An upper and lower bound estimation interval early warning method based on shore bridge health monitoring data

Abstract

The application discloses a kind of upper and lower limit estimation interval early warning method based on shore bridge health monitoring data, and relates to the technical field of shore bridge structure health monitoring early warning, the method comprises: shore bridge girder installs acceleration sensor, carries out structure monitoring, obtains monitoring data;Extract the time domain feature of health monitoring data, construct time domain feature interval early warning model based on upper and lower limit estimation method, obtain initial dynamic threshold value;Optimize interval early warning model parameter, obtain optimal early warning interval;According to the trend change of the time domain feature of real-time monitoring data in optimal early warning interval, realize real-time dynamic early warning.The method of the application can capture abnormal signs of shore bridge structure in time, realize real-time monitoring and dynamic early warning of the health state of shore bridge structure, effectively improve the accuracy and timeliness of abnormal state early warning of shore bridge structure.

Description

Technical Field

[0001] This invention relates to the field of quay crane structural health monitoring and early warning technology, specifically to an early warning method based on upper and lower bound estimation intervals of quay crane health monitoring data. Background Technology

[0002] Quay cranes (or quay cranes for short) are core equipment in port loading and unloading operations. Operating at the quayhead, they handle the loading and unloading of container ships, and their efficiency determines the overall level of container terminal operations. The health of the quay cranes directly determines their safe operation. Conducting structural health monitoring of quay cranes, capturing abnormal signs in real time, and issuing timely warnings of structural anomalies are crucial for maintaining the structural safety of quay cranes, reducing the incidence of accidents, and minimizing economic losses and casualties. However, to date, there is a lack of complete fault or damage sample data in quay crane structural health monitoring, resulting in insufficient timeliness and low accuracy in issuing warnings of abnormal structural conditions. Summary of the Invention

[0003] This invention discloses an early warning method for estimating the upper and lower bounds of quay crane health monitoring data. The purpose is to solve the technical problem that the current quay crane structural health monitoring data lacks complete fault or damage sample data, which leads to insufficient timeliness and low accuracy of early warning.

[0004] To address the above technical problems, this invention provides a method for early warning based on upper and lower bound estimation intervals of quay bridge health monitoring data, characterized by the following specific steps:

[0005] Step S1: Install acceleration sensors at different locations on the main girder structure of the quay crane to monitor the structure at different locations and collect structural vibration monitoring data;

[0006] Step S2: Extract time-domain feature indicators of structural health status based on vibration monitoring data, construct a time-domain feature interval early warning model using the upper and lower bound estimation method based on correlation vector machine, and obtain the initial dynamic early warning threshold of time-domain features;

[0007] Step S3: Use the particle swarm optimization algorithm to optimize the parameters of the interval early warning model, optimize the initial early warning interval, and then obtain the optimal early warning interval;

[0008] Step S4: Determine the structural state based on the trend changes of time-domain characteristic indicators within the optimal warning interval, thereby achieving dynamic early warning of structural anomalies.

[0009] Preferably, in step S1, MEMS fiber optic acceleration sensors are installed at different measuring points on the main girder structure of the quay crane to obtain structural health monitoring data covering one working cycle of the quay crane from the acceleration sensors. The sampling length of the monitoring signal should include changes in the time domain characteristics.

[0010] Preferably, in step S2, the acquired health monitoring data is preprocessed for noise reduction using the moving average method, and then the preprocessed monitoring dataset is subjected to time-domain feature extraction to establish an upper and lower bound estimation interval early warning model based on the correlation vector machine; the time-domain feature sequence is divided into training and testing sets in a 3:2 ratio, and the model is trained and tested to predict the upper and lower thresholds of the early warning interval.

[0011] Further, in step S2, the calculation window width for extracting the time-domain feature indicators of structural health status is set to 200 data points (2s), and the mean, standard deviation, kurtosis, and skewness of the training and test sets are calculated respectively. The calculation formulas for each indicator are as follows:

[0012] average value:

[0013] Standard deviation:

[0014] kurtosis:

[0015] Skewness:

[0016] x i : Vibration data at time i, N: amount of data collected, M: average value of vibration acceleration amplitude of quay crane structure, S: standard deviation of vibration acceleration amplitude of quay crane structure, K: kurtosis of vibration acceleration amplitude of quay crane structure, D: skewness of vibration acceleration amplitude of quay crane structure.

[0017] Furthermore, in step S2, based on the upper and lower bound estimation (LUBE) interval prediction framework, an upper and lower bound estimation interval early warning model based on correlation vector machine is established. Two correlation vector machine (RVM) models are used to train the upper and lower thresholds of the early warning interval of the time-domain feature index of structural health status to obtain the initial dynamic early warning interval.

[0018] Furthermore, in step S2, the kernel functions of both RVM models are radial basis function (RBF) kernel functions, the input variables for training both RVM models are the same time-domain features extracted within a given time length, and the outputs of the model training are the predicted upper and lower thresholds of the warning interval, respectively.

[0019] Preferably, in step S3, the quality of the constructed initial warning interval needs to be evaluated using relevant indicators. A higher quality warning interval is one that can cover a greater number of target values ​​within a potentially narrow prediction interval. The performance evaluation indicators for the established initial warning interval are: Predicted Interval Coverage Probability (PICP), Normalized Average Predicted Interval Width (NMPIW), and Coverage Width Criterion (CWC). The formulas for these interval performance evaluation indicators are as follows:

[0020] Predicted interval coverage probability:

[0021] N: The length of the sample data in the input dataset. The coverage of the i-th prediction interval;

[0022] Normalized average prediction interval width:

[0023] ρ: The range of input sample data, ρ = x max -x min , The lower bound of the i-th prediction interval The upper bound of the i-th prediction interval, α: confidence level;

[0024] Coverage width criteria:

[0025] Confidence level, υ: penalty factor, γ: step function.

[0026] Using CWC as the loss function for training the RVM model and the RVM kernel width as the optimization variable, the Particle Swarm Optimization (PSO) algorithm is employed to continuously optimize the RVM kernel parameters. By minimizing the loss function CWC, the optimal kernel width is obtained, leading to the optimized RVM model. The upper and lower thresholds of the warning interval are then updated using the two optimized RVM models to optimize the initially constructed dynamic warning interval. The model outputs are the optimized upper and lower thresholds of the warning interval, respectively.

[0027] Preferably, in step S4, for new monitoring data acquired in real time, the time-domain features of the new monitoring data are extracted and input into the trained interval early warning model. The model then determines whether the newly input time-domain features exceed the upper and lower limits of the early warning interval based on their trend changes. If the input time-domain feature sequence exceeds the upper and lower thresholds of the early warning interval, or shows a trend exceeding the upper and lower limits of the early warning interval, it indicates an abnormal structural state, and an early warning is issued.

[0028] This invention discloses an early warning method for upper and lower bound estimation intervals based on quay crane health monitoring data. It utilizes the core idea of ​​upper and lower bound estimation methods to construct early warning intervals for time-domain characteristic indicators of structural health status, and establishes an early warning system based on particle swarm optimization-correlation vector machine (PSO-VR). By establishing the optimal early warning interval, it achieves early warning of structural anomalies. Within the framework of upper and lower bound estimation, this method uses two Restricted Vector Machine (RVM) models to construct the upper and lower bounds of the time-domain characteristic early warning intervals. The PSO algorithm is used to optimize the RVM kernel parameters, and the quality of the constructed early warning intervals is evaluated using interval performance indices PICP, NMPIW, and CWC. This invention can better capture changes in the structural state of quay cranes, achieving early warning of abnormal structural states. It solves the technical problem of untimely early warning of abnormal structural states in quay cranes when actual monitoring data damage information is incomplete, thus improving the accuracy and timeliness of early warning of structural anomalies. Attached Figure Description

[0029] Figure 1 This is a flowchart of an early warning method for estimating the upper and lower bounds of a quay bridge health monitoring data.

[0030] Figure 2 This is a schematic diagram of the installation of the quay bridge structure monitoring sensor in an embodiment of the present invention;

[0031] Figure 3 These are four temporal feature sequence diagrams of the sample dataset in this embodiment of the invention;

[0032] Figure 4 This is a diagram of the upper and lower bound estimation interval early warning model based on particle swarm optimization-correlation vector machine in an embodiment of the present invention;

[0033] Figure 5 This is a diagram showing the construction results of early warning intervals for different time-domain characteristics under structural health conditions in this embodiment of the invention: Figure 5 (a) shows the result of constructing the early warning interval for skewness. Figure 5 (b) shows the result of constructing the kurtosis warning interval. Figure 5 (c) represents the result of constructing the warning interval for the standard deviation. Figure 5 (d) represents the result of constructing the warning interval based on the mean.

[0034] Figure 6 This is a diagram showing the early warning results of abnormal conditions of the quay bridge structure in this embodiment of the invention: Figure 6 (a) represents the warning interval for the constructed mean. Figure 6 (b) represents the constructed warning interval for the standard deviation. Figure 6 (c) represents the constructed skewness warning interval. Figure 6 (d) represents the early warning range for the constructed kurtosis. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in further detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of the invention.

[0036] This invention discloses an early warning method for estimating upper and lower bounds of quay bridge health monitoring data. The method flow is as follows: Figure 1 As shown, it includes:

[0037] Step S1: Install acceleration sensors at different locations on the main girder structure of the quay crane to monitor the structure at different locations and collect structural vibration monitoring data;

[0038] Step S2: Extract time-domain feature indicators of structural health status based on vibration monitoring data, construct a time-domain feature interval early warning model using the upper and lower bound estimation method based on correlation vector machine, and obtain the initial dynamic early warning threshold of time-domain features;

[0039] Step S3: Use the particle swarm optimization algorithm to optimize the parameters of the interval early warning model, optimize the initial early warning interval, and then obtain the optimal early warning interval;

[0040] Step S4: Determine the structural state based on the trend changes of time-domain characteristic indicators within the optimal warning interval, thereby achieving dynamic early warning of structural anomalies.

[0041] The following example uses monitoring data of a quay crane structure on a certain day to verify the effectiveness and correctness of the method of this invention, in conjunction with the accompanying drawings.

[0042] Step S1: Install MEMS fiber optic acceleration sensors at different measuring points on the main girder structure of the quay crane, with a sampling frequency of 100 Hz. Figure 2 This is a schematic diagram of the sensor layout for the quay crane. A location near the sea-side gate frame, namely measurement point V5, was selected. 17 minutes of health monitoring data, totaling 102,000 data points, were collected from this point. The sampling length of the monitoring signal should include changes in the time domain characteristics.

[0043] Step S2: The acquired health monitoring data undergoes noise reduction preprocessing using a moving average method. Temporal features are extracted from the preprocessed monitoring dataset. The extracted temporal feature sequences are then divided into training and testing sets in a 3:2 ratio. The calculation window width for the temporal feature indicators is set to 200 data points (2 seconds). The mean, standard deviation, kurtosis, and skewness of the training and testing sets are calculated respectively, resulting in the temporal feature sequences of the dataset as shown below. Figure 3 As shown. The calculation formulas for each indicator are as follows:

[0044] average value:

[0045] Standard deviation:

[0046] kurtosis:

[0047] Skewness is:

[0048] x i : Vibration data at time i, N: amount of data collected, M: average value of vibration acceleration amplitude of quay crane structure, S: standard deviation of vibration acceleration amplitude of quay crane structure, K: kurtosis of vibration acceleration amplitude of quay crane structure, D: skewness value of vibration acceleration amplitude of quay crane structure.

[0049] Based on the upper and lower bound estimation (LUBE) interval prediction framework, two correlation vector machine (RVM) models are used to construct the upper and lower thresholds of the warning interval for the time-domain characteristic indicators of structural health status, thus establishing an upper and lower bound estimation interval warning model based on correlation vector machine. Figure 4 As shown, both RVM models use radial basis function (RBF) kernel functions, and the training set is used for model training and testing. The input variables for training both RVM models are the same time-domain features extracted within a given time period. The outputs of the two RVM models are the upper and lower thresholds of the warning interval, respectively, to obtain the initial dynamic warning interval.

[0050] Step S3: The quality of the constructed initial warning interval needs to be evaluated using relevant indicators. A higher quality warning interval is one that can cover a greater number of target values ​​within a potentially narrow prediction interval. The Coverage Width Criterion (CWC), Predicted Interval Coverage Probability (PICP), and Normalized Average Predicted Interval Width (NMPIW) are used as performance evaluation indicators to assess the quality of the established initial warning intervals. The formulas for these interval performance evaluation indicators are as follows:

[0051] Predicted interval coverage probability:

[0052] N: The length of the sample data in the input dataset. The coverage of the i-th prediction interval;

[0053] PICP is directly related to the quality of the constructed prediction interval (PI) and measures the reliability of the interval. That is, the larger the PICP value, the more target values ​​the PI contains, and the higher the reliability of the PI.

[0054] Normalized average prediction interval width:

[0055] ρ: The range of input sample data. The lower bound of the i-th prediction interval The upper bound of the i-th prediction interval, α: confidence level;

[0056] Coverage width criteria:

[0057] Confidence level, υ: penalty factor, γ: step function.

[0058] Using CWC as the loss function for training the RVM model and the RVM kernel width as the optimization variable, the Particle Swarm Optimization (PSO) algorithm is employed to continuously optimize the RVM kernel parameters. By minimizing the loss function CWC, the optimal kernel width is obtained, leading to the optimized RVM model. The upper and lower thresholds of the warning interval are then updated using the two optimized RVM models to optimize the initially constructed dynamic warning interval. The model outputs are the optimized upper and lower thresholds of the warning interval, respectively.

[0059] Step S4: For new monitoring data acquired in real time, extract the time-domain features of the new monitoring data, input them into the trained interval early warning model, and determine whether the new input time-domain features exceed the upper and lower limits of the early warning interval based on the trend changes. Figure 5 The results of constructing warning intervals for different time-domain characteristics under structural health conditions show that the warning intervals constructed by the training set can well cover the mean, standard deviation, skewness and kurtosis. The dynamic changes of the upper and lower limits of the warning intervals of the test set are consistent with the changing trends of the actual characteristics. They do not exceed the warning intervals or show a trend of exceeding the warning intervals, indicating that the structural state is normal and no warning is issued.

[0060] To verify the early warning effect of the proposed method under structural damage conditions, a finite element model of the quay crane structure was established, and structural damage to the main beam was simulated by reducing element stiffness. A moving load was used to simulate the movement of a trolley carrying a load on the main beam, and transient analysis was performed on the finite element model of the quay crane structure. The moving load was used as input to the transient analysis, and acceleration response data at the same locations as the actual monitoring points were obtained. The time-domain feature sequence of the response data under damaged conditions was extracted and input into a trained interval early warning model to obtain the trend changes of the time-domain feature indicators within the optimal early warning interval. Figure 6 As shown, when structural anomalies occur, the feature sequences in different time domains all begin to deviate from the center of the warning interval. Many scatter plot values ​​of kurtosis and skewness gradually exceed the warning interval threshold. Some scatter plot values ​​of the mean and standard deviation also exceed the warning interval threshold, indicating an abnormal structural state and triggering an early warning. This demonstrates that the method of the present invention can better capture changes in structural state and effectively achieve early warning of abnormal conditions in quay bridge structures.

[0061] The above are merely preferred embodiments of the present invention and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should recognize that any equivalent substitutions and obvious changes made based on the description and illustrations of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for early warning based on upper and lower bound estimation intervals of quay bridge health monitoring data, characterized in that, Includes the following steps: Step S1: Install acceleration sensors at different locations on the main girder structure of the quay crane to monitor the structure at different locations and collect structural vibration monitoring data; Step S2: Extract time-domain feature indicators of structural health status based on vibration monitoring data, construct a time-domain feature interval early warning model using the upper and lower bound estimation method based on correlation vector machine, and obtain the initial dynamic early warning threshold of time-domain features; Based on the upper and lower bound estimation (LUBE) interval prediction framework, an upper and lower bound estimation interval early warning model based on correlation vector machine is established. Two correlation vector machine (RVM) models are used to train the upper and lower thresholds of the early warning interval for the time domain feature indicators of structural health status. The kernel functions of the two correlation vector machine models are both radial basis (RBF) kernel functions. The input variables for model training are the same time domain features extracted within a given time length. The outputs of model training are the predicted upper and lower thresholds of the early warning interval, respectively, to obtain the initial dynamic early warning interval. Step S3: Use the particle swarm optimization algorithm to optimize the parameters of the interval early warning model, optimize the initial early warning interval, and then obtain the optimal early warning interval; Step S4: Determine the structural state based on the trend changes of time-domain characteristic indicators within the optimal warning interval, thereby achieving dynamic early warning of structural anomalies.

2. The method for early warning of upper and lower bounds estimation intervals based on quay bridge health monitoring data according to claim 1, characterized in that, In step S1, MEMS fiber optic acceleration sensors are installed at different measuring points on the main girder structure of the quay crane. Structural health monitoring data covering one working cycle of the quay crane is obtained from the quay crane acceleration sensors. The sampling length of the monitoring signal should include changes in the time domain characteristics.

3. The method for early warning based on upper and lower bound estimation intervals of quay bridge health monitoring data according to claim 1, characterized in that, In step S2, the acquired health monitoring data is preprocessed for noise reduction using the moving average method, and then the preprocessed monitoring dataset is subjected to time-domain feature extraction to establish an upper and lower bound estimation interval early warning model based on the correlation vector machine. The time-domain feature sequence is divided into training and testing sets in a 3:2 ratio, and the model is trained and tested to predict the upper and lower thresholds of the early warning interval.

4. The method for early warning based on upper and lower bound estimation intervals of quay bridge health monitoring data according to claim 3, characterized in that, In step S2, the calculation window width for the time-domain feature indicators of the structural health monitoring data is set to 200 data points. The mean, standard deviation, kurtosis, and skewness of the training and test sets are calculated respectively. The calculation formulas for each indicator are as follows: average value: Standard deviation: kurtosis: Skewness: : Vibration data at time i, N: amount of data collected, M: average value of vibration acceleration amplitude of quay crane structure, S: standard deviation of vibration acceleration amplitude of quay crane structure, K: kurtosis of vibration acceleration amplitude of quay crane structure, D: skewness of vibration acceleration amplitude of quay crane structure.

5. The method for early warning based on upper and lower bound estimation intervals of quay bridge health monitoring data according to claim 1, characterized in that, In step S3, the quality of the constructed initial warning interval needs to be evaluated using relevant indicators. A higher quality warning interval is one that can cover a greater number of target values ​​within a potentially narrow prediction interval. The performance evaluation indicators for the established initial warning interval are: Predicted Interval Coverage Probability (PICP), Normalized Average Predicted Interval Width (NMPIW), and Coverage Width Criterion (CWC). The formulas for these interval performance evaluation indicators are as follows: Predicted interval coverage probability: N: The length of the sample data in the input dataset. : The coverage of the i-th prediction interval; Normalized average prediction interval width: : The range of input sample data , : The lower bound of the i-th prediction interval : The upper bound of the i-th prediction interval Confidence level; Coverage width criteria: Confidence level : Punishment factor Step function.

6. The method for early warning of upper and lower bound estimation intervals based on quay bridge health monitoring data according to claim 5, characterized in that, In step S3, CWC is used as the loss function for training the correlation vector machine model, and the RVM kernel width is used as the optimization variable. The particle swarm optimization (PSO) algorithm is used to continuously optimize the RVM kernel parameters. By minimizing the loss function CWC, the optimal kernel width is obtained, and thus the optimized correlation vector machine model is obtained. The upper and lower thresholds of the warning interval are updated using the two optimized correlation vector machine models to optimize the initially constructed dynamic warning interval. The model outputs are the upper and lower thresholds of the optimized warning interval, respectively.

7. The method for early warning based on upper and lower bound estimation intervals of quay bridge health monitoring data according to claim 1, characterized in that, In step S4, new monitoring data is acquired in real time, and the time-domain features of the new monitoring data are extracted and input into the trained interval early warning model. The trend change of the newly input time-domain features is used to determine whether it exceeds the upper and lower limits of the early warning interval. If the input time-domain feature sequence exceeds the upper and lower thresholds of the early warning interval, or shows a trend of change exceeding the upper and lower limits of the early warning interval, it indicates that the structural state is abnormal and an early warning is issued.

Images