the statistical analysis results of long-term monitoring data of similar bridge structures at similar measuring point positions, it is generally considered that under normal traffic and environmental loads, for healthy structural components, the mean square deviation of their strain signals should be maintained at a low level, lower than a specific limit value. According to historical data analysis, the upper limit of the 95% confidence interval of the strain mean square deviation at this measuring point position in the stable state is approximately . To leave a certain margin, the threshold is set to . Compare the mean square deviation calculated from the samples with this preset threshold . This result indicates that within the selected initial 10-second time period, the sensor signal fluctuation is within the normal and acceptable range, and the data quality is suitable for calculating the reference. Therefore, calculate the arithmetic mean of this initial time period (i.e., the first 100 smoothed data points) as the reference strain value, and the calculation result is . Set this calculated mean value as the reference value of sensor FBG-A03 . If in the comparison step, the calculated mean square deviation is higher than the threshold , for example, the calculated value is , it is determined that the data fluctuates too much in the initial 10 seconds, which may include short-term abnormal disturbances or the structure is in an unstable response stage and is not suitable for direct use in calculating the benchmark. At this time, the system automatically extends the selected time period length, expands the sample range to 30 seconds from the start of the sequence (i.e., the first 300 data points), and recalculates its arithmetic mean based on these 300 smoothed strain data. The mean calculated based on the longer time sample is set as the benchmark value. , through the above steps, the benchmark strain load parameters of sensor FBG-A03 are finally generated. , this parameter will be used for subsequent offset analysis.

[0025] Please refer to Figure 3 , the offset gradient module includes: The strain window truncation sub-module calls the real-time strain values within adjacent time windows, divides the sampled data according to the window time length, intercepts the difference between the maximum and minimum strain values within each window, and arranges the differences in the order of window numbers to generate a time window strain set; The strain window truncation sub-module receives the original strain sequence generated by the real-time data acquisition sub-module , (Note: The original sequence is used here instead of the smoothed sequence to capture faster strain changes), and it is segmented according to the set time window length. The window time length is set to 1 second, corresponding to sampling points. Starting from the beginning of the sequence, it is successively divided into continuous non-overlapping time windows. The first window contains the 10 original strain data with timestamps from to με. For the data in this window, find its maximum strain value and minimum strain value , calculate the difference between the two to obtain the strain difference of the first window , then process the second window, which contains the data with timestamps from to to με, calculate the difference between its maximum and minimum values , perform the same operation on all subsequent 1-second windows in the sequence, calculate the strain extreme differences within each window, and arrange these calculated differences in chronological order of the windows to form a new sequence, called the time window strain set , the first few values of this sequence are με, and this sequence reflects the severity of strain fluctuations within each 1-second time period. με, this sequence reflects the severity of strain fluctuations within each 1-second time period.

[0026] The gradient change calculation sub-module calculates the absolute value of the difference change rate between adjacent windows based on the time window strain set, sorts the absolute values, extracts the top 10% of the candidate offsets, counts the number of windows in which the candidate offsets continuously appear, compares the number with the preset offset persistence threshold, filters out the continuous segments that meet the conditions, and generates an offset gradient sequence. The gradient change calculation sub-module is based on the time window strain set generated by the strain window intercepting sub-module με, calculates the change rate of the strain difference between adjacent windows. Since the length of each window is 1 second, the difference change rate between adjacent windows is the absolute value obtained by subtracting the difference of the previous window from the difference of the next window (unit: με / s). The calculation process is as follows: The first change rate , The second change rate , The third change rate , The fourth , The fifth , The sixth , The seventh , The eighth , The ninth , Calculating in this way, an absolute value sequence of change rates is obtained με / s. Assuming that a total of 60 windows are processed and 59 change rate values are obtained, these 59 change rate values are sorted in descending order, and the top 10% of the sorted results are selected as candidate offsets, that is, the first largest change rate values. If the top 5 values after sorting are με / s (the values here are only for illustration and are not directly corresponding to the G sequence calculated earlier), next, the system checks whether these 5 candidate offset values form a continuously occurring segment in the original change rate sequence and counts the number of windows in which they continuously appear. Set an offset persistence threshold . The setting of this threshold is to filter out isolated large change rate values caused by single impacts or short-term noises, requiring significant offsets to have a certain duration. According to the analysis of the vibration characteristics of such bridges, the persistence threshold is set to 3 consecutive windows (that is, the duration reaches ). It is checked and found that the change rate value (assuming it is the fifth candidate offset) continuously appears at the 4th, 5th, and 6th positions in the sequence (that is, , where the calculation result of the previous G sequence is adjusted here to construct continuous instances), the number of consecutive occurrences is 3. Compare this consecutive count of 3 with the threshold for comparison, . Since the persistence condition is met, this continuous segment formed by the difference changes in the 4th, 5th, and 6th windows is identified. If the consecutive occurrence counts of other candidate offset values (such as 2.5, 2.3, 2.1, 1.9 με / s) in the sequence are all less than 3, they are not selected. Through this step, an offset gradient sequence is generated, which records the detailed information of all high-gradient change segments that meet the persistence condition, including the starting window number, the number of continuous windows, and the corresponding change rate value.

[0027] The reverse node identification sub-module calls the offset gradient sequence and the reference strain load parameters, extracts the positive and negative direction signs of each offset segment in the sequence, compares the number of positive and negative alternations of the direction signs of adjacent segments, counts the node positions where the directions are continuously opposite three times, and verifies the reverse of the node positions with the load direction in the reference parameters to generate directionally continuously reverse nodes.

[0028] The reverse node identification sub-module calls the offset gradient sequence generated by the gradient change calculation sub-module (which contains the identified offset segments that meet the persistence condition, such as the segment with a change rate of 1.7 με / s formed by the 4th to 6th windows above), and combines it with the reference strain load parameters generated in Paragraph 1 to further analyze the strain change direction characteristics of these high-gradient segments. Extract the first identified offset segment (the 4th, 5th, and 6th windows, corresponding to the time to ), check the main change trend of the values in the original strain sequence relative to the reference value . Specifically, check the values from to . It is found that the strain value continuously increases from about 155 με to 168 με, and most of the time (or the average value) in this segment is significantly higher than the reference value . Therefore, it is determined that the direction of this offset segment is positive (+), indicating an increasing strain trend. Subsequently, assume that the system identifies another offset segment that meets the conditions at a subsequent time (such as the 10th to 12th windows, time to ). Check the corresponding original strain data and find that its value continuously decreases from 160 με to 145 με, significantly lower than the reference value . Determine its direction to be negative (-), indicating a decreasing strain trend. Then, later (such as the 15th to 17th windows, time to )(+) Another positive offset segment is identified, thus obtaining a sequence recording the directions of consecutive significant offset events, in the form of [+,-,+,-,-,+,…]. The module then checks whether there is a pattern of three consecutive opposite directions in this direction sequence, that is, looking for subsequences such as “+,-,+” or “-,+,-”. In the currently obtained sequence [+,-,+], the “+,-,+” pattern is found. The occurrence of this pattern means that the structure has experienced rapid and directionally alternating strain changes at the measurement point location, indicating an oscillatory behavior rather than a monotonic increase or decrease in load. The system records the time window ranges corresponding to the three consecutive offset segments that make up this pattern (i.e., the combined time period of the 4th - 6th window, the 10th - 12th window, and the 15th - 17th window), and defines these identified time periods (or the key turning points therein) that exhibit the characteristic of three consecutive direction reversals as “directionally consecutive reverse nodes”. These nodes themselves represent dynamic response events different from the stable reference state with specific oscillatory characteristics, and generates a set of directionally consecutive reverse nodes containing the time position information of these nodes for subsequent rhythm analysis.

[0029] Please refer to Figure 4 , the rhythm risk module includes: The mutation parameter extraction sub - module calls the bearing capacity offset gradient sequence and the strain mutation time points within the reverse node detection load cycle, extracts the difference in strain values between adjacent mutation time points, calculates the time interval between mutation time points, records the positive and negative signs of the mutation direction parameters, and stores the difference, interval, and direction parameters in association according to the mutation order to generate a set of mutation parameters; The mutation parameter extraction sub - module calls Figure 3 the offset gradient sequence output by the offset gradient module in (including high - gradient offset segment information) and the set of directionally consecutive reverse nodes output by the reverse node recognition module (marking the time regions where the “+,-,+” or “-,+,-” pattern occurs), focuses on these time periods identified as significant changes or oscillatory behaviors, extracts key mutation parameters from them, and defines the mutation time points as the starting points and ending points of each identified significant offset segment. For the first positive offset segment (the 4th - 6th window) identified previously, extract its starting point (corresponding to the beginning of the 4th second, i.e., ), and the ending point (corresponding to the end of the 6th second, i.e., ) of the original strain value . For the second negative offset segment (the 10th - 12th window), extract the starting point ( ) strain value and the ending point ( ) strain value , for the third positive offset segment (windows 15 - 17), extract the starting point ( ) strain value and the ending point ( ) strain value , calculate the difference in strain values within each mutation segment and the time interval , and record the direction symbol for this segment , the specific calculation is as follows: for the first mutation segment: , (spanning 3 one - second windows), direction ; for the second mutation segment: , , direction ; for the third mutation segment: , , direction ; organize and store these calculated parameters {difference , interval , direction } in the chronological order of the mutation events to form a mutation parameter set, such as .

[0030] Based on the mutation parameter set, the rhythm feature analysis sub - module statistically analyzes the mean of the time intervals between adjacent mutation time points and the fluctuation range of the differences, calculates the proportion of the number of times the direction parameters are continuously consistent, compares the mean of the intervals with a preset rhythm period threshold, screens out the mutation segments with a fluctuation range exceeding the threshold, and integrates the mean of the intervals, the fluctuation range, and the direction consistency ratio to generate a rhythm feature sequence; The rhythm feature analysis sub - module receives the mutation parameter set generated by the mutation parameter extraction sub - module , and conducts statistical analysis on it to mine potential rhythm - related risk features. First, calculate the arithmetic mean of the time intervals of all mutation events , and calculate the fluctuation range of the strain differences , defined as the difference between the maximum and minimum values in the dataset . Assuming that after analyzing the entire dataset (including mutation events), the mean of the time intervals is , and the fluctuation range of the differences is . Second, calculate the proportion of the number of times the direction parameters are continuously consistent (i.e., continuously appear as "+", "+" or "-", "-") to the total number of pairs of adjacent events ( ) , if in 50 events (49 pairs of adjacent events), 15 consecutive same-direction cases are observed, then , next, introduce a preset rhythm period threshold , the setting of this threshold needs to be based on the research of the dynamic characteristics of the monitored bridge structure, especially its main natural vibration frequency or the response period that is easily induced by external excitations (such as vortex-induced vibration under a specific wind speed, regular vehicle flow). If the analysis shows that the period of a certain key vibration mode of the bridge is about 6 seconds, a threshold interval around this period can be set, such as seconds, compare the calculated average time interval with this threshold interval, , indicating that the average time interval of the observed strain mutation events falls within the sensitive period interval of the structure, which suggests that there may be a risk of resonance or near-resonance. At the same time, introduce a difference fluctuation range threshold , this threshold is determined based on the fatigue performance data of the materials used in the bridge structure (such as steel, concrete) or the limitations on the stress / strain change amplitude under cyclic loads in relevant design codes. Set , indicating that if strain fluctuations exceeding this amplitude occur repeatedly in a short period, attention needs to be paid. Compare the calculated fluctuation range with the threshold , , indicating that the severity of the strain change exceeds the warning level. Finally, integrate these analysis results: the average time interval (falling within the sensitive interval), the fluctuation range (exceeding the threshold), and the direction consistency ratio of 30.6% (reflecting a certain degree of non-fully random fluctuation) to generate a rhythm feature sequence, which is a quantitative description of the rhythmic risk of the structure response during this period.

[0031] The jump map generation sub-module calls the rhythm feature sequence, sorts the mutation segments in descending order according to the fluctuation range, maps the sorting result to the time axis, assigns node weights according to the direction consistency ratio, constructs a three-dimensional grid topological relationship of time-difference-direction, converts the grid data into a format compatible with the map rendering interface, and generates a strain jump rhythm map.

[0032] The jump map generation sub-module calls the rhythm feature sequence output by the rhythm feature analysis sub-module and the original mutation parameter set, aiming to display the analyzed risk information in an intuitive graphical way. First, based on the absolute value of the strain difference in the mutation parameter set, sort all mutation events in descending order. For example, the absolute value of 15 is greater than the absolute value of 13, then the former is ranked in front. Arrange each sorted mutation event according to its occurrence time point (the start time can be used or the central time ) is mapped onto the time axis (X-axis) of the atlas, and the strain difference is mapped onto the Y-axis, and the direction can be represented by the Z-axis coordinate (+1 or -1) or color (warm color indicates positive and cold color indicates negative). To reflect the importance or risk level of different mutation events, weights are assigned to each node mapped onto the atlas , and the weight can be directly taken as the absolute value of the strain difference, that is . Then the weight of the first mutation event is 13, the second is 15, and the third is 10. In this way, a three-dimensional scatter plot basis containing time, difference, direction, and weight information is formed. Further, connecting lines can be drawn between the mutation events at adjacent time points to form a time evolution path, constructing a three-dimensional grid topological relationship of time-difference-direction. Finally, the grid data containing node positions ( ), node weights and optional connection information is converted into a data format that can be directly read and parsed by a standard three-dimensional graphics rendering engine (such as the VTK library, OpenGL interface, or the Three.js library on the web side), usually a JSON array or a specific geometric file format. For example, the JSON format can be [{time:t_start1,delta_strain:13,direction:1,weight:13},{time:t_start2,delta_strain:-15,direction:-1,weight:15},...]. Through this process, a strain jump rhythm atlas that intuitively reflects the magnitude, direction, occurrence time, and potential rhythm of strain mutation events is generated.

[0033] Please refer to Figure 5 , and the threshold determination module includes: The hierarchical parameter alignment sub-module calls the mutation time difference parameter of the current level and the strain jump rhythm atlas of the previous level, aligns the mutation time points of the two levels along the time axis, calculates the absolute value of the time stamp difference at the corresponding time points, and adjusts the time window length. The adjusted time difference is associated and stored according to the level number to generate a cross-level time difference sequence; The hierarchical parameter alignment sub-module aims to compare the strain response synchronization between different structural levels or regions. It calls the strain jump rhythm atlas data generated by the current analysis level (defined as the main cable saddle region, marked as L2) and the previous level (defined as the main cable mid-span region, marked as L1) used as a reference respectively through Figure 4 the process. These two atlases contain the time stamps of the strain mutation events identified in their respective regions and (where k and m are the indices of the mutation events in the L1 and L2 spectra respectively), the goal of the module is to find pairs of cross-level mutation events that are close enough in time, and set an initial time matching window width , the setting of this width needs to consider the physical propagation time of the signal in the structure and the small time differences that may be introduced by sensor sampling and processing. Based on the distance between the L1 and L2 regions (about several hundred meters) and the stress wave propagation speed of the structural material (steel) (about 5000 m / s), the propagation time is estimated to be on the order of 0.1 seconds. Considering the data processing delay, set seconds, the system traverses each mutation event in L1 , and searches for mutation events in the spectral data of L2 that satisfy the absolute value of the difference in timestamps is less than , that is seconds. If the number of event pairs matched through this window width is not sufficient for effective statistics (less than the preset minimum sample size, such as 10 pairs), the system can appropriately widen the window width, such as adjusting it to 0.8 seconds. Conversely, if there are too many matching pairs and it may introduce noise, then tighten the window. For all successfully matched event pairs , calculate the absolute value of their exact timestamp difference , and associate this time difference with the corresponding event information (including their respective strain differences and directions ) for storage, forming a sequence that records the time differences of cross-level synchronous events. The data structure is [{t_L1:time1_k,strain_L1:de_L1k,dir_L1:s_L1k,t_L2:time2_m,strain_L2:de_L2m,dir_L2:s_L2m,time_diff:diff_km},...], generating a cross-level time difference sequence.

[0034] Table 2 Example of cross-level matching mutation event pairs: ; As shown in Table 2, some synchronous mutation event pairs between the L1 and L2 levels found through time window matching and their related parameters are listed, including their respective timestamps, strain differences, directions, and the time difference between the two.

[0035] Based on the cross-level time difference sequence, the synchronous threshold matching sub-module extracts the current level's strain difference threshold and direction synchronization rate parameters, compares the absolute value of the strain difference at the corresponding time point in the previous level with the current threshold, calculates the direction synchronization rate as the percentage of the number of times in the same direction to the total number of times, screens the time points where the strain difference is lower than the threshold and the synchronization rate is higher than the preset synchronization rate threshold, and generates a threshold matching node set; The synchronization threshold matching sub-module utilizes the cross-level time difference sequence and associated data generated by the hierarchical parameter alignment sub-module for more stringent synchronization screening. The aim is to identify event pairs that are not only close in time but also highly consistent in terms of strain amplitude and change direction. The strain difference threshold of the current level L2 is introduced. And the direction synchronization rate threshold , The setting of which needs to consider that the L2 level (main cable saddle) as a key stress area usually allows a more stringent strain fluctuation range than the L1 level (mid-span). According to the detailed finite element analysis results of the saddle component and the fatigue check requirements in the design specifications, the strain difference threshold of L2 is set to , and the direction synchronization rate threshold is used to ensure that the identified synchronization is systematic rather than accidental. It requires that the directions of most synchronous events must be the same and is set to a relatively high percentage, such as . First, the system calculates the actual direction synchronization rate of all matched event pairs in the entire cross-level time difference sequence. The calculation method is: count the number of event pairs with the same direction (i.e., ), then . Assuming there are 100 pairs of matching points in Table 2 and subsequent data, and 85 pairs have the same direction, the calculated . Compare this actual synchronization rate with the threshold: . If the global synchronization rate requirement is met, then, for each matched event pair , check whether the absolute value of the strain difference of the upper level L1 is less than the threshold of the current level L2. The purpose of this condition is to find those events that may not have reached the warning line of the L2 level at the L1 level itself but are worthy of attention because of their high synchronization with L2 (close in time and high direction consistency). This may indicate that the risk is transferring from L1 to L2 or strong coupled vibrations occur in the two regions. Taking the first row of Table 2 as an example, the L1 strain difference is 55 με, . If this condition is met and assuming the global synchronization rate is also satisfied, then the event pair (k, m) is preliminarily screened out. Perform this check on all event pairs to generate a set of event pairs that only satisfy (and globally satisfy ), which is called the threshold matching node set.

[0036] The risk node screening sub-module calls the threshold matching node set, counts the number of nodes with consecutive cross-level time differences less than the preset alignment threshold, calculates the product of the time span and the direction synchronization rate of the consecutive node segments as the risk weight, screens the node segments with the top 20% of the risk weight values, integrates the node coordinates and weight parameters, and generates a bearing capacity risk node set.

[0037] The risk node screening sub-module receives the threshold matching node set output by the synchronization threshold matching sub-module and performs the last step of screening to pinpoint the most risky and continuously synchronized structural response areas, introducing a more stringent cross-level time difference alignment threshold , which represents a highly synchronized state where events occur almost instantaneously between two levels. Its setting is based on eliminating the time difference even after a small propagation delay and focusing on the tightest coupled responses. Set seconds. The system checks whether there are multiple consecutive event pairs in the threshold matching node set whose cross-level time differences are all continuously less than . In Table 2 and subsequent data, it is found that the time differences of the event pairs in rows 1, 2, 4, and 5 are all 0.1 second, less than 0.2 second, and they occur continuously in time (only the non-conforming point in row 3 is in between), forming a continuous segment of length 4 (assuming they are adjacent in the original data, or "continuous" can be defined to allow a few non-conforming points in between, but the overall trend is continuous). Calculate the duration span of this continuous segment , that is, the average timestamp of the last event pair in the segment minus the average timestamp of the first event pair, seconds, and calculate the local direction synchronization rate within this segment . This segment contains 4 event pairs, among which the directions of pairs 1, 2, 4, and 5 are all the same (+vs+,-vs-,+vs+,-vs-), and the direction of pair 3 (assuming it is within the segment) is different. If only the 4 pairs that meet seconds are counted, then the synchronization rate is . Multiply these two metrics to get the risk weight of this continuous segment , where is used in its decimal form as the multiplier, that is, . Calculate the risk weights for all identified continuous segments that meet , and then sort all the calculated risk weight values in descending order, and screen out the top 20% of the continuous segments with the highest risk weights. Suppose a total of 10 such continuous segments are identified, then select the ones with the highest risk weights , assuming a total of 10 such continuous segments are identified, then select the top For each paragraph, finally, integrate the spatial coordinate ranges of the nodes (sensors) involved in the two highest-risk paragraphs (including the relevant sensors at levels L1 and L2) and their respective calculated risk weights. Generate a final bearing capacity risk node set, which indicates the areas in the structure that require the most attention and exhibit continuous cross-level highly synchronous responses.

[0038] Please refer to Figure 6 The atlas reconstruction module includes: The node parameter parsing sub-module calls the coordinates of the bearing capacity risk node set, extracts the distance between adjacent nodes and the included angle of the direction vectors, filters out the node pairs with the included angle difference less than the direction consistency threshold, stores the direction vector parameters associated by node numbers, counts the number of adjacent directions that meet the threshold for each node, and generates a node direction parameter set. The node parameter parsing sub-module receives Figure 5 The generated bearing capacity risk node set, which contains the information of the selected high-risk areas, specifically a series of risk nodes (i.e., sensors participating in the formation of high-risk continuous segments) and their associated spatial coordinates and risk weights. This sub-module aims to analyze the geometric and directional relationships within and between these risk nodes. First, extract the spatial coordinates of all nodes in the risk node set and calculate the distances between adjacent nodes in space. The adjacent relationship can be determined based on a preset distance threshold (such as meters) or based on the structural topology connection relationship. At the same time, it is necessary to determine the main strain change direction of each risk node during the risk period, which can be obtained by analyzing the original strain data sequence of the node in the corresponding high-risk segment. Calculate the first principal direction of its principal component analysis (PCA) or calculate the time-averaged direction of the strain gradient to obtain the direction vector of each node. Set a direction consistency threshold to determine whether the response directions between adjacent nodes are convergent. This threshold is set based on engineering experience, believing that under the action of the same stress field, the strain direction deviation of adjacent points should not be too large. Set For each pair of spatially adjacent risk nodes (i,j), calculate the included angle between their direction vectors and Filter out all node pairs (i,j) that satisfy , which means that the strain change directions of nodes i and j are highly consistent. Then, for each risk node i, count how many adjacent nodes j satisfy the above direction consistency condition. This number is recorded as the adjacent direction density of node i. Finally, the coordinates and risk weights of each risk node , the calculated direction vector and the number of its neighbors with the same direction Integrate and store them to generate a node direction parameter set.

[0039] Table 3 Example of the risk node direction parameter set: ; As shown in Table 3, some risk nodes and their parsed parameters are given, including the position, the weight of the risk segment to which they belong, the calculated principal strain direction vector, and the number of neighbors with the same direction (density).

[0040] The clustering level division sub-module is based on the node direction parameter set, calculates the density of each node's adjacent directions, compares the density with a preset core density threshold, marks the nodes with a density higher than the threshold as core nodes, and at the same time expands adjacent nodes centered on the core nodes, merges the node groups whose continuous direction angles meet the threshold, sorts the groups according to the density from high to low to divide the safety levels, and generates the clustering group levels; The clustering level division sub-module is based on the node direction parameter set generated by the node parameter parsing sub-module, organizes the risk nodes that are spatially adjacent and have the same response direction into groups through a clustering analysis method, and divides the levels according to the risk levels of the groups. First, use the number of neighbors with the same direction of the node (i.e., the direction density) as a key indicator, and set a core density threshold , this threshold defines the minimum requirement to become a clustering core, aiming to ensure that the core nodes represent the center of a region with significant collaborative responses. According to the expectation of the density of the risk region, set , traverse all risk nodes, and mark the nodes that meet as core nodes. In Table 3, the of node A03 meets the condition and is marked as a core node. Next, take all the marked core nodes (such as A03) as starting points, and use the idea similar to density clustering (such as neighborhood expansion in DBSCAN) to construct groups: starting from a core node, find all its spatially adjacent and directionally consistent (i.e., the included angle ) neighbor nodes, and add these neighbor nodes to the current group; if the neighbor node itself is also a core node, recursively add its neighbors to the group; if the neighbor node is not a core node, but it is spatially adjacent and directionally consistent with a certain node in the current group, also add it to the group. Repeat this process until no new nodes can be added to any existing group. In this way, the risk nodes that are connected to each other (through the relationship of proximity and the same direction) are divided into different clustering groups. Then, it is necessary to evaluate the overall risk level of each formed group, calculate the comprehensive risk index of each group, and this index can use the risk weights of all nodes within the group Defined by the average value, or the number of core nodes included in the group, or the spatial volume covered by the group, etc. Here, the average value of the node risk weights within the group is used as the indicator , after calculating the values for all groups , sort these groups in descending order according to the values. According to the sorting results, divide the groups into different risk levels. For example, divide the groups with the top 10% values into the highest risk level (Level 1), the next 20% into the second highest risk level (Level 2), and the rest into the lower risk level (Level 3) to generate the final clustering group level information, which includes the node list of each group and the risk level number assigned to this group.

[0041] The security topology rendering sub-module calls the clustering group level, maps the security level number to the node coordinates, constructs a three-dimensional grid topology relationship, assigns a preset color gradient for coloring and rendering according to the level number, superimposes the direction vector arrow mark, and converts the grid data into a format that can be parsed by the graphics interface to generate a visual detection result.

[0042] The security topology rendering sub-module calls the clustering group level information generated by the clustering level division sub-module, and presents the analyzed risk distribution and level results in a three-dimensional visualization form. First, associate the risk level number (such as 1, 2, 3) to which each risk node belongs with its spatial coordinates. Based on these node coordinates with level attributes and their proximity relationships (or group membership relationships) among them, construct a three-dimensional geometric model. This model can be a node-link diagram, or an isosurface or volume rendering generated based on the node positions to represent the risk area. To visually distinguish different risk levels, set a set of color mapping schemes, map the risk level numbers to the preset color gradients, and agree to use eye-catching red for rendering risk level 1 (highest risk), orange for risk level 2, and yellow for risk level 3. Apply these colors to the corresponding nodes or regions in the three-dimensional model to achieve visual differentiation of risk levels. Further, at the centroid position of each node or each group, superimpose an arrow symbol. The direction of the arrow is set to the main strain change direction of this node or group (i.e., the direction vector calculated in the node parameter parsing or the average direction of the vectors within the group), and the length or thickness of the arrow can be related to the risk weight of the node or the comprehensive risk indicator of the group Proportional to more richly display risk information. Finally, the constructed 3D scene data containing information such as geometric shapes, color coding, and direction arrows will be exported into a general file format that can be loaded and interactively browsed in professional visualization software or web-based graphics libraries, such as the VTK format (.vtk), the glTF format (.gltf), or the OBJ format (.obj), to generate the final visualization detection result atlas for engineers to analyze and make decisions.

[0043] The above is only the preferred embodiment of the present invention and does not limit the present invention in other forms. Any person skilled in the art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A building bearing capacity detection system based on the Internet of Things, characterized in that: The system comprises: The strain load module is used to collect the real-time strain value of the building node through the distributed strain sensor, perform sliding average filtering to generate the reference strain load parameter, and transmit the real-time strain value and the reference strain load parameter to the offset gradient module; An offset gradient module is used to call a gradient tracking algorithm, calculate the maximum strain offset based on the real-time strain value in the adjacent time window, generate a bearing capacity offset gradient sequence, identify the direction continuous reverse node in combination with the reference strain load parameter, and transmit the bearing capacity offset gradient sequence and the reverse node to the rhythm risk module; The rhythm risk module is used to detect the strain mutation time points within the load cycle according to the bearing capacity offset gradient sequence and the reverse node, extract the strain difference, time interval and direction parameters of adjacent mutation time points, construct a strain jump rhythm map, mark the abnormal load section, and transmit the strain jump rhythm map to the threshold judgment module.

2. The building bearing capacity detection system based on the Internet of Things according to claim 1 is characterized in that: The reference strain load parameters are specifically the filtered mean, variance, and range; the bearing capacity offset gradient sequence includes the maximum offset, gradient direction, and time window number; the strain jump rhythm spectrum includes the time difference of the mutation point, the amplitude difference, and the directional continuity parameter.

3. The building bearing capacity detection system based on the Internet of Things according to claim 2 is characterized in that: The strain load module includes: The real-time data acquisition submodule monitors the output signal of the distributed strain sensor, intercepts the strain sampling value at the corresponding timestamp of the building node, stores it according to the sensor number and spatial coordinate association, extracts the data of all sampling points within the fixed time period at the current moment, arranges them in chronological order as a continuous strain sequence, and generates the original strain sequence; The dynamic filtering processing submodule sets a sliding window of fixed width based on the original strain sequence, successively intercepts the data in the window according to the preset time step, performs arithmetic mean operation on the strain value in each window, replaces the window starting point position value with the window mean, traverses all windows to complete the mean replacement, and generates a smooth strain sequence; The benchmark parameter generation submodule calls the smoothed strain sequence, selects the values ​​in the initial time period, calculates the mean square error and fluctuation amplitude, compares the mean square error with the preset strain fluctuation threshold, and if the mean square error is lower than the threshold, sets the mean of the initial time period data as the benchmark value; if it is higher than the threshold, the time period is extended and the mean is recalculated to generate the benchmark strain load parameters.

4. The building bearing capacity detection system based on the Internet of Things according to claim 3 is characterized in that: The offset gradient module includes: The strain window interception submodule calls the real-time strain value in the adjacent time window, divides the sampled data according to the window time length, intercepts the difference between the maximum and minimum strain values ​​in each window, arranges the difference in the order of the window numbers, and generates a time window strain set; The gradient change calculation submodule calculates the absolute value of the difference change rate between adjacent windows based on the time window strain set, sorts the absolute values ​​and extracts the top 10% candidate offsets, counts the number of windows where the candidate offsets appear continuously, compares the number with the preset offset continuity threshold, selects the continuous segments that meet the conditions, and generates an offset gradient sequence; The reverse node identification submodule calls the offset gradient sequence and the reference strain load parameters, extracts the positive and negative direction signs of each offset segment in the sequence, compares the number of positive and negative alternations of the direction signs of adjacent segments, counts the node positions with three consecutive opposite directions, reversely verifies the node positions and the load directions in the reference parameters, and generates nodes with continuous reverse directions.

5. The building bearing capacity detection system based on the Internet of Things according to claim 4 is characterized in that: The rhythm risk module includes: The mutation parameter extraction submodule calls the bearing capacity offset gradient sequence and the reverse node to detect the strain mutation time point within the load cycle, extracts the strain value difference between adjacent mutation time points, calculates the time interval between mutation time points, records the positive and negative signs of the mutation direction parameter, associates and stores the difference, interval and direction parameter according to the mutation order, and generates a mutation parameter set; The rhythm feature analysis submodule, based on the mutation parameter set, counts the mean time intervals and the fluctuation range of the difference between adjacent mutation time points, calculates the proportion of times when the direction parameters are continuously consistent, compares the mean interval with the preset rhythm cycle threshold, screens the mutation segments whose fluctuation range exceeds the threshold, integrates the mean interval, fluctuation range and direction consistency ratio, and generates a rhythm feature sequence; The transition spectrum generation submodule calls the rhythm feature sequence, sorts the mutation segments from high to low according to the fluctuation range, maps the sorting results to the time axis, assigns node weights according to the direction consistency ratio, constructs a three-dimensional grid topological relationship of time-difference-direction, converts the grid data into a format compatible with the graph rendering interface, and generates a strain transition rhythm spectrum.

6. The building bearing capacity detection system based on the Internet of Things according to claim 5 is characterized in that: The system further comprises: A threshold determination module is used to use a dynamic time warping algorithm to compare the mutation time difference, strain difference threshold and direction synchronization rate of the strain jump rhythm map of the current level and the previous level, screen cross-level synchronous mutation nodes to generate a bearing capacity risk node set, and pass the bearing capacity risk node set to the map reconstruction module; A graph reconstruction module is used to call a density clustering algorithm to sort the directional consistency parameters between nodes through the coordinates of the bearing capacity risk node set, generate a bearing capacity distribution topology graph reflecting the safety level, and output a visual detection result; The bearing capacity risk node set includes risk level, synchronization timestamp, and direction consistency coefficient; the bearing capacity distribution topology map specifically includes security level label, node space coordinates, and cluster path number.

7. The building bearing capacity detection system based on the Internet of Things according to claim 6 is characterized in that: The threshold determination module comprises: The hierarchical parameter alignment submodule calls the mutation time difference parameters of the strain jump rhythm map of the current level and the previous level, aligns the mutation time points of the two levels according to the time axis, calculates the absolute value of the timestamp difference of the corresponding time points, and adjusts the time window length. The adjusted time difference is stored in association with the level number to generate a cross-level time difference sequence; The synchronization threshold matching submodule extracts the strain difference threshold and direction synchronization rate parameters of the current level based on the cross-level time difference sequence, compares the absolute value of the strain difference at the corresponding time point of the previous level with the current threshold, calculates the direction synchronization rate as the percentage of the number of times in the same direction to the total number of times, selects the time points where the strain difference is lower than the threshold and the synchronization rate is higher than the preset synchronization rate threshold, and generates a threshold matching node set; The risk node screening submodule calls the threshold matching node set, counts the number of nodes whose cross-level time difference is continuously less than the preset alignment threshold, calculates the product of the time span and the directional synchronization rate of continuous node segments as the risk weight, screens the node segments with the top 20% risk weight values, integrates the node coordinates and weight parameters, and generates a bearing capacity risk node set.

8. The building bearing capacity detection system based on the Internet of Things according to claim 7 is characterized in that: The graph reconstruction module includes: The node parameter parsing submodule calls the coordinates of the bearing capacity risk node set, extracts the angle between the adjacent node spacing and the direction vector, selects the node pairs whose angle difference is less than the direction consistency threshold, stores the direction vector parameters by node number, counts the number of adjacent directions of each node that meet the threshold, and generates a node direction parameter set; The clustering hierarchical division submodule counts the density of the neighboring directions of each node based on the node direction parameter set, compares the density with the preset core density threshold, marks the nodes with density higher than the threshold as core nodes, expands the adjacent nodes with the core nodes as the center, merges the node groups whose direction angles continuously meet the threshold, divides the security level by group density from high to low, and generates a clustering group level; The security topology rendering submodule calls the cluster group level, maps the security level number to the node coordinates, constructs a three-dimensional grid topology relationship, assigns a preset color gradient according to the level number for coloring and rendering, superimposes a direction vector arrow mark, converts the grid data into a format that can be parsed by a graphical interface, and generates a visual detection result.

Images