Load effect separation method for special bridge for track

Abstract

The invention discloses a load effect separation method for a special bridge for a track. The method comprises the following specific steps: 1, identifying train passing stages; 2, separating train load and temperature load effect data by adopting a nonlinear model; 3, removing data of a temperature load effect by adopting a self-adaptive algorithm with a memory effect; 4, processing the monitoring data by adopting a multi-scale singular value decomposition (MS-SVD) method; 5, performing optimal estimation on temperature and train load effect data separation by using a joint distribution model based on a probability theory; and 6, describing the time dynamic change of the train load and the temperature load through the state transition matrix. According to the method, the accuracy of the monitoring data is improved, the train load effect and the temperature load effect in the bridge health monitoring data are effectively distinguished through a self-adaptive data separation technology, interference of slow change factors such as a degradation effect is avoided, and the accuracy of a bridge structure health assessment result is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of bridge health monitoring, and particularly to a method for separating load effects of a special railway bridge. Background Art

[0002] In a bridge health monitoring system, the structure response information obtained in real time usually contains the superposition of various load effects, mainly including train load effects, temperature load effects, and deterioration effects, etc. These response data reflect the overall performance of the bridge structure under different conditions. Deterioration effects, such as concrete shrinkage and creep, steel corrosion, etc., are a slow and continuous process. Although they have a long-term impact on the bridge health condition, the changes are relatively small in the short term, so they can usually be ignored. In contrast, there are significant differences in signal frequency and amplitude between train load effects and temperature load effects. Train load effects are manifested as high-frequency, short-time peak amplitudes, which result from the instantaneous forces when the train passes through the bridge; while temperature load effects gradually accumulate with the change of external climate conditions, presenting a low-frequency, long-term and relatively stable response. This difference in frequency and amplitude provides the possibility for separating load effects.

[0003] However, there are still certain limitations in the prior art when dealing with the separation of train load and temperature load effects. Due to the complexity of the bridge structure and the possible noise and abnormal fluctuations in the monitoring data, it is difficult to achieve accurate separation of load effects only through the distinction based on response characteristics, thus affecting the accuracy of data analysis. Especially in actual operation, environmental interference and data missing further increase the difficulty of separation. Therefore, how to accurately distinguish train load effects from temperature load effects under complex bridge load conditions has become a key issue in bridge structure health assessment and early damage detection. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention provides a method for separating load effects of a special railway bridge, which can accurately separate train load effects and temperature load effects from the structure response information obtained in real time by a bridge health monitoring system, so as to improve the accuracy of bridge structure health assessment and the ability of early damage detection. The specific technical solutions are as follows:

[0005] A method for separating load effects of a special railway bridge, characterized in that the specific steps are as follows:

[0006] Step 1: Identify the stage when the train passes by, and mark the data affected by the train load on the bridge;

[0007] Step 2: Use a non-linear model to separate the train load and temperature load effect data;

[0008] Step 3: Use an adaptive algorithm with a memory effect to remove the data of the temperature load effect;

[0009] Step 4: Use the multi-scale singular value decomposition (MS-SVD) method to process the monitoring data and improve the separation effect;

[0010] Step 5: Use a joint distribution model based on probability theory to perform an optimal estimation of the separation of the temperature and train load effect data;

[0011] Step 6: Describe the time dynamic changes of the train load and temperature load through the state transition matrix, predict the response of the future bridge under different load conditions, and provide long-term health monitoring data support.

[0012] As an optimization: Specifically, in Step 1, before data processing and load effect separation, first identify the time period when the train passes by detecting the change in displacement data within a set time. Identify the time period when the train passes by the detection method of the displacement change amount in each set time. When the change amount of displacement within a set time exceeds the preset threshold ΔD, it is considered that the train has passed. After marking the data affected by the train load, separate the train load and temperature load effects based on the difference in response characteristics, and further calculate the change amount of displacement within each set time:

[0013] C(t) = D end - D start

[0014] where C(t) is the change per unit time, D end is the monitoring data at the end of this unit time, D start is the monitoring data at the start of this unit time. Define a change amount threshold ΔD. When the absolute value of the displacement change amount |C(t)| ≥ ΔD within a set time, it is considered that the train has passed.

[0015] As an optimization: Specifically, in Step 2, for the superposition of the train load effect and the temperature load effect, use a non-linear model. As a combination of multi-variable functions, the monitoring data is expressed as:

[0016] y(t) = g(T(t), L(t)) + h(T(t), L(t))

[0017] where g(T(t), L(t)) is a non-linear function representing the coupling effect between temperature and train load, and is expressed using a quadratic polynomial:

[0018] g(T(t), L(t)) = a 0 + a 1 T(t) + a 2 L(t) + a 3 T(t)·L(t)

[0019] Among them, a 0 ~a 3 is a set of coefficients representing the influence on the objective function; a 0 : constant term, representing the base value or offset of the function g(T(t), L(t)) when both the temperature T(t) and the load L(t) are zero; a 1 : coefficient of the temperature T(t), representing the degree of influence of temperature on g(T(t), L(t)); a 2 : coefficient of the load L(t), representing the degree of influence of the load on g(T(t), L(t)); a 3 : interaction coefficient of the temperature and the load T(t)·L(t), representing the influence of the interaction effect between the temperature and the load on g(T(t), L(t));

[0020] And h(T(t), L(t)) represents the higher-order non-linear relationship between the two, adopting the form of a cubic polynomial function, specifically:

[0021] h(T(t), L(t)) = b 0 +b 1 T(t) 2 +b 2 L(t) 3

[0022] Among them, b 0 ~b 2 is a set of coefficients representing the influence on the objective function; b 0 : constant term, representing the base value or offset of the function h(T(t), L(t)) when both the temperature T(t) and the load L(t) are zero; b 1 : coefficient related to the square, representing the degree of influence of the temperature square term on h(T(t), L(t)); b 2 : coefficient related to the cube of the load L(t), representing the degree of influence of the load on h(T(t), L(t)).

[0023] As an optimization: specifically, in the process of separating the load effect, an algorithm with a memory effect is introduced to dynamically adjust the parameters, and a historical data weight decay factor λ is defined to update the calculation parameters of the load response, so as to gradually weaken the influence of past data on the current response;

[0024] The specific algorithm is:

[0025]

[0026] Among them, the decay factor λ ∈ (0, 1) controls the decay rate of historical data; w(t) is the parameter vector (or weight vector) at time t; the learning rate η is used to adjust the update step to ensure that the model reaches the optimal state during the dynamic adaptation process; At time t, the gradient of the loss function E(w(t)) with respect to the parameter vector w(t);

[0027] The calculation formula of the mean square error E(w) is as follows:

[0028]

[0029] Among them, N represents the number of monitoring data, which is the total number of observed data points; y i is the true data or observed value of the i-th monitoring point; w is the parameter vector of the model, which can be the weight coefficient and reflects the influence degree of different factors in the model on the result; x i is the input feature vector of the i-th observed data, which is the sensor reading or environmental variable related to this data point; w T x i is the model prediction value, which is obtained by linearly combining the parameter w and the input feature x i as a result.

[0030] As an optimization: specifically, in step four, to improve the separation effect, the multi-scale singular value decomposition (MS-SVD) method is adopted to decompose the data matrix Y at different time scales. Suppose the data matrix is divided into multiple sub-matrices Y i by time, then each sub-matrix is subjected to singular value decomposition:

[0031] Y i = U i Σ i V ii T

[0032] Among them, Y i represents the original data matrix or system output matrix, representing the i-th sample or observed data; U i represents the left singular vector matrix, representing the spatial pattern of the data; ∑ i represents the singular value matrix, representing the main components or weights of the features of the data; V i T represents the transpose of the right singular vector matrix, representing the time or direction pattern of the data;

[0033] Then, the singular value matrices ∑ i of all sub-matrices are combined to obtain signal components at different time scales, and more refined separation is achieved through threshold screening.

[0034] As an optimization: Step five is specifically to further process noise and uncertainty. Using a joint distribution model based on probability theory, assuming that the randomness of the temperature load effect and the train load effect is represented as a joint distribution p(T(t), L(t)), and combining the observation noise ∈, the probability model of the monitoring data is expressed as:

[0035] p(y(t)|T(t), L(t)) = p(y(t) - g(T(t), L(t)) - h(T(t), L(t))|0, σ 2 )

[0036] where y(t) represents the observed value at time t; T(t) represents the temperature value at time t; L(t) represents the load at time t; G(T(T), L(t)) represents a function related to the temperature T(t) and the load L(t), representing the influence of temperature and load on the structure; h(T(t), L(t)) represents another function related to the temperature T(t) and the load L(t), representing other factors, including the influence of noise on the response.

[0037] As an optimization: Step six is specifically to assume that the train load effect L(t) and the temperature load effect T(t) respectively follow their own state transition matrices P L and P T , then the state transition of the entire system can be expressed as:

[0038]

[0039] where represents the Kronecker product. By analyzing the long-term stable state of the Markov process, the response behavior of the future bridge under different load effects can be predicted.

[0040] The beneficial effects of the present invention are as follows:

[0041] 1. Improve the accuracy of monitoring data: By using the adaptive data separation technology, the present invention effectively distinguishes the train load effect and the temperature load effect in the bridge health monitoring data, avoids the interference of slow-changing factors such as degradation effects, and significantly improves the accuracy of the bridge structure health assessment results.

[0042] 2. Enhance the ability to detect early damage: By effectively separating different load effects, the present invention can improve the sensitivity of detecting early damage to the bridge structure, help to timely discover potential problems of the bridge, and prevent the further deterioration of the structure.

[0043] 3. Application of Adaptive Data Processing Technology: The present invention introduces an innovative algorithm with a memory effect, which dynamically adjusts current parameters through the weight decay mechanism of historical data, making it more flexible in the face of complex load conditions. At the same time, combined with the gradient descent method for optimization adjustment, it further improves the accuracy of data processing:

[0044]

[0045] Where λ is the historical data decay coefficient, which enhances the system's memory and processing ability for historical data.

[0046] 4. Strong Stability: The method of the present invention has strong anti-noise ability. By using the method based on multi-scale singular value decomposition (MS-SVD) to decompose and process signals, it further improves the stability when separating different load effects:

[0047] Y i =U i Σ i V ii T

[0048] The decomposition of the sub-matrix Y i helps to analyze the load effect on different time scales, ensuring the stable operation of the system.

[0049] 5. Wide Applicability: This method is especially suitable for the health monitoring of rail transit bridges, and can finely analyze the train load and temperature load effects in complex environments, with high generality. Brief Description of the Drawings

[0050] Figure 1 is the measured value of the vertical displacement of the main girder in the embodiment of the present invention.

[0051] Figure 2 is the data of the change in the vertical displacement of the main girder under the train load in the embodiment of the present invention.

[0052] Figure 3 is the vertical displacement of the main girder under the temperature load in the embodiment of the present invention.

[0053] Figure 4 is the flow chart of the present invention. Detailed Description of the Preferred Embodiment

[0054] The following elaborates on the preferred embodiments of the present invention in detail with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making the protection scope of the present invention more clearly defined.

[0055] As Figures 1 to 4As shown in the figure: In the working conditions of this embodiment, the main bridge of this bridge adopts a semi-floating system cable-stayed bridge with five spans of high and low towers and double cable planes. The longitudinal layout is 34.5 + 180.5 + 480 + 215.5 + 94.5 = 1005.0 m, and the cross-sectional layout is: 2.8 m (cable area and inspection path) + 16.2 m (track area) + 2.8 m (cable area and inspection path) = 21.8 m (excluding fairings). The main beam adopts a steel-concrete composite beam type, with a beam height of 3.3 m. The main structure of the steel beam is made of Q345qD, and the precast bridge deck is made of C60 concrete. The main tower adopts a portal frame structure composed of three upper, middle and lower cross beams. The height of Tower P2 is 158 m, and the height of Tower P3 is 227 m. The material is C50 concrete. There are 16 pairs of stay cables on the P2 tower side and 27 pairs of stay cables on the P3 tower side. The stay cables adopt finished cables with Φ7.0 mm galvanized high-strength low-relaxation parallel steel wires and HDPE sheaths, with a tensile strength of not less than 1770 MPa and a tensile elastic modulus of not less than 195 GPa. Figure 1 The original monitoring data is presented, showing the vertical displacement response of the main beam of a certain bridge under different load conditions. It can be seen that the data simultaneously includes the superposition of train load effects and temperature load effects, resulting in an increase in the complexity of the response signal and making it impossible to clearly distinguish each load effect.

[0056] A method for separating load effects of a special bridge for tracks, the specific steps are as follows:

[0057] S1. Collection and processing of the vertical displacement data of the main beam. Taking the vertical displacement data of the main beam of a certain bridge as an example, the system collects the response data of the main beam under different environmental and load conditions through the data collected on November 1, 2023 (08:00 - 20:00). First, the displacement data collected by the sensor is segmented and calculated according to each minute time period, and the displacement difference at the beginning and end of each minute is calculated. When the absolute value of this difference is greater than or equal to ΔD, the system marks this time period as the time period when the train passes. Subsequently, the marked time periods are used to separate and analyze the displacement data, so as to screen out the data time periods affected by the train load.

[0058] S2. Preliminary determination of the train load effect: According to the collected displacement data, for the superposition of the train load effect and the temperature load effect, a non-linear model is adopted and regarded as a combination of multi-variable functions. The monitoring data can be expressed as:

[0059] y(t) = g(T(t), L(t)) + h(T(t), L(t))

[0060] Among them, g(T(t), L(t)) is a non-linear function representing the coupling effect between temperature and train load effects, which can be expressed by a quadratic polynomial, such as:

[0061] g(T(t), L(t)) = a 0 + a 1T(t) + a 2 L(t) + a 3 T(t)·L(t)

[0062] The h(T(t), L(t)) represents the higher-order non-linear relationship between the two, and a cubic polynomial or a more complex functional form can be adopted. Specifically:

[0063] h(T(t), L(t)) = b 0 + b 1 T(t) 2 + b 2 L(t) 3

[0064] The non-linear model captures the complex relationship between temperature and train load effects better through the above quadratic and cubic polynomial forms. Through this model, the system can accurately separate the data that may be affected by train loads under different load conditions, thus providing more accurate support for subsequent load effect analysis.

[0065] S4. Process the monitoring data through the multi-scale singular value decomposition (MS-SVD) method. Decompose the data matrix Y i at different time scales, and screen out the signal components of different load effects through singular value analysis to ensure the fineness and stability of the final separation effect.

[0066] S5. To address the problems of noise and data uncertainty, a probability theory model is adopted to describe the load effects. By maximizing the joint probability distribution, the separation effect of train and temperature load effects can be optimized, thus improving the reliability of data analysis.

[0067] S6. Finally, describe the time dynamic changes of train loads and temperature loads through the state transition matrix. The system state transition is: This model can effectively predict the responses of the future bridge under different load conditions and provide long-term health monitoring data support.

[0068] Taking the vertical displacement data of the bridge main girder in this embodiment as an example, the system collects the response data of the main girder under different environmental and load conditions through the data from November 1, 2023 (8:00 - 20:00), as Figure 1 shown. According to the foregoing method, the following processing is performed on this data:

[0069] 1. Preliminary determination of train load effects: According to the collected displacement data, use the non-linear model in step 2 to preliminarily determine the train load effects and screen out the data that may be affected by train loads.

[0070] 2. Finite element model calculation verification: Use the finite element model of the bridge to calculate the theoretical response under the action of train loads, and confirm that the calculation results are consistent with the monitoring data. As Figure 2 shown, after adopting the nonlinear load effect separation method, the train load effect data extracted. In data processing, we used a nonlinear model based on temperature and train loads and successfully separated the high-frequency short-time vibration response caused by train loads. The characteristics of the train load effect are high-frequency instantaneous impacts, and this part of the data reflects the vertical displacement response of the main girder when the train passes by.

[0071] 3. Removal of temperature load effect: By introducing an adaptive algorithm to process the train load effect data and optimizing it in combination with historical data, the interference of the temperature load effect can be effectively removed, thus realizing the effective separation of load effects. This process ensures that the impact of train loads on bridge health monitoring data can be accurately identified and analyzed, thereby improving the accuracy and reliability of bridge structure health assessment.

[0072] 4. Display of separation effect: After data processing, the temperature load effect is separated, and finally the corrected bridge vertical displacement data is obtained. As Figure 3 shown, the temperature load effect data separated from the original data. The temperature load effect has the characteristics of low-frequency and stable changes. As time goes by, the vertical displacement of the main girder of the bridge is gradually affected by temperature changes. During the data separation process, this long-term changing temperature response is extracted through the temperature load model.

Claims

1. A method for separating load effects of a dedicated rail bridge, characterized in that: The specific steps are: Step 1: Identify the train passing stage and mark the bridge train load impact data; Step 2: Separate the train load and temperature load effect data using a nonlinear model; Step 3: Use an adaptive algorithm with memory effect to remove the data of temperature load effect; Step 4: Use the multi-scale singular value decomposition (MS-SVD) method to process the monitoring data to improve the separation effect; Step 5: Use the joint distribution model based on probability theory to optimally estimate the temperature and train load effect data separation; Step 6: Describe the temporal dynamic changes of train load and temperature load through the state transfer matrix, predict the future response of the bridge under different load conditions, and provide long-term health monitoring data support.

2. The method for separating the load effect of a dedicated rail bridge according to claim 1, characterized in that: Specifically, the step 1 is as follows: before data processing and load effect separation, the time period during which the train passes is first identified by detecting the change in displacement data within a set time, and the time period during which the train passes is identified by detecting the displacement change amount within each set time. When the displacement change amount within a set time exceeds a preset threshold value ΔD, it is considered that the train has passed; after marking the data affected by the train load, the train load and temperature load effects are separated based on the difference in response characteristics, and the displacement change amount within each set time is further calculated: C(t)=D end -D start Among them, C(t) is the change per unit time, D end is the monitoring data at the end of this unit time, D start For the monitoring data at the beginning of this unit time, a change threshold ΔD is defined. When the absolute value of the displacement change within a set time |C(t)|≥ΔD, it is considered that the train has passed.

3. The method for separating the load effect of a dedicated rail bridge according to claim 1, characterized in that: Specifically, the second step is to use a nonlinear model as a combination of multivariable functions for the superposition of train load effect and temperature load effect, and the monitoring data is expressed as: y(t)=g(T(t),L(t))+h(T(t),L(t)) Among them, g(T(t),L(t)) is a nonlinear function representing the coupling between temperature and train load effect, which is expressed by a quadratic polynomial: g(T(t),L(t))=a0+a1T(t)+a2L(t)+a3T(t)·L(t) Among them, a0~a3 are a set of coefficients, which represent the influence on the objective function; a0: constant term, which represents the base value or offset of the function g(T(t), L(t)) when both temperature T(t) and load L(t) are zero; a1: coefficient of temperature T(t), which represents the influence of temperature on g(T(t), L(t)); a2: coefficient of load L(t), which represents the influence of load on g(T(t), L(t)); a3: interaction coefficient of temperature and load T(t)·L(t), which represents the influence of the interaction effect between temperature and load on g(T(t), L(t)); h(T(t), L(t)) represents the high-order nonlinear relationship between the two, in the form of a cubic polynomial function, specifically: h(T(t),L(t))=b0+b1T(t) 2 +b2L(t) 3 Among them, b0~b2 are a set of coefficients, which represent the influence on the objective function; b0: constant term, which represents the base value or offset of the function h(T(t), L(t)) when both the temperature T(t) and the load L(t) are zero; b1: square correlation coefficient, which represents the influence of the square temperature term on h(T(t), L(t)); b2: cubic correlation coefficient of the load L(t), which represents the influence of the load on h(T(t), L(t)).

4. The method for separating the load effect of a dedicated rail bridge according to claim 1, characterized in that: Specifically, the step three is to introduce an algorithm with a memory effect in the process of load effect separation, dynamically adjust parameters, and define a historical data weight attenuation factor λ to update the calculation parameters of the load response so as to gradually reduce the influence of past data on the current response; The specific algorithm is: Among them, the decay factor λ∈(0,1) controls the decay speed of historical data; w(t) is the parameter vector (or weight vector) at time t; the learning rate η is used to adjust the update step to ensure that the model reaches the optimal state during the dynamic adaptation process; At time t, the gradient of the loss function E(w(t)) with respect to the parameter vector w(t); The calculation formula of mean square error E(w) is as follows: Where N represents the number of monitoring data, which is the total number of observed data points; y i is the real data or observation value of the i-th monitoring point; w is the parameter vector of the model, which can be a weight coefficient, reflecting the influence of different factors in the model on the results; x i is the input feature vector of the i-th observation data, which is the sensor reading or environmental variable associated with the data point; w T x i To predict the value of the model, we combine the parameter w with the input feature x i The result of linear combination.

5. The method for separating the load effect of a dedicated rail bridge according to claim 1, characterized in that: Specifically, in order to improve the separation effect, the multi-scale singular value decomposition (MS-SVD) method is used to decompose the data matrix Y at different time scales. The data matrix is ​​divided into multiple sub-matrices Y according to time. i , then each submatrix is ​​subjected to singular value decomposition: Y i =U i ∑ i V i T Among them, Y i represents the original data matrix or system output matrix, which represents the data of the i-th sample or observation; U i represents the left singular vector matrix, representing the spatial pattern of the data; ∑ i represents the singular value matrix, which represents the weights of the main components or features of the data; V i T represents the transpose of the right singular vector matrix, representing the temporal or directional pattern of the data; Then for all submatrices, the singular value matrix ∑ i By combining them, signal components at different time scales are obtained, and threshold screening is used to achieve finer separation.

6. The method for separating the load effect of a dedicated rail bridge according to claim 1, characterized in that: Specifically, step five further processes noise and uncertainty, uses a joint distribution model based on probability theory, assumes that the randomness of the temperature load effect and the train load effect is expressed as a joint distribution p(T(t), L(t)), and combines the observation noise ∈, then the probability model of the monitoring data is expressed as: p(y(t)|T(t),L(t))=p(y(t)-g(T(t),L(t))-h(T(t),L(t))|0,σ 2 ) Among them, y(t): represents the observed value at time t; T(t): represents the temperature value at time t; L(t): represents the load at time t; g(T(t), L(t)): represents a function related to temperature T(t) and load L(t), representing the influence of temperature and load on the structure; h(T(t), L(t)): represents another function related to temperature T(t) and load L(t), representing the influence of other factors, including noise on the response.

7. The method for separating the load effect of a dedicated rail bridge according to claim 1, characterized in that: Specifically, the step 6 is: assuming that the train load effect L(t) and the temperature load effect T(t) follow their respective state transfer matrices P L and P T , then the state transition of the entire system can be expressed as: in represents the Kronecker product. By analyzing the long-term stable state of the Markov process, the response behavior of the future bridge under different load effects can be predicted.

Images