† Corresponding author. E-mail:
A promising tool to detect micro-cracks in plate-like structures is used for generating higher harmonic Lamb waves. In this paper, a method combining nonlinear S0 mode Lamb waves with time reversal to locate micro-cracks is presented and verified by numerical simulations. Two different models, the contact acoustic nonlinearity (CAN) model and the Preisach–Mayergoyz (PM) model, are used to simulate a localized damage in a thin plate. Pulse inversion method is employed to extract the second and fourth harmonics from the received signal. Time reversal is performed to compensate the dispersion of S0 mode Lamb waves. Consequently, the higher harmonics generated from the damaged area can be refocused on their source. By investigating the spatial distribution of harmonic wave packets, the location of micro-cracks will be revealed. The numerical simulations indicate that this method gives accurate locations of the damaged area in a plate. Furthermore, the PM model is proved to be a suitable model to simulate the micro-cracks in plates for generation of higher harmonics.
The detection of incipient structural damages is an essential issue in industry because if not accurately detected and dealt with, even micro-cracks may subsequently lead to a critical failure of the structure. When dealing with large-scale damaged structures such as composite plates in aeronautical and aerospace industry, nonlinear Lamb wave inspection is a promising candidate because it combines the long inspection range of guided waves with the sensitivity to even micro-cracks of nonlinear inspection.
Laboratory experiments[1–4] have demonstrated the capability of nonlinear Lamb waves to detect micro-cracks. Both analytical and numerical studies have been conducted to have a deeper insight into the basic mechanism of this method. A first step is the generation of cumulative higher harmonics based on classical nonlinear theory.[5–9] The cumulative effect provides an improved sensitivity to second harmonics whose amplitude is usually very small, which eventually facilitates the detection of micro-cracks such as fatigue damages in metal[1] and impact damages in composite.[3] Additionally, the non-classical nonlinearity has also been considered but mainly from the numerical perspective. The contact analysis is widely used in simulations[3,10,11] because it provides a straightforward description of the wave-damage interaction as well as remarkable higher harmonic generation. The signal analysis has also played an important role in nonlinear Lamb wave inspection to reveal the damage information contained in time-domain signals. In previous works, two methods are mainly involved, the short-time Fourier transform[12,13] and wavelet transform.[14,15] The signal features including amplitude and mode of higher harmonics are extracted to investigate the presence, orientation, and severity of micro-cracks.[10,16–18]
However, certain conditions of the cumulative effect are sometimes hard to fulfill because they are satisfied by limited mode pairs at particular frequencies. Additionally, many studies focus on the propagation characteristics of nonlinear Lamb waves and they are restricted to the second harmonics. Although the existence of a micro-crack can be revealed, its position remains unknown. Moreover, higher harmonics, which contain crucial information of the micro-cracks, are not fully utilized as most of them are omitted in the analysis.
In this paper, the localized damage in a thin plate is detected by the combination of nonlinear S0 Lamb waves and time reversal. A direct illustration of its location is given. The contact acoustic nonlinearity (CAN) model and the Preisach–Mayergoyz (PM) model are both used to model the localized damage. The second and fourth harmonics are extracted from the received signals by pulse inversion (PI). The location of the damaged area is revealed by investigating the distribution of harmonic wave packets after time reversal. Simulation results indicate that this method is feasible and reliable to locate a damaged area in a thin metallic plate.
Lamb waves are guided waves propagating in a solid plate or layer considering stress-free boundary conditions and the propagation of Lamb waves is governed by the Rayleigh–Lamb equation.[17] By numerically solving the equation, the group velocity dispersion curves for a 2-mm thick aluminum plate with Young’s modulus 69 GPa, density 2700 kg/m3, and Poisson’s ratio 0.33 are plotted in Fig.
In a nonlinear medium, the response of a mono-frequency excitation contains higher harmonic frequencies, which are integer multiples of the excitation frequency. In previous researches, the generation of higher harmonics in micro-damaged waveguides is explained by CAN,[10,11,17,20] where the clapping of interfaces of a micro-structural crack as well as the friction between them lead to asymmetrical dynamic responses and further to higher harmonic generation. CAN is emphasized not only because it is predominant in the case of delamination in composites[20] but it is also effective in the evaluation of fatigue cracks in metal. A breathing crack with contact analysis is implemented in the following simulations to describe this mechanism. This modeling method is termed “CAN model” in this paper.
Higher harmonic generation is also introduced by the PM model in this paper. Compared with the CAN model, the PM model is the representation of localized material damages at the mesoscopic level. It describes the nonlinear hysteretic behavior of fatigued metallic samples induced by micro-cracks.[21,22] It is a multiscale model taking the summation of the strain contribution of a large number of elementary hysteretic elements into consideration. The stress–strain relation of each hysteretic element of the PM model considered here is plotted in Fig.
Both CAN model and PM model are used in the following simulations. The center of the damaged area is changed to different positions and the location results using the method proposed in this paper are obtained respectively. By comparing the results, two different models of a damaged area responsible for higher harmonic generation are investigated and compared.
The numerical simulations based on nonlinear elastic response and time reversal are performed using ABAQUS/Explicit. As is schematically shown in Fig.
The simulations are performed according to the following steps: (i) S0-mode Lamb wave is excited in a micro-damaged waveguide and the response signal is recorded at the receiver. (ii) Higher harmonic components induced by the damaged area are extracted and time reversed as the reemitted signal. (iii) Backward propagation of the reemitted signal is performed on an intact waveguide and the signals acquired at the receivers in the inspection region are analyzed.
During the forward propagation, the signal is excited at x = −100 mm and the corresponding response is received at x = 100 mm. The reemission is performed at the same position (x = 100 mm). The reemitted signal is the received signal of the forward propagation processed by PI and time reversal operator. The inspection region of interest is at the center of the upper surface of the intact plate with a size of 100 mm (x = −50 mm to 50 mm). Receivers are set in this region and more dense receivers are placed around the position where the crack is expected for precise location.
Because of the multimodal and dispersive nature of Lamb waves, an accurate determination of higher harmonic components from a time domain signal can be difficult. Compared with short-time Fourier transform whose results strongly depend on setting experience-based parameters, the wavelet transform is a consistent and robust tool once an appropriate mother wavelet is chosen.[3,26] Therefore, in this paper, the further signal processing for higher harmonic generation is done by the wavelet transform and the Morlet wavelet is employed as the mother wavelet.[3,17] The peak magnitude of the wavelet coefficient at a given frequency is proportional to the amplitude of the raw signal at this frequency,[26] hence the higher harmonics can be represented by corresponding wavelet coefficients.
The generation of higher harmonics complicates the nonlinear Lamb wave propagation problem because more frequencies mean more excited Lamb wave modes. However, if an S0 mode wave is excited at low frequencies in a nonlinear waveguide, it is possible that some of the induced higher harmonics are still S0 mode and they contain the majority of the harmonic energy. The realization of this S0–S0 pair will greatly simplify the problem. Therefore, S0-mode waves are used in this paper and the excitation of the forward propagation is realized by a hamming windowed tone burst consisting of 5 cycles with the central frequency of 200 kHz.
Furthermore, to ensure the existence of only symmetric modes in the waveguide, symmetric excitation is performed in the simulations of both forward and backward propagations. By simultaneously exciting two sources placed on the double surfaces with the same signal, symmetric mode Lamb waves will be enhanced while antisymmetric waves will be suppressed. In our simulations, the point force method[27] is used to model the excitation source of Lamb waves. Moreover, the receiver is modeled by a point on the upper surface and the x-direction normal stress σxx at this point is recorded as the receiving signal.[10,23]
With nonlinear analysis as a pretreatment of time reversal, only higher harmonic Lamb waves are expected to be time reversed and reemitted to the waveguide. However, because of the multimodal and dispersive nature of Lamb waves, the extraction of Lamb waves at a certain frequency implicates some difficulties. To overcome this problem, the filtering operation called PI is used. Within a linear medium, the phase inversion of a pulsed excitation signal leads to exact inverted phase response signals. However, it is not the case in a nonlinear medium. It has been proved mathematically that when two phase-inversed excitation signals propagate in a nonlinear medium, in the sum of two corresponding response signals, all of the odd harmonic components including the fundamental component will be cancelled out while all the even harmonic components will be reserved.[28]
Although PI filtering method has been investigated for bulk waves,[22] its validation for Lamb wave simulations is essential. In a preliminary simulation, the forward propagation is performed twice using excitation signals with opposite phases (0° and 180°) respectively and the nonlinearity is introduced by CAN model with oCAN at (0, 0) mm. The response signals are SIG+ and SIG – respectively. Due to the generation of higher harmonics, the sum signal of SIG+ and SIG–, SUM, is a non-zero signal (Fig.
PM model is a phenomenological model for micro-damaged materials and has been reported to generate remarkable higher harmonics in bulk wave simulations.[22,23,25,29] The S0-mode Lamb waves used here resemble the displacement field of the simple axial wave[30] and show particle displacements mostly in the x direction.[27] Therefore, higher harmonic generation is expected in nonlinear Lamb wave simulations with the PM model. This is verified by a simulation of forward propagation using PM model with oPM at (0, 0) mm. Figure
Due to the symmetric excitation performed in both forward and backward propagations, the higher harmonics are constricted to symmetric modes. Furthermore, it is shown by Fig.
A preliminary simulation of the backward propagation is performed using time reversed SUM in Fig.
To locate the damaged area, consider the signal u(x, t) acquired at each receiver in the inspection region during the backward propagation:
The propagation of P1 and P2 are investigated to locate the damaged area. Noticing that the S0–S0 pair used here does not satisfy the certain conditions of cumulative higher harmonic generation, the higher harmonics are only induced by the localized damage. The damaged area are located by determining the position of EMAX. To obtain good location results, Δ f = 50 kHz are chosen to calculate e(x, t).
Two groups of simulations are conducted based on the process aforementioned wherein two different models of the damaged area are used, respectively. In the first group of simulations, the damaged area of a plate is modeled by the CAN model. The center coordinate of the crack oCAN is set to (–20, 0) mm, (–10, 0) mm, (0, 0) mm, (10, 0) mm, (20, 0) mm, respectively. In the second group of simulations, the CAN model is replaced by the PM model. Similarly, the center coordinate of the local defect oPM is set to (–20, 0) mm, (–10, 0) mm, (0, 0) mm, (10, 0) mm, (20, 0) mm, respectively.
The location results of the CAN model and the PM model are shown in Fig.
Table
When propagating in a damaged waveguide, Lamb waves are distorted by the damaged area. As a result, the primary excited frequency as well as higher harmonic frequencies exist in the response signal. In this paper, to overcome the multimodal and dispersive problem of Lamb waves, the symmetric excitation is used in both forward and backward propagations and the wavelet analysis is employed in signal processing. The primary excitation signal is carefully chosen so that the higher harmonics of interest are S0 mode. Pulse inversion method is used to extract the harmonic components without changing their distribution in time domain. Therefore, the time reversal is performed only to compensate the group velocity dispersion. The spatial distribution of harmonic wave packets is investigated in the backward propagation and the convergence of two chosen wave packets is proposed as an indicator of the damaged area. Two different models of the localized damage are used in the simulations. It is demonstrated that the PM model is a reasonable model in nonlinear Lamb wave research. Furthermore, the numerical results make evident that the proposed method is promising in detecting and locating micro-cracks in plate-like structures. Further experimental investigations will be performed to confirm the observed phenomena in the future.
[1] | |
[2] | |
[3] | |
[4] | |
[5] | |
[6] | |
[7] | |
[8] | |
[9] | |
[10] | |
[11] | |
[12] | |
[13] | |
[14] | |
[15] | |
[16] | |
[17] | |
[18] | |
[19] | |
[20] | |
[21] | |
[22] | |
[23] | |
[24] | |
[25] | |
[26] | |
[27] | |
[28] | |
[29] | |
[30] |