Multiscale analysis procedures for composites often involve coupling the macroscale (e.g., structural) and meso/microscale (e.g., ply, constituent) levels. These procedures are often computationally inefficient and thus are limited to coarse subscale discretizations. In this work, various computational strategies were employed to enhance the efficiency of multiscale analysis procedures. An ensemble averaging technique was applied to stochastic microscale simulation results based on the generalized method of cells (GMC) to assess the discretization required in multiscale models. The procedure was shown to be applicable for micromechanics analyses involving both elastic materials with damage and viscoplastic materials. A trade-off in macro/microscale discretizations was assessed. By appropriately discretizing the macro/microscale domains, similar predicted strengths were obtained at a significantly less computational cost. Further improvements in the computational efficiency were obtained by appropriately initiating multiscale analyses in a macroscale domain. A stress-based criterion was used to initiate lower length scale GMC calculations at macroscale finite element integration points without any a priori knowledge of the critical regions. Adaptive multiscale analyses were 30% more efficient than full-domain multiscale analyses. The GMC sacrifices some accuracy in calculated local fields by assuming a low-order displacement field. More accurate microscale behavior can be obtained by using the highidelity GMC (HFGMC) at a significant computational cost. Proper orthogonal decomposition (POD) order-reduction methods were applied to the ensuing HFGMC sets of simultaneous equations as a means of improving the efficiency of their solution. A Galerkin-based POD method was used to both accurately and efficiently represent the HFGMC micromechanics relations for a linearly elastic E-glass/epoxy composite for both standalone and multiscale composite analyses. The computational efficiency significantly improved as the repeating unit cell discretization increased (10-85% reduction in computational runtime). A Petrov-Galerkin-based POD method was then applied to the nonlinear HFGMC micromechanics relations for a linearly elastic E-glass/elastic-perfectly plastic Nylon-12 composite. The use of accurate order-reduced models resulted in a 4.8-6.3x speedup in the equation assembly/solution runtimes (21-38% reduction in total runtimes). By appropriately discretizing model domains and enhancing the efficiency of lower length scale calculations, the goal of performing highidelity multiscale analyses of composites can be more readily realized.

Plants and bacteria both have specialized light-harvesting pigment-protein complexes, composed of a network of chromophores encompassed by a protein scaffold, that are involved in photosynthesis. While chromophore, as well as protein, composition and arrangement vary in these light-harvesting complexes, chromophores transfer energy as molecular excitation energy through their complex multi-chromophoric network with near perfect efficiency. Understanding the efficiency of this excitation energy transfer process has been the focus of many interdisciplinary studies. By elucidating the mechanisms involved in efficient excitation energy transfer in biological systems, we are able to guide the design of novel organic materials for their application in photovoltaic systems. Interdisciplinary studies of light-harvesting biological systems leverage advanced spectroscopic techniques and theoretical models to help explain the interaction be- tween excited electronic states. Difficulties in assigning the origin of spectral features in spectroscopy experiments arise from both homogeneous and inhomogeneous effects. Various computational studies have been able to provide theoretical models that help disentangle these effects and provide insight into the origin of some these spectral features. In this work, we present a computational approach that is used to calculate an ensemble of model Hamiltonians for a light-harvesting pigment-protein complex found in algae. To verify the reliability of our model, we compare various computed spec- tra with experimental measurements. Next, we extend our computational approach for parameterizing an ensemble of Hamiltonians for two configurationally unique or- ganic dimers. Finally, we examine the error of some of the approximations made while partitioning “system” and “bath” degrees of freedom when computing molecu- lar properties. Using these methods we are able to provide mechanistic interpretations and explanations of spectral signatures observed in various linear and nonlinear ex- perimental spectra.

Cette thèse concerne l’étude de la propagation d’ondes acoustiques dans des structures hétérogènes. Le but essentiel de ces travaux est de confronter des résultats d’expériences numériques effectuées dans le domaine physique (espace, temps) à des prédictions analytiques pour la propagation des ondes de surface SH le long d’un demi-espace stratifié périodique produisant des spectres discontinus de dispersion pour les ondes, ainsi que pour la diffusion multiple dans des milieux aléatoires inclusionnaires (fissures, cavités). Le code numérique FDTD développé lors de cette étude a permis, en autres choses, de corroborer quantitativement les fenêtres spectrales théoriques d’existence des ondes de surface dans les demi-espaces périodiques,ainsi que de montrer des zones de validité fréquentielles des approches analytiques de diffusion multiple concernant les propriétés effectives de milieux aléatoires. / The study is concerned with acoustic waves in elastic media with a different nature of in homogeneity consisting in either periodically continuous or piece wise variation of material properties, or in random sets of defects embedded into a homogeneous matrix, with a given statistical distribution. The scope of problems is topical in non-destructive testing and other applications of ultrasound.Theoretical methods describing involved acoustic phenomena (complex dispersion features, coherent wave in random media, ensemble average techniques) often rely on certain a priori assumptions which render numerical verification especially important.The thesis presents results of analytical modelling of the propagation of surface acoustic waves along periodic half-space, for which the dispersion spectrum is rather complex (discontinuous spectrum of propagation for the surface waves). A 2nd order FDTD numerical code has been developed in order to perform numerical experiments in the space and time domains, and to corroborate the analytical predictions in the frequency domain. A good agreement of simulated results with analytical modelling demonstrates applicability and consistency of the numerical tool. Finally, the code has been used for extracting numerically the coherent wave regime (mean wave over ensemble averaging of the positions of scatterers) for the acoustic propagation in different types of populations of randomly distributed scatterers. The results indicate ranges of validity of some multiple scattering analytical techniques.

<div> This thesis proposes an artificial neural network (NN) model to predict fuel consumption in heavy vehicles. The model uses predictors derived from vehicle speed, mass, and road grade. These variables are readily available from telematics devices that are becoming an integral part of connected vehicles. The model predictors are aggregated over a fixed distance traveled (i.e., window) instead of fixed time interval. It was found that 1km windows is most appropriate for the vocations studied in this thesis. Two vocations were studied, refuse and delivery trucks.</div><div><br></div><div> The proposed NN model was compared to two traditional models. The first is a parametric model similar to one found in the literature. The second is a linear regression model that uses the same features developed for the NN model.</div><div><br></div><div> The confidence level of the models using these three methods were calculated in order to evaluate the models variances. It was found that the NN models produce lower point-wise error. However, the stability of the models are not as high as regression models. In order to improve the variance of the NN models, an ensemble based on the average of 5-fold models was created. </div><div><br></div><div> Finally, the confidence level of each model is analyzed in order to understand how much error is expected from each model. The mean training error was used to correct the ensemble predictions for five K-Fold models. The ensemble K-fold model predictions are more reliable than the single NN and has lower confidence interval than both the parametric and regression models.</div>