AI-driven vibration monitoring device and beam drop risk early warning system for curved bridges

Abstract

This application relates to the field of intelligent sensing system monitoring technology, specifically to an AI-driven vibration monitoring device for curved bridges and a beam-fall risk early warning system. The device includes an intelligent sensing system and a wireless communication unit, used to collect data on the reaction forces and vibrations generated by the bridge supports and transmit them to the beam-fall risk early warning system. The system performs anomaly analysis on the reaction force and vibration data of the curved bridge, using historical data to predict the probability of risk during the current vehicle load, thus achieving vibration monitoring and early warning for curved bridges. This application aims to improve the accuracy of vibration monitoring and early warning for curved bridges.

Description

Technical Field

[0001] This application relates to the field of intelligent sensing system monitoring technology, specifically to an AI-driven curved bridge vibration monitoring device and a beam drop risk early warning system. Background Technology

[0002] Curved bridges play a vital role in modern transportation infrastructure, but their complex structure and stress characteristics expose them to significant safety risks during operation. Bridge vibration is a crucial factor affecting bridge safety; excessive vibration can lead to structural damage. Specifically, curved beam bridges are prone to bending-torsional coupling effects under external loads. This complex spatial stress state can cause supports to detach or lift, easily inducing lateral buckling or overturning of the beam, increasing the risk of beam collapse. This risk increases significantly, especially under extreme events such as earthquakes, ship collisions, and floods. Therefore, vibration monitoring of curved bridges is essential.

[0003] During operation, bridges are subjected to both static and dynamic loads, with vehicle loads being one of the most common dynamic loads. When vehicles cross a bridge, the interaction between the bridge and the vehicle generates a vehicle-bridge coupling response. Moving vehicles continuously impact the bridge, causing vibration and fatigue problems. Compared to traditional straight beam bridges, curved beam bridges experience more complex stresses, especially noticeable at curves, and the stress state of the bridge varies at different vehicle speeds. Conventional methods fail to adequately consider the vehicle-bridge coupling response caused by the curvature of curved bridges during monitoring, making it difficult to accurately assess the overall stability of the results and resulting in insufficient accuracy in vibration monitoring and early warning for curved bridges. Summary of the Invention

[0004] To address the aforementioned technical issues, this application provides an AI-driven vibration monitoring device for curved bridges and a beam drop risk early warning system. The specific technical solution adopted is as follows:

[0005] In a first aspect, one embodiment of this application provides an AI-driven vibration monitoring device for curved bridges, which includes an intelligent sensing system and a wireless communication unit.

[0006] Among them, the pressure sensor is used to acquire the reaction force data generated by the support of the curved bridge on the bridge, and the reaction force data is the reaction force generated by the support on the bridge.

[0007] A low-frequency accelerometer is used to acquire vibration simulation signals of curved bridge supports.

[0008] The signal processing unit, connected to the low-frequency acceleration sensor, is used to filter, amplify, and convert the vibration analog signal to digital, and output vibration data.

[0009] The wireless communication unit is connected to the intelligent sensing system and is used to send the reaction force data and vibration data to the beam drop risk early warning system in real time for data analysis, processing and early warning.

[0010] A power supply is provided to power the intelligent sensing system and the wireless communication unit.

[0011] Secondly, another embodiment of this application also provides a beam drop risk early warning system. This system is built into the aforementioned curved bridge vibration monitoring device and is used to receive and process reaction force data and vibration data sent by the wireless communication unit. The system implements the following steps:

[0012] For each support, the reaction force data and vibration data under each vehicle load are as follows:

[0013] After curve fitting using reaction force data, the peak characteristics on the curve are used to determine the overall waveform anomaly coefficient of the peak; and the local peak values ​​of the reaction force data within the corresponding peak width range of the peak are extracted, and the standard deviation and summation of the Euclidean distance between them and the corresponding values ​​on the fitted curve are calculated to determine the first anomaly coefficient of the reaction force data.

[0014] The vibration data segment corresponding to the vehicle load is obtained by analyzing whether the mean value of the vibration data within the preset sliding window exceeds a preset threshold; the coefficient of variation of the peak sequence and valley sequence in the vibration data segment after difference, as well as the correlation coefficient between the peak sequence and valley sequence, are extracted to determine the second anomaly coefficient of the vibration data.

[0015] By using the correlation coefficient between the first and second anomaly coefficients of the same support under N consecutive vehicle loads, the mean difference between the correlation coefficients of the target outer ring support and its relative inner ring support and relative outer ring support is calculated, and the load influence difference coefficient of each outer ring support under each vehicle load is obtained.

[0016] By combining the first and second anomaly coefficients and the load effect difference coefficient, the significance coefficient for each vehicle load application is determined;

[0017] The risk probability of the current vehicle load is predicted by using the significance coefficients of all previous historical vehicle loads, in order to determine whether to issue an early warning.

[0018] Preferably, the overall waveform anomaly coefficient of the wave peak is determined by the product of the absolute values ​​of kurtosis and skewness within the peak width range of the fitted curve.

[0019] Preferably, the peak is the maximum value among all maxima on the fitted curve.

[0020] Preferably, the first anomaly coefficient is determined by the product of the overall waveform anomaly coefficient and the local jitter significance value.

[0021] Preferably, the method for obtaining the vibration data segment is as follows: if the average value of the data within the window is greater than a preset threshold, then the data point at the center of the window is taken as valid vibration data, and continuous valid vibration data is taken as a vibration data segment.

[0022] Preferably, the method for determining the second abnormal data is as follows: calculating the mean coefficient of variation of the sequence after differentiating the peak sequence and the valley sequence; and taking the product of the mean coefficient of variation and the correlation coefficient as the second abnormal coefficient.

[0023] Preferably, the method for obtaining the peak sequence and valley sequence is as follows: the peak and valley values ​​of the vibration data are extracted by an automatic multi-scale peak finding algorithm, and the absolute values ​​of the obtained peak and valley values ​​are arranged in ascending order of time to obtain the sequences, which are respectively denoted as the peak sequence and valley sequence.

[0024] Preferably, the method for determining the significance coefficient for each vehicle load application is as follows:

[0025] Calculate the mean value of the load influence difference coefficient of all outer ring supports under each vehicle load;

[0026] Calculate the average of the first and second anomaly coefficients for all supports under each vehicle load.

[0027] The product of the two means is used as the significance coefficient for each vehicle load application.

[0028] Preferably, the risk probability is predicted using a trained LSTM neural network model.

[0029] This application has at least the following beneficial effects:

[0030] This application proposes an AI-driven vibration monitoring device and a beam-fall risk early warning system for curved bridges. Through an intelligent sensing system, it deeply analyzes the asymmetric anomalies of vibration data at various support locations of the curved bridge under vehicle loads, considering the random fluctuations in amplitude due to transient impacts, as well as the anomalies of overall waveform changes and severe vibrations in the support reaction forces. Furthermore, it considers the differences in load influence at different support locations of the curved bridge and calculates the corresponding significance coefficients. Its advantage lies in fully considering the vehicle-bridge coupling response characteristics caused by the curved structure of the curved bridge using an intelligent sensing system, and using AI-driven technology combined with the obtained feature values ​​for beam-fall risk early warning, which helps to compensate for the insufficient accuracy of vibration monitoring and early warning for curved bridges. Attached Figure Description

[0031] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 A flowchart illustrating a beam drop risk early warning system provided in one embodiment of this application;

[0033] Figure 2 This is a plan view of the support location of a curved bridge section according to an embodiment of this application. Detailed Implementation

[0034] Example 1

[0035] In the early warning of beam collapse risks on curved bridges, the bearing reaction force directly reflects the stress state of the bridge under the dynamic influence of vehicle loads and other factors, while vibration data can reflect the dynamic response characteristics of the bridge. By monitoring these data in real time and using sensors installed at key parts of the bridge, real-time status information of the bridge can be captured.

[0036] Because the risk of beam collapse is higher at the curves of curved bridges, this application installs a vibration monitoring device on each support in the curve section of the curved bridge to collect real-time reaction force and vibration data of the supports, with the data acquisition frequency set at 500 Hz. The reaction force data refers to the reaction force generated by the supports on the bridge when subjected to dynamic loads.

[0037] One embodiment of this application provides an AI-driven curved bridge vibration monitoring device, which includes an intelligent sensing system and a wireless communication unit.

[0038] The intelligent sensing system consists of a pressure sensor, a low-frequency acceleration sensor, and a signal processing unit.

[0039] Among them, the pressure sensor is used to acquire the reaction force data generated by the support of the curved bridge on the bridge, and the reaction force data is the reaction force generated by the support on the bridge.

[0040] A low-frequency accelerometer is used to acquire simulated vibration signals of curved bridge bearings. Since curved bridges have large spans and generally low natural frequencies, this application employs an accelerometer with excellent low-frequency performance to receive simulated vibration signals from bridge bearings.

[0041] The signal processing unit, connected to the low-frequency acceleration sensor, is used to filter, amplify, and convert the vibration analog signal to digital, and output vibration data.

[0042] The wireless communication unit is connected to the intelligent sensing system and is used to send the reaction force data and vibration data to the beam drop risk early warning system in real time for data analysis, processing and early warning.

[0043] A power supply is provided to power the intelligent sensing system and the wireless communication unit.

[0044] Example 2

[0045] As attached Figure 1 As shown in the flowchart of the beam drop risk early warning system, another embodiment of this application also provides a beam drop risk early warning system. This system is built into the curved bridge vibration monitoring device and is used to receive and process reaction force data and vibration data sent by the wireless communication unit. The system implements the following steps:

[0046] S1. Using the curve fitting characteristics of the reaction force data, determine the overall waveform anomaly coefficient of the peak; and extract the local peak values ​​of the reaction force data within the peak width range corresponding to the peak, calculate the standard deviation and cumulative sum of the Euclidean distance between the values ​​and the corresponding time values ​​on the fitted curve, so as to determine the first anomaly coefficient of the reaction force data.

[0047] Curved bridges differ from straight bridges. Curved beam bridges are prone to bending-torsional coupling effects under external loads. This complex spatial stress state can cause the supports to detach or lift, easily inducing lateral buckling or overturning of the beam and increasing the risk of beam collapse. Furthermore, vehicle speed is a key factor affecting the bridge's vibration response, especially under the bending-torsional coupling characteristics of curved bridges, where the impact of vehicle speed may be more significant. Centrifugal force and lateral swaying force from vehicles make curved beam bridges more susceptible to lateral displacement. Under vibration loads, if the lateral displacement of the beam exceeds the bearing capacity of the supports, the beam may slip off the support pads. Therefore, this application analyzes the stress state and vibration characteristics of curved bridges under the influence of vehicle speed to provide a warning of beam collapse risk.

[0048] Speeding on bridges is becoming increasingly common, potentially causing localized damage or destruction to curved bridge structures. Taking a particular support as an example, when subjected to vehicle loads, the detected reaction force exhibits a pattern of gradually increasing, peaking, and then gradually decreasing, displaying a Gaussian distribution. Furthermore, the contact between the wheel and the bridge deck is not entirely rigid, exhibiting some elasticity and relative motion, which causes the support's reaction force to continuously change, resulting in severe vibrations during this process. When localized damage exists in the bridge structure, as vehicle speed increases, the overall waveform of the reaction force data becomes steeper, its symmetry decreases, and the irregularity of the severe vibrations intensifies, exhibiting more localized spikes and abrupt changes.

[0049] Therefore, to obtain the overall waveform variation characteristics, this application first uses a quadratic polynomial fitting technique to obtain the fitting curve of the reaction force data. Then, it calculates the maximum value among the maxima of the fitting curve and takes the obtained maximum value point as the peak point of the overall waveform. The peak width of each peak point of the overall waveform is then obtained. Taking a certain peak as an example, the kurtosis and skewness of the corresponding fitting curve within its peak width range are calculated. The calculation of kurtosis and skewness is a well-known technique, and the specific process will not be elaborated further. The larger the obtained kurtosis, the sharper the waveform of the corresponding peak in the reaction force data; the larger the absolute value of the obtained skewness, the more asymmetrical the waveform of the corresponding peak in the reaction force data.

[0050] Then, the product of the absolute values ​​of kurtosis and skewness corresponding to a certain peak is taken as the overall waveform anomaly coefficient of that peak position, denoted as A. The larger the A is, the more obvious the overall waveform steepness and asymmetry of the reaction force data are.

[0051] Taking this wave peak as an example, in order to obtain the severe fluctuation characteristics of the reaction force data within its peak width range, this application uses an automatic multi-scale peak finding algorithm to obtain all peak values ​​of the reaction force data within the corresponding peak width range, i.e., local peak values ​​in the reaction force data. The stronger the irregularity of the severe fluctuation of the reaction force data, the more uneven the distribution of the obtained peak values ​​near the fitting curve.

[0052] Therefore, the Euclidean distance between each peak point and the fitted curve at the same time is calculated. Then, the standard deviation and sum of all Euclidean distances are calculated. The standard deviation reflects the irregularity of local jitter in the reaction force data, and the sum reflects the severity of local jitter in the reaction force data. The product of the standard deviation and the sum is then taken as the significance value of the irregularity and severity of local jitter in the reaction force data, denoted as B. The larger the B, the stronger the irregularity of the severe jitter in the reaction force data, and the greater the amplitude of the jitter.

[0053] Therefore, based on the overall waveform characteristics and local vibration characteristics of the reaction force data, the first anomaly coefficient of the reaction force data under vehicle load is calculated, and its formula is as follows: The larger the value of C, the more obvious the abnormal characteristics of the overall waveform and local jitter of the corresponding reaction force data.

[0054] S2, obtain the vibration data segment corresponding to the vehicle load by analyzing whether the mean value of the vibration data within the preset sliding window exceeds the preset threshold; extract the coefficient of variation of the peak sequence and valley sequence in the vibration data segment after difference, as well as the correlation coefficient between the peak sequence and valley sequence, to determine the second anomaly coefficient of the vibration data.

[0055] Furthermore, if a section of the bridge structure is damaged, the amplitude of the bearing vibration under vehicle load initially increases and then decreases. However, the faster the vehicle travels, the more random the amplitude changes due to transient impacts become, and the more pronounced the asymmetric vibration characteristics become. Generally, the amplitude of the curved bridge bearings is extremely small when there is no vehicle load. To extract vibration data under vehicle load, this application obtains the vibration data segment corresponding to the vehicle load by calculating whether the mean value of the vibration data within a sliding window exceeds a preset threshold. In this embodiment, the sliding step size is set to 1, the window length to 21, and the preset threshold to 0.05 m / s². If the average value of the data within the window is greater than the set threshold, the data point at the center of the window will be taken as the valid vibration data, and the continuous valid vibration data will be taken as a vibration data segment.

[0056] Based on the extracted vibration data segments, this application utilizes an automatic multi-scale peak finding algorithm to extract the peak and valley values ​​of the vibration data. The absolute values ​​of the obtained peaks and valleys are arranged in ascending order by time, resulting in sequences denoted as the peak sequence and valley sequence, respectively. The Spearman correlation coefficient between the two sequences is calculated as a significance value of the vertical symmetry of the vibration data, denoted as D. The larger the value of D, the more pronounced the vertical symmetry of the vibration data segment.

[0057] To obtain the amplitude fluctuation characteristics caused by transient impacts, taking the peak values ​​of vibration data as an example, since the amplitude has a certain time-varying characteristic (increasing first and then decreasing), the peak sequence also has a corresponding trend. To reduce the influence of the peak values' own trend characteristics, we first obtain the first-order difference sequence of the peak values, and then calculate the coefficient of variation of this first-order difference sequence. The same steps are used for the valley data to obtain the corresponding coefficient of variation. The mean of the obtained coefficients of variation is taken as the randomness coefficient of amplitude fluctuations caused by transient impacts in the vibration data, denoted as E. The obtained E reflects the random characteristics of amplitude fluctuations caused by transient impacts in the vibration data.

[0058] Therefore, the formula for calculating the second anomaly coefficient of the vibration data of bridge bearings due to transient impact, including random amplitude fluctuations and asymmetric upper and lower characteristics, is as follows: The larger the value of G, the more significant the vibration data of the support is due to the random fluctuation of amplitude caused by transient impact and the asymmetric abnormal characteristics.

[0059] S3. Using the correlation coefficient between the first and second anomaly coefficients of the same support under N consecutive vehicle loads, calculate the mean difference between the correlation coefficients of the target outer ring support and its relative inner ring support and relative outer ring support, and obtain the load influence difference coefficient of each outer ring support under each vehicle load.

[0060] Under vehicle loads, the variation characteristics of the bearing reaction force and the vibration response of curved bridges are highly consistent over time. However, the worse the bridge structure condition, the stronger the positive correlation between the first anomaly coefficient of the reaction force data under vehicle loads and the second anomaly coefficient of the vibration data due to the random fluctuation of amplitude caused by transient impacts and the asymmetric characteristics of the upper and lower parts of the structure.

[0061] Therefore, taking a specific bearing as an example, this application first obtains the first and second anomaly coefficients calculated after the i-th and the preceding N-1 vehicle load applications, where N is an integer within the range [20, 30]. Then, the Spearman correlation coefficient between the obtained N first anomaly coefficients and N second anomaly coefficients is calculated. This Spearman correlation coefficient is used as the positive correlation value between the reaction force data and vibration data corresponding to the i-th vehicle load application of the bearing, denoted as... The result The larger the value, the stronger the positive correlation between the changes in the relevant abnormal characteristics of the reaction force data and the changes in the relevant abnormal characteristics of the vibration data.

[0062] Furthermore, when a vehicle travels to the middle of a curved bridge, the load on the bridge supports is greater than when entering or exiting the curve. In addition, the load on the outer ring of the bridge is greater than that on the inner ring, which can easily lead to vehicle-bridge coupling response and increase the risk of beam collapse.

[0063] In this embodiment, a plan view of the support locations in the curve section of the curved bridge is shown below. Figure 2 As shown.

[0064] exist Figure 2 In the diagram, each mark represents a support. Support 2-1 is located on the outer ring and in the middle of the curved bridge, and therefore experiences a greater vehicle load compared to supports 1-1 and 2-2. Furthermore, the greater the degree of damage to the bridge structure, the more significant the comprehensive abnormal characteristics of the supports and the differences in load influence between supports become.

[0065] In view of this, based on the above characteristics, the mean value of the difference between the positive correlation values ​​of each outer ring support and its position relative to the inner ring support and its position relative to the outer ring support during the i-th vehicle load is calculated, and is used as the load influence difference coefficient of the outer ring support during the i-th vehicle load.

[0066] In this application, the relative inner ring support and relative outer ring support positions refer only to the front, back, left, and right adjacent positions of the supports. Specifically, taking support 2-1 as an example, support 1-1 and this position belong to the outer ring, but support 1-1 is closer to the outer side of the curved section of the curved bridge than support 2-1, so support 1-1 is regarded as the relative outer ring support of support 2-1; support 2-2 and support 2-1 belong to the same support structure of the bridge and are both located in the middle position, but support 2-2 is closer to the inner ring of the bridge, so support 2-2 is regarded as the relative inner ring support of support 2-1.

[0067] Thus, the load influence difference coefficient of each outer ring support under the i-th vehicle load was obtained.

[0068] S4, combining the first and second anomaly coefficients and the load influence difference coefficient, determines the significance coefficient for each vehicle load application.

[0069] Furthermore, a significance coefficient is constructed based on the overall abnormal state of the vibration and reaction force of the corresponding curved bridge supports under the i-th vehicle load, as well as the differences in load influence between supports. The formula is as follows: .in, Let be the significance coefficient under the i-th vehicle load. Let be the mean of the load influence difference coefficients of all outer ring supports under the i-th vehicle load. This reflects the differences in load effects caused by the bending structure of curved bridges. Let be the average of the first and second anomaly coefficients for all supports under the i-th vehicle load. This reflects the overall degree of anomaly in the vibration response and reaction force changes of the curved bridge supports. The results... The larger the value, the more significant the structural risk of the curved bridge under the i-th vehicle load.

[0070] S5 uses the significance coefficients of all previous historical vehicle loads to predict the risk probability of the current vehicle load, in order to determine whether to issue an early warning.

[0071] This application analyzes in depth the random fluctuations and asymmetric anomalies of vibration data at various support locations of curved bridges under the influence of vehicle loads due to transient impacts, as well as the overall waveform changes and severe shaking anomalies of support reaction forces. Furthermore, considering the differences in load influence at different support locations of curved bridges, the application calculates the significance coefficients of the overall abnormal state of vibration and reaction forces and the differences in load influence between supports. These significance coefficients reflect the degree of structural risk associated with the existence of curved bridges.

[0072] This application utilizes an LSTM neural network model to predict the risk of beam falling. The optimizer for this network model is set to Adam, and the loss function is the cross-entropy loss function.

[0073] Specifically, based on previously collected vibration and reaction force data, the aforementioned feature extraction and model training are performed. During the operation of the curved bridge, sensors installed at various support locations collect vibration and reaction force data in real time, and extract the salient coefficients of these data. These salient coefficients are then input into the trained LSTM model, which outputs the risk probability of the current vehicle load. A preset warning threshold of 0.6 is used to evaluate the prediction results. If the predicted risk probability exceeds the warning threshold, a warning signal is triggered, promptly alerting relevant personnel to take appropriate measures, such as restricting vehicle traffic, conducting bridge inspections and maintenance, to prevent beam collapse accidents.

[0074] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not invented in this application.

[0075] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A beam-falling risk early warning system, characterized in that, The beam drop risk early warning system is built into the curved bridge vibration monitoring device. It is used to receive and process the reaction force data and vibration data sent by the wireless communication unit in the curved bridge vibration monitoring device. The device includes an intelligent sensing system and a wireless communication unit. The intelligent sensing system consists of a pressure sensor, a low-frequency acceleration sensor, and a signal processing unit. The pressure sensor is used to acquire data on the reaction force generated by the curved bridge bearing on the bridge, and the reaction force data is the reaction force generated by the bearing on the bridge. The low-frequency acceleration sensor is used to acquire the vibration simulation signal of the curved bridge bearing. The signal processing unit, connected to the low-frequency acceleration sensor, is used to filter, amplify, and convert the vibration analog signal to digital, and output vibration data. The wireless communication unit is connected to the intelligent sensing system and is used to send the reaction force data and vibration data to the beam drop risk early warning system in real time for data analysis, processing and early warning. A power supply is provided to power the intelligent sensing system and the wireless communication unit. The beam drop risk early warning system implements the following steps: For each support, the reaction force data and vibration data under each vehicle load are as follows: Curve fitting is performed using reaction force data to obtain the peak characteristics on the curve. The maximum value among all maxima on the fitted curve is taken as the peak corresponding to the peak point. The overall waveform anomaly coefficient of the peak is determined by the product of the absolute values ​​of kurtosis and skewness within the peak width range of the fitted curve. Local peak values ​​of the reaction force data within the corresponding peak width range of the peak are extracted, and the standard deviation and sum of the Euclidean distance between them and the corresponding time values ​​on the fitted curve are calculated. The product of the overall waveform anomaly coefficient, the standard deviation, and the sum is used as the first anomaly coefficient of the reaction force data. The vibration data segment corresponding to the vehicle load is obtained by analyzing whether the mean value of the vibration data within the preset sliding window exceeds a preset threshold; the peak sequence and valley sequence in the vibration data segment are obtained, the coefficient of variation of the sequence after the peak sequence is differentially divided, the coefficient of variation of the sequence after the valley sequence is differentially divided, and the Spearman correlation coefficient between the peak sequence and the valley sequence are obtained, and the product of the Spearman correlation coefficient and the mean of the coefficient of variation is used as the second anomaly coefficient of the vibration data. By using the correlation coefficient between the first and second anomaly coefficients of the same support under N consecutive vehicle loads, the mean difference between the correlation coefficients of the target outer ring support and its relative inner ring support and relative outer ring support is calculated, and the load influence difference coefficient of each outer ring support under each vehicle load is obtained. By combining the first and second anomaly coefficients and the load effect difference coefficient, the significance coefficient for each vehicle load application is determined; The risk probability of the current vehicle load is predicted by using the significance coefficients of all historical vehicle loads prior to the current vehicle load, in order to determine whether to issue an early warning.

2. The beam drop risk early warning system as described in claim 1, characterized in that, The method for obtaining the vibration data segment is as follows: if the average value of the data within the window is greater than a preset threshold, then the data point at the center of the window is taken as valid vibration data, and continuous valid vibration data is taken as a vibration data segment.

3. The beam drop risk early warning system as described in claim 1, characterized in that, The method for determining the second anomaly coefficient is as follows: calculate the mean coefficient of variation of the sequence after differentiating the peak sequence and the trough sequence; and take the product of the mean coefficient of variation and the correlation coefficient as the second anomaly coefficient.

4. The beam drop risk early warning system as described in claim 3, characterized in that, The method for obtaining the peak sequence and valley sequence is as follows: the peak and valley values ​​of the vibration data are extracted by an automatic multi-scale peak search algorithm, and the absolute values ​​of the obtained peak and valley values ​​are arranged in ascending order of time. The resulting sequences are denoted as the peak sequence and valley sequence, respectively.

5. The beam drop risk early warning system as described in claim 1, characterized in that, The method for determining the significance coefficient for each vehicle load application is as follows: Calculate the mean value of the load influence difference coefficient of all outer ring supports under each vehicle load; Calculate the average of the first and second anomaly coefficients for all supports under each vehicle load. The product of the two means is used as the significance coefficient for each vehicle load application.

6. The beam drop risk early warning system as described in claim 1, characterized in that, The risk probability is predicted using a trained LSTM neural network model.

Images