High-speed railway bridge support state real-time evaluation method based on BiLSTM and train-induced displacement

Abstract

The invention discloses a high-speed railway bridge support state real-time evaluation method based on BiLSTM and vehicle-induced displacement. The method comprises the steps that vehicle-induced bridge displacement data of a support in a healthy state is obtained through numerical simulation or field actual measurement of a vehicle-rail bridge coupling system; training a benchmark BiLSTM model to establish bridge displacement dynamic association; a real-predicted signal energy ratio index is constructed based on a bridge displacement measured value and a predicted value, and statistical distribution of the real-predicted signal energy ratio index in a healthy state is used as a support health reference; inputting the actually measured bridge displacement in an unknown support state into the reference BiLSTM model, and contrastively analyzing the distribution characteristics of the TPSR relative to the health reference to realize qualitative and quantitative evaluation of the support state; the weak damage identification sensitivity is improved through equalization resampling, and multi-support state synchronous evaluation is realized in combination with a support vector machine. The bridge displacement response in the operation period is directly utilized, special detection equipment does not need to be arranged on the support, and the method has the advantages of being real-time in evaluation, light in scheme and high in anti-jamming capability.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of bridge health monitoring, and relates to a bridge support state evaluation technology, in particular to a high-speed railway bridge support state real-time evaluation method based on BiLSTM and vehicle-induced displacement. BACKGROUND

[0002] With the rapid development of high-speed railway technology, the load borne by high-speed railway bridges is increasingly diversified and complicated. The support connects the upper structure and the lower structure of the bridge and is an important force-bearing and force-transferring component. Meanwhile, the support is a key and weak part that is prone to damage. The damage of the support directly changes the boundary conditions of the bridge, thereby affecting the dynamic response of the bridge. Meanwhile, the vibration of the bridge caused by high-speed trains is reacted on the train on the bridge through the vehicle-track-bridge coupling system, thereby affecting the running stability and safety of the train. Therefore, real-time evaluation of the service state of the bridge support is of great significance to the safety of high-speed railway operation.

[0003] The current methods for evaluating the state of the high-speed railway bridge support can be divided into three categories: (1) manual inspection method (such as the knife blade method), which has been applied in the identification of diseases such as support voiding and eccentric pressure, but has limitations such as limited types of diseases that can be identified, high detection risk, and poor timeliness; (2) image evaluation method based on computer vision technology, which can continuously monitor the support through a camera fixedly installed at the bottom of the bridge or can conduct mobile inspection through a drone and a robot, and can intuitively identify the specific damage type of the bridge support, but cannot quantify the damage degree of the support, and the position of the bridge support is hidden, which requires a high working environment and high performance of the detection equipment; (3) dynamic inversion evaluation method based on train or bridge vibration data, which indirectly evaluates the state of the support by extracting and analyzing damage-sensitive features in the vibration signal, but relies on detection trains or special detection data collected near the support, has a high data acquisition cost, and is based on global dynamic characteristics such as frequency and mode shape, which is not sensitive to early and weak stiffness changes of the support, and the early warning is lagging.

[0004] The existing support state evaluation methods have three limitations: (1) in terms of deployment mode and timeliness, they mainly rely on offline and periodic manual inspection or special detection, which has high monitoring cost, slow data update, and is difficult to support real-time evaluation and continuous tracking of the state of the high-speed railway bridge support during operation; (2) in terms of technical kernel, there is a lack of universal index that can uniformly represent the two typical damages of stiffness reduction and abnormal increase of the high-speed railway bridge support, making it difficult to achieve accurate qualitative and quantitative evaluation; (3) in terms of evaluation ability, although the existing methods can identify significant damage at a macro level, they are generally not sensitive to early and weak stiffness changes of the support, making it difficult to achieve early warning and accurate quantification of damage, and the evaluation and early warning are lagging. SUMMARY

[0005] Purpose of the invention: In order to overcome the shortcomings of existing technologies, such as reliance on specialized testing, insensitivity to minor damage, and lack of unified quantitative indicators, this invention provides a real-time assessment method for the condition of high-speed railway bridge bearings based on BiLSTM and vehicle-induced displacement. It has the advantages of real-time assessment process, lightweight monitoring scheme, and outstanding anti-interference ability. It can reliably perform qualitative and quantitative assessments of high-speed railway bridge bearings, providing effective technical support for the safe operation of high-speed railway lines.

[0006] Technical Solution: To achieve the above objectives, this invention provides a real-time evaluation method for the bearing status of high-speed railway bridges based on BiLSTM and vehicle-induced displacement, comprising the following steps:

[0007] Step S1: Obtain vehicle-induced bridge displacement data under healthy support conditions through numerical simulation of the vehicle-rail-bridge coupling system or on-site measurement;

[0008] Step S2: Based on the vehicle-induced bridge displacement data from Step S1, build and train a benchmark BiLSTM deep learning model to establish the dynamic correlation between the displacements of different measuring points of the bridge under the condition of healthy supports, and realize bridge displacement prediction.

[0009] Step S3: Based on the vehicle-induced bridge displacement data from Step S1 and the bridge displacement prediction data from Step S2, establish the statistical characteristics of the True-to-Predicted Signal Energy Ratio (TPSR) of the measured and predicted bridge displacement values ​​under the support health condition, and use it as a health benchmark.

[0010] Step S4: Input the measured bridge displacement under unknown conditions into the benchmark BiLSTM deep learning model established in step S2. By analyzing the distribution characteristics of the TPSR index relative to the health benchmark, a qualitative and quantitative assessment of the support condition is achieved.

[0011] Step S5: Mean-resample the TPSR index generated by batch train events to reduce its random fluctuations and improve the sensitivity to identify minor damage.

[0012] Step S6: Combine the evaluation indicators of multiple supports into a joint feature vector, input it into the support vector machine model, and realize the synchronous evaluation of the state of multiple supports.

[0013] Furthermore, in step S1, the bridge displacement measuring points caused by the vehicle include three quarter points on the main beam.

[0014] Further, step S2 includes:

[0015] Step S21: Normalize and preprocess the bridge displacement data collected in step S1 and divide it into training set, test set and validation set, which are used for model training, establishing health benchmark and early stop judgment, respectively.

[0016] Step S22: Build and train the BiLSTM deep learning model, using the mid-span displacement of the bridge as the model input and the left and right quarter-span displacements as the model output.

[0017] Step S23: Set an early stopping mechanism for the BiLSTM deep learning model training process. Calculate the validation set loss function after each training cycle. End model training when the loss no longer decreases after n consecutive validations.

[0018] Furthermore, in step S3, an evaluation benchmark for the health status of the bearings is established using the test set from step S21. Specifically, the mid-span displacement of the bridge in the test set is input into the trained BiLSTM model to obtain the predicted values ​​of the left and right quarter-span displacements. The TPSR index under the health status is calculated based on the actual values ​​and the predicted values. The distribution characteristics of the TPSR index of a batch of train events are statistically analyzed, and the health benchmark for evaluating the bearing status is used as the basis for evaluating the health status of the bearings.

[0019] Further, step S4 includes:

[0020] Step S41: Using the measured mid-span displacement of the bridge under unknown support conditions as input, call the benchmark BiLSTM model established in step S22 to predict the displacement of the left and right quarter spans of the bridge.

[0021] Step S42: Calculate the TPSR evaluation index based on the actual and predicted values ​​of the displacement of the left and right quarter spans of the bridge;

[0022] Step S43: Based on the trend and absolute value of the TPSR index distribution relative to the health baseline, qualitatively determine the type of support damage and quantitatively assess the degree of damage.

[0023] Furthermore, the mathematical expression for the true-to-predicted signal energy ratio (TPSR) of the support condition assessment index in step S42 is as follows:

[0024]

[0025] in, and These represent the actual and predicted values ​​of the bridge displacement, respectively; n is the number of sample points.

[0026] Furthermore, in step S43, when the TPSR index distribution is generally to the right of the healthy baseline, it is determined that the support has a decrease in stiffness caused by material performance degradation (such as rubber aging, steel corrosion, etc.); when the TPSR distribution is generally to the left of the healthy baseline, it is determined that the support has an abnormal increase in stiffness caused by mechanical hard contact (such as loss of mobility caused by jamming of the upper and lower support plates, etc.).

[0027] Furthermore, in step S43, the support damage state D can be quantitatively assessed using the following formula:

[0028]

[0029] in, The mean TPSR index generated by a batch of train events under unknown conditions; f represents the mean of the TPSR index under healthy baseline conditions; f is the index offset. The calibration function relating to the change in support stiffness can be a linear function, a polynomial function, or a piecewise function, depending on the actual situation, and can be obtained through numerical simulation or experimental calibration.

[0030] Furthermore, in step S5, the TPSR indices calculated from batch train events are resampled using mean resampling to obtain TPSR-m indices with smaller variance and more concentrated distribution, thereby improving the sensitivity for identifying minor damages. The mean resampling process involves resampling from the original sample set consisting of N single train event TPSR indices. In this process, R random samplings with replacement are performed, with m samples drawn each time and their mean calculated. This mean is defined as the first resampling index TPSR-m. After R resampling operations, a new set consisting of R TPSR-m indices is obtained. ;Y R The variance compared to X N The variance is reduced by 1 / m;

[0031] The mathematical expression for the TPSR-m index is:

[0032]

[0033] in, The TPSR-m index is obtained from the r-th resampling. Let be the i-th sample randomly drawn from the set during the r-th resampling.

[0034] Furthermore, in step S6, the left and right support state evaluation indicators are combined into a multi-dimensional joint feature vector, and the joint feature vector is input into the trained support vector machine classification model, and the classification model outputs the synchronous evaluation results of the left and right support states.

[0035] Furthermore, the selection of evaluation indicators in step S6 should be determined in conjunction with the requirements for identifying minor damage to the bearings and the data quality: if multiple bearings are simultaneously evaluated for obvious damage and the monitoring data is less affected by noise and other factors, the TPSR calculated in step S4 can be used directly as the evaluation indicator; if it is necessary to improve the ability to identify early minor damage to the bearings, or if the monitoring data is significantly affected by random fluctuations, then the TPSR-m obtained after mean resampling in step S5 should be used as the evaluation indicator.

[0036] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0037] 1. Lightweight monitoring scheme and high evaluation efficiency. This invention only requires the deployment of displacement sensors at three quarter points of the bridge without the need for specialized testing equipment on the supports. Based on bridge displacement data generated by batch train events, real-time evaluation of the support status can be achieved, providing a feasible technical approach for lightweight monitoring and evaluation.

[0038] 2. Sensitive damage identification and clear type differentiation. Based on the statistical characteristics of the TPSR index of batch train events, the present invention can distinguish between two typical types of damage: stiffness reduction and abnormal increase, based on the trend of its change relative to the support health benchmark. The absolute value can quantify the degree of damage. Combined with the mean resampling strategy, the sensitivity of identifying weak damage is further improved.

[0039] 3. Accurate and robust multi-support state assessment. This invention employs an assessment strategy that combines joint indicators with support vector machine classification, effectively improving the accuracy and robustness of simultaneous multi-support state assessment. Attached Figure Description

[0040] Figure 1 This is a schematic flowchart of the method of the present invention;

[0041] Figure 2 This is a diagram showing the arrangement of bridge displacement sensors according to an embodiment of the present invention;

[0042] Figure 3 Here is a diagram of the BiLSTM model structure;

[0043] Figure 4 This is a training loss curve of the BiLSTM model in an embodiment of the present invention;

[0044] Figure 5 The following is a statistical chart of single support damage status assessment indicators according to an embodiment of the present invention, wherein (a) is working condition BD1; (b) is working condition BD2; (c) is working condition BD3; and (d) is working condition BD4.

[0045] Figure 6 This is the single-support condition evaluation result of an embodiment of the present invention;

[0046] Figure 7 The following are the results of the multi-support status assessment in the embodiments of the present invention, wherein (a) is the status assessment result of the left support; and (b) is the status assessment result of the right support. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0048] like Figure 1 As shown, this embodiment provides a real-time evaluation method for the bearing status of high-speed railway bridges based on BiLSTM and vehicle-induced displacement, including the following steps:

[0049] Step S1: Obtain the reference data of vehicle-induced bridge displacement under the healthy condition of the supports through numerical simulation of the vehicle-rail-bridge coupling system.

[0050] like Figure 2 As shown, in this embodiment, the bridge span is 32 m, and the initial stiffness of the left and right supports is... Displacement sensors were placed at three quarter points of the bridge, spaced 8 m apart. These three displacement measurement points were designated DM1 (left quarter point), DM2 (mid-span), and DM3 (right quarter point). Through numerical simulation of the vehicle-track-bridge coupled system, considering excitations from different track irregularities, 1000 independent calculations were performed to construct a benchmark dataset covering the dynamic displacement response of the bridge under healthy bearing conditions.

[0051] Step S2: Build and train a benchmark BiLSTM deep learning model to establish the intrinsic dynamic relationship between the displacements of different measuring points of the bridge under the condition of healthy supports.

[0052] Step S2 specifically includes the following process:

[0053] Step S21: Data preprocessing and dataset partitioning

[0054] The displacement time history data obtained in step S1 is normalized using Min-Max. The displacement sequence of the mid-span measuring point DM2 is used as the model input, and the displacement sequences of the left and right quarter points DM1 and DM3 are used as the model output. The 1000 samples are divided into training, testing, and validation sets in a 5:3:2 ratio. The training set is used for model training, the testing set is used to establish a benchmark for supporting health status assessment, and the validation set is used for early stopping judgment in model training.

[0055] Step S22: BiLSTM model construction and training.

[0056] like Figure 3As shown, a bidirectional long short-term memory network model is constructed, and its core structure includes:

[0057] Input layer: Receives normalized DM2 measurement point displacement time series data.

[0058] BiLSTM layer: In this embodiment, a 1-layer BiLSTM network with 64 neuron units is used to capture the temporal dependence features in the displacement response from both positive and negative directions.

[0059] Fully connected layer: The high-dimensional features extracted by the BiLSTM layer are mapped to two output channels, corresponding to the displacement prediction values ​​of DM1 and DM3 respectively.

[0060] Output layer (regression layer): Outputs the predicted displacement time history of measurement points DM1 and DM3.

[0061] Step S23: Model training and setting up an early stopping mechanism.

[0062] The Adam optimizer is used, with mean squared error as the loss function for model training. An early stopping mechanism is introduced during training: after each training epoch, the model's loss on the validation set is calculated. Training terminates when the validation set loss no longer decreases for five consecutive training epochs. This effectively prevents overfitting. After training, the baseline BiLSTM deep learning model is saved.

[0063] like Figure 4 As shown, the training loss function of the BiLSTM model converges after 235 iterations, indicating that the BiLSTM model accurately establishes the dynamic correlation of bridge displacement under healthy support conditions.

[0064] Step S3: Establish the statistical distribution characteristics of the True-Predicted Signal Energy Ratio (TPSR) of the bridge displacement under the condition of bearing health, as a health benchmark.

[0065] The test set in step S21 is used to establish an evaluation benchmark for the bearing health status. Specifically, 300 sets of bridge mid-span displacements from the test set are input into the BiLSTM model trained in step S22 to obtain predicted values ​​for the left and right quarter-span displacements. Based on the predicted and actual values, the TPSR index under the health status is calculated. By statistically analyzing the TPSR index of a batch of train events, the distribution characteristics of the TPSR index are used as the health benchmark for evaluating the bearing status. For ease of distinction, the health benchmark for the bearing status in this embodiment is denoted as condition BD0.

[0066] Step S4: Input the measured bridge displacement under unknown conditions into the benchmark BiLSTM deep learning model established in step S2. By analyzing the statistical distribution characteristics of the TPSR index relative to the health benchmark, the qualitative and quantitative assessment of the support status is achieved.

[0067] Step S41: Using the measured mid-span of the bridge under unknown support conditions as input, call the benchmark BiLSTM deep learning model established in step S22 to predict the displacement of the left and right quarter spans of the bridge.

[0068] To verify the effectiveness and reliability of the method of this invention in qualitative and quantitative assessment of the bearing condition of high-speed railway bridges, the single bearing damage conditions considered in this embodiment are shown in Table 1. In these conditions, the stiffness of the left bearing decreases sequentially from BD1 to BD3, indicating a decrease in stiffness caused by material performance degradation (such as rubber aging and steel corrosion). In condition BD4, the stiffness of the left bearing is infinite, indicating an abnormal increase in support stiffness due to mechanical hard contact (such as loss of mobility caused by jamming of the upper and lower bearing plates).

[0069] Table 1 Single Support Damage Conditions

[0070]

[0071] Step S42: Calculate the TPSR evaluation index based on the actual and predicted values ​​of the displacement of the left and right quarter spans of the bridge.

[0072] In this embodiment, 300 sets of bridge displacement data are collected for each evaluation state. The displacement data of measuring point DM2 is input into the BiLSTM model trained in step S22 to obtain the predicted displacement values ​​of the left and right quarter points DM1 and DM3. Subsequently, based on the actual values ​​and model prediction values ​​of the displacements of DM1 and DM3, the TPSR evaluation index is calculated. The mathematical expression of the bearing condition evaluation index TPSR is:

[0073]

[0074] in, and These represent the actual and predicted values ​​of the bridge displacement, respectively; n is the number of sample points.

[0075] Step S43: Based on the trend and absolute value of the TPSR index relative to the health baseline, qualitatively determine the type of support damage and quantitatively assess the degree of damage.

[0076] The bearing damage state D can be quantitatively assessed using the following formula:

[0077]

[0078] in, The mean TPSR index generated by a batch of train events under unknown conditions; f represents the mean of the TPSR index under healthy baseline conditions; f is the index offset. The calibration function relating to the change in support stiffness can be a linear function, a polynomial function, or a piecewise function, depending on the actual situation, and can be obtained through numerical simulation or experimental calibration.

[0079] The distribution characteristics specifically refer to the distribution histogram of TPSR generated by batch train events. In this embodiment, the distribution characteristics are as follows: Figure 5 As shown.

[0080] like Figure 5 As shown, the TPSR distribution characteristics of the left bearings under different bearing damage conditions exhibit significant differences: the TPSR distribution of conditions BD1-BD3 shifts to the right relative to the healthy baseline (BD0), and the degree of shift increases as the bearing stiffness decreases; while condition BD4 shifts to the left. Therefore, when the TPSR distribution is generally to the right of the healthy baseline, it can be determined that the bearing stiffness has decreased; when the TPSR distribution is generally to the left of the healthy baseline, it can be determined that the bearing stiffness has abnormally increased. For the method of this invention, only the bridge displacement data during the passage of a batch of trains is needed to perform qualitative and quantitative assessments of the bearing condition.

[0081] Step S5: Mean-resample the TPSR index generated by batch train events to reduce its random fluctuations and improve the sensitivity to identify minor damage.

[0082] In actual operation, high-speed railway bridges experience complex load patterns and random disturbances due to track irregularities. Furthermore, bridge displacement data is susceptible to noise. These factors may mask the effects of bearing damage, causing overlap between the TPSR index distribution and the health baseline. This leads to a decrease in the reliability of statistical distribution-based damage assessment methods and their ability to identify early, minor damage. The method of this invention performs mean resampling on the TPSR indices calculated from batch train events to obtain a TPSR-m index with smaller variance and a more concentrated distribution, thus improving the sensitivity for identifying minor damage. The mean resampling process involves resampled from an original sample set consisting of TPSR indices from N single train events. In this process, R random samplings with replacement are performed, with m samples drawn each time and their mean calculated. This mean is defined as the first resampling index TPSR-m. After R resampling operations, a new set consisting of R TPSR-m indices is obtained. ;Y R The variance compared to X N The variance is reduced by 1 / m. The mathematical expression for the TPSR-m index is:

[0083]

[0084] in, The TPSR-m index is obtained from the r-th resampling. Let be the i-th sample randomly drawn from the set during the r-th resampling.

[0085] like Figure 6 As shown, after mean-based resampling, the boundaries between different support damage conditions are clear. Therefore, for the method of this invention, a reliable assessment of the support condition can be achieved simply by comparing the resampling index TPSR-m under the condition to be evaluated with the pre-established health baseline distribution.

[0086] In this embodiment, N is 300, m is 30, and R is 20. It can be understood that the values ​​of m and R are a preferred configuration determined by the total sample size N in this embodiment, aiming to achieve a balance between effectively smoothing random fluctuations and preserving impaired characteristics. Those skilled in the art can adaptively adjust the values ​​of m and R according to the actual total amount of monitored data, data fluctuation characteristics, and specific requirements for indicator stability. For example, when the data volume is larger or the noise is more significant, the value of m can be appropriately increased to enhance the smoothing effect; to obtain a more stable statistical distribution, the value of R can also be increased. Such parameter adjustments based on the same resampling principle all fall within the protection scope of this invention.

[0087] Step S6: Combine the evaluation indicators of multiple supports into a joint feature vector, and input it into the support vector machine classification model to achieve synchronous evaluation of the state of multiple supports.

[0088] Because damage to one abutment will cause changes in the stress on its adjacent abutments and accelerate the damage development process of those abutments, multiple abutment damage is more common in actual bridge operation. The method of this invention combines the left and right abutment condition assessment indices into a two-dimensional joint feature vector T, T=[TPSR] L ,TPSR R The joint feature vector is then input into the trained support vector machine classification model, which outputs the synchronous evaluation results of the left and right support states.

[0089] In this embodiment, the Monte Carlo method is used to uniformly and randomly sample the stiffness of the left and right supports from the discrete set {0.4k, 0.6k, 0.8k, k, Inf}, generating 1000 sets of data and calculating the corresponding T. The 1000 sets of samples T are then divided into training and test sets in a 7:3 ratio to train the support vector machine.

[0090] like Figure 7 As shown, the accuracy of the assessment results for both the left and right supports reached 99.3%, with only a small number of misjudgments between the normal state and minor damage of the supports. Therefore, for the method of the invention, reliable assessment of the state of multiple supports can be achieved simply by inputting the joint index into a trained support vector machine.

[0091] The above embodiments are only used to illustrate the design concept and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, all equivalent changes or modifications made based on the principles and design ideas disclosed in the present invention are within the protection scope of the present invention.

Claims

1. A high-speed railway bridge support state real-time evaluation method based on BiLSTM and vehicle-induced displacement, characterized in that, Comprise the following steps: Step S1: Obtain the vehicle-induced bridge displacement data under the health state of the support through numerical simulation or field measurement of the vehicle-track-bridge coupling system; Step S2: Based on the vehicle-induced bridge displacement data of step S1, build and train a benchmark BiLSTM deep learning model to establish the dynamic correlation between the displacements of different measuring points of the bridge under the health state of the support, and realize the prediction of the bridge displacement; Step S3: Based on the vehicle-induced bridge displacement data of step S1 and the bridge displacement prediction data of step S2, establish the statistical characteristics of the true-predicted signal energy ratio (TPSR) index of the measured and predicted values of the bridge displacement under the health state of the support, and use it as the health benchmark; Step S4: Input the measured bridge displacement under the unknown state into the benchmark BiLSTM deep learning model established in step S2, and realize the qualitative and quantitative evaluation of the support state by analyzing the distribution characteristics of the TPSR index relative to the health benchmark; Step S5: Resample the TPSR index generated by a batch of train events by averaging to reduce its random fluctuations and improve the recognition sensitivity to weak damage; Step S6: Combine the evaluation indexes of multiple supports into a joint feature vector and input it into a support vector machine model to realize the synchronous evaluation of multiple support states.

2. The real-time evaluation method of high-speed railway bridge support state based on BiLSTM and vehicle-induced displacement according to claim 1, characterized in that, The vehicle-induced bridge displacement measuring points in step S1 include three quarter points of the main girder. 3.The real-time evaluation method of high-speed railway bridge support state based on BiLSTM and vehicle-induced displacement according to claim 1, characterized in that, The step S2 comprises: Step S21: Normalize the bridge displacement data collected in step S1 and divide it into training set, test set and validation set for model training, health benchmark establishment and early stopping judgment; Step S22: Build and train a BiLSTM deep learning model, input the bridge mid-span displacement as the model input, and output the left and right quarter-span displacement as the model output; Step S23: Set an early stopping mechanism for the BiLSTM deep learning model training process, calculate the validation set loss function after each training period, and end the model training when the loss validation does not decrease continuously for n times.

4. The real-time evaluation method of high-speed railway bridge support state based on BiLSTM and vehicle-induced displacement according to claim 3, characterized in that, In step S3, the evaluation benchmark of the support in the health state is established based on the test set of step S21, specifically: input the bridge mid-span displacement in the test set into the trained BiLSTM model to obtain the predicted values of the left and right quarter-span displacement, and calculate the TPSR index under the health state based on the true value and the predicted value. The distribution characteristics of the index are used as the health benchmark for evaluating the state of the support.

5. The real-time evaluation method of high-speed railway bridge support state based on BiLSTM and vehicle-induced displacement according to claim 4, characterized in that, The step S4 comprises: Step S41: Input the measured bridge mid-span displacement under the unknown support state to call the benchmark BiLSTM model established in step S22 to predict the left and right quarter-span displacement of the bridge; Step S42: Calculate the TPSR evaluation index based on the true value and the predicted value of the left and right quarter-span displacement of the bridge; Step S43: Qualitatively judge the damage type of the support and quantitatively evaluate the damage degree according to the change trend and absolute value of the TPSR index distribution relative to the health benchmark.

6. The real-time evaluation method of high-speed railway bridge support state based on BiLSTM and vehicle-induced displacement according to claim 5, characterized in that, In step S42, the mathematical expression of the support state evaluation index true-predicted signal energy ratio is: ； wherein, and are the true and predicted values of the bridge displacement, respectively; n is the number of sample points.

7. The real-time evaluation method of high-speed railway bridge support state based on BiLSTM and vehicle-induced displacement according to claim 5, characterized in that, When the TPSR index distribution is overall rightward compared with the healthy benchmark in the step S43, it is determined that the stiffness of the support is reduced; and when the TPSR distribution is overall leftward compared with the healthy benchmark, it is determined that the stiffness of the support is abnormally increased.

8. The real-time evaluation method of high-speed railway bridge support state based on BiLSTM and vehicle-induced displacement according to claim 5, characterized in that, The support damage state D is quantitatively evaluated by the following formula in the step S43: ； wherein, is the mean of the TPSR index generated by the batch train event in the unknown state; is the mean of the TPSR index in the healthy baseline state; f is the index offset is the calibration function between the change in support stiffness.

9. The real-time evaluation method of high-speed railway bridge support state based on BiLSTM and vehicle-induced displacement according to claim 1, characterized in that, The TPSR index calculated in the batch train event in the step S5 is subjected to mean value resampling to obtain a TPSR-m index with smaller variance and more concentrated distribution, so as to improve the identification sensitivity to weak damage; the mean value resampling process is as follows: from an original sample set composed of N single train event TPSR indexes , R times of random sampling with replacement are performed, m samples are extracted each time, and the mean value is calculated, and the mean value is defined as a resampling index TPSR-m; after R times of resampling, a new set composed of R TPSR-m indexes is obtained ; Y R The variance of X N is reduced by 1 / m compared with the variance of X. The mathematical expression of the TPSR-m index is as follows: ； wherein, TPSR-m is the TPSR-m index obtained for the rth resampling; is the ith sample randomly drawn from the set in the rth resampling.

10. The real-time evaluation method of high-speed railway bridge support state based on BiLSTM and vehicle-induced displacement according to claim 1, characterized in that, In the step S6, the left and right support state evaluation indexes are combined into a multi-dimensional joint feature vector, and the joint feature vector is input into the trained support vector machine classification model, and the synchronous evaluation results of the left and right support states are output by the classification model.

Images