1. Introduction
Engineered wood or mass timber products have been gaining prominence in the construction industry [
1]. The resilience of tall timber buildings recently became self-evident through a series of shake table tests on a 10-story timber building, which withstood a simulated 7.7 earthquake without major structural damages [
2]. Compared with conventional solid wood, engineered wood offers greater design flexibility on the dimensions of manufactured structural components [
3]. Moreover, as alternative construction materials to steel and concrete, mass timber products feature a higher strength-to-weight ratio [
1,
4] with a reduced carbon footprint. Layered engineered wood, including cross-laminated timber (CLT), mass ply panel (MPP), laminated veneer lumber (LVL), and glue-laminated timber (glulam), is among the most important and popular mass timber product category designs. With the increasing adoption of mass timber products, there is a potential rising demand for nondestructive evaluation (NDE) and quality assurance (QA) on layered engineered wood-based structural components, which have been challenging tasks due to their intrinsic structural complexity and heterogeneity.
Take the laminated veneer lumber or LVL as an example. LVL has been adopted in the design and construction of beams, columns, walls, and bridge decks [
4,
5,
6,
7,
8] for its high strength and reliable mechanical properties. During its production, natural wood logs are cut into thin slices (veneers), dried, coated with adhesive, stacked together, and finally hot pressed to form a solid sheet. Natural wood has cellulose fibers in their natural alignment along the longitudinal direction (long grains); the radial direction goes from the center of the wood towards the bark; and the tangential direction is around the circumference of the log. LVL’s material properties, such as strength, stiffness, and thermal expansion, can vary significantly depending on the direction in which they are measured. Specifically, we assumed that LVL and layered engineered wood, in general, can be effectively modeled as an orthotropic material with nine independent elastic constants [
9,
10]. Furthermore, layered engineered wood shares origins with natural wood, and it is not without imperfections, including delamination, knots, and fiber defects, leading to variations in strength parameters [
4,
11]. Understanding the mechanical properties of layered engineered wood is key to material behavior prediction and design [
12], but it is often challenging and limited by data scarcity [
4].
Many research efforts have been invested on destructive and nondestructive tests for determining the elastic constants of solid woods [
13,
14,
15], which could be applicable to layered engineered wood materials. For instance, compression tests on Scots pine wood yielded a 6 × 6 matrix of elastic constants with a 39% coefficient of variance for the shear modulus [
13]. Similar studies on Eucalyptus Globulus wood also showed considerable variations, where the ultrasound tests provided parameter estimations closest to ground truth [
14]. Dackermann et al. [
15] employed static and ultrasound velocity testing to determine elastic constants in all directions. While elastic moduli along the longitudinal direction showed a reasonable agreement, measured values of Poisson’s ratio exhibited a 31% coefficient of variance, highlighting the challenges of measurement. Aira et al. [
13] pointed out that ultrasound velocity testing is generally favored for its efficiency over laborious static bending and structural testing, and compression tests could lead to large variations (coefficient of variance or CoV) in the measured mechanical properties of wood materials. Therefore, alternative NDE methods for characterizing mechanical properties in orthotropic composites are reviewed, which can be applicable to engineered wood. Impulse- or impact-based vibration testing [
16], exemplified by Gibson [
17], utilizes broadband impulse vibration to determine elastic moduli and damping factors in fiber composite polymer materials. Considering multiple vibrational modes improved the accuracy of estimating elastic constants. Vibration testing also has the potential for detecting defects through natural frequency shifts. A typical vibrational test involves capturing the first five natural frequencies using three accelerometers and the Fast Fourier Transform (FFT) [
17]. Researchers have also demonstrated the effectiveness of scanning laser Doppler vibrometry (SLDV) [
18] on inspecting inlaid wood and easel paintings [
19] based on vibration and ultrasound. Moreover, Josifovski et al. [
20] reported the feasibility of using ultrasound and X-rays for in situ evaluation of timber structures. On the other hand, guided wave-based approaches provide another avenue whose analytical models, though complex, can describe wave propagation in anisotropic materials [
21]. Morandi et al. [
22] demonstrated that it is feasible to generate and measure guided wave dispersion in cross-laminated timber plates, where a bending mode was clearly identified.
Integrating experimental guided wave measurements with numerical models and optimization techniques is crucial for solving inverse problems and determining the mechanical properties of complex waveguides [
23]. Both 2D and 3D finite element methods offer solutions for guided wave propagation [
24,
25,
26]. Galerkin’s principle and energy velocities have been employed, revealing forward and backward waves in composite materials [
27]. Fully discretized 3D finite element models were developed to understand dispersive guided wave propagation through commercial FEM software, such as Ansys 5.3 [
28]. Furthermore, by assuming an analytical formulation along the wave propagation direction and only fully discretizing the 2D cross-section of a waveguide, the Semi-Analytical Finite Element (SAFE) method lends itself to compute wave dispersions with greatly reduced computational expenses [
29,
30,
31]. More recently, researchers developed an extremely powerful tool that can be easily implemented in commercial FEM software for wave dispersion calculation leveraging Floquet periodicity [
32]. Moreover, optimization algorithms have been developed for material characterization based on guided wave models and measurements. Zhu et al. [
23] utilized an improved genetic algorithm (GA) to predict material properties, emphasizing the importance of proper wave mode selection for optimization. Their study achieved material property estimation accuracy of under 1%. Similar experiments involved composite plates, employing non-linear optimization to estimate material properties, achieving accuracy within a 10% range [
27]. Cui et al. [
29] used simulated annealing for anisotropic and quasi-isotropic laminates and analyzed wave modal sensitivities. Rautela et al. [
33] developed a 1D Convolutional Neural Network for composite material properties, maintaining accuracy at around 1% with rapid training. Vishnuvardhan et al. [
34] optimized all nine elastic constants of graphite–epoxy composite plates via GA, highlighting sensitivity analysis and achieving promising errors of approximately 1%. Genetic algorithms can generally support a performance similar to particle swarm optimization and outperform simulated annealing [
35], and they are capable of handling complex multi-dimensional and multi-modal inverse problems [
36]. Guided wave measurements incorporating computational optimization through GA offer precise material parameter estimation with enhanced accuracy and consistency compared to destructive testing, which enables material characterization and structural condition assessment.
Prior research mainly focused on using destructive tests and computational simulations to characterize the mechanical properties of natural wood [
13,
14,
15,
37], identifying wave dispersion in mass timber plates [
22], and structural health monitoring based on hygeothermal, static, and dynamic behavior of mass-timber buildings [
38]. In this study, the very first framework to determine orthotropic elastic constants of layered engineered wood based on guided wave dispersion was proposed. Its feasibility was evaluated by determining orthotropic elastic constants of an LVL bar in a nondestructive manner. We acquired experimental dispersion curve data from the LVL bar, developed finite element models to understand the specific wave propagation behavior, and employed a genetic algorithm to optimize elastic constants by bridging experimental and computational approaches. This approach enhances predictive accuracy for engineered wood behavior, benefiting material scientists, engineers, and industry stakeholders. It promotes efficient and sustainable use of engineered wood, contributing to green and resilient structures.
4. Conclusions
This study presented the first comprehensive investigation of material characterization for orthotropic layered engineered wood structures using multi-modal guided waves and the genetic algorithm. The laminated veneer lumber, or LVL, derived from natural solid wood, is susceptible to internal defects and delamination, making it imperative to comprehend their material properties for potential NDE and quality assurance of engineered wood components and structures. Our investigation encompassed both destructive compression tests and nondestructive guided wave measurements on LVL samples, with the aim of inferring orthotropic material properties. To achieve this, a sophisticated multi-parameter optimization technique, namely the genetic algorithm, was developed, which was tailored to determine the elastic constants of engineered wood by carefully defining an appropriate fitness function. The sensitivity analysis of dispersion curves to variations in material properties was systematically examined, leading to the precise definition of a fitness function capable of quantifying prediction errors. By minimizing fitness values, a set of optimized material parameters that closely matched the dispersion curves were derived based on actual material properties. Further performance evaluation using a larger sample size of compressive testing is needed to ensure the accuracy and reliability of the optimization process.
Our current study primarily focused on the determination of material properties of layered engineered woods, employing an inversion method to optimize the nine sets of independent material properties to best fit the dispersion curves obtained from guided waves. The experimental setup, however, was designed to excite vertical flexural modes, prompting us to optimize parameters sensitive to these modes. While the specific mode family lacks sensitivity to some of the elastic constants, our study successfully optimized a subset of the nine elastic constants. It is essential to note that our objective was to develop a method employing guided waves and the GA optimization tool for estimating material properties, which need to undergo further verification before their application in the structural design process. Future research endeavors can extend the optimization scope to include not only the elastic constants but also the viscoelastic properties of engineered wood. In our study, the low-frequency spectrum of guided waves was studied, which are typically less prone to attenuation. However, it is essential to account for the viscoelastic effect inherent in wood characteristics. The elasticity matrix, composed of elastic constants, is intricate, and incorporating viscoelasticity into the optimization technique can facilitate the optimization of the attenuation coefficient of engineered wood. While the feasibility of the proposed framework has been verified in this study, statistical modeling and analysis on the framework’s performance is necessary to make informed decisions for NDE and QA purposes. This holistic approach would enable a more comprehensive characterization of the material, dispelling the assumption of negligible viscoelastic effects and thus enhancing the accuracy of material property determination for practical applications in structural design and engineering.