2.1. Selection of Excitation Signals
An appropriate excitation signal is an important guarantee for successful damage detection. A guided wave propagating along a hollow cylindrical shell can be divided into the longitudinal mode (L mode), torsional mode (T mode), and flexural mode (F mode), respectively. The decomposed modes of the guided waves can be controlled by a special dispersion equation which can be obtained from the Navior dynamic equations [
19]. The derived dispersion equation is always a transcendental equation which can only be solved by a numerical method [
22]. According to the material and dimension conditions of the pipe, the frequency dispersion curves can be obtained from a software package, named PCDISP, written in the Matlab environment, and freely available to be adapted to particular circumstances [
37], as shown in
Figure 2. The curves shown in
Figure 2 are only the part from zero to 100 kHz for the guided waves propagating along the steel pipe with an outside diameter of 70 mm and wall thickness of 4 mm. It is obviously observed that the guided waves have typical properties of multiple modes and frequency dispersion effects.
To simplify the process of damage identification and increase the identification result precision, an L(0, 2) mode guided wave with the central frequency of 70 kHz is selected as the excitation signals. This is because (1) L mode guided waves are usually sensitive to circumferential cracks of pipes; (2) the group velocity of L(0, 2) mode has the maximum value near 70 kHz, as shown in
Figure 2a, having a chance to firstly reach and be detected by sensors, which will significantly lower the identification difficulties; and (3) the selected mode in the range has a relatively weak frequency dispersion effect which is beneficial for the velocity evaluation during a damage locating process by using the time-of-flight (TOF) method. The three issues mentioned here are suggested in the paper as the rule for the guided waves selection in pipe structure damage identifications.
A five-peak impulse wave has some advantages such as a narrow pulse width and less frequency dispersion. Therefore, an L(0, 2) mode five-peaks impulse wave with the center frequency of 70 kHz and the amplitude of 5 V, which is modulated by a HANNING window function, is chosen as the detection signal in this paper. The excitation signals in the time domain and frequency domain are shown in
Figure 3, respectively.
Figure 3a shows the normalized waveform in the time domain, and
Figure 3b represents the corresponding wave in the frequency domain.
2.3. Selection of Analysis Setup and Time Increment
The analysis step and time increment have a great impact on the numerical results using FEM. For the issue of the guided wave-based pipe structural numerical simulation, the selection of the analysis step and time increment depends not only on the properties of the structure and damages, but also on the detection signal sampling interval and frequency. As for a larger analysis step and time increment, although the FEA calculation is time-saving, the selected detection signal might be insensitive to relative tiny damages, indicating an unexpected or even a wrong evaluation result. As for a much smaller analysis step and time increment, although it is efficient for tiny damages identification, a higher time consumption might increase the cost of applying the SHM technology in engineering. Therefore, as the dynamic implicit analysis step of the ABAQUS/Standard module is being selected to analyze the issue of pipe structure damage identification coupled with piezoceramic patches, time step and increment are selected based on prior published work and study to ensure convergence. According to the Newmark time increment scheme [
38], the maximum time increment should be less than 1/20 of the period of the excitation signal corresponding to the highest frequency, as shown in Equation (1)
where ∆
t means the time increment and
fmax represents the highest frequency of the excitation signal. In this paper, the highest frequency of the excitation signal is 70 kHz, so it can be obtained that the time increment should be less than 0.7 µm when employing Equation (1). Considering the accuracy of the analytical results, 0.1 µm is chosen as the time increment.
The output settings in ABAQUS are mainly composed of field output settings and history output settings. The simulation analysis results in the paper are selected from time history curves by extracting voltage and displacement signals in the corresponding nodes in the model.
2.5. Simulation Result Analysis
Generally, the simulations are divided into two cases according to whether damages occur or not. One case is a healthy one, which has no damages along the pipe. The other is a damaged one, where artificial damages with a certain level are placed on the given positions. The electrical analog signals for the healthy pipe structure model extracted from a piezoelectric receiving element potential are shown in
Figure 5. In
Figure 5, the signal packages from left to right are the received initial signal and the boundary reflection signals, respectively. It can be obtained from the result that as a signal is propagating in a healthy pipe structure, the guided wave for the same excitation central frequency will be observed for a phenomenon of decreasing amplitudes because of the energy dissipation with an increase of propagation distances and reflection times, accompanying the phenomenon of an increasing signal package length in the time domain. For each of the received signal packages, a more than five-peak impulse is clearly observed due to the boundary superposition phenomenon. This is because the PZT sensing array is so close (50 mm) to the end of the pipe structure that the later part of the sensor signal might have a superposition with the boundary reflected front part of the same sensor signal.
In order to verify whether the excitation signal is proximate to the expected guided wave with an L(0, 2) mode in the pipe detection simulation, three kinds of typical displacements along axial, radial, and circumferential directions at the given nodes are analyzed, respectively. Firstly, the axial, radial, and circumferential displacement curves are extracted and compared, as shown in
Figure 6.
Figure 6 shows that axial displacement amplitudes of the signal package at the same node in the pipe model are greater than the other two directional ones, indicating that the displacement distribution approximately complies with the L mode guided wave.
Secondly, by calculating the time difference between the initial received signal wave package and the echo signal wave one, the propagation time of the guided waves in the pipe can be calculated, and according to the position of the nodes in the pipe, the propagation group velocities of the guided waves in the pipe can be obtained, shown in
Table 5.
It can be observed that the calculated group velocities values are basically the same as the theoretical value of the L(0, 2) mode guided wave at the center frequency of 70 kHz shown in
Figure 2a. It is indirectly verified that the excitation signal in this simulation has approximate properties of the L(0, 2) mode guided wave.
Meanwhile, according to the calculated values of the wave group velocities shown in
Table 5, the group velocities of the guided wave gradually decrease for a small certain extent (from 0.149% to 3.619%) with an increase of the signal reflection time and propagation distances, which also shows that there is an energy attenuation during guided wave propagation.
As the detection signal is propagating along the pipe and reflecting at the ends, the echo signal usually becomes more complicated. By analyzing the displacement time history at given nodes for the healthy pipe, it is found that the initial received signal and the echo one at the end have a phenomenon of a waveform superposition and a signal package time width broadening. It is preliminarily speculated that the superposition of the received signal occurs at areas near the ends, and signal package time width broadening comes from the frequency dispersion in the propagation process. In order to clearly explain the phenomenon, the displacement time history responses at the nodes with distances away from the excitation end of 0 mm, 1000 mm, and 2000 mm are extracted and analyzed, respectively, as shown in
Figure 7.
Figure 7a shows the layout of three observation points located at 0 mm, 1000 mm, and 2000 mm from the left end.
Figure 7b indicates the observed signal at the point of 0 mm. The first signal package represents the initial excitation signal and the second one is the received echo signal from the right end.
Figure 7c shows the signal propagating through the point of 1000 mm. The first and third main signal packages represent the initial excitation signal and the reflection one from the right end, respectively, and the second main one represents that from the left end echo.
Figure 7d denotes the signal travelling through the point of 2000 mm. The first main signal package is the initial excitation signal and the second main one is from the left end reecho. Through the displacement comparison at difference nodes shown in
Figure 7, it can be concluded that the incident wave and boundary reflection ones will occur as a superposition phenomenon in the close end (connection) area.
In order to determine the superposition area range, a new observation point of a signal wavelength away from the left end is selected. The node displacement of the initial received signal and its reflected signal starts clearly separating, as shown in
Figure 8.
The identification of the signal separation range due to the boundary reflection superposition is important for the layout of PZT transducer arrangement. Appropriately arranging the transducer array to protect the signal from superposition may not only make installing PZT arrays much easier, but also increase the damage identification precision. In
Figure 8a, the distance between observation nodes when the initial signal is separated and the connection (or end) place of the pipe is 305 mm. In
Figure 8b, the similar distance for the echo signal is 415 mm. It is found that the distance for signal packages starting separation is closely associated with the signal wavelength. Equation
shows how to calculate the wavelength of an excitation signal, where
λ means the signal wavelength,
cL represents the longitudinal wave velocity in elastic medium, and
T denotes the period for the excitation signal.
From the above Equation, the distance between the node when the initial signal is separated and the connection (or end) of the pipe is 0.8λ, which means that the initial wave is prone to superposition within the range of 0.8λ away from the pipe connection (or end). Meanwhile, it can be also obtained that the echo wave is prone to superposition within the range of 1.1λ away from the pipe connection. Therefore, to avoid the superposition of the incident signal at the connection of pipe structures, the distance between the signal transducer and the connection (or end) of the pipe should be greater than the wavelength of the exciting signal.
According to above analysis results, in order to avoid superposition phenomena in the connection of a pipe, excited piezoceramic elements are arranged at a distance of 20 mm from the connection of the pipe, and the received piezoceramic elements are placed at a distance of 300 mm from the connection of the pipe, and the layout of the piezoelectric elements is shown in
Figure 9.
The received voltage signal for the healthy pipe in accordance with the new piezoelectric elements layout is shown in
Figure 10.
After rearranging the excited elements and the received elements, the initial signal and its boundary reflection signal have been separated for the first and second packages from the left of the pipe, and so are the echo signals for the third and fourth packages. In this way, the difficulty of result analysis due to the superposition of signals is efficiently avoided, which will lay a foundation for the layout of the transducers and the analysis of the subsequent damaged pipes.