Mathematical Problems in Engineering

Volume 2015, Article ID 260641, 8 pages

http://dx.doi.org/10.1155/2015/260641

## High-Order Spectral Finite Elements in Analysis of Collinear Wave Mixing

^{1}School of Civil Engineering, Southwest Jiaotong University, Chengdu, Sichuan 610031, China^{2}Key Laboratory of Highway Engineering of Sichuan Province, Southwest Jiaotong University, Chengdu, Sichuan 610031, China^{3}College of Aerospace Engineering, Chongqing University, Chongqing 400044, China^{4}Department of Hydraulic Engineering, Tsinghua University, Beijing 100084, China

Received 8 October 2015; Revised 25 November 2015; Accepted 26 November 2015

Academic Editor: Roman Lewandowski

Copyright © 2015 Changfa Ai et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

#### Abstract

Implementing collinear wave mixing techniques with numerical methods to detect acoustic nonlinearity due to damage and defects is of vital importance in nondestructive examination engineering. However, numerical simulations in existing literatures are often limited due to the compromise between computational efficiency and accuracy. In order to balance the contradiction, spectral finite element (abbreviated as SFE) with 3 × 3 and 8 × 6 nodes is developed to simulate collinear wave mixing for 1D and 2D cases in this study. The comparisons among analytical solutions, experiments, finite element method (FEM), and spectral finite element method are presented to validate the feasibility, efficiency, and accuracy of the proposed SFEs. The results demonstrate that the proposed SFEs are capable of increasing computational efficiency by as much as 14 times while maintaining the same accuracy in comparison with FEM. In addition, five 3 × 3 nodes’ SFEs or one 8 × 6 nodes’ SFE per the shortest wavelength is sufficient to capture mixing waves. Finally, the proposed 8 × 6 nodes’ SFE is recommended for collinear wave mixing to detect damage, which can offer more accuracy with similar efficiency compared to 3 × 3 nodes’ SFE.

#### 1. Introduction

Recently, wave mixing techniques with noncollinear [1, 2] and collinear [3–6] incident waves have been used to detect the change of material nonlinearity caused by plasticity and fatigue damage. The techniques are less sensitive to the measurement system nonlinearity and can detect the nonlinearity of the mixing zone instead of the average value during incident wave propagation. The nonlinear wave mixing techniques have some unique advantages compared to traditional nonlinear wave techniques. For instance, frequencies can be selected according to the requirements of users, which can avoid unwanted harmonics typically generated by a number of electronic components in the measurement system. Besides, researchers can scan over the regions of interest directly by controlling the wave mixing locations.

Collinear wave mixing techniques were studied analytically, numerically, and experimentally by Tang et al. [3–5, 7] and Chen et al. [6, 7]. Chen et al. [6] derived a set of necessary and sufficient conditions for generating resonant waves by two propagating time-harmonic plane waves and obtained closed-form analytical solutions to resonant waves generated by two collinearly propagating sinusoidal pulses. Numerical simulations based on finite element method and experimental measurements using one-way mixing were conducted. However, waveforms and propagation rules of mixing waves can be obtained from the analytical solutions [6] only for plane wave (1D) case in the semi-infinite domain when resonant conditions are strictly satisfied. In realistic experimental measurements, the limitations, such as frequency deviation due to inaccurate acoustic velocity of material or quasi-collinear wave generation, could result in imperfect resonant conditions when using collinear wave mixing techniques. Therefore, it is necessary to simulate collinear wave mixing with numerical methods, which are the extension and supplement to the analytical solutions and the experiments.

For the past two decades, various numerical methods including finite difference method [8, 9], finite element method [10, 11], boundary element method [12, 13], finite strip element method [14, 15], mass-spring lattice model [16, 17], and local interaction simulation approach [18, 19] have been applied to simulate wave propagation. Finite element method requires strict rules for spatial and temporal discretization to study the interaction of waves, which can cause numerical problems in the cases of high frequencies and great dimensions [20–22]. No less than 20 first-order 4-node reduced elements per wavelength of the highest frequencies are required to capture nonlinear interaction [23]. Therefore, finite element method is limited due to the contradiction between accuracy and efficiency for high frequency wave propagation. However, the orthogonal polynomials-based spectral finite element method [24–26] is much more suitable for analyzing wave propagation in structures with complex geometry. This method is characterized by high-order orthogonal polynomials as approximation functions with diagonal mass matrix obtained naturally. More recently, spectral finite element method was used to simulate wave propagation in structures. Wave propagation in 1D structures, such as rod and beam, was investigated by some researches [27–31]. Numerical simulations of transverse wave propagation in a composite plate were presented by Kudela et al. [32]. Komatitsch et al. [33] applied a spectral element method based upon a conforming mesh of quadrangles and triangles to the problem of 2D elastic wave propagation. Rekatsinas et al. [34] developed a time-domain spectral finite element for improving the efficiency of numerical simulations of guided waves in laminated composite strips. The applications of 3D spectral finite element to wave propagation problems [35–39] were also investigated in many fields.

However, the applications of spectral finite element method for wave mixing techniques have not been widely reported in literatures so far. In this paper, two types of spectral finite elements are developed to simulate collinear wave mixing for damage detection. Results from analytical solutions, experiments, FEM, and SFE are compared to validate the feasibility, efficiency, and accuracy of the proposed SFEs. Finally, the comparison between two types of spectral finite elements is investigated by considering the contradiction between accuracy and efficiency.

#### 2. Theory for Collinear Mixing of Wave Pulses

Consider a homogeneous solid with quadratic nonlinearity; the displacement equations of motion in one dimension can be written as [6]where and are the two components of the displacement and and are the longitudinal and transverse phase velocities, respectively. and are called, respectively, the longitudinal and transverse acoustic nonlinearity parameters. and are the Lamé constants, is the mass density, and , , and are the Murnaghan third-order elastic constants.

When a longitudinal wave pulse and a transverse wave pulse are both emitted at and propagate in the positive -direction, it is called one-way mixing. When the transverse pulse is emitted at and propagates in the positive -direction, while the longitudinal pulse is emitted at and propagates in the negative -direction, it is called two-way mixing. A resonant transverse wave will be generated and propagate in the opposite direction of the primary transverse wave, if resonance conditions for one-way mixing and for two-way mixing can be satisfied, respectively. The frequencies of resonant waves for both cases are , where and and are the circular frequencies of the longitudinal and transverse waves, respectively.

The signal received at for one-way mixing can be expressed as

The signal received at for two-way mixing can be expressed asThe detailed expressions of (2) and (3) are given in [6].

#### 3. Formulation of 2D Spectral Finite Element

The spectral finite element formulation [27, 33, 39] process of the stiffness and mass matrices is similar to the traditional finite element formulation. The domain in 2D is firstly meshed to a number of nonoverlapping quadrilaterals. The quadrilateral spectral finite element defined on the domain is subsequently mapped from the physical coordinate to the reference domain with using invertible local mapping . In the reference domain , a set of local shape functions are defined consisting of Lagrange polynomials of degree . The local nodes , , are defined as Gauss-Lobatto-Legendre (GLL) points which are roots of the equationwhere denotes the first derivative of the Legendre polynomial of degree .

The displacement field within the quadrangular element can be approximated aswhere denotes nodal degrees of freedom and denotes the th Lagrange interpolation at GLL points . is equal to 1 at and 0 at all other points . From this definition, a fundamental property appears, , where denotes the Kronecker delta.

Similar to the traditional finite element method, the element matrices and and the vector are formulated numerically as follows:where is termed the material stiffness matrix, is a distributed load, and is the Jacobian associated with the mapping from the element to the reference domain. The superscript represents the matrix of the element in the local coordinates both in linear and in nonlinear conditions.

The quadratic nonlinear elastic constitutive relation [40] including second-order constants and third-order constants is used, which is expressed using Voigt’s notation , :where is the Lagrangian or Green strain, , .

The matrix is connected with approximated strains:where denotes a differential operator matrix

The quadrature weights , which are independent of the element, are determined by

The element matrices are assembled to the global coordinate system and finally a modeling problem of wave propagation is reduced to a well-known ordinary differential equation, which can be written aswhere is the global mass matrix, is the global damping matrix, and is the global stiffness matrix. They are all the functions of displacements in the nonlinear conditions. is a vector of the time-dependent excitation signal.

Because of the excellent property of the SFE with the diagonal mass matrix, it is especially suitable for explicit scheme (central difference method) to discretize the second-order ordinary differential equation in time. Therefore, based on the finite element method software ABAQUS, two types of 2D spectral finite elements (3 × 3 and 8 × 6 nodes) are developed via the user defined element subroutine of explicit solver (VUEL) using FORTRAN language to simulate collinear wave mixing. The scheme of the spectral finite element with 3 × 3 nodes for 2-order Legendre polynomial both in and in direction is shown in Figure 1(a). Similarly, the spectral finite elements with 8 × 6 nodes for 7-order Legendre polynomial in direction and 5-order Legendre polynomial in direction are shown in Figure 1(b). The flowchart of VUEL subroutine is listed in Figure 2.