A detection data acquisition system and method for bridge support design

Abstract

The present application relates to the technical field of bridge detection, and discloses a detection data collection system and method for bridge support design, which comprises a sliding trend prediction model with a disturbance weight gate structure, historical sliding trajectory data and pressure gradient data are fused, the time sequence modeling capability of the nonlinear response of the sliding support under the vibration state is effectively enhanced, the perception accuracy of the model to the short-time disturbance and the inertial inversion behavior is improved, then, a disturbance change rate sequence is constructed based on the predicted sliding trend sequence, a disturbance score factor is calculated by using the trend fluctuation amplitude and the number of direction reversal, and the disturbance time window is dynamically marked accordingly, thereby solving the problem that there is no dynamic division means for the disturbance behavior of the support in the prior art, finally, different filtering processing modes are adopted for the disturbance time window and the non-disturbance time window respectively, and the real-time strain data is processed differently, thereby improving the availability of the strain data under the low-frequency disturbance background.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bridge detection, more particularly, the present application relates to a detection data acquisition system and method for bridge support design. BACKGROUND

[0002] In large bridge engineering, the bridge structure is affected by wind load, traffic load, temperature change and structure vibration, etc. during construction and operation, and produces dynamic response, which further affects its long-term safety performance and service state. Therefore, monitoring and evaluation of the vibration state of the bridge structure become the key means to ensure the safe operation of the bridge. The existing bridge vibration monitoring system usually relies on distributed strain gauges, accelerometers or fiber optic sensors, etc. to obtain real-time stress, strain, displacement and other multi-source data of the structure under load, and combines signal processing and state recognition algorithms to analyze the response characteristics and evolution law of the structure, so as to realize comprehensive evaluation of the bridge operation state, damage trend and bearing capacity.

[0003] However, during the construction of high pier bridges or the early service stage, the sliding support is often in a temporary positioning state, and is easily affected by concrete shrinkage, pier top vibration or intermittent construction load, which may cause slight deviation or nonlinear sliding behavior, resulting in rapid changes in the stress state of the support connection part. In addition, due to the interference of intermittent load (such as construction equipment start-stop and short-time traffic flow), a large amount of low-frequency disturbance is often mixed in the structure response signal, which may cause non-stationary noise to the strain data. The existing technology relies on static strain filtering or structure modal monitoring method, lacks predictive modeling and disturbance dynamic partitioning strategy for sliding support behavior, and thus the stress change data of the bearing part is difficult to be stably captured under the influence of unstable temporary positioning of the sliding support and low-frequency disturbance, and the abnormal signal is difficult to be identified and suppressed in time, which affects the stability and precision of the structure health monitoring system.

[0004] In view of the above problems, the present application provides a detection data acquisition system and method for bridge support design. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, the present application provides a detection data acquisition system and method for bridge support design.

[0006] To achieve the above object, the present application provides the following technical scheme:

[0007] In the first aspect, a detection data acquisition method for bridge support design is provided, comprising:

[0008] The historical sliding trajectory data and the historical pressure gradient data of the sliding support under vibration state are obtained, and a sliding trend prediction model with disturbance weight gating structure is constructed based on the historical sliding trajectory data, the contact pressure gradient data and a preset LSTM model.

[0009] inputting the real-time slip trajectory data and the real-time pressure gradient data at the current moment into the slip trend prediction model to obtain a slip trend sequence;

[0010] calculating a disturbance change rate sequence corresponding to the slip trend sequence, marking a prediction time axis of the slip trend sequence based on the disturbance change rate sequence to obtain a target time window, and the target time window including a disturbance time window and a non-disturbance time window;

[0011] obtaining real-time strain data, performing first filtering processing on the real-time strain data in the disturbance time window, and performing second filtering processing on the real-time strain data in the non-disturbance time window, wherein a collection time axis of the real-time strain data corresponds to the prediction time axis of the slip trend sequence.

[0012] In some embodiments, the method for obtaining the historical slip trajectory data of the slip support in the vibration state includes:

[0013] a first inertial measurement unit and a second inertial measurement unit are respectively arranged at the top of the pier and the bottom of the slip support to collect acceleration data in the vibration state;

[0014] performing time alignment processing on the acceleration data, and performing filtering and denoising processing on the time-aligned acceleration data, and the filtering and denoising processing includes removing high-frequency interference components and low-frequency drift components;

[0015] integrating the filtered acceleration data to obtain a first displacement response sequence of the top of the pier and a second displacement response sequence of the bottom of the slip support;

[0016] based on the first displacement response sequence and the second displacement response sequence, calculating a response difference value sequence, the response difference value sequence representing the slip trajectory change of the slip support in the vibration state, and taking the response difference value sequence as the historical slip trajectory data.

[0017] In some embodiments, the method for constructing a slip trend prediction model with a disturbance weight gate structure based on the historical slip trajectory data, the contact pressure gradient data and a preset LSTM model includes:

[0018] based on the historical slip trajectory data and the historical pressure gradient data, respectively constructing a first input sequence and a second input sequence;

[0019] inputting the first input sequence into a main sequence encoding channel of the LSTM model to obtain a displacement change channel;

[0020] The second input sequence is input into the preset disturbance identification network, a disturbance intensity sequence is extracted, a disturbance volatility is calculated based on the disturbance intensity sequence, the disturbance volatility is input into a gate residual channel of an LSTM model, a disturbance adjustment channel is obtained, and the disturbance volatility is a first-order difference variance of the disturbance intensity sequence;

[0021] Based on an asymmetric residual coupling mode, the displacement change channel and the disturbance adjustment channel are fused to generate a fusion state vector, the fusion state vector is input into a memory unit of the LSTM model for state updating, and a slip trend prediction model is obtained.

[0022] In some embodiments, based on historical slip trajectory data and historical pressure gradient data, the method for constructing the first input sequence and the second input sequence respectively includes:

[0023] The historical slip trajectory data is subjected to vibration cycle identification processing to obtain inflection point features of the historical slip trajectory data at cycle boundaries, and based on the inflection point features, a main displacement segment sequence is extracted from the historical slip trajectory data, and the main displacement segment sequence is encoded into the first input sequence according to a continuous time index;

[0024] The historical pressure gradient data is subjected to synchronous segment extraction processing to obtain a pressure disturbance response segment corresponding to a time axis of the main displacement segment sequence, and a maximum gradient amplitude and a gradient consistency index of the pressure disturbance response segment in each slip cycle are calculated;

[0025] The maximum gradient amplitude and the gradient consistency index are encoded into a disturbance intensity structure sequence on the time axis, and the disturbance intensity structure sequence is subjected to time sequence embedding processing to obtain the second input sequence.

[0026] In some embodiments, the inflection point features include acceleration extreme points and local curvature extreme points, and the method for extracting the main displacement segment sequence from the historical slip trajectory data based on the inflection point features includes:

[0027] The acceleration extreme points are taken as first inflection point indexes, and the local curvature extreme points are taken as second inflection point indexes;

[0028] The first inflection point indexes and the second inflection point indexes are subjected to set fusion to form a joint inflection point sequence;

[0029] A time interval between every two adjacent indexes in the joint inflection point sequence is determined as a candidate segment;

[0030] For each candidate segment, a displacement amplitude of the candidate segment is calculated;

[0031] If the displacement amplitude is less than a preset amplitude threshold, the current candidate segment is deleted; otherwise, a number of speed reversals in the candidate segment is calculated;

[0032] If the number of speed reversals is less than the preset threshold, the current candidate segment is deleted; otherwise, the candidate segment is included in the main displacement segment sequence.

[0033] In some embodiments, the method for encoding the maximum gradient amplitude and the gradient consistency index into a sequence of perturbation intensity structures on the time axis, and performing temporal embedding processing on the sequence of perturbation intensity structures to obtain the second input sequence comprises:

[0034] The maximum gradient amplitude is taken as the pressure intensity dimension of the perturbation intensity structure, the gradient direction consistency index is taken as the direction consistency dimension of the perturbation intensity structure, and the time index corresponding to the pressure perturbation response segment is combined as the time dimension of the perturbation intensity structure, to construct the perturbation intensity structure in a ternary structure form;

[0035] All the perturbation intensity structures are arranged in an increasing order of time index to form a sequence of perturbation intensity structures;

[0036] Each perturbation intensity structure in the sequence of perturbation intensity structures is normalized, and a temporal position vector is added to each perturbation intensity structure based on a preset position encoding scheme;

[0037] The sequence of perturbation intensity structures with time encoding is input into a preset temporal embedding network, a nonlinear activation operation is performed, and a corresponding second input sequence is obtained.

[0038] In some embodiments, the method for calculating a corresponding perturbation change rate sequence from the slip trend sequence comprises:

[0039] The slip trend sequence is divided into a plurality of slip trend subsequences according to a preset rule;

[0040] In each slip trend subsequence, the slip trend value corresponding to the first time point of the subsequence is selected as a trend start value, the slip trend value corresponding to the last time point of the subsequence is selected as a trend end value, the median time index of the subsequence is calculated based on the time index sequence of the subsequence, and the slip trend value corresponding to the median time index is extracted as a trend median value;

[0041] The first change difference value between the trend start value and the trend median value, and the second change difference value between the trend median value and the trend end value are calculated, and the two-segment change index is determined based on the first change difference value and the second change difference value.

[0042] In some embodiments, the method for determining the two-segment change index based on the first change difference value and the second change difference value comprises:

[0043] The first change amplitude value and the second change amplitude value are weighted and summed as the trend fluctuation amplitude of the current slip trend subsequence;

[0044] Calculate the amplitude product of the first change amplitude value and the second change amplitude value, and take the sign information of the amplitude product as a trend change indicator;

[0045] Combine the trend fluctuation amplitude and the trend change indicator to form a double-segment change indicator corresponding to the current slip trend subsequence.

[0046] In some embodiments, the method for marking the prediction time axis of the slip trend sequence based on the disturbance change rate sequence includes:

[0047] Performing sliding time window division on the disturbance change rate sequence according to a fixed length to obtain a plurality of disturbance change rate subsequences;

[0048] Normalizing all trend fluctuation amplitude values in the current disturbance change rate subsequence based on a range normalization rule to obtain a normalized fluctuation amplitude sequence;

[0049] Based on the trend change indicators in the current disturbance change rate subsequence, the number of sign switching is counted as the number of direction reversals, and the direction reversal refers to the difference between adjacent two trend change indicators;

[0050] Based on the normalized fluctuation amplitude sequence, the corresponding fluctuation average amplitude is calculated;

[0051] The fluctuation average amplitude and the number of direction reversals are multiplied to obtain a disturbance score factor corresponding to the current disturbance change rate subsequence;

[0052] Determine whether the disturbance score factor is greater than a preset score threshold, if yes, mark the sliding time window as a disturbance time window, if not, mark the sliding time window as a non-disturbance time window.

[0053] In a second aspect, a bridge support design detection data acquisition system is provided, which is used to implement the bridge support design detection data acquisition method described above, and includes:

[0054] The construction module is used to obtain historical slip trajectory data and historical pressure gradient data of the slip support under vibration state, and construct a slip trend prediction model with disturbance weight gate structure based on the historical slip trajectory data, the contact pressure gradient data and a preset LSTM model;

[0055] The prediction module is used to input real-time slip trajectory data and real-time pressure gradient data at the current time into the slip trend prediction model to obtain a slip trend sequence;

[0056] The segmentation module is used to calculate the corresponding perturbation rate of change sequence based on the slip trend sequence, mark the prediction time axis of the slip trend sequence based on the perturbation rate of change sequence, and obtain the target time window, which includes the perturbation time window and the non-perturbation time window.

[0057] Filtering module: Used to acquire real-time strain data, perform a first filtering process on the real-time strain data within the disturbance time window, and perform a second filtering process on the real-time strain data within the non-disturbance time window. The acquisition time axis of the real-time strain data corresponds to the prediction time axis of the slip trend sequence.

[0058] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0059] This invention constructs a slip trend prediction model with a perturbation weighted gating structure, integrating historical slip trajectory data and pressure gradient data. This effectively enhances the time-series modeling capability for the nonlinear response of slip supports under vibration conditions and improves the model's perception accuracy for short-term perturbations and inertial reversal behavior. Subsequently, a perturbation change rate sequence is constructed based on the predicted slip trend sequence. The perturbation scoring factor is calculated using the trend fluctuation amplitude and the number of direction reversals, and the perturbation time window is dynamically marked accordingly. This solves the problem of the lack of dynamic segmentation methods for support perturbation behavior in existing technologies. Finally, different filtering methods are used for perturbation time windows and non-perturbation time windows to differentiate real-time strain data. This not only improves the usability of strain data under low-frequency perturbation backgrounds but also enhances the accuracy and stability of the structural health monitoring system in capturing abnormal strain responses. Attached Figure Description

[0060] Figure 1 This is a flowchart illustrating a method for collecting test data for bridge bearing design according to the present invention.

[0061] Figure 2 This is a schematic diagram comparing the predicted slip trend sequence with the actual trajectory in this invention;

[0062] Figure 3 This is a heatmap showing the time-series distribution of the perturbation scoring factors in this invention;

[0063] Figure 4 This is a schematic diagram of the structure of a bridge bearing design detection data acquisition system according to the present invention. Detailed Implementation

[0064] For the purposes of making the objects, technical solutions, and advantages of the present application clearer, the present application is further described in detail below with reference to specific embodiments and with reference to the drawings, in the following detailed description, numerous specific details are set forth in order to provide a thorough understanding of the described exemplary embodiments, however, it will be apparent to those skilled in the art that the described embodiments can be practiced without some or all of these specific details, in other exemplary embodiments, well-known structures are not described in detail to avoid unnecessarily obscuring the concepts of the present disclosure, it should be understood that the specific embodiments described herein are merely used to explain the present application and are not used to limit the present application, and the various aspects described in the embodiments can be combined arbitrarily without conflict.

[0065] Embodiment 1

[0066] Please refer to Figure 1 The present embodiment discloses a bridge support design detection data acquisition method, which comprises the following steps:

[0067] S10: Obtain historical sliding trajectory data and historical pressure gradient data of the sliding support in the vibration state, and construct a sliding trend prediction model with a disturbance weight gate structure based on the historical sliding trajectory data, the contact pressure gradient data, and a preset LSTM model;

[0068] In the present embodiment, the sliding support in the vibration state refers to a support device that has a slight sliding, local relative displacement, or intermittent contact force change between the support body and the upper / lower connecting structure due to reasons such as foundation disturbance, bridge pier vibration, intermittent load action, or wind-induced vibration during bridge construction or operation, in this state, the sliding support shows an enhanced short-time displacement response, intensified pressure gradient fluctuation, and reduced structural connection stiffness, which is easy to cause disturbance stress fluctuation of the bearing structure, therefore, real-time modeling and dynamic prediction of the sliding behavior are needed to assist in identifying structural strain abnormalities caused by the sliding support.

[0069] It can be understood that the historical sliding trajectory data refers to a relative displacement sequence of the sliding support body or its contact boundary position changing with time in the vibration state, which is usually recorded continuously by a displacement sensor (such as a laser displacement meter, an LVDT, or a capacitive displacement meter) disposed near the sliding surface of the support at a fixed sampling period, forming a time-displacement pair, the historical pressure gradient data refers to a data sequence of the spatial distribution difference of the unit area normal force between the support and the upper or lower structure of the bridge changing with time during the operation of the support, which is usually obtained by a multi-point distributed pressure sensor (such as a piezoelectric film or a fiber optic pressure array) disposed on the contact surface of the support, with a unit of kPa or MPa.

[0070] It should be noted that the historical sliding trajectory data and the historical pressure gradient data can be obtained in the above manner, or can be obtained in the following manner. The method for obtaining the historical sliding trajectory data of the sliding support in the vibration state comprises:

[0071] A first inertial measurement unit and a second inertial measurement unit are arranged at the top of the pier and the bottom of the sliding support respectively, for collecting acceleration data in the vibration state;

[0072] The acceleration data is subjected to time alignment processing, and the acceleration data after time alignment is subjected to filtering and denoising processing, which includes removing high-frequency interference components and low-frequency drift components;

[0073] The filtered acceleration data is subjected to integral operation respectively, to obtain a first displacement response sequence of the top of the pier and a second displacement response sequence of the bottom of the sliding support;

[0074] Based on the first displacement response sequence and the second displacement response sequence, a response difference sequence is calculated, which represents the change of the sliding trajectory of the sliding support in the vibration state, and the response difference sequence is taken as the historical sliding trajectory data.

[0075] In the embodiment, the first inertial measurement unit collects acceleration data of the top of the pier along the sliding direction, and the second inertial measurement unit collects acceleration data of the bottom of the sliding support along the sliding direction. The above-mentioned removing high-frequency interference components and low-frequency drift components means using a low-pass filtering algorithm to remove high-frequency components in the acceleration data caused by electromagnetic interference and environmental mechanical impact, and based on sliding average, zero initial velocity constraint or high-pass filtering algorithm to suppress low-frequency drift error in the integral process, to ensure that the obtained displacement response sequence can accurately reflect the true sliding behavior in the vibration state. The low-pass filtering algorithm can be a Butterworth low-pass filtering algorithm, and the high-pass filtering algorithm can be a Butterworth high-pass filtering algorithm.

[0076] In the embodiment, the integral operation on the filtered acceleration data specifically comprises:

[0077] First integral processing:

[0078] The filtered acceleration data of the top of the pier and the bottom of the sliding support is subjected to a number of numerical integrations respectively, to obtain corresponding velocity response sequences;

[0079] The integral operation adopts a complex trapezoidal method or a Runge-Kutta method, and a zero initial velocity boundary condition is introduced to avoid integral starting point deviation;

[0080] Second integral processing:

[0081] Based on the obtained velocity response sequence, a second integral is performed to obtain the first displacement response sequence at the top of the pier and the second displacement response sequence at the bottom of the sliding support.

[0082] Methods for calculating the response difference sequence based on the first displacement response sequence and the second displacement response sequence include:

[0083] The first displacement response sequence at the top of the pier and the second displacement response sequence at the bottom of the sliding support are calculated point-to-point on the same time axis to obtain the response difference sequence.

[0084] This embodiment collects acceleration data by setting inertial measurement units at the top of the pier and the bottom of the sliding bearing, and performs time alignment, filtering and noise reduction, and integration processing to obtain a displacement response sequence. Then, the response difference sequence constructed based on the displacement response difference characterizes the change of the sliding trajectory of the sliding bearing under vibration, forming historical sliding trajectory data that can be obtained without contacting the sliding surface. This method is suitable for complex structural environments where it is difficult to deploy sensors on the bearing surface or where temporary positioning errors exist, thus improving the feasibility and robustness of data acquisition.

[0085] The logic for calculating the displacement response difference is as follows:

[0086] ;

[0087] In the formula, Indicates the first The displacement response difference under each time index reflects the relative displacement change between the top of the pier and the bottom of the sliding bearing within the same time period. This represents the filtered acceleration signal at the top of the bridge pier at time τ. This represents the filtered acceleration signal at the bottom of the sliding support at time τ; double integral. and The acceleration signal is integrated twice to obtain the displacement response sequence. The upper and lower limits of integration are both from 0 to t, indicating that the integration is carried out within a unified time interval.

[0088] Furthermore, this invention obtains the historical strain response sequence of the strain sensor installed on the support base, and combines multi-scale autoregressive analysis and strain-stress mapping method to convert the residual components into historical pressure gradient data, thereby constructing a reconstruction path that reflects the trend of normal stress change in the contact area. This technical mechanism eliminates the dependence on traditional pressure sensors and is suitable for gradient reconstruction needs under non-contact, non-constant support conditions.

[0089] Methods for constructing slip trend prediction models with perturbation weight gating structures based on historical slip trajectory data, contact pressure gradient data, and pre-defined LSTM models include:

[0090] based on the historical slip trajectory data and the historical pressure gradient data, a first input sequence and a second input sequence are respectively constructed;

[0091] The first input sequence is input into a main sequence encoding channel of the LSTM model to obtain a displacement change channel;

[0092] The second input sequence is input into a preset disturbance identification network to extract a disturbance intensity sequence, a disturbance volatility rate is calculated based on the disturbance intensity sequence, and the disturbance volatility rate is input into a gate residual channel of the LSTM model to obtain a disturbance adjustment channel. The disturbance volatility rate is a first-order difference variance of the disturbance intensity sequence.

[0093] Based on an asymmetric residual coupling mode, the displacement change channel and the disturbance adjustment channel are fused to generate a fusion state vector, the fusion state vector is input into a memory unit of the LSTM model for state updating to obtain a slip trend prediction model.

[0094] It should be noted that the main sequence encoding channel refers to a main path channel for receiving the first input sequence constructed based on the historical slip trajectory data and performing sequence modeling through the LSTM model. The channel relies on a long short-term memory structure developed in time synchronization to extract the time sequence evolution characteristics of the slip trajectory and output a time sequence vector sequence representing the slip displacement change trend, providing a basic channel output for subsequent state fusion. The disturbance identification network refers to a subnetwork structure for receiving the second input sequence constructed based on the historical pressure gradient data and performing disturbance amplitude identification and disturbance time sequence feature extraction. The network performs local disturbance mutation mining and global disturbance intensity trend coding through multiple layers of convolution, residual enhancement, or gate attention mechanism, and outputs the disturbance intensity sequence as the adjustment input of the gate residual channel, aiming to extract the normal stress fluctuation characteristics of the contact area to assist in correcting the state updating path of the slip trend prediction model.

[0095] The gate residual channel is a dynamic weight adjustment structure arranged in parallel with the LSTM main sequence encoding channel. The channel receives the disturbance intensity sequence or its derivative quantity (such as the disturbance volatility rate) output by the disturbance identification network, and maps the disturbance signal to a gate value residual, thereby controlling the state updating amplitude and forgetting ratio of the memory unit, improving the response ability of the model to sudden disturbances. The asymmetric residual coupling mode refers to a set of non-equal weight fusion strategies or gate multiplicative combination modes when fusing the outputs of the main sequence encoding channel and the gate residual channel, to emphasize the dominant role of the slip trend channel in the overall state evolution, while introducing the disturbance modulation signal of the disturbance adjustment channel as a secondary correction vector for dynamically adjusting the activation strength and prediction bias of the memory state. This mode avoids the problem of disturbance signal dominating the instability of the main sequence, and improves the prediction stability and anti-interference robustness of the model in the non-contact support case.

[0096] The method for constructing the first input sequence and the second input sequence based on historical slip trajectory data and historical pressure gradient data comprises:

[0097] The historical slip trajectory data is subjected to vibration cycle identification processing to obtain inflection point features of the historical slip trajectory data at cycle boundaries, and a main displacement segment sequence is extracted from the historical slip trajectory data based on the inflection point features, and the main displacement segment sequence is encoded into the first input sequence according to a continuous time index;

[0098] Synchronous segment extraction processing is performed on the historical pressure gradient data to obtain pressure disturbance response segments corresponding to a time axis of the main displacement segment sequence, and a maximum gradient amplitude and a gradient consistency index of the pressure disturbance response segments in each slip cycle are calculated;

[0099] The maximum gradient amplitude and the gradient consistency index are encoded into a disturbance intensity structure sequence on the time axis, and the disturbance intensity structure sequence is subjected to time sequence embedding processing to obtain the second input sequence.

[0100] It should be noted that the inflection point features include but are not limited to acceleration extreme points and local curvature extreme points, the acceleration extreme points refer to time points at which a second derivative (i.e., slip acceleration) of a slip trajectory reaches a local maximum or minimum, representing a starting position of inertial reversal of a support structure under inertial disturbance, and the local curvature extreme points refer to positions at which a curvature (or a first derivative change rate) of the slip trajectory reaches an extreme value within a unit time, representing a high disturbance point of a trajectory mutation, the historical slip trajectory data is subjected to vibration cycle identification processing, a second difference of the slip trajectory data is obtained to identify acceleration extreme points (i.e., local maximum or minimum acceleration points) as inertial reversal starts, and at the same time, curvature calculation is performed on the original trajectory data based on a sliding window to identify positions at which a local curvature change rate reaches an extreme value as significant points of trajectory disturbance, the extraction methods of the acceleration extreme points and the curvature extreme points are common signal feature recognition technologies in the field and belong to the prior art, and thus will not be described herein.

[0101] Taking the inflection point features including the acceleration extreme points and the local curvature extreme points as examples, the method for extracting the main displacement segment sequence from the historical slip trajectory data based on the inflection point features comprises:

[0102] The acceleration extreme points are taken as first inflection point indexes, and the local curvature extreme points are taken as second inflection point indexes;

[0103] The first inflection point indexes and the second inflection point indexes are subjected to set fusion to form a joint inflection point sequence;

[0104] A time interval between every two adjacent indexes in the joint inflection point sequence is determined as a candidate segment;

[0105] For each candidate segment, calculate the displacement amplitude of the candidate segment;

[0106] If the displacement amplitude is less than a preset amplitude threshold, delete the current candidate segment; otherwise, calculate the number of speed reversals in the candidate segment;

[0107] If the number of speed reversals is less than a preset reversal number threshold, delete the current candidate segment; otherwise, include the candidate segment in the main displacement segment sequence.

[0108] The method for set fusion of the first inflection point index and the second inflection point index to form a joint inflection point sequence includes: performing a time axis merging and deduplication processing on the two index sets, reordering according to the time stamp order corresponding to each inflection point, and ensuring that the fused joint inflection point sequence is strictly arranged in time increasing order, thereby defining the candidate segment interval formed between adjacent inflection points.

[0109] The method for calculating the displacement amplitude of the candidate segment includes: extracting the displacement values corresponding to the start time and end time of the candidate segment, calculating the absolute difference value as the displacement amplitude of the segment, and the method for calculating the number of speed reversals in the candidate segment includes: performing first-order difference on the displacement data sequence in the segment to obtain a speed sequence, and counting the number of speed sign changes, thereby obtaining the number of speed reversals, which can reflect the oscillation behavior in the slip process.

[0110] It should be noted that the gradient consistency index mentioned above can be calculated by calculating the proportion of the symbol with the largest proportion in the direction encoding sequence as the gradient direction consistency index, the maximum gradient amplitude and the gradient consistency index are encoded into a perturbation intensity structure sequence on the time axis, and the method for obtaining the second input sequence includes:

[0111] The maximum gradient amplitude is taken as the pressure intensity dimension of the perturbation intensity structure, the gradient direction consistency index is taken as the direction consistency dimension of the perturbation intensity structure, and the time index corresponding to the pressure perturbation response segment is combined as the time dimension of the perturbation intensity structure, and the perturbation intensity structure in the form of a three-element structure is constructed;

[0112] Arrange all the perturbation intensity structures in ascending order of time index to form a perturbation intensity structure sequence;

[0113] Normalize each perturbation intensity structure in the perturbation intensity structure sequence, and add a time sequence position vector to each perturbation intensity structure based on a preset position encoding scheme;

[0114] Input the perturbation intensity structure sequence with time encoding into a preset time sequence embedding network, perform a nonlinear activation operation, and obtain the corresponding second input sequence.

[0115] It should be noted that the position coding scheme can be an existing sine-cosine position coding or anchor point alignment coding. Taking the anchor point alignment coding as an example, the time index of each perturbation intensity structure is normalized with respect to the position of the anchor point in the period, and the normalized time position is assigned a period alignment feature coding vector. The time sequence embedding network can be an existing one-dimensional convolution embedding network. Taking the one-dimensional convolution embedding network as an example, a plurality of one-dimensional convolution kernels with different receptive fields are used to extract features of the perturbation intensity structure sequence. The local correlation pattern of the perturbation intensity and the direction consistency feature in the time domain is captured through convolution operation, and the convolution result is output as the embedding representation of each time in the second input sequence after nonlinear activation and normalization processing.

[0116] A specific example is as follows:

[0117] Suppose that a pressure perturbation response segment in a vibration period consists of 5 time points, and the corresponding maximum gradient amplitudes are 4.2, 3.8, 5.1, 4.5, and 4.7 kPa, respectively. The direction coding sequence in this period is +1, +1, +1, -1, +1. Where “+1” represents gradient rise, and “-1” represents gradient fall.

[0118] Calculate the gradient direction consistency index:

[0119] In this sequence, “+1” appears 4 times, accounting for 80%, so the consistency index is 0.8.

[0120] Construct the perturbation intensity structure:

[0121] For each time point, the maximum gradient amplitude value is combined with the consistency index 0.8 to form a triple such as (4.2, 0.8, t1), (3.8, 0.8, t2)...(4.7, 0.8, t5), to form a perturbation intensity structure sequence.

[0122] This embodiment is aimed at the engineering background that the sliding support generates non-periodic, low-frequency perturbation and intermixed inertia reversal behavior in the vibration state. A pressure perturbation structure construction method based on inflection point feature extraction main displacement segment sequence and gradient consistency coding is proposed, which realizes significant structure reconstruction of input data. By constructing the main displacement segment sequence and aligning it with the corresponding pressure perturbation response segment, not only the non-key segments are effectively eliminated, but also the time consistency mapping between displacement change and perturbation intensity is realized.

[0123] Further, in constructing the disturbance intensity structure from the pressure disturbance response segment, a two-dimensional feature code with maximum gradient amplitude and consistent direction is used, and an anchor alignment type time sequence position code and a one-dimensional convolution embedding network are introduced, which improves the perception ability of the model to periodic disturbance patterns. This improvement significantly improves the structural expression ability of the prediction input without introducing complex hardware or additional sensors, enhances the sensitivity and generalization ability of the slip trend prediction model to structural disturbances, and is an important optimization means for actual engineering deployment.

[0124] The method for calculating the disturbance volatility based on the disturbance intensity sequence comprises:

[0125] Perform a sliding window segmentation operation on the disturbance intensity sequence, set a fixed length local window to slide along the time axis step by step, and extract the disturbance intensity subsequence in each window;

[0126] Perform a first-order difference operation on each disturbance intensity subsequence to obtain a disturbance intensity change sequence;

[0127] Calculate the first-order difference variance based on the disturbance intensity change sequence, and take the first-order difference variance as the disturbance volatility corresponding to the current window.

[0128] The method for inputting the fusion state vector into the memory unit of the LSTM model to update the state and obtaining the slip trend prediction model comprises:

[0129] The fusion state vector is input as the state at the current time into the preset LSTM model, spliced with the hidden state at the last time, and then input into the LSTM model for calculating the activation paths of the forget gate, input gate and output gate to obtain the corresponding gate weight;

[0130] The degree of preservation of the memory state at the last time is controlled based on the forget gate weight, and the update amplitude of the current fusion state vector to the memory state is adjusted based on the input gate weight;

[0131] The current memory state and the output state obtained after the above processing are taken as the output results of the LSTM unit at the current time to obtain the slip trend prediction model.

[0132] It can be understood that the control of the retention degree of the memory state of the previous moment based on the forgetting gate weight refers to that in the slip trend prediction model, the LSTM unit filters the memory state of the previous moment through the forgetting gate, the forgetting gate generates a weight vector consistent with the dimension of the memory state according to the current input, the value is between 0 and 1, which is used to adjust whether to retain the past memory information dimension by dimension, when the weight of a certain dimension is close to 1, it means that the historical slip trend information of this dimension is completely retained; when the weight is close to 0, it means that the information is effectively inhibited. This mechanism helps the model to actively forget irrelevant historical information and avoid state drift caused by long-term accumulation of invalid disturbance.

[0133] The adjustment of the update amplitude of the current fusion state vector to the memory state based on the input gate weight refers to that when the LSTM model encodes the current slip trend feature, the input gate determines the writing degree of each dimension information in the current fusion state vector to the memory state. The input gate exists in the form of a gate vector output by a sigmoid function, which controls which information in the fusion state vector is introduced into the new memory state. After the fusion state vector is transformed by tanh, it is multiplied by the input gate weight element by element to generate an update amount, and is superimposed with the old memory state processed by the forgetting gate. This processing method ensures that the disturbance adjustment signal is only introduced when it has time series stability or trend, thereby improving the accuracy and anti-interference ability of the slip trend prediction.

[0134] S20: input the real-time slip trajectory data and real-time pressure gradient data at the current moment into the slip trend prediction model to obtain a slip trend sequence;

[0135] In this embodiment, the slip trend sequence represents the displacement change trend of the slip support in the future time steps. The slip trend prediction model based on the real-time slip trajectory data and real-time pressure gradient data at the current moment is used to predict the slip trend sequence, which is output as a continuous vector with time series structure. Specifically:

[0136] The predicted value corresponding to each time step represents the possible displacement state of the slip support relative to the initial reference position at that time point;

[0137] The slip trend sequence as a whole reflects whether the slip behavior tends to be stable, reverse, intensify or weaken in the short term;

[0138] The slip trend sequence can be used as a slip warning parameter in the bridge health monitoring system to assist in judging whether the structure has abnormal displacement accumulation, stress relaxation or potential instability risk.

[0139] As shown in Figure 2 Figure 2 ​The comparison between the sliding trend sequence output by the sliding trend prediction model constructed in the embodiment and the actual sliding trajectory is shown in the figure. The blue solid line in the figure represents the displacement response sequence obtained by integrating the acceleration data obtained based on the inertial measurement unit. The actual trajectory of the sliding trend is constructed by calculating the difference between adjacent displacement points. The orange dashed line is the predicted sliding trend sequence output after fusing the main sequence encoding channel and the gating residual channel in the present application, which is used to depict the trend response change of the sliding support under disturbance conditions.

[0140] The red background area in the figure is the disturbance time window, i.e. the disturbance score factor is calculated through the disturbance change rate sequence, and the disturbance response time period is determined after comparing with the preset score threshold. As can be seen, within the disturbance time window, the predicted sliding trend sequence and the actual trajectory change remain good consistency, which shows that the disturbance gating structure based on asymmetric residual coupling in the embodiment can accurately identify the short-time displacement response mutation section and has strong sliding trend modeling and prediction ability.

[0141] It should be noted that, in order to realize the sensitivity of the disturbance score factor index to the unstable section of the sliding trend in the embodiment, the difference between the actual and predicted sliding trend sequences is shown in the figure, and the matching degree of the score factor and the trend error change in the plurality of disturbance windows is calculated respectively, and the calculation index is as follows: Figure 2

[0142] ;

[0143] In the formula, MAE represents the mean absolute error, N represents the number of samples, represents the actual sliding trend value at the th time point, represents the predicted sliding trend value at the th time point, represents the absolute value operation, represents the average processing of the error sum.

[0144] In the embodiment, in order to verify the enhancement effect of the proposed gating residual structure on the prediction accuracy of the sliding trend under different interference conditions, a "model prediction error comparison table under different interference intensities" as shown in the table is constructed. The mean square error (MSE) index of the "no gating structure" and the "introduced gating structure" two model structures for sliding trend prediction under the four interference conditions of "no disturbance, slight disturbance, moderate disturbance and strong disturbance" is calculated respectively. By comparing the error difference of the prediction results of the two structures under the same interference intensity, the improvement amplitude of the model in the anti-interference prediction performance after introducing the gating structure is analyzed.

[0145] Model prediction error comparison table under different interference intensities​

[0146]

[0147] As can be seen from the table, with the gradual increase of interference intensity, the prediction errors of the two types of structures show an upward trend, but the prediction error of the model introducing the gating structure is always significantly lower than that of the model without introducing the structure, where the prediction error is reduced from 0.021 to 0.014 under the “slight disturbance” scenario, and from 0.036 and 0.052 to 0.018 and 0.029 under the “moderate disturbance” and “strong disturbance” conditions, respectively, with a maximum error reduction of more than 44.2%, indicating that the proposed gating residual structure can effectively adjust the influence of disturbance input on the memory state update path, significantly improve the trend tracking stability and anti-interference prediction ability of the model under high disturbance scenarios, and verify the robustness advantage of the model in the slip trend prediction task.

[0148] S30: calculating a disturbance change rate sequence corresponding to the slip trend sequence, marking a prediction time axis of the slip trend sequence based on the disturbance change rate sequence, and obtaining a target time window, the target time window including a disturbance time window and a non-disturbance time window;

[0149] The method for calculating the disturbance change rate sequence corresponding to the slip trend sequence comprises:

[0150] Dividing the slip trend sequence into a plurality of slip trend subsequences according to a preset rule;

[0151] In each slip trend subsequence, selecting a slip trend value corresponding to a first time point of the subsequence as a trend start value, selecting a slip trend value corresponding to an end time point of the subsequence as a trend end value, calculating a median time index of the subsequence based on a time index sequence of the subsequence, and extracting a slip trend value corresponding to the median time index as a trend median value;

[0152] Calculating a first change difference value between the trend start value and the trend median value, and a second change difference value between the trend median value and the trend end value, and determining a two-segment change index based on the first change difference value and the second change difference value;

[0153] Arranging the two-segment change index in time sequence to obtain the disturbance change rate sequence.

[0154] The method for determining the two-segment change index based on the first change difference value and the second change difference value comprises:

[0155] Weighted sum of the first change amplitude value and the second change amplitude value as the trend fluctuation amplitude of the current slip trend subsequence;

[0156] Calculating the amplitude product of the first change amplitude value and the second change amplitude value, and taking the sign information of the amplitude product as the trend change index.

[0157] The trend fluctuation amplitude and the trend change index are combined to form a double-segment change index corresponding to the current sliding trend subsequence.

[0158] For example, assume that the sliding trend values of the current sliding trend subsequence are as follows (each value corresponds to a time point):

[0159] Time point t1 t2 t3 t4 t5;

[0160] Trend value V1 V2 V3 V4 V5;

[0161] Numerical value 2 5 4 3 6;

[0162] The trend start value, the trend middle value, and the trend end value are constructed.

[0163] Trend start value = V1 = 2 (value corresponding to t1);

[0164] Trend end value = V5 = 6 (value corresponding to t5);

[0165] Middle time index = t3 (middle position), and trend middle value = V3 = 4;

[0166] Step 2: Calculate the first change difference and the second change difference.

[0167] First change difference = trend middle value - trend start value = 4 - 2 = +2;

[0168] Second change difference = trend end value - trend middle value = 6 - 4 = +2;

[0169] Calculate the double-segment change index.

[0170] Trend fluctuation amplitude:

[0171] First change amplitude value = |+2| = 2;

[0172] Second change amplitude value = |+2| = 2;

[0173] Weighted sum (assuming equal weights) = 2 + 2 = 4;

[0174] Trend change direction flag:

[0175] Amplitude product sign = sign(+2 x +2) = positive;

[0176] That is, the change directions are consistent (both rising);

[0177] Final double-segment change index:

[0178] The combined form can be: [fluctuation amplitude = 4, direction flag = +1].

[0179] The embodiment divides the sliding trend sequence into sliding trend subsequences, extracts trend start value, trend middle value and trend end value respectively, constructs double-segment change difference index between trend segments, and then calculates trend fluctuation amplitude and direction change mark respectively to form a combined double-segment change index. The index can not only depict the nonlinear evolution characteristics of local trend, but also jointly depict trend mutation and continuity. The method can avoid the structural omission problem of traditional judgment based on single slope or sliding difference, thereby improving the construction accuracy of the disturbance change rate sequence, realizing higher precision marking of the disturbance time window in the prediction time axis, and having obvious engineering application value.

[0180] The method for marking the prediction time axis of the sliding trend sequence based on the disturbance change rate sequence includes the following steps.

[0181] Performing sliding time window division on the disturbance change rate sequence according to a fixed length to obtain a plurality of disturbance change rate subsequences;

[0182] Performing normalization processing on all trend fluctuation amplitude values in the current disturbance change rate subsequence based on a range normalization rule to obtain a normalized fluctuation amplitude sequence;

[0183] Based on the trend change index in the current disturbance change rate subsequence, the number of sign switching is counted as the number of direction reversals. The direction reversal refers to that adjacent two trend change indexes are different.

[0184] Based on the normalized fluctuation amplitude sequence, the corresponding fluctuation average amplitude is calculated.

[0185] The fluctuation average amplitude and the number of direction reversals are multiplied to obtain a disturbance score factor corresponding to the current disturbance change rate subsequence.

[0186] Judging whether the disturbance score factor is greater than a preset score threshold. If yes, the sliding time window is marked as a disturbance time window. If not, the sliding time window is marked as a non-disturbance time window.

[0187] The range normalization rule refers to linearly scaling a set of numerical values according to the difference between the maximum value and the minimum value, so as to normalize the numerical values to a specified interval (usually 0, 1). The core of the rule is to maintain the relative size relationship of the original data while eliminating the influence of different scales on subsequent calculations. The range normalization rule is the prior art, and this embodiment will not make too much repetition. It should be noted that the disturbance score factor is composed of the product of the trend fluctuation amplitude and the number of direction reversals. The former reflects the intensity of the trend change, and the latter reflects the instability of the trend direction. The factor comprehensively considers the strength and repeatability of the trend change, can more accurately depict the mutation characteristics of the potential disturbance event, and is helpful to define the time interval of the significant turning point of the sliding behavior.

[0188] In this embodiment, the disturbance score factor is constructed by introducing the combination of the fluctuation average amplitude and the number of direction reversals, which significantly improves the ability to capture the characteristics of "intensity + direction variability" of the trend change. Further, by comparing the score factor with the preset threshold, the sliding trend prediction time axis is finely marked, the disturbance window caused by local trend mutation or trend repeated oscillation can be automatically identified, the structural omission problem caused by relying on a specific change amplitude or a single slope threshold is avoided, and the accuracy and stability of the disturbance time window identification are improved.

[0189] As shown in Figure 3 , the horizontal axis is the time index number in the sliding trend prediction time sequence, the vertical axis is the disturbance feature channel contained in the disturbance intensity structure, such as the maximum gradient amplitude, the gradient direction consistency index, etc., and the color scale represents the disturbance score value of the corresponding feature channel at each time point. The deeper the color scale, the more intense the trend change in the time period, and the higher the disturbance probability.

[0190] The figure is used to assist in the disturbance identification of the sliding trend sequence and the division of the disturbance time window, and serves as the basis for the generation of the disturbance section in Figure 2 . By discriminating the time sequence aggregation area of the disturbance score value, the disturbance section index range can be dynamically output, and the automatic labeling of the disturbance section and the non-disturbance section of the sliding trend sequence can be realized.

[0191] For example, in order to further verify the applicability of the definition method of the disturbance score factor in the actual sliding trend prediction, five representative sliding trend time windows are selected, and the trend fluctuation amplitude, the number of direction reversals and the corresponding disturbance score factor are calculated, and the calculation formula is as follows:

[0192] ;

[0193] ;

[0194] In the formula, the disturbance score factor is calculated according to the following formula: , where the disturbance score factor is calculated according to the following formula: a trend fluctuation amplitude in a time window, a trend value representing a start position of the window, a trend value representing a middle position of the window, a trend value representing an end position of the window, an absolute value operation, a disturbance score factor corresponding to a first a number of direction reversals in the window.

[0195] It should be noted that the above calculation is applied to five time windows, and the experimental data results are shown in the following table:

[0196] Disturbance score factor experimental data table

[0197] Window number Trend start value Trend mid value Trend end value Mean fluctuation amplitude Direction reversal count Disturbance score factor W1 4.6468 5.3081 3.9322 1.019 1 1.019 W2 1.3132 1.9316 1.9126 0.319 2 0.637 W3 7.7491 6.4796 6.1376 0.806 1 0.806 W4 3.9019 3.4046 2.7216 0.590 1 0.590 W5 5.1311 4.5589 3.8366 0.647 2 1.294

[0198] As can be seen from the table results, window W5 has the highest disturbance score factor (1.294), the corresponding number of direction reversals is 2, and the average fluctuation amplitude is 0.647, indicating that there is significant reversal and continuous fluctuation in the trend change process, which is consistent with the typical disturbance behavior characteristics; under the condition that the number of direction reversals is also 2, window W2 has a significantly lower disturbance score (0.637) due to a fluctuation amplitude of only 0.319, indicating that the score factor can effectively distinguish the difference between amplitude-dominant and direction-dominant disturbances; the rest such as W1, W3, and W4 are single direction reversals, and the scoring results are highly consistent with the fluctuation amplitudes, further verifying the response sensitivity and segment difference description ability of the disturbance score factor in combination with the trend change intensity and instability characteristics, proving that the factor has practical application value and discrimination advantage in trend prediction disturbance window identification tasks.

[0199] To verify the robustness and optimal setting interval of the proposed disturbance score factor in identifying abnormal segments of sliding trends under different threshold settings, a "score threshold sensitivity test table" is constructed, six cases are set for the score threshold from 0.3 to 0.8, the disturbance identification accuracy and disturbance identification recall rate under each threshold are counted, to evaluate the influence of score factor threshold adjustment on the system identification performance, and to assist in determining the best threshold range.

[0200] Score threshold sensitivity test table

[0201] Score threshold Disturbance identification accuracy Disturbance identification recall 0.3 0.86 0.87 0.4 0.89 0.88 0.5 0.91 0.89 0.6 0.92 0.90 0.7 0.88 0.86 0.8 0.85 0.81

[0202] ​It should be noted that from the data results in the table, it can be seen that the score threshold reaches the optimal recognition effect at 0.6, the accuracy is 0.92, and the recall rate is 0.90, which is significantly better than the remaining threshold combinations; when the threshold is lower than 0.4 or higher than 0.7, the recognition performance decreases, specifically, the recall rate decreases (such as 0.8, the recall rate is 0.81), or the accuracy decreases (such as 0.3, the accuracy is only 0.86), indicating that the sensitivity of the score factor in this range has a nonlinear response characteristic. Considering the accuracy and coverage, the score threshold is set in the [0.5, 0.6] interval, which can effectively improve the positioning accuracy and stability of the slip trend disturbance segment, and has good engineering adaptability and parameter adjustment space.

[0203] S40: Obtain real-time strain data, perform first filtering processing on the real-time strain data in the disturbance time window, and perform second filtering processing on the real-time strain data in the non-disturbance time window, wherein the collection time axis of the real-time strain data corresponds to the prediction time axis of the slip trend sequence.

[0204] In this embodiment, the real-time strain data refers to continuous time sequence data collected by strain sensors arranged on the support body or its connecting structure (such as the bottom of the bridge pier or the end of the bridge beam) during the operation of the slip support, which specifically represents the local deformation response of the structure under the action of micro-slip or vibration disturbance. This data is usually obtained at a set sampling frequency (such as 100 Hz, 500 Hz), reflecting the micro stress response characteristics of the bridge structure due to stress changes.

[0205] It should be noted that the first filtering processing refers to enhanced filtering operation performed on the real-time strain data identified in the disturbance time window, which aims to retain the strain change signal as much as possible while suppressing local high-frequency noise. The first filtering processing can be band-pass filtering processing or wavelet denoising processing. The second filtering processing refers to the smoothing filtering operation performed on the real-time strain data in the non-disturbance time window, which aims to suppress the micro signal fluctuation and extract the structure static strain baseline or background trend. The second filtering processing can be a sliding average filtering or low-pass filtering algorithm.

[0206] The embodiment fuses historical sliding trajectory data and pressure gradient data by constructing a sliding trend prediction model with a disturbance weight gating structure, effectively enhances the time sequence modeling capability of the nonlinear response of the sliding support under vibration state, improves the perception accuracy of the model to short-time disturbance and inertial reversal behavior, then constructs a disturbance change rate sequence based on the predicted sliding trend sequence, calculates the disturbance score factor using the trend fluctuation amplitude and the number of direction reversal, and dynamically marks the disturbance time window accordingly, thereby solving the problem of lacking dynamic division means for support disturbance behavior in the prior art, finally, different filtering processing methods are used for the disturbance time window and the non-disturbance time window respectively, and the real-time strain data is processed differently, which not only improves the availability of strain data under low-frequency disturbance background, but also enhances the capture accuracy and stability of the structural health monitoring system to abnormal strain response.

[0207] Embodiment 2

[0208] Please refer to Figure 4 Based on the same inventive concept, the embodiment discloses a bridge support design detection data acquisition system, and the details of the embodiment can be referred to the description of the related part in Embodiment 1. The system comprises:

[0209] The construction module is configured to obtain historical sliding trajectory data and historical pressure gradient data of the sliding support under vibration state, and construct a sliding trend prediction model with a disturbance weight gating structure based on the historical sliding trajectory data, the contact pressure gradient data and a preset LSTM model.

[0210] The prediction module is configured to input real-time sliding trajectory data and real-time pressure gradient data at the current time into the sliding trend prediction model to obtain a sliding trend sequence.

[0211] The division module is configured to calculate a corresponding disturbance change rate sequence according to the sliding trend sequence, mark a prediction time axis of the sliding trend sequence based on the disturbance change rate sequence, and obtain a target time window, wherein the target time window comprises a disturbance time window and a non-disturbance time window.

[0212] The filtering module is configured to obtain real-time strain data, perform first filtering processing on the real-time strain data in the disturbance time window, and perform second filtering processing on the real-time strain data in the non-disturbance time window, wherein a collection time axis of the real-time strain data corresponds to a prediction time axis of the sliding trend sequence.

[0213] The detailed description set forth above in connection with the appended drawings describes examples and does not represent all the examples that can be implemented or that are within the scope of the claims. The terms "example" and "exemplary" used herein means "serving as an example, instance, or illustration," and not "preferred or advantageous over other examples." The detailed description set forth above in connection with the appended drawings describes examples and does not represent all the examples that can be implemented or that are within the scope of the claims.

[0214] References to "one embodiment" or "an embodiment" throughout this specification mean that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the application, and the appearance of the phrases "in one embodiment" or "in an embodiment" in various places throughout this specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Moreover, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0215] It should also be noted that these embodiments can be described as a process that is depicted as a flow diagram, a structure diagram, or a block diagram. Although a flow diagram can describe operations as a sequential process, many of the operations can be performed in parallel, or concurrently, or in any suitable order. In addition, the order of the operations can be re-arranged.

Claims

1. A method for acquiring test data for bridge bearing design, characterized in that, include: Historical sliding trajectory data and historical pressure gradient data of sliding supports under vibration conditions are obtained. Based on the historical sliding trajectory data, contact pressure gradient data and a preset LSTM model, a sliding trend prediction model with perturbation weight gating structure is constructed. The real-time slip trajectory data and real-time pressure gradient data at the current moment are input into the slip trend prediction model to obtain the slip trend sequence; The corresponding perturbation rate of change sequence is calculated based on the slip trend sequence. The prediction time axis of the slip trend sequence is marked based on the perturbation rate of change sequence to obtain the target time window, which includes the perturbation time window and the non-perturbation time window. Real-time strain data is acquired, and the real-time strain data within the disturbance time window is subjected to a first filtering process, while the real-time strain data within the non-disturbance time window is subjected to a second filtering process. The acquisition time axis of the real-time strain data corresponds to the prediction time axis of the slip trend sequence. Methods for constructing slip trend prediction models with perturbation weight gating structures based on historical slip trajectory data, contact pressure gradient data, and pre-defined LSTM models include: Based on historical slip trajectory data and historical pressure gradient data, a first input sequence and a second input sequence are constructed respectively; The first input sequence is input into the main sequence encoding channel of the LSTM model to obtain the displacement change channel; The second input sequence is input into the preset disturbance recognition network to extract the disturbance intensity sequence. The disturbance volatility is calculated based on the disturbance intensity sequence. The disturbance volatility is input into the gated residual channel of the LSTM model to obtain the disturbance adjustment channel. The disturbance volatility is the first difference variance of the disturbance intensity sequence. Based on the asymmetric residual coupling method, the displacement change channel and the disturbance adjustment channel are fused to generate a fused state vector. The fused state vector is then input into the memory cell of the LSTM model for state update, resulting in a slip trend prediction model.

2. The method for collecting test data for bridge bearing design according to claim 1, characterized in that, Methods for obtaining historical slip trajectory data of sliding supports under vibration conditions include: A first inertial measurement unit and a second inertial measurement unit are respectively installed at the top of the pier and the bottom of the sliding bearing to collect acceleration data under vibration conditions. Time alignment is performed on the acceleration data, and then filtering and denoising are performed on the time-aligned acceleration data. The filtering and denoising process includes removing high-frequency interference components and low-frequency drift components. The filtered acceleration data are integrated to obtain the first displacement response sequence at the top of the pier and the second displacement response sequence at the bottom of the sliding support. Based on the first displacement response sequence and the second displacement response sequence, the response difference sequence is calculated. The response difference sequence characterizes the change of the sliding trajectory of the sliding support under vibration state, and the response difference sequence is used as historical sliding trajectory data.

3. The method for collecting test data for bridge bearing design according to claim 1, characterized in that, The methods for constructing the first input sequence and the second input sequence based on historical slip trajectory data and historical pressure gradient data include: Vibration period identification processing is performed on historical slip trajectory data to obtain inflection point features at the period boundary of historical slip trajectory data. Based on the inflection point features, the main displacement segment sequence is extracted from historical slip trajectory data, and the main displacement segment sequence is encoded as the first input sequence according to continuous time index. Synchronous segment extraction processing is performed on historical pressure gradient data to obtain pressure disturbance response segments corresponding to the time axis of the main displacement segment sequence, and the maximum gradient amplitude and gradient consistency index of the pressure disturbance response segments in each slip cycle are calculated. The maximum gradient magnitude and gradient consistency index are encoded into a perturbation intensity structure sequence on the time axis. The perturbation intensity structure sequence is then subjected to time-series embedding processing to obtain the second input sequence.

4. The method for acquiring test data for bridge bearing design according to claim 3, characterized in that, The inflection point features include acceleration extrema and local curvature extrema. Methods for extracting main displacement segment sequences from historical slip trajectory data based on inflection point features include: The acceleration extremum is used as the first inflection point index, and the local curvature extremum is used as the second inflection point index. The first inflection point index and the second inflection point index are fused together to form a joint inflection point sequence; The time interval between every two adjacent indices in the joint inflection point sequence is determined as a candidate segment; For each candidate segment, calculate the displacement amplitude of the candidate segment; If the displacement amplitude is less than the preset amplitude threshold, delete the current candidate segment; otherwise, calculate the number of velocity reversals within the candidate segment. If the number of velocity reversals is lower than the preset reversal number threshold, the current candidate segment is deleted; otherwise, the candidate segment is included in the main displacement segment sequence.

5. The method for acquiring test data for bridge bearing design according to claim 4, characterized in that, The method of encoding the maximum gradient magnitude and gradient consistency index into a perturbation intensity structure sequence on the time axis, and then performing time-series embedding processing on the perturbation intensity structure sequence to obtain the second input sequence includes: The maximum gradient magnitude is used as the pressure intensity dimension of the perturbation intensity structure, the gradient direction consistency index is used as the direction consistency dimension of the perturbation intensity structure, and the time index combination corresponding to the pressure perturbation response segment is used as the time dimension of the perturbation intensity structure, thus constructing a perturbation intensity structure in the form of a ternary structure. Arrange all disturbance intensity structures in ascending order of time index to form a disturbance intensity structure sequence; Normalize each perturbation intensity structure in the perturbation intensity structure sequence, and add a temporal position vector to each perturbation intensity structure based on a preset position encoding scheme; The perturbation intensity structure sequence with time encoding is input into a preset temporal embedding network, and a nonlinear activation operation is performed to obtain the corresponding second input sequence.

6. The method for acquiring test data for bridge bearing design according to claim 1, characterized in that, Methods for calculating the corresponding rate of change of disturbance sequence based on slip trend sequence include: The slip trend sequence is divided into multiple slip trend sub-series according to preset rules; In each slip trend subsequence, the slip trend value corresponding to the first time point of the subsequence is selected as the trend start value, and the slip trend value corresponding to the last time point of the subsequence is selected as the trend end value. The median time index is calculated based on the time index sequence of the subsequence, and the slip trend value corresponding to the median time index is extracted as the trend median. Calculate the first change difference between the trend start value and the trend median, and the second change difference between the trend median and the trend end value. Determine the two-segment change index based on the first change difference and the second change difference.

7. The method for acquiring test data for bridge bearing design according to claim 6, characterized in that, Methods for determining two-segment change indicators based on the first change difference and the second change difference include: The first change magnitude value and the second change magnitude value are weighted and summed to obtain the trend fluctuation magnitude of the current slip trend subsequence; Calculate the product of the first and second magnitude changes, and use the sign information of the product as an indicator of trend change. By combining the trend fluctuation amplitude with the trend change indicator, a two-segment change indicator corresponding to the current slip trend subsequence is formed.

8. The method for acquiring test data for bridge bearing design according to claim 7, characterized in that, Methods for marking the prediction time axis of a slip trend sequence based on the perturbation rate of change sequence to obtain the target time window include: The perturbation rate of change sequence is divided into multiple perturbation rate of change subsequences by performing a sliding time window of fixed length; Based on the range normalization rule, all trend fluctuation amplitude values ​​in the current perturbation rate of change subsequence are normalized to obtain the normalized fluctuation amplitude sequence. Based on the trend change indicators in the current perturbation change rate subsequence, the number of times the statistical sign changes is used as the number of directional reversals. A directional reversal means that two adjacent trend change indicators are different. Based on the normalized fluctuation amplitude sequence, calculate the corresponding average fluctuation amplitude; The average amplitude of the fluctuation is multiplied by the number of direction reversals, and the result is used as the perturbation score factor corresponding to the current perturbation rate of change subsequence. Determine whether the disturbance scoring factor is greater than the preset scoring threshold. If so, mark the sliding time window as a disturbance time window; otherwise, mark the sliding time window as a non-disturbance time window.

9. A bridge bearing design testing data acquisition system, used to implement the bridge bearing design testing data acquisition method according to any one of claims 1-8, characterized in that, include: The building module is used to acquire historical slip trajectory data and historical pressure gradient data of the sliding support under vibration conditions, and to build a slip trend prediction model with perturbation weight gating structure based on historical slip trajectory data, contact pressure gradient data and preset LSTM model. Prediction module: Used to input real-time slip trajectory data and real-time pressure gradient data at the current moment into the slip trend prediction model to obtain the slip trend sequence; The segmentation module is used to calculate the corresponding perturbation rate of change sequence based on the slip trend sequence, mark the prediction time axis of the slip trend sequence based on the perturbation rate of change sequence, and obtain the target time window, which includes the perturbation time window and the non-perturbation time window. Filtering module: Used to acquire real-time strain data, perform a first filtering process on the real-time strain data within the disturbance time window, and perform a second filtering process on the real-time strain data within the non-disturbance time window. The acquisition time axis of the real-time strain data corresponds to the prediction time axis of the slip trend sequence.

Images