Method for automatically detecting free vibration response of high-speed railway bridge for modal identification

The presented invention belongs to the field of structural health monitoring, and relates to a method for automatically detecting a free vibration response of a high-speed railway bridge after a train passing which can be applied in engineering structure modal identification.

In the structural health monitoring of bridges, evaluating the structural overall state of the based on real-time changes in modal parameters is a feasible idea. But the bridges in practice are subject to complex and ever-changing load, extracting the structural modal parameters accurately is difficult. The commonly used modal parameter identification methods, such as the stochastic subspace method, the natural excitation technology combined with the eigensystem realization algorithm, are stable and reliable but require the excitation subject to Gauss stationary white noise characteristics. Some novel time-frequency modal identification methods, such as Hilbert-Huang transform and continuous wavelet transform, can solve the non-stationary excitation to some extent, but there are also some problems in practical application, such as the difficult determination of the algorithm parameters and the unsatisfactory effect of the modal separation. In the previous modal analysis of bridge structures, it is generally analyzed by selecting a long-time vibration response sequence to ensure that the environmental excitation (including wind load, train load, etc.) has asymptotically stationary characteristics. But the longer the vibration response time, the lower the computational efficiency of the identification algorithm. In addition, the change of ambient temperature will cause the change of material properties and restraint internal forces of bridge structures, which will be reflected as the slow time-varying characteristics of structural modes. The modal parameters solved by the long time vibration response sequence can not effectively show the slow time-varying characteristics of the structure. Therefore, a method of modal identification based on the short-time response sequence is needed. An excellent vibration characteristic of high-speed railway bridges is that the short time vibration response after a train passing is a free vibration response with only the dynamic characteristics of the structure retained. Because of the large vibration amplitude caused by the strong impact force of the vehicle load, the high signal-to-noise ratio of the free vibration response after a train passing will be led to. Meanwhile, when the short time vibration response is used for modal identification, the structural characteristics will not change. Thus, the free vibration response data segment can be used to identify modal parameters with high accuracy under low computational burden. However, the forced vibration response of the train passing through the bridge has serious non-stationary characteristics. If the forced vibration components are mixed into the free vibration data segment, the accuracy of modal parameter identification will be reduced. Therefore, it is necessary to detect the free vibration data segment accurately. The traditional detection method based on manual selection is not suitable for long-term real-time extraction of modal parameters. In addition, it is difficult to accurately distinguish the forced vibration responses from the free vibration responses depending on the recorded vehicle information due to the short time of vehicle passing through the bridge. Therefore, it is of great engineering significance to develop a method for automatically detecting the free vibration response data after a train passing.

The objective of the presented invention is to provide an automatic method for detecting the free vibration data segment of the bridge structures, which can solve the problems that the large artificial workload and the unsatisfactory accuracy in modal parameter identification by the free vibration responses of high-speed railway bridges after the train passing.

The technical solution of the present invention is as follows:

A method for automatically detecting the free vibration response segment of the high-speed railway bridge after the train passing is proposed. The test response to be decomposed is pre-selected based on the maximum of the time instants corresponding to the absolute maximum value of the response vector at each measuring point. Then the iterative variational modal decomposition is used to decompose the test response to obtain the single-degree-of-freedom modal response. Further, the instantaneous envelope difference of the modal response is taken as the characteristic of the structural response at each time instant. According to the characteristic that the envelope amplitude of the single-degree-of-freedom free vibration modal response is decreasing, the structural response corresponding to the negative instantaneous envelope difference is marked as the decay vibration and the longest decay vibration sequence is determined as the free vibration response segment. Finally, the determined free vibration response segment is utilized in the eigensystem realization algorithm for modal identification to verify the feasibility of the proposed method.

The procedures of automatically detecting the structural free vibration response data segment are as follows:

The acceleration responses at different measuring points are given as y(t)=[y 1 (t),y 2 (t), . . . ,y s (t)] T ,t=Δt,2Δt, . . . ,NΔt, where Δt is the sampling time interval; N is the number of samples; s is the number of measuring points and the superscript T indicates transposition.

Calculate the time instant corresponding to the absolute maximum value of the response vector at each measuring point i (i=1,2, . . . ,s) as

t i = arg ⁢ ⁢ max Δ ⁢ ⁢ t ≤ t ≤ N ⁢ ⁢ Δ ⁢ ⁢ t ⁢  y i ⁡ ( t )  .

Then the measuring point corresponding to the maximum of the time instants t, (i=1,2, . . . ,s) is obtained as

h = arg ⁢ ⁢ max 1 ≤ i ≤ s ⁢  t i  ,

where ∥ represents the absolute value. Then the response sequence y h (t h ), t=t h ,t h +Δt, . . . ,NΔt is taken as the test response to be decomposed.

Decompose the pre-selected test response by variational mode decomposition with the component number of 2 to obtain the modal response, as follows:

min { y h , q } , { ω q } ⁢ { ∑ q = 1 2 ⁢  ∂ t ⁢ [ ( δ ⁡ ( t ) + j π ⁢ ⁢ t ) * y h , q ⁡ ( t ) ] ⁢ e - i ⁢ ⁢ ω q ⁢ t  2 2 } ⁢ ⁢ s . t . ⁢ ∑ q = 1 2 ⁢ y h , q ⁡ ( t ) = y h ⁡ ( t ) ( 1 )

where δ means the Dirichlet function; j is the imaginary unit; ∂ t represents the gradient function with respect to t; e is the Euler number; π is the circumference ratio; ∥∥ 2 denotes the 2-norm of a vector; * represents the convolution; ω q is the central angular frequency of the component y h,q (t); y h,q (t) is the q(q=1,2) component decomposed from the test response y h (t).

Decompose the test responses y h (t) into two components y h,1 [1] (t) and y h,2 [1] (t). Then calculate the difference of the central angular frequencies as Δω [1] =|Δ 1 [1] −ω 2 [1] |. If the angular frequency difference is greater than 0.01 times of the fundamental frequency of the structure, i.e., Δω [1] >2πf min /100, calculate the component energy as

e q [ 1 ] = ∑ t ⁢ ( y h , q [ 1 ] ⁡ ( t ) ) 2 ,

(q=1,2) and the component y h,{tilde over (q)} [1] (t),

q ~ = arg ⁢ ⁢ max q = 1 , 2 ⁢  e q [ 1 ] 

with higher energy are updated as the test signal to be decomposed. Then the variational mode decomposition is used again to decompose the component y h,{tilde over (q)} [1] (t). The above process will be repeated r times until the angular frequency difference of two components after the r-th decomposition satisfies Δω [r] ≤2πf min /100, which means that the two components are with the identical frequency. The component y h,{tilde over (q)} [1] (t) which has larger energy is deemed as the modal response with the angular frequency ω{tilde over (q)} [r] , labelled as x(t)≡ {tilde over (q)} [r] (t).

Extend the modal response x(t) as x e (t), and the Hilbert transform is performed on the continuation modal response x e (t) to obtain its envelope amplitude a e (t) as:

The envelope a(t) corresponding to the modal response x(t) is intercepted from the envelope a e (t) of the continuation modal response x e (t), and the instantaneous amplitude difference is calculated as Δa(t)=a(t+1)−a(t). According to the characteristic that the envelope amplitude of the free vibration modal response is decreasing, the vibration feature corresponding to the time instants of Δa(t)≤0 is marked as “decay vibration” and expressed in 1. Otherwise, if Δa(t)>0, the vibration feature is marked as “non-decay vibration” and expressed in 0. Thus, the vibration feature from the selected time instants t=[t h ,t h +Δt, . . . ,NΔt] will be labelled as a set of 0 or 1. Choose the time instants [t h +wΔt,t h +(w+1)Δt, . . . ,t h +κΔt] corresponding to the longest vibration feature sequence which are continuously marked as 1. And extract the structural vibration responses y(t), t=t h +wΔt, . . . ,t h +κΔt as the free vibration response segment {tilde over (y)}(k), k=1,2, . . . κ−w+1. A simple example as: For the sequence [0,1,1,0,1,0,0,1,1,1,1,1], the longest sub-sequence [1,1,1,1,1] appears at the position of the w=8 element to the κ=12 element which are in correspondence with the time instants t=[t h +8Δt, . . . ,t h +12Δt]. Then the free vibration responses are y(t), t=t h +8Δt, . . . ,t h +12Δt.

Eigensystem realization algorithm with data correlation is used to identify modal parameters. First, construct the Hankel matrix H(k) by the obtained free vibration responses {tilde over (y)}(k) as

Then build the correlation function matrix S=H(l)(0) T by the Hankel matrix. Implement eigensystem realization algorithm for the correlation function matrix to solve the modal parameters, including the structural frequency, the damping ratio and the mode shape vector.

The advantage of the invention is that the free vibration data segment is separated from the vibration responses of the high-speed railway bridges by means of the amplitude decreasing characteristic of the modal response, which can calculate modal parameters with low computational burden and high precision. Meanwhile, to avoid introducing the empirical parameters into the free vibration detection which will hinder the real-time modal analysis, the iterative variational mode decomposition is adopted to extract modal responses, so as to automatically detect the free vibration segment.

FIG. 1 presents the numerical example model.

FIG. 2 shows the extraction process of modal responses by iterative mode decomposition.

FIG. 3 presents the detected free vibration data segment.

The present invention is further described below in combination with the technical solution.

The numerical example of a simply supported beam model is employed, as shown in FIG. 1 . The length of each element is 10 m. Based on the idea of static condensation, only the vertical displacement of the structure is considered and the torsional displacement is ignored. The element stiffness after static condensation is k 1 =k 2 =k 3 =k 4 =k 5 =k 6 =100 N/m. The mass is m 1 =m 2 =m 3 =m 4 =m 5 =1.5 kg. Rayleigh damping is adopted with the mass matrix coefficient α=0.0446 and the stiffness matrix coefficient β=0.0013. The constant loads F 1 (t)=F 2 (t)=2×10 3 with the spatial interval of 0.05 m move from the left to the right of the beam at a speed of 1 m/s. The 175 s acceleration responses at each node position are collected at a sampling frequency of 20 Hz.

The procedures are described as follows:

The acceleration responses collected at each measuring point is given as y(t)=[y 1 (t),y 2 (t), . . . ,y s (t)] T , where the sampling time interval is Δt=1/f s =0.05; the number of samples is N=3501; the number of measuring points is s=5.

Calculate the time instant corresponding to the absolute maximum of the response vector at each measuring point i (i=1,2, . . . ,s) as

t i = arg ⁢ ⁢ max Δ ⁢ ⁢ t ≤ t ≤ N ⁢ ⁢ Δ ⁢ ⁢ t ⁢  y i ⁡ ( t )  .

Then the measuring point corresponding to the maximum of the time instants t i , i=1,2, . . . ,5, is obtained by

h = arg ⁢ ⁢ max 1 ≤ i ≤ s ⁢  t i 

as h=1 with t h =t 1 =1022 Δt=51.1. Then the response sequence y 1 (t),t=t 1 , . . . ,NΔt is taken as the test response to be decomposed.

The test response y 1 (t) is decomposed by the variational mode decomposition in Eq. (1) where the number of components is fixed as 2. After the first decomposition, two components y 1,1 [1] (t) and y 1,2 [1] (t) with the central angular frequencies ω 1 [1] and ω 2 [1] are obtained, as shown in FIG. 2 . The fundamental frequency is determined as f min =0.6727 Hz in accordance with the power spectrum of the vibration responses. Since Δω [1] >2πf min /100, the energy of each component is calculated as

e 1 [ 1 ] = ∑ t ⁢ ( y 1 , 1 [ 1 ] ⁡ ( t ) ) 2 ⁢ ⁢ and ⁢ ⁢ e 2 [ 1 ] = ∑ t ⁢ ( y 1 , 2 [ 1 ] ⁡ ( t ) ) 2 .

Since e 1 [1] >e 2 [1] , the component y 1,1 [1] (t) is considered as the new test signal to be decomposed by the variational mode decomposition. Repeat the above process until the angular frequency difference Δω [4] =|ω 1 [4] −ω 2 [4] |<2πf min /100. Since e 1 [4] >e 2 [4] , the component y 1,1 [4] (t) is selected as the modal response, which will be labelled as x(t)≡y 1,1 [4] (t).

Extend the modal response x(t) to obtain the continuation signal x e (t) to avoid the edge effect in the subsequent transform. Then the Hilbert transform is used to obtain the envelope amplitude of the continuation signal x e (t) as a e (t), as Eq. (2).

The envelope a(t) corresponding to the modal response x(t) is intercepted from the envelope a e (t) of the continuation modal response x e (t), and the instantaneous amplitude difference is calculated as Δa(t)=a(t+1)−a(t). According to the characteristic that the envelope amplitude of the free vibration modal response is decreasing, the vibration feature corresponding to the time instants of Δa(t)≤0 is marked as “decay vibration” and expressed in 1. Otherwise, if Δa(t)>0, the vibration feature is marked as “non-decay vibration” and expressed in 0. Thus, the vibration feature from the selected time instants t=[t h ,t h +Δt, . . . ,NΔt] will be labelled as a set of 0 or 1. Choose the time instants [t h +wΔt,t h +(w+1)Δt, . . . ,t h +κΔt] corresponding to the longest vibration feature sequence which are continuously marked as 1. And extract the structural vibration responses y(t), t=t h +wΔt, . . . ,t h +κΔt as the free vibration response segment {tilde over (y)}(k), k=1,2, . . . κ−w+1, as shown in FIG. 3 .

The eigensystem realization algorithm with data correlation is performed on the detected free vibration response segment. First, the obtained free vibration responses are utilized to construct the Hankel matrix H(k) as Eq. (3). Set k=l=30, g=N b −2 l+1, and build the correlation function matrix as S=H(l)H(0) T . Implement eigensystem realization algorithm for the correlation function matrix to solve the first 5 modal parameters of the structure, which are consistent with the numerical solution. The identified frequencies are f 1 =0.6727 Hz, f 2 =1.2995 Hz, f 3 =1.8378 Hz, f 4 =2.2508 Hz, f 5 =2.5104 Hz and the identified damping ratios are d 1 =0.8000%, d 2 =0.8000%, d 3 =0.9384%, d 4 =1.0706%, d 5 =1.1596%. As a comparison, the vibration responses corresponding to the time instants t=t h , . . . ,NΔt are implemented into the eigensystem realization algorithm to solve the first 5 modal parameters, which are a little dissimilarity with the numerical solution. The estimated frequencies are f′ 1 =0.6712 Hz, f′ 2 =1.2963 Hz, f′ 3 =1.8413 Hz, f═ 4 =2.2513 Hz, f′ 5 =2.5132 Hz. The estimated damping ratios are d′ 1 =0.7621%, d′ 2 =2.1911%, d′ 3 =0.6523%, d′ 4 =0.8205%, d′ 5 =1.3894%.