A hanging basket retraction safety detection system

Abstract

This invention relates to the field of monitoring technology, specifically to a hanging basket retraction safety detection system. The system includes: a dynamic safety baseline construction module, a real-time state vector deviation calculation module, a risk trend gradient calculation module, and a comprehensive risk level determination module. In this invention, a safety baseline vector is generated by calculating the average value of pressure and strain. This is combined with real-time data to construct a real-time vector, forming baseline deviation and risk gradient. This enables dynamic monitoring of the hanging basket retraction process, providing real-time feedback on changes in the state of the support structure and key connection parts, especially the rate of change of pressure and strain at connection points in load-bearing areas. This reflects hidden risks and potential problems. Kalman filtering optimizes monitoring accuracy and response speed, resulting in more precise comprehensive risk level determination. The risk level signal can guide operators to intervene in a timely manner, improving safety and operational efficiency, reducing human error interference, and providing a more reliable risk warning method than manual inspection.

Description

Technical Field

[0001] This invention relates to the field of monitoring technology, and in particular to a hanging basket retraction safety detection system. Background Technology

[0002] The field of monitoring technology involves the real-time acquisition, transmission, and recording of key status parameters in engineering facilities, production processes, and public safety environments. Its core aspects include continuous observation and data retention of structural morphology, operational status, and environmental conditions. A systematic monitoring system is formed through sensing, data acquisition, and remote transmission. This technology encompasses safety monitoring during engineering construction, status monitoring of operating equipment, and early warning recording of abnormal situations, serving as a crucial supporting technology for comprehensive engineering safety management. Among these technologies, the traditional hanging basket retraction safety monitoring system refers to a device used during bridge cantilever construction to monitor the status of the hanging basket's support structure and connections to ensure safety during movement or retraction. This typically involves installing physical testing tools such as mechanical force gauges, displacement indicators, or wire rulers to obtain the force and displacement during the hanging basket's retraction process, supplemented by regular manual inspections and recording.

[0003] Existing technologies rely on tools such as mechanical force gauges, displacement indicators, or wire rulers to obtain information on the force and displacement of the hanging basket. This requires manual periodic inspections and recording, which introduces significant human error and monitoring lag. Especially in complex working environments or where frequent manual inspections are inconvenient, it is impossible to grasp the status of various parts of the hanging basket in real time, easily missing some hidden problems or potential risks. This leads to an inability to take timely countermeasures, increasing safety hazards during operations. Furthermore, it is impossible to effectively process data in complex environments, which significantly affects the accuracy and timeliness of monitoring results and impacts the overall safety management effectiveness of the project. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the existing technology and to propose a hanging basket retraction safety detection system.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a hanging basket retraction safety detection system, comprising:

[0006] The dynamic safety baseline construction module calculates the average value of pressure and strain measurements, combines the calculated average pressure and strain values ​​to generate a safety baseline vector, and transmits the safety baseline vector to the real-time state vector deviation calculation module.

[0007] The real-time state vector deviation calculation module constructs a real-time vector based on the safety baseline vector and the real-time collected pressure and strain measurements, calculates the weighted absolute difference and generates the baseline deviation, and transmits the baseline deviation and the real-time vector to the comprehensive risk level determination module and the risk trend gradient calculation module, respectively.

[0008] The risk trend gradient calculation module, based on the real-time vector, processes the data queue using the Kalman filter algorithm, calculates the time change rate of the state values ​​at the beginning and end of the queue to generate a risk gradient, and transmits the risk gradient to the comprehensive risk level determination module.

[0009] The comprehensive risk level determination module combines the baseline deviation and the risk gradient as a feature vector, inputs it into the support vector machine model to obtain the classification result, obtains the limit violation flag, determines the risk level based on the classification result and the limit violation flag, and generates a signal.

[0010] As a further aspect of the present invention, the safety baseline vector includes the reference pressure value and reference strain value of each measuring point, and the risk gradient includes the pressure change rate of the core load-bearing part and the strain change rate of the key connection point.

[0011] As a further aspect of the present invention, the dynamic security baseline construction module includes:

[0012] The data acquisition submodule acquires the pressure and strain measurements at each measuring point within a specified time period and establishes a raw data set containing all measurements.

[0013] The mean calculation submodule calls the original data set to calculate the arithmetic mean of all pressure measurements and the arithmetic mean of all strain measurements, generating the pressure average and strain average.

[0014] The baseline vector generation submodule combines the average pressure and the average strain to generate a safe baseline vector.

[0015] As a further aspect of the present invention, the real-time state vector deviation calculation module includes:

[0016] The real-time vector construction submodule acquires real-time pressure and strain measurements, arranges them according to the same dimensions and order as the safety baseline vector, and establishes a real-time vector.

[0017] The weighting factor allocation submodule presets an independent weighting factor for each element in the real-time vector based on the importance and data sensitivity of each measurement point, and generates a weight matrix.

[0018] The deviation calculation submodule calls the safety baseline vector, the real-time vector, and the weight matrix to calculate the weighted sum of the absolute values ​​of the differences between corresponding elements of the vectors, and generates the baseline deviation.

[0019] As a further aspect of the present invention, the risk trend gradient calculation module includes:

[0020] The data queue management submodule stores the continuously generated real-time vectors in chronological order to form a data queue with a fixed length.

[0021] The state filtering processing submodule calls the data queue and uses the Kalman filter algorithm to iteratively process each real-time vector in the queue to filter out measurement noise and predict the state at the next moment, thereby obtaining a filtered state sequence.

[0022] The gradient calculation submodule extracts the first and last state values ​​of the filtered state sequence, calculates their rate of change over time, and generates a risk gradient.

[0023] As a further aspect of the present invention, the comprehensive risk level determination module includes:

[0024] The feature vector fusion submodule takes the baseline deviation and the risk gradient as two independent feature dimensions, performs normalization processing, and combines them into a feature vector.

[0025] The model classification processing submodule inputs the feature vector into the pre-trained support vector machine model, calculates the classification result through the model's decision function, and generates the classification result.

[0026] The risk level synthesis submodule acquires the independent limit-crossing flag, determines the risk level and generates a corresponding warning signal based on the classification result and the limit-crossing flag according to the preset decision rules.

[0027] As a further aspect of the present invention, the specific process by which the deviation calculation submodule calculates the baseline deviation is as follows: obtaining the safe baseline vector, the real-time vector, and the weight matrix;

[0028] The baseline deviation is calculated using the following formula: ;

[0029] in, This represents the final baseline deviation. This represents the total number of measurement point-related parameters contained in the real-time vector and the safety baseline vector. Index representing the parameters of the measuring point. Represents the weight matrix with respect to the first... The weighting coefficients corresponding to the parameters of each measuring point Represents the first in the real-time vector The element values ​​corresponding to the parameters of each measuring point Represents the first in the security baseline vector The element values ​​corresponding to the parameters of each measuring point.

[0030] As a further aspect of the present invention, the specific process of the gradient calculation submodule calculating the risk gradient is as follows: obtaining the head state value and tail state value of the filtered state sequence;

[0031] The risk gradient is calculated using the following formula: ;

[0032] in, This represents the final generated risk gradient. This represents the filtered state sequence at the current time. The estimated value of the tail state, Represents the filtered state sequence in The estimated state of the head of the queue at time [time]. This represents the fixed length of the data queue, that is, the number of real-time vectors contained in the queue. This represents the time interval between two consecutive real-time data acquisitions.

[0033] As a further aspect of the present invention, the support vector machine model employs a radial basis kernel function for nonlinear mapping processing;

[0034] In the model classification processing submodule, the feature vector is mapped to a high-dimensional feature space through the radial basis kernel function, and the optimal classification hyperplane is found in the high-dimensional feature space.

[0035] The radial basis kernel function is calculated according to the following formula: ;

[0036] in, The kernel function calculation result represents the kernel function between two different feature vectors input to the support vector machine model. and represents any two feature vectors that are used as input. Represents kernel parameters, This represents the square of the Euclidean distance between the two eigenvectors. Represents the natural exponential function;

[0037] The kernel parameters The classification results are determined by cross-validation training on multiple sets of typical working condition data during the historical hanging basket retraction process. The classification results include three levels: safety, early warning, and alarm.

[0038] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0039] In this invention, a safety baseline vector is generated by calculating the average value of pressure and strain, and a real-time vector is constructed by combining real-time data to form the baseline deviation and risk gradient. This method enables real-time, dynamic monitoring of the hanging basket contraction process, providing a more accurate risk assessment than traditional static monitoring. It provides real-time feedback on the state changes of the hanging basket support structure and key connection parts, especially the pressure change rate of the core load-bearing parts and the strain change rate of the connection points, which can effectively reflect hidden risks and potential problems. By processing the data queue through Kalman filtering, the monitoring accuracy and response speed are further optimized, and the comprehensive risk level judgment is more accurate. The timely feedback of risk level signals can guide operators to make timely interventions when abnormal situations occur, effectively improving safety and work efficiency, reducing the interference of human error on the detection results, and providing a more reliable risk early warning means than manual inspection. Attached Figure Description

[0040] Figure 1 This is a flowchart illustrating the overall structure of the hanging basket retraction safety detection system of the present invention.

[0041] Figure 2 This is a flowchart illustrating the dynamic security baseline construction process of this invention.

[0042] Figure 3 This is a flowchart of the real-time state vector deviation calculation process of the present invention;

[0043] Figure 4 This is a flowchart of the risk trend gradient calculation process of the present invention;

[0044] Figure 5 This is a flowchart for determining the overall risk level of this invention. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this invention clearer, the software-based technical solution is described in detail below with reference to system architecture diagrams and embodiments. It should be understood that the specific embodiments described herein are only for explaining the technical solutions of this invention and do not constitute a limitation on the scope of protection.

[0046] In the description of this invention, the system architecture relationships or data processing flows indicated by terms such as "layer," "module," "interface," "data flow," "client," and "server" are all defined based on the architecture diagram or flowchart corresponding to the embodiments. This way of describing is only used to clearly illustrate the logical relationships between the elements in the technical solution, and not to limit the physical deployment form. The term "multiple" includes two or more technical units, including but not limited to multiple data nodes, processing threads, service instances, or functional components and other scalable elements. The specific number is determined according to the actual business scenario and needs to be specifically specified.

[0047] Please see Figure 1 and Figure 2 The present invention provides a technical solution: a hanging basket retraction safety detection system comprising:

[0048] The dynamic safety baseline construction module calculates the average value of pressure and strain measurements, combines the calculated average pressure and strain values ​​to generate a safety baseline vector, and transmits the safety baseline vector to the real-time state vector deviation calculation module.

[0049] The safety baseline vector includes the reference pressure and reference strain values ​​for each measuring point;

[0050] The dynamic safety baseline construction module includes: a data acquisition submodule, which acquires the pressure and strain measurements of each measuring point within a specified time period and establishes an original data set containing all measurements;

[0051] The mean calculation submodule calls the original data set and calculates the arithmetic mean of all pressure measurements and the arithmetic mean of all strain measurements, generating the pressure average and strain average.

[0052] The baseline vector generation submodule combines the average pressure and average strain to generate a safe baseline vector.

[0053] The specific implementation process of the dynamic safety baseline construction module is as follows: First, the data acquisition submodule is executed, taking a segmental precast glued assembly external prestressed continuous box girder bridge of a certain model as the application object. Pressure sensors and strain gauges are deployed at its core load-bearing parts and key connection points. Specifically, pressure measuring points 1 and 2 are set at the connection between the lower chord of the main truss and the bottom formwork support beam, and HPS-P20 pressure sensors are installed at these two locations respectively; strain measuring point 1 is set at the connection node between the front end hanger of the main truss and the upper beam, and strain measuring point 2 is set at the anchorage connection area between the rear end main truss and the completed beam segment, and BMS120-3AA strain gauges are installed at these two locations respectively. After the hanging basket is assembled and placed under no-load static conditions, the data acquisition system is started, the acquisition time period is set to 3600 seconds, and the data acquisition frequency is 1 / 60 Hz, that is, the pressure and strain values ​​of all measuring points are acquired every 60 seconds. The acquisition process lasts for one hour, and a total of 60 sets of measurement data are obtained. The collected pressure measurements are in megapascals (MPa), and the strain measurements are in microstrain (με). Through this process, a raw data set containing all 240 measurements (60 sets x (2 pressure measurement points + 2 strain measurement points)) is established. This set is stored in the system's database, and some data is shown in Table 1 below.

[0054] Table 1: Initial Status Monitoring Data of Hanging Basket

[0055] ;

[0056] As shown in Table 1, this table records the original measurement data of each measuring point within a specified time period, providing a data basis for subsequent calculations.

[0057] Subsequently, the mean calculation submodule is executed, and the system retrieves the previously established raw data set from the database. For the pressure measurements, the 60 pressure measurements collected at pressure point 1 within one hour are summed, and then the sum is divided by 60 to calculate the arithmetic mean of pressure point 1. Similarly, the same calculation process is performed on the measurements at pressure point 2, strain point 1, and strain point 2 to generate their respective arithmetic means. For example, if the sum of the 60 pressure measurements at pressure point 1 is 301.2 MPa, then its average pressure is... MPa; if the sum of the 60 pressure readings at pressure point 2 is 301.8 MPa, then its average pressure is MPa; if the sum of the 60 strain measurements at strain point 1 is 9024 με, then its average strain is με; if the sum of the 60 strain measurements at strain point 2 is 9036με, then its average strain is με.

[0058] Finally, the baseline vector generation submodule is executed, which arranges the four average values ​​calculated in the previous step in the order of [average value of pressure measuring point 1, average value of pressure measuring point 2, average value of strain measuring point 1, average value of strain measuring point 2] to generate a four-dimensional safety baseline vector. The final generated security baseline vector is: This vector is passed to the real-time state vector deviation calculation module.

[0059] Please see Figure 1 and Figure 3 The real-time state vector deviation calculation module constructs a real-time vector based on the safety baseline vector and the real-time collected pressure and strain measurements. It calculates the weighted absolute difference and generates the baseline deviation, and then transmits the baseline deviation and the real-time vector to the comprehensive risk level determination module and the risk trend gradient calculation module, respectively.

[0060] The real-time state vector deviation calculation module includes: a real-time vector construction submodule, which acquires real-time pressure and strain measurements, arranges them according to the same dimensions and order as the safety baseline vector, and establishes a real-time vector;

[0061] The weighting factor allocation submodule presets an independent weighting factor for each element in the real-time vector based on the importance and data sensitivity of each measurement point, and generates a weight matrix.

[0062] The deviation calculation submodule calls the safety baseline vector, real-time vector and weight matrix, calculates the weighted absolute value sum of the differences between corresponding elements of the vector, and generates the baseline deviation.

[0063] The specific process of calculating baseline deviation by the deviation calculation submodule is as follows: obtain the safe baseline vector, real-time vector and weight matrix;

[0064] Calculate the baseline deviation using the following formula: ;

[0065] in, Represents the final baseline deviation. This represents the total number of measurement point-related parameters contained in the real-time vector and the safety baseline vector. Index representing the parameters of the measuring point. Represents the weight matrix and the first The weighting coefficients corresponding to the parameters of each measuring point Represents the first in the real-time vector The element values ​​corresponding to the parameters of each measuring point Represents the first in the safety baseline vector The element values ​​corresponding to the parameters of each measuring point.

[0066] The specific implementation process of the real-time state vector deviation calculation module is as follows: First, the real-time vector construction submodule is executed. This submodule obtains the current time (denoted as ) from the data acquisition system. Real-time pressure and strain measurements at (time point). For example, at At that moment, the system collected readings of 5.15 MPa at pressure measuring point 1, 5.18 MPa at pressure measuring point 2, 162.1 με at strain measuring point 1, and 165.3 με at strain measuring point 2. These real-time measurements were then compared with the aforementioned safety baseline vector. The same dimensions and measurement point order are used to arrange the data, i.e., [real-time value of pressure measurement point 1, real-time value of pressure measurement point 2, real-time value of strain measurement point 1, real-time value of strain measurement point 2], thus constructing a real-time vector. In this example, the established real-time vector is... .

[0067] Next, the weighting factor allocation submodule is executed to preset an independent weighting factor for each element in the real-time vector. The weighting factor settings are based on finite element analysis of the hanging basket structure and statistical analysis of historical failure data. The analysis shows that the lower chord of the main truss (corresponding to pressure measuring points 1 and 2) is the most important load-bearing structure of the hanging basket, and its pressure change directly reflects the overall load-bearing state of the structure. To determine the specific weight values, a series of loading experiments were conducted: 0% to 150% of the design load was applied to the hanging basket model, and the load level at which the data of each measuring point reached the warning limit was recorded. The experimental results show that the second derivative of the rate of change of the pressure measuring point first exceeded 0.01 when the load reached 85% of the design load, while the strain measuring point only showed a similar change at 105% of the design load. Based on this experimental data, the weight of each measuring point is calculated according to the following formula: .

[0068] In the above formula, Representing the Weighting coefficients for each measurement point; This represents the total number of measuring points, which is 4 in this case. and For the index of the measurement point; Representing the The percentage of load level when a measuring point reaches the warning limit; This represents the adjustment coefficient set based on the importance of the structural part where the measuring point is located.

[0069] The adjustment coefficients in the above formula are set as follows: Based on the finite element analysis results, the stress concentration factor of the lower chord of the main truss is 1.8, while the stress concentration factors of the front hanger connection point and the rear anchor point are 1.1 and 1.0, respectively. Adjustment coefficients are then set for pressure measuring points 1 and 2 accordingly. Set an adjustment coefficient for strain measurement point 1. Set adjustment coefficients for strain measurement point 2 Substitute the experimental data and adjustment coefficients into the formula to calculate: the weight of pressure measuring point 1. The numerator is The weight of pressure measurement point 2 The numerator is Weight of strain measurement point 1 The numerator is The weight of strain measurement point 2 The numerator is The sum of all numerator terms is The weight calculation results for each measuring point are as follows: , , , Therefore, the weight matrix is ​​generated. .

[0070] Subsequently, the deviation calculation submodule is executed, which calls the previously generated safety baseline vector. Real-time vector and weight matrix The system is based on the formula. Calculate baseline deviation .

[0071] The formula is explained in detail below: This represents the final baseline deviation. The total number of measurement point-related parameters contained in the representative vector, in this embodiment, . These are the indices of the measuring point parameters, increasing from 1 to... , Corresponding to pressure measuring point 1, Corresponding to pressure measuring point 2, Corresponding strain measurement point 1, Corresponding strain measurement point 2. Represents the weight matrix and the first The weighting coefficients corresponding to the parameters of each measurement point are determined by the weighting factor allocation submodule. Represents the first in the real-time vector The element value corresponding to the parameter of the _th measuring point, i.e. the _th Real-time measurement values ​​of each measuring point. Represents the first in the safety baseline vector The element value corresponding to the parameter of the _th measuring point, i.e. the _th The baseline value for each measuring point. The operation calculated the first The absolute difference between the real-time value and the benchmark value of each measuring point. The operation multiplies the absolute difference by the corresponding weighting coefficient. The final summation operation... The weighted absolute differences of all measuring points are summed to generate a numerical value. .

[0072] Substitute the specific numerical values ​​into the formula to calculate:

[0073] ;

[0074] ;

[0075] ;

[0076] Calculated baseline deviation The value is 4.2892. The warning threshold for baseline deviation is set to 3.0, and the alarm threshold is set to 6.0. The current calculated result of 4.2892 is greater than the warning threshold of 3.0 but less than the alarm threshold of 6.0, indicating that the current state of the hanging basket has deviated from the normal safe range and has entered the warning state. This baseline deviation... With real-time vector They will be transmitted to the comprehensive risk level determination module and the risk trend gradient calculation module, respectively.

[0077] The baseline deviation warning threshold is set at 3.0, and the alarm threshold at 6.0. The determination process is as follows: Baseline deviation D values ​​for 100 historical hanging basket construction conditions were collected, along with their corresponding risk levels (safe, warning, alarm) assessed jointly by three independent structural safety experts. Receiver Operating Characteristic (ROC) curve analysis was used to construct ROC curves to distinguish between "safe" and "warning" states, and between "warning" and "alarm" states. Based on the principle of maximizing the Youden Index (Youden Index = sensitivity + specificity - 1), the optimal tangent point on the first ROC curve corresponds to a D value of 3.0, which is therefore set as the warning threshold; the optimal tangent point on the second ROC curve corresponds to a D value of 6.0, which is therefore set as the alarm threshold.

[0078] Please see Figure 1 and Figure 4 The risk trend gradient calculation module, based on real-time vectors, processes the data queue using the Kalman filter algorithm, calculates the time change rate of the state values ​​at the beginning and end of the queue to generate a risk gradient, and transmits the risk gradient to the comprehensive risk level determination module.

[0079] The risk gradient includes the rate of pressure change in the core load-bearing components and the rate of strain change at key connection points;

[0080] The risk trend gradient calculation module includes: a data queue management submodule, which stores continuously generated real-time vectors in chronological order to form a data queue with a fixed length;

[0081] The state filtering submodule calls the data queue and uses the Kalman filter algorithm to iteratively process each real-time vector in the queue, filtering out measurement noise and predicting the state at the next moment, thus obtaining the filtered state sequence.

[0082] The gradient calculation submodule extracts the first and last state values ​​of the filtered state sequence, calculates the rate of change of the two over time, and generates the risk gradient.

[0083] The specific process of calculating the risk gradient in the gradient calculation submodule is as follows: obtain the head state value and tail state value of the filtered state sequence;

[0084] Calculate the risk gradient using the following formula: ;

[0085] in, This represents the final generated risk gradient. Represents the filtered state sequence at the current time. The estimated value of the tail state, Represents the filtered state sequence in The estimated state of the head of the queue at time [time]. This represents the fixed length of the data queue, that is, the number of real-time vectors contained in the queue. This represents the time interval between two consecutive real-time data acquisitions.

[0086] The specific implementation process of the risk trend gradient calculation module is as follows: First, the data queue management submodule is executed. This submodule receives and stores the continuously generated real-time vectors transmitted by the real-time state vector deviation calculation module in chronological order. The system sets a fixed length for the data queue. The value is 10, meaning the queue always maintains the 10 most recent real-time vectors. The time interval between two adjacent real-time acquisitions. The interval is set to 5 seconds. When a new real-time vector is generated, it is added to the tail of the queue, while the oldest vector at the head of the queue is removed, maintaining the queue length unchanged. For example, at the current time... The data queue stores data from... Time's up Ten real-time vectors at time points.

[0087] Next, the state filtering submodule is executed. This submodule calls the aforementioned data queue and iteratively processes the 10 real-time vectors in the queue using the Kalman filter algorithm. The processing is not a simple algorithm call; instead, filtering is performed on each of the four components of each vector (two pressure values ​​and two strain values). Taking pressure measurement point 1 as an example, the system establishes a state equation. and observation equations .in, yes The real stress level at any given moment It is the state transition matrix (set to 1 to indicate a uniform change model). It is process noise. yes The observed value at time (i.e., the pressure value in the real-time vector). This is the observation matrix (set to 1). This is measurement noise. The process noise covariance is determined to be Q = 0.0001, and the measurement noise covariance is R = 0.0025. Filtering starts from the head vector and applies it to each time step. (from arrive The process involves two steps: prediction and update. First, the state at the current time is predicted based on the estimate from the previous time step. Then, the predicted value is corrected using the actual measured value at the current time step and the Kalman gain, resulting in the optimal state estimate for the current time step. This process is repeated iteratively until all 10 vectors in the queue have been processed, resulting in a sequence of 10 filtered state vectors, denoted as the filtered state sequence. Some data is shown in Table 2 below.

[0088] The process noise covariance Q = 0.0001 and the measurement noise covariance R = 0.0025 were determined as follows: The measurement noise covariance R was determined by continuously collecting 1000 sets of sensor readings under the unloaded and static state of the basket, calculating the variance of these 1000 sets of static data samples, and using this variance as an estimate of R, resulting in R = 0.0025. The process noise covariance Q was determined by experimenting with 50 different Q values ​​in the interval [1e-6, 1e-2] with a logarithmic step size. Each Q value was applied to the Kalman filtering process of the historical dynamically loaded data sequence, and the root mean square error between the filtered result and the true value (measured by a high-precision laser displacement gauge) was calculated. The Q value that minimized the root mean square error was selected as the final parameter, determined to be Q = 0.0001.

[0089] Table 2: Real-time data and filtered status values ​​of pressure measuring point 1:

[0090] ;

[0091] As shown in Table 2, this table lists some of the original measurements in the queue and the state estimates after Kalman filtering. The filtered data sequence is smoother.

[0092] Subsequently, the gradient calculation submodule is executed to extract the head-of-line state value (i.e., ...) from the filtered state sequence. The state value at the moment and the state value at the tail of the queue (i.e., the current state value) (State value at any given time). In this embodiment, the queue length... Time interval Seconds. Assume the estimated state of the head of pressure measurement point 1 in the filtered state sequence. The estimated value of the tail state is 5.123 MPa. The pressure at pressure measuring point 2 is 5.146 MPa; the head value at pressure measuring point 2 is 5.151 MPa, and the tail value is 5.178 MPa; the head value at strain measuring point 1 is 160.5 με, and the tail value is 163.2 με; the head value at strain measuring point 2 is 163.0 με, and the tail value is 166.4 με. The system is based on the formula... Calculate the risk gradient.

[0093] The formula is explained in detail below: The final generated risk gradient is a vector whose components correspond to the average rate of change of parameters at each measurement point within a specific time window. This represents the estimated tail state of the filtered state sequence at the current time. This represents the state estimate of the filtered state sequence at the head of the queue. (Numerator) The total change in state estimates over the entire data queue time span was calculated. (Denominator) This represents the total time length covered by the data queue, which in this embodiment is... Seconds. The logic of the entire division operation lies in calculating the change in state per unit time, i.e., the rate of change.

[0094] Calculate the parameters for each measuring point separately:

[0095] Risk gradient of pressure measurement point 1 MPa / s;

[0096] Risk gradient at pressure measurement point 2 MPa / s;

[0097] Risk gradient of strain measurement point 1 με / s;

[0098] Risk gradient of strain measurement point 2 με / s.

[0099] The final generated risk gradient is a vector. This risk gradient The data was then transmitted to the comprehensive risk level assessment module. The results showed that all monitoring parameters exhibited an upward trend, with the normalized value of the strain rate of change being greater than that of the normalized value of the pressure rate of change.

[0100] Please see Figure 1 and Figure 5 The comprehensive risk level determination module combines baseline deviation and risk gradient as feature vectors, inputs them into a support vector machine model to obtain classification results, obtains the limit violation flag, determines the risk level based on the classification results and the limit violation flag, and generates a signal.

[0101] The comprehensive risk level determination module includes a feature vector fusion submodule, which combines baseline deviation and risk gradient as two independent feature dimensions after normalization into a single feature vector.

[0102] The model classification processing submodule inputs the feature vectors into a pre-trained support vector machine model, calculates the classification results through the model's decision function, and generates classification results.

[0103] The risk level synthesis submodule obtains independent over-limit flags, determines the risk level and generates a corresponding early warning signal based on preset decision rules, combined with the classification results and over-limit flags;

[0104] The support vector machine model uses a radial basis kernel function for nonlinear mapping.

[0105] In the model classification processing submodule, the feature vectors are mapped to a high-dimensional feature space through the radial basis kernel function, and the optimal classification hyperplane is found in the high-dimensional feature space.

[0106] The radial basis function kernel is calculated according to the following formula: ;

[0107] in, The kernel function calculation result represents the kernel function between two different feature vectors input to the support vector machine model. and This represents any two feature vectors used as input. These represent the kernel parameters used to control the range of action of the radial basis function. This represents the square of the Euclidean distance between two eigenvectors. Represents the natural exponential function;

[0108] nuclear parameters The classification results are determined by cross-validation training on multiple sets of typical working condition data during the historical hanging basket contraction process. The classification results include three levels: safety, early warning, and alarm.

[0109] The specific implementation process of the comprehensive risk level determination module is as follows: First, the feature vector fusion submodule is executed, which receives the baseline deviation from the aforementioned module. and risk gradient vector The two inputs are then normalized. Normalization is based on preset maximum reference values. The maximum reference values ​​are set as follows: baseline deviation 10.0, pressure change rate 0.01 MPa / s, and strain change rate 0.5 με / s. The normalization process involves dividing each original value by its corresponding maximum reference value.

[0110] Normalized baseline deviation ;

[0111] Normalized risk gradient ;

[0112] Normalized and All components are combined into a five-dimensional feature vector:

[0113] .

[0114] The maximum reference values ​​used in the above normalization were determined as follows: By analyzing all monitoring data from 50 completed hanging basket construction projects in the historical database, the maximum values ​​of each parameter (baseline deviation, rate of change at each measuring point) throughout the entire construction lifecycle were extracted, forming multiple sets of maximum values. The 99th percentile of the maximum value set for each parameter was used as the maximum reference value for that parameter. Using this method, the maximum reference value for baseline deviation was determined to be 10.0, the maximum reference value for pressure change rate was 0.01 MPa / s, and the maximum reference value for strain change rate was 0.5 με / s.

[0115] Next, the model classification processing submodule is executed to process the fused feature vectors. The input is fed into a pre-trained Support Vector Machine (SVM) model. This model uses a radial basis function (RBF) kernel for non-linear mapping. The model receives feature vectors... Then, the classification is calculated using an internal decision function. The core of this function is to use a kernel function to calculate the relationship between the input vector and all support vectors, and then perform a weighted summation using Lagrange multipliers, ultimately outputting a classification result. In this embodiment, the classification result is defined as one of three levels: safe, warning, or alarm. The radial basis function is based on the formula... Perform the calculation.

[0116] The formula is explained in detail below: The kernel function calculation result represents the kernel function between two different feature vectors input to the support vector machine model. and These are any two feature vectors used as input. It represents the square of the Euclidean distance between two eigenvectors, calculated as the sum of the squares of the differences between corresponding elements of the two vectors. It is a positive kernel parameter. This represents the natural exponential function. The logic of the entire formula lies in nonlinearly transforming the Euclidean distance between two input vectors using an exponential function.

[0117] nuclear parameters The determination process is as follows: A sample set containing 1000 sets of historical data is collected, each set containing a feature vector and its corresponding actual risk level (manually labeled by experts). A 10-fold cross-validation method is used to train the model to find the optimal model. Value. Tested. The value range is The step size is 0.01. For each The value was used to repeat the training and validation process 10 times, and the average classification accuracy was calculated. Experimental data showed that when At that time, the model achieved a peak average classification accuracy of 97.2%. Therefore, the kernel parameters used in the model were ultimately determined. .

[0118] With the current feature vector A support vector belonging to the "early warning" class For example, calculate its kernel function value:

[0119] Euclidean distance squared ;

[0120] Kernel function value .

[0121] Since the kernel function value is close to 1, it indicates that the current state is similar to a typical "warning" state. After calculation with all support vectors and discrimination by the decision function, the classification result output by the SVM model is "warning".

[0122] Finally, the risk level synthesis submodule is executed, which obtains an independent over-limit flag. The over-limit flag is set according to a fixed safety threshold. If the real-time value of any pressure measuring point exceeds 10 MPa or the real-time value of any strain measuring point exceeds 1000 με, the over-limit flag is set to "yes". In this embodiment, the real-time measurements do not exceed this absolute threshold, so the over-limit flag is "no". According to the preset decision rule: if the over-limit flag is "yes", the risk level is directly determined as "alarm"; if the over-limit flag is "no", the classification result of the SVM model is used. Since the current over-limit flag is "no" and the SVM output is "warning", the system finally determines the risk level as "warning" and generates a corresponding warning signal, such as activating the yellow rotating warning light on site and emitting a 1Hz buzzer.

[0123] The above-mentioned fixed safety threshold is set based on the following: The threshold is determined according to the hanging basket design specifications and the mechanical properties of the materials. The lower chord of the main truss is made of Q345 steel, with a yield strength of 345MPa. According to the "Steel Structure Design Standard" (GB50017-2017), its design strength value is 215MPa. The design strength value is converted to the pressure sensor reading in this embodiment, and a safety factor of 1.5 is considered, i.e. MPa. To allow for margin, the absolute alarm threshold for the pressure measuring point is set to 10 MPa. Similarly, based on the allowable strain value of the material of the connecting component where the strain gauge is located and applying a safety factor, the absolute alarm threshold for the strain measuring point is set to 1000 με.

[0124] The above embodiments illustrate preferred embodiments of the present invention. Any equivalent adjustments to the technical solution based on software engineering methods are within the scope of protection, including but not limited to: implementing algorithm logic using different programming languages, refactoring functional modules into services, adjusting data interaction protocols, and optimizing resource scheduling strategies. Any implementation scheme derived from reasonable modifications to the data processing flow, service call chain, or system architecture layer without departing from the core technology of the present invention should be considered within the scope of protection defined by the claims of the present invention.

Claims

1. A hanging basket retraction safety detection system, characterized in that, The system includes: The dynamic safety baseline construction module calculates the average value of pressure and strain measurements, combines the calculated average pressure and strain values ​​to generate a safety baseline vector, and transmits the safety baseline vector to the real-time state vector deviation calculation module. The real-time state vector deviation calculation module constructs a real-time vector based on the safety baseline vector and the real-time collected pressure and strain measurements, calculates the weighted absolute difference and generates the baseline deviation, and transmits the baseline deviation and the real-time vector to the comprehensive risk level determination module and the risk trend gradient calculation module, respectively. The real-time state vector deviation calculation module includes: The real-time vector construction submodule acquires real-time pressure and strain measurements, arranges them according to the same dimensions and order as the safety baseline vector, and establishes a real-time vector. The weighting factor allocation submodule presets an independent weighting factor for each element in the real-time vector based on the importance and data sensitivity of each measurement point, and generates a weight matrix. The deviation calculation submodule calls the safety baseline vector, the real-time vector, and the weight matrix to calculate the weighted sum of the absolute values ​​of the differences between corresponding elements of the vectors, and generates the baseline deviation. The risk trend gradient calculation module, based on the real-time vector, processes the data queue using the Kalman filter algorithm, calculates the time change rate of the state values ​​at the beginning and end of the queue to generate a risk gradient, and transmits the risk gradient to the comprehensive risk level determination module. The risk trend gradient calculation module includes: The data queue management submodule stores the continuously generated real-time vectors in chronological order to form a data queue with a fixed length. The state filtering processing submodule calls the data queue and uses the Kalman filter algorithm to iteratively process each real-time vector in the queue to filter out measurement noise and predict the state at the next moment, thereby obtaining a filtered state sequence. The gradient calculation submodule extracts the first and last state values ​​of the filtered state sequence, calculates the rate of change of the two over time, and generates a risk gradient. The comprehensive risk level determination module combines the baseline deviation and the risk gradient as a feature vector, inputs it into the support vector machine model to obtain the classification result, obtains the limit violation flag, determines the risk level based on the classification result and the limit violation flag, and generates a signal.

2. The hanging basket retraction safety detection system according to claim 1, characterized in that, The safety baseline vector includes the reference pressure value and reference strain value of each measuring point, and the risk gradient includes the pressure change rate of the core load-bearing part and the strain change rate of the key connection point.

3. The hanging basket retraction safety detection system according to claim 1, characterized in that, The dynamic security baseline construction module includes: The data acquisition submodule acquires the pressure and strain measurements at each measuring point within a specified time period and establishes a raw data set containing all measurements. The mean calculation submodule calls the original data set to calculate the arithmetic mean of all pressure measurements and the arithmetic mean of all strain measurements, generating the pressure average and strain average. The baseline vector generation submodule combines the average pressure and the average strain to generate a safe baseline vector.

4. The hanging basket retraction safety detection system according to claim 1, characterized in that, The comprehensive risk level determination module includes: The feature vector fusion submodule takes the baseline deviation and the risk gradient as two independent feature dimensions, performs normalization processing, and combines them into a feature vector. The model classification processing submodule inputs the feature vector into the pre-trained support vector machine model, calculates the classification result through the model's decision function, and generates the classification result. The risk level synthesis submodule acquires the independent limit-crossing flag, determines the risk level and generates a corresponding early warning signal based on the preset decision rules, the classification result and the limit-crossing flag.

5. The hanging basket retraction safety detection system according to claim 1, characterized in that, The specific process by which the deviation calculation submodule calculates the baseline deviation is as follows: obtaining the safe baseline vector, the real-time vector, and the weight matrix; The baseline deviation is calculated using the following formula: ; in, This represents the final baseline deviation. This represents the total number of measurement point-related parameters contained in the real-time vector and the safety baseline vector. Index representing the parameters of the measuring point. Represents the weight matrix with respect to the first... The weighting coefficients corresponding to the parameters of each measuring point Represents the first in the real-time vector The element values ​​corresponding to the parameters of each measuring point Represents the first in the security baseline vector The element values ​​corresponding to the parameters of each measuring point.

6. The hanging basket retraction safety detection system according to claim 1, characterized in that, The specific process of the gradient calculation submodule in calculating the risk gradient is as follows: obtaining the head state value and tail state value of the filtered state sequence; The risk gradient is calculated using the following formula: ; in, This represents the final generated risk gradient. This represents the filtered state sequence at the current time. The estimated value of the tail state, Represents the filtered state sequence in The estimated state of the head of the queue at time [time]. This represents the fixed length of the data queue, that is, the number of real-time vectors contained in the queue. This represents the time interval between two consecutive real-time data acquisitions.

7. The hanging basket retraction safety detection system according to claim 4, characterized in that, The support vector machine model uses a radial basis kernel function for nonlinear mapping processing. In the model classification processing submodule, the feature vector is mapped to a high-dimensional feature space through the radial basis kernel function, and the optimal classification hyperplane is found in the high-dimensional feature space. The radial basis kernel function is calculated according to the following formula: ; in, The kernel function calculation result represents the kernel function between two different feature vectors input to the support vector machine model. and represents any two feature vectors that are used as input. Represents kernel parameters, This represents the square of the Euclidean distance between the two eigenvectors. Represents the natural exponential function; The kernel parameters The classification results are determined by cross-validation training on multiple sets of typical working condition data during the historical hanging basket retraction process. The classification results include three levels: safety, early warning, and alarm.

Images