Heterogeneous materials are vital for both the modern engineer and inquisitive scientist alike. They make up a vital material class that can either form inevitably as a result of material processing (such as crystallization in metals) or can be intentionally designed for to gain desirable properties (such as anisotropy in composites). As such, due to their prevalence and applicability to various engineering problems, predicting their behavior has become a topic of great interest in the solid mechanics community over the last few decades.

One particularly pressing challenge is performing fast mechanical simulations along multiple length scales in heterogeneous materials. To this end, the reduced order method known as Self-Consistent Clustering Analysis (SCA) has been proposed as an effective means of striking a balance between accuracy and efficiency. Underpinning this is SCA's ability to decompose a domain into clusters in a preliminary offline stage (learning), efficiently reducing the problem's degrees of freedom. It has been shown to be remarkably accurate in predicting plasticity without significantly degrading accuracy compared to other methods, such as the finite element method. However, the offline stage requires a quantity known as the Cluster Interaction Tensor (CIT) to be computed, whose computational complexity scales quadratically with the number of clusters, thus causing a bottleneck in the method. Additionally, the CIT is recomputed during the online stage (prediction) in the recently proposed extension called adaptive Self-consistent Clustering Analysis (ASCA), which further stresses the need to speed up their computation.

To address this limitation, a data-driven surrogate model is proposed to compute the CIT efficiently. The behavior of the local CIT components is first analyzed, from which it is concluded that a surrogate model shall be developed to predict upper off-diagonal terms. This decision is made due to the bi-modal nature of certain components in the tensor and the quadratic scaling behavior of off-diagonal terms versus the linear scaling of diagonal ones. Following that, a sensitivity study shows how various magnitudes of noise affect the homogenized response's accuracy and convergence performance. Furthermore, it is shown that the solution accuracy and convergence behavior are degraded when the number of clusters is increased. With a proper understanding of the CIT's behavior and its function within SCA, the surrogate can be created.

The surrogate's feasibility is first shown using the ResNet-18 architecture, a CNN model derived from computer vision. It is demonstrated that ResNet-18 can make accurate predictions for a range of different microstructural parameters. This includes the number of clusters and samples in the dataset's distribution and out of distribution. Composite RVEs are used as out-of-distribution samples to test the robustness of the surrogate to realistic inputs. An attempt is made to improve the model's performance by training it on an unbalanced dataset, and although it improves the overall CIT prediction in specific regimes, the resulting stress-strain generalizability is degraded. Subsequently, a lighter version of ResNet-18, known as ResNet-lite, is tested and shown to give faster predictions than the baseline method. This, however, comes at a cost to the accuracy of the solution.

Additionally, deemed as a critical aspect of the study, the efficient generation of a large representative training dataset is discussed. A novel method for generating clustered microstructures efficiently using gradient noise is introduced. By leveraging datasets derived using this method, the surrogate can learn the fundamental interactions needed to make accurate predictions on realistic, out-of-distribution samples. The need for a large dataset is further emphasized using a dataset size sensitivity study.

Neural networks have made significant progress in domains like image recognition and natural language processing. However, they encounter the challenge of catastrophic forgetting in continual learning tasks, where they sequentially learn from distinct datasets. Learning a new task can lead to forgetting important information from previous tasks, resulting in decreased performance on those earlier tasks. This issue is further intensified in dynamic scenarios where the task sequence varies unpredictably. To address this problem, architectural methods have been developed to modify a neural network's structure, creating or adapting subnetworks to retain task-specific knowledge and mitigate catastrophic forgetting. However, these solutions can lead to network saturation, where the accumulation of task-specific adaptations hampers the network's ability to learn new tasks. This research aims to address the problem of network saturation by developing innovative methods that enable neural networks to maintain high performance across both existing and new tasks in continual learning scenarios. Eventually, the new model improved its learning ability on new tasks in the presence of an allowable forgetting, while demonstrating better overall learning ability.

This research explores the integration of Neural Architecture Search with Optical Neural Networks to optimize the efficiency and performance in traditional visual image classification tasks. The study introduces a new approach that applies Neural Architecture Search, a technique traditionally used to optimize the performance of Artificial Neural Networks, to the field of Optical Neural Networks.

Snow is a natural hazard to human life and infrastructure. This motivates current research efforts to understand the granular material. The material point method models snow as a continuum. Application length scales range from the microstructural level to full scale avalanches. This conventional numerical method relies on solely spatially local information to make local updates. The recent graph neural network machine learning model is shown to include both local and global information in making local updates. This model’s promising attribute motivates its use to replace the conventional snow simulation method. However, it is uncertain if current graph neural network applications to learn physical simulations truly learn the underlying physics. This work is inspired by the finite element community's patch-test proposed in the 1960s. This insight is used to reimagine the means a graph neural network model is evaluated. Through this novel evaluation choice, may the model be investigated on the core properties of numerical methods. Further, a state-of-the-art graph neural network model is improved to utilize unnormalized features and targets in making stable predictions. Future research recommends these machine learning models in this application make architecture design choices such that the core properties of conventional numerical methods are met.

Mycelium based composites (MBC) have revolutionised the field of material production, where a living fungus is employed to overgrow and bind lignocellulosic substrate materials together. Their bio-based nature, low embodied energy, and biodegradability, mark their great potential to reduce the increasing pressure conventional materials put on the environment. Unfortunately, the material's applications are limited due to its low mechanical properties, that are equivalent to those of foams or natural fibre boards. Furthermore, where the material is primarily used and studied in its heat-treated non-living form, harnessing the biological power of the fungus shows to give the material self-healing and sensing capabilities. The required hydrolysed state is however shown to further decrease its mechanical properties, and yet hardly evaluated in literature. Especially the mechanics of the mycelium-lignocellulose interface are inadequately studied, and important to evaluate for a broadening of the living composite material's applications.

In this study we present the development of a double cantilever beam (DCB) test according to ASTM D5528, setup to quantify the interlaminar fracture toughness (G_I) of Ganoderma lucidum (G. lucidum) grown in between wood veneers. The data was evaluated using an analytical approach based on a decrease in the beams' compliance. The specimens were fabricated with the use of additive manufacturing, allowing precise control over the placement of the fungus and its provided nutrients. Its growth behaviour into the substrate was qualitatively assessed through optical microscopy and scanning electron microscopy (SEM), and its digestive ability on the surface by fourier-transform infrared spectroscopy (FTIR).

A mycelium-laden ink presented by Gantenbein et al. (2023) was reproduced with a 75% lower agar content to serve a stable source of mycelium on the DCB specimens [1]. A growth period between 3 and 4 weeks from the printed hydrogel was required for substantial mycelium-substrate binding, which stabilised after 4 weeks of growth. Furthermore, the provision of malt extract (ME) was required, but not needed to be higher than 5% of the ink's weight. The G_I was in these conditions reported to be 1.83 J/m² on hornbeam veneers, where a maximum value of 3.46 J/m² was reached. Variable growth generated substantially different mechanical properties, which resulted in the lower G_I of mycelium grown on beech and spruce samples, caused by the use of an older fungal inoculum of a different reference plate. A stronger binding on beech than spruce did suggest a more important role of the substrate chemistry than its density. The mycelium always showed cohesive failure, showing the low G_I to not result from poor substrate adhesion, but rather from the mechanics of the hyphae network developed. Microscopy and FTIR evaluations showed the ability of G. lucidum to digest the hornbeam's lignin. The full depths of hornbeam and beech veneer substrates with 5 and 3 mm thickness were colonised, where vessel elements served as the main pathway for the hyphae.

This study was, to the best of our knowledge, the first to isolate the binding behaviour of a living mycelium in a mode-I loading condition. It thereby provides valuable new insights in the field of living MBC, and contributes to the further development of this advanced, eco-friendly, and living material. The setup can now be utilised to study the binding behaviour of different fungi on substrates of varying chemistry and porosity. Furthermore, aiming to elucidate the use of additive manufacturing in the composite production, this study opened up pathways into controlling the composite's properties in the future.

[1] Silvan Gantenbein et al. “Three-dimensional printing of mycelium hydrogels into living complex materials”. In: Nature Materials 22.1 (2023), pp. 128–134.

Delamination is a significant failure mode that has been the subject of extensive research in laminated structures. The interface in assembly structures often represents the susceptibly weak link and necessitates careful consideration to guarantee structural stability. It is of utmost significance to understand the behaviour of the delamination phenomenon and precisely evaluate the fracture toughness (resistance to delamination) in composites. Numerous test techniques have been established over time, some of which have been recognized and adopted as standards. The complexity associated with the standardized tests (crack tip monitoring, design configuration) and the limitations in the scope of testing various interface configurations have led to favoring of alternative testing methods in recent years. This thesis primarily aims at exploiting the new potentials of the Mandrel Peel Test method by incorporating the novel “Multi-Mandrel Concept” in determining the fracture toughness of the Thermoplastic Composites. Mandrel Peel test, a modified adaptation of the standard 90-degree peel test, presents as a promising technique for the delamination fracture testing of composites by determining the energy necessary for peeling off a flexible adherend (generally referred to as the peel-arm) from a rigid substrate. In this thesis, the objective was to further the understanding of the Mandrel Peel test, investigate the sensitivity of specimen configuration and testing parameters on different thermoplastic materials systems, and
evaluate the suitability of the Mandrel Peel test method for establishing a relationship between mandrel roller size and mixed mode fracture behavior. Comparative assessments were conducted between the multi-mandrel radii peel test and standardized tests, including DCB, ENF, and MMB tests. Experimental tests, fractography studies, and numerical validation were performed on UD GF/PP and UD CF/PPS composites.

The results show that the average fracture toughness values are slightly higher and more consistent for 2-ply peel arm specimens compared to 1-ply peel arm specimens. This is attributed to the microstructure (matrix and fiber distribution) at the interface and its influence on damage mechanism and fracture propagation behavior. The thickness of the ply also affects delamination propagation behavior and depends on the microstructure, with thicker plies exhibiting non-uniform fiber and matrix distribution. The experimental investigation reveals that the fracture toughness values increase as the mandrel
radius increases for UD GF/PP and UD CF/PPS composites with a 0|0 interface. The fracture toughness of CF/PPS peel specimens with a 0|90 interface shows limited variation
with respect to the mandrel radius. The relationship between mandrel radius and fracture toughness may not follow a linear trend, and it can depend on other factors such as material thickness, interface orientation, and fracture energy at pure loading modes. Comparative assessments between standardized tests and mandrel peel tests show similar trends in fracture toughness values with increasing mode mixture and mandrel roller radii.
Fractography analysis provides valuable insights into the fracture behavior of tested materials, and a correlation is identified between the failure modes observed in mixed-mode bending (MMB) tests and mandrel peel tests. The mandrel peel test enables establishing a relatively straight crack front during the delamination propagation, controlled by the mandrel’s kinematics, resulting in a more uniform distribution of strain energy across the specimen’s width, unlike standard tests.
The findings suggest that utilizing the multi-mandrel concept in assessing mixed-mode fracture toughness offers the possibility of extrapolating pure mode I and II fracture toughness values. This approach presents a viable tool for characterizing the fracture energy of composites with non-zero fiber-oriented interfaces that cannot be effectively assessed using classical tests. The results support the claim that mandrel roller size influences the degree of mode-mixity and suggests considering mandrel size when assessing mode-mixity in fracture phenomena.
Overall, this study provides valuable insights into the sensitivity of specimen configuration and testing parameters, the influence of mandrel roller size on mixed-mode fracture behavior, and the potential benefits of the mandrel peel test for delamination studies. By considering these findings, further advancements can be made in understanding and characterizing fracture properties in hybrid material systems.


Keywords: Delamination, fracture toughness, mixed-mode fracture, DCB, ENF,
MMB, Mandrel Peel test, thermoplastic composite, GF/PP, CF/PPS, Multi-
Mandrel Concept, peel arm, mandrel roller, mode mixity

The two most common Post Consumer Recycled (PCR) plastics, isotactic polypropylene (iPP) and high density polyethylene (HDPE), differ in composition and mechanical behavior when compared to their virgin counterparts. This thesis focuses on understanding and modeling the mechanical performance of these two PCR plastics separately. Within this context, the present work implements three finite strain thermoelasto-viscoplastic constitutive models developed by Johnsen et al., Mirkhalaf et al. and Anand et al. proposed in the literature to predict the behavior of PCR-iPP and PCR-HDPE. The models are compared and further developed to take into account the effects of recycling. All the models depend on the fully implicit return mapping algorithm and associated state update procedures.

Given the complexity of the models, this thesis proposes the use of Bayesian optimization to facilitate the material parameter calibration when provided with the experimental data. A two-step procedure is proposed where first the models are calibrated for yielding, and then for post yielding behavior (strain softening and orientational hardening).

The models are assessed considering different experimental tests, including standardized specimens with different radius of curvature. A simple modification is suggested to capture the strain hardening response at large deformations accurately. This work concludes that the model developed by Mikhalaf et al. is capable of accurately reproducing the experimental results obtained in the validation experiments of PCR-PP where as the modified Anand model is capable of accurately reproducing the experimental results of PCR-PE.

Bio-based composites have been a viable material choice in aerospace, automobile and construction industries over the past few decades. From day-to-day products like spoons and chairs to the construction of rocket parts, bio-based composites find their applications due to their mechanically enhanced, cost-effective and lightweight structures, with a reduced carbon footprint. This study collaborates with NPSP, one of the companies leading the way towards a circular economy.
However, the manufacturing and testing of bio-based composites is time and resource expensive. Moreover, exploring all natural fillers and fibres ratios is not feasible experimentally. Optimization and analytical models are two potential approaches that accelerate the search towards an optimal bio-based composite recipe when combined with bio-based materials research. A data-scarce Bayesian optimization model was already developed to research the composition of bio-based composites. The proof-of-concept program adjusts the natural materials’ weight ratios to optimize toward user-defined mechanical properties. The objective of this study is to experimentally investigate if the Bayesian Optimization model works by varying the objective functions and adapting the model according to the clients’ needs. By exploiting the machine learning model at an initial stage, the purpose is to test how the model reacts to different objective functions defined at high weight values and observe if the model can generate optimized recipes with better mechanical properties than the defined training set. For this study, four different fillers (calcite, lignocellulosic filler 1, lignocellulosic filler 2, waste-based filler) and two different fibres (flax, bamboo) are used, with calcite as the reference filler. A primary goal to be achieved by NPSP is to reduce or eliminate calcite as the primary filler due to its high density and brittle nature. Promising results have been obtained within two iterations of using the model for the different filler/fibre systems used.
Additionally, the thesis also investigates the rules of mixing for multi-component systems to develop analytical models that predict the mechanical properties of the composite. The Lewis-Nielsen model and the Cox -Krenchel theory have been used to compare theoretical values with experimental data points. The initial binary phase curve fitting followed by applying cascading to obtain ternary and quaternary phase mechanical properties provides approximate results for most recipes made. Finally, recommendations to improve both models have been covered, and a potential to combine both approaches has been shown.

Designing and modeling compatibilized polymer blends require accurate interface model. In addition, it is possible that crazing occurs during failure of the interfaces leading to large deformation prior to complete failure and therefore must be accounted for by the interface model.

A preliminary literature review showed that existing formulations for the large
deformation account for the nonlinearity by adjusting the assumptions. For example, Van den Bosch et al. (2007) proposed redefining the local basis at each integration point. In contrast, Reinoso and Paggi (2014) argued that this did not account for geometric nonlinearity and proposed including the first derivative of rotation vector in the finite element equation. However, the mixed-mode responses of the models were not characterized and validated thoroughly. Therefore, we studied the response from standardized mixed-mode tests to compare the large displacement formulation with the commonly used small-displacement formulation of the cohesive zone model.

The standardized tests used by Moreira et al. (2020) for characterizing the mixed-mode behavior of ductile interfaces inspired the tests used in this study. When implemented with the BK criteria, the mode partitioning method used by Moreira et al. (2020) results in a wider spread of the mean predicted mode ratio and the mean predicted fracture toughness. However, the corresponding mode ratio predictions are similar when the predicted fracture toughness is close to expected. Therefore, while the power-law is better for implementing the mode partitioning method, we can use the predictions from the mode partitioning method implemented with the power-law to find the BK parameter.

Further, simulating the mixed-mode fracture tests with properties presented by
Moreira et al. (2020) showed that bulk materials with high modulus or stiffness, such as carbon fiber reinforced plastics, do not undergo nonlinear deformation to require the large deformation formulation. However, for bulk materials with lower modulus or stiffness, the responses of the two formulations in question are different. Additionally, for a given load case, a cohesive zone with anisotropic fracture properties experiences a different mode ratio when implemented with the large displacement formulation than the small displacement formulation. Moreover, more significant influence of the stronger mode on the load case results in greater difference between the mode ratio experienced at the interface. Nevertheless, the large displacement formulation is also applicable for stronger mode-II interfaces. However, a further investigation involving physical experiments is required to compare the response of the two formulations to the behavior of real material systems.

Conventional Topology Optimization (TO) enables the inverse design of nanophotonic structures by specifying the objective and constraints without a predefined topological concept. Yet, extreme scenarios such as the design of a lightsail pose challenges that require new solutions. Here, a convolutional neural network (CNN) based TO methodology is extended to optimize a two-dimensional photonic crystal used to design a lightsail that aims to reach the nearest star (Alpha Centauri) within 20 years by achieving 20% of the speed of light. The CNN-TO performance is compared to a more conventional method of moving asymptotes (MMA) based TO by optimizing a photonic crystal unit-cell for the 2016 Starshot Initiative parameters. The CNN-TO requires up to 40% fewer iterations than MMA-TO to reach better performance under different operational conditions. The generated design turned out to be easy to fabricate, allowing them to be produced with optical lithography. Additionally, a study regarding the design challenges of the lightsail has been performed, which resulted in an optimization considering the functionality of the sail. Additionally, the study showed the sensitivity of the resulting design to varying objectives and materials. Therefore, underlining the necessity of considering multiple operating conditions (e.g. laser alignment and cooling) within the design process.

The superior nature of long fiber reinforced thermoplastics (LFTs) in terms of properties such as specific modulus, strength, and excellent impact resistance, along with their lightweight properties, ease of processability and recyclability make them one of the most competitive materials leading to their growth in various fields of applications.
With the growing rise of LFTs, the research on their mechanical properties and their processing has grown as well. Injection molding method has been one of the most widely applied production method for LFTs processing. Research has shown that the mechanical edge offered by the LFT products obtained from injection molding is dependent on the lengths of the fibers in the final product. This makes it important to understand the breakage of fibers during processing.
This work addresses the breakage of fiber in injection molding by conducting a thorough review of the industrial, experimental and theoretical research done in the field. Following this, simulation approach has been undertaken to understand the fiber breakage in geometries which represent slightly complex shapes as compared to the conventionally used shapes. Based on the theoretical research, the state-of-the-art models have been reviewed and compared in their ability to produce reliable fiber breakage results.
The results showed the influence of rheology and geometry on fiber breakage in simulations, where shear rates led to higher fiber breakage and increasing viscosity led to a slight reduction in breakage. Further, the simulation study provided inconsistent results with variation in geometry, and a need for further fine tuning of simulation parameters was observed. The theoretical models applied in the study gave reliable results in terms of the trends of the fiber breakage, with a novel model, called Bechara model, showing acceptable and more time efficient results, in comparison with the currently applied commercial model, Phelps model.

The accuracy of hot press forming process simulations with unidirectional fiber reinforced thermoplastics is not at the desired level. Fundamental knowledge about the interactions between adjacent plies is needed to enhance predictive quality. Several mechanisms can be distinguished during hot press forming of composites. This thesis focuses on the inter-ply friction behaviour which is the resistance against inter-ply slip. The main variable investigated in this study is temperature.

In this research, an extensive friction characterization with UD C/LM-PAEK is conducted at temperatures ranging from 300 to 365 ◦C. The neat matrix material has been studied with DSC and rheometry experiments. In general, a peak response can be seen during start-up in a friction characterization experiment. This peak, or overshoot, progresses towards a steady state friction response after a slip distance of several mm. Reducing the temperature showed similar effects to increasing the sliding velocity in a ply-ply slip system. The peak during start-up increases in magnitude while the steady state response remains approximately constant. Indications of flow induced crystallisation have been observed during friction characterization around the melting point of the material. The timetemperature-superposition principle has been applied to experimental friction data. This enabled to predict the duration of the transition of peak friction response towards a steady state. Several modelling efforts have been compared to the experimental data. The accuracy of the model predictions is similar between 315 and 365 ◦C. Influences of flow induced crystallisation impede the reliability of the specific models around and below the melting point.

The research lead to useful insights in the friction behaviour at relatively low temperatures. Further research is required on the field of flow induced crystallisation for a better understanding of its role in the friction response. Further study with other materials is needed to validate the application of the time-temperature-superposition principle to predict the speed of the transition of peak friction response towards steady state.

Polymers have a long history of development and their application widely supports the normal operation of industry and society. Yet, the environmental consequences of using virgin plastics demand new sustainable solutions. In this quest, being capable of predicting the mechanical behavior of post-consumer recycled plastics is essential to support the adoption of these materials in engineering applications. This literature review focuses on recycled polypropylene and its mechanical viability when used in closures, a common part in packaging industry. Recycled polypropylene’s performance is found to be dependent on crystallinity, which in turn is dependent on the length of the polymer chains, isotacticity and other co-polymerization segments. This document also reviews constitutive models applicable to predict the behavior of closures via the the finite element method. A suitable thermo-viscoplastic model is selected, setting the stage for the finite element analyses to be conducted during the dissertation work. The application of recycled polypropylene is challenging due to its low mechanical properties. In the thesis, we demonstrate the possible aspects which decide the properties of post consumer recycled polypropylene (PCR-PP), including molecular weight, tacticity, interface structure and degradation. After discussing the material properties, the focus is to testify the application of PCR-PP in closures for particular functionality. For active closures, the weakest structure is the hinge, at the middle of lid. To verify if the material is applicable to active closures. Finite element analysis are conducted to help develop new products in a more efficient way, and also reduce the plastic waste. In the thesis, the explicit dynamic method and the static general method are applied with different boundary conditions in the model to simulate the differentiation of closures including manufacturing conditions. The explicit dynamic simulation involves higher computational cost, and due to the existence of inertial effect, the observed reaction force is significantly higher than it should be when the mass scaling is over a certain region. Considering an implicit static analysis was found to be proper. Once the loading condition was simplified and by carefully choosing an propriate thermo-visco-plastic constitutive model, the finite element analysis can predict the snap through behavior of the closure. The snapthrough behavior also shows that the equivalent plastic strain at the hinge middle is higher than the maximum tensile strain of polypropylene(QCP-300P) under room temperature. However, when subjected to 69 degree Celsius, the closure is expected to deform without failing. This result demonstrate the necessity of bending the closures right after injection molding, which fit with the experimental result.

Suspended microchannel resonators are a kind of mechanical resonator with an embedded microchannel inside. This is done as for some samples it is convenient or even necessary to put the sample into a fluid. The sample can then travel through the channel and its mass can be determined. The channel also simplifies placement of the samples compared to regular nanomechanical resonators. By having this channel embedded within the resonator, it is then possible to place the resonator as a whole in a vacuum and so eliminate medium losses. The main purpose of these suspended microchannel resonators is mass spectrometry which has already been performed on viruses and biological cells and so could potentially be very useful for the medical field as well. Especially if suspended microchannel resonators can be used to measure individual proteins which has not been done before with these resonators. It would also be interesting to use these kinds of resonators in quantum mechanical experiments. This means that the performance of these suspended microchannel resonators needs to be improved.

In order to accomplish this, a new kind of suspended microchannel resonator is introduced. The determination of these designs is strictly FEM based. The new resonator design, which is a doubly clamped beam, is made from pre-stressed silicon nitride which enables high Q-factors caused by an effect known as dissipation dilution. Dissipation dilution means diluting the energy losses of the system by increasing the stored energy which is increased by the initial stress in the silicon nitride. To increase this effect, the resonator is tapered towards the center. This is called strain engineering. Silicon nitride has been used in suspended microchannel resonators before but it did not result in higher Q-factors. Possibly due to clamping losses. The designs presented here offer a solution to that in the form of soft-clamping. This reduces the curvature near the boundaries to a minimum which means that the clamping losses are also reduced or eliminated. The combined effect of these features results in very high Q-factors as has already been reported for nanomechanical resonators. Through machine learning optimization it is possible to achieve 6E12 Hz in Qf product required for quantum mechanical experiments at room temperature and more than one order of magnitude improvement in mass sensitivity.

Algorithmic optimization is a viable tool for solving complex materials engineering issues. In this study, a data-scarce Bayesian optimization model was developed to research the composition of bio-based composites. The proof-of-concept program adjusts the natural materials' weight ratios to optimize towards user-defined mechanical properties. Preliminary results show that the bio-composites proposed by the program had improved properties compared to existing bulk-moulding compounds. However, the algorithm choice is often arbitrary or based on anecdotal evidence. In parallel, this thesis proposed a data-driven framework for general data-scarce optimization problems to adapt the meta-heuristic during optimization. Guided by the 'No Free Lunch' theorem, we verified that the effectiveness over a selection of algorithms is dependent on problem-specific features and convergence. This effectiveness was captured in a unique identifier metric by optimizing a generated training set of optimization problems. The average solution quality was improved by combining several meta-heuristics in series, based on these problem-specifics. During the optimization of problems in the testing set, the same unique identifier was constructed at predefined stages in the optimization process. Subsequently, the problem was classified, and the meta-heuristic was adapted to the best-performing algorithm based on similar training samples. Experiments with various classifiers and a different number of predefined assessment stages were performed. Results show that the data-driven heuristic decision strategy outperformed the individual optimizers on the testing set. Despite the use of binarization techniques, the classification accuracy was heavily influenced by the imbalanced training set. In terms of computational resources, the various adaptions of the data-driven heuristic strategy are 2.5 times faster in runtime compared to the best-performing meta-heuristic Bayesian Optimization. Lastly, the framework was benchmarked against the 'learning to optimize' study and shows excellent performance on the logistic regression problem compared to the autonomous optimizer. In conclusion, it has been shown that even with the limited information of black-box optimization problems, data-driven optimization effectively improves the current standard of materials engineering processes.

Inverse design with topology optimization has followed the same computational
graph for decades. The unknown material density is distributed within a domain,
a computational analysis predicts the response of that design and its derivative
with respect to the unknown, and this information is used by a chosen gradient­
based optimization algorithm to find the next design iteration until it reaches an
optimum (local or global). Recently, however, a counter­intuitive strategy was pro­
posed which augments the computational graph by including a neural network in
between the response prediction (computational analysis) and the generation of a
new design (image). This shifts the optimization problem from its original space
to the weight space of the neural network. Yet, this indirect optimization pro­cess was shown elsewhere to outperform conventional topology optimization for a
large number of structural compliance problems – at least when choosing a par­ticular convolutional neural network (part of a U-­Net) and a particular optimizer
(L­-BFGS). This investigation provides quantitative and qualitative arguments that
justify why these choices are successful, concluding that the line­-search compo­nent of L-­BFGS is key to traversing the reparameterized objective (loss) landscape
and quickly reaching good solutions in “flat” regions of the landscape. Importantly,
these topology optimization problems are not stochastic which make them different
from the majority of conventional deep learning applications, favoring the use of
line-­search. Similarly, although to lesser extent, the approximation of the Hessian
provided by L-­BFGS helps moving more effectively within the flat regions by rescal­ing the gradients, the quality of the approximation is less relevant. Together with
the deep image prior effect associated to deep learning, these arguments explain
the early success of the neural reparameterization strategy in topology optimiza­tion, in spite of the non­-convex objective landscape distortion that they introduce
even for landscapes that were originally convex.

The existing drug development process is economically and scientifically challenging. It fails to efficiently emulate human physiology in-vitro with the current pre-clinical studies which includes in-vitro cell culture models and animal testing. Organ-on-Chip (OoC) technology aims to recreate an in-vivo like micro environment to investigate drug response more effectively. There are ongoing attempts to fabricate OoC technology as a single-platform microdevice to minimize its reliance on external components. In this perspective, the functionality and throughput of this technology can be improved. One such novel approach is addition of an ionic electroactive polymer (iEAP) actuated diaphragm micropump. The primary aim of this thesis project was to determine the suitable dimensions of a micro cantilever iEAP, specifically Ionic polymer metal composite (IPMC) to generate appropriate flow rate for the projected diaphragm micropump. In addition to that, dynamics of the IPMC cantilever actuator was examined in dry environment. To achieve this the actuator tip - force, tip-displacement and longevity tests were performed. The results at macroscopic scale were tentatively explained with molecular characteristics of the material. As a result, it was shown that an IPMC cantilever actuator of millimetric size possesses viscoelastic properties and classical mechanical theories cannot be used to validate the experimental results. Secondly, the actuation results for 0.1 and 1 Hz align with the input driving frequency. The IPMC cantilever of length 7 mm generates the maximum tip-force of 0.138 mN and it is suggested to be used as a diaphragm actuator for the upcoming micropump.

Various engineering applications rely on efficient, high performance materials to overcome design challenges. This high performance can be achieved by engineering micro-heterogenous materials also known as composites. Since the behavior of composites relies heavily on micro-scale interactions between different components, modeling macrostructures with fully-represented microscopic geometry is needed. Thus, the standard finite element modeling approach becomes impractical. Computational homogenization, also known as concurrent finite element analysis (FE$^2$), is a method that is employed to model materials with distinct multi-scaled structure. FE$^2$ employs the concept of embedding a representative volume element (RVE), at each integration point of the macro-scale problem and obtaining the macroscopic constitutive behavior through homogenization, thus bypassing the need to develop a macro-scale constitutive model. Although it succeeds in upscaling the microscopic material behavior accurately, this method comes with the major drawback of being computationally expensive due to its nested structure. Developing methods to bypass the aforementioned computational bottleneck of FE$^2$ is an ongoing research endeavor. Employing machine learning algorithms to create surrogate constitutive models for microscopic behavior is one possible approach. However, creating surrogate models is not an easy task. A thoroughly collected training set is needed for the surrogate model to be representative. Thus, investigation of surrogate model creation strategies with machine learning while trying to reduce the computational burden of this \textit{offline} training process is a compelling area of study. Gaussian Process Regression (GPR) is a probabilistic machine learning model. It can be utilized to create surrogate constitutive models effectively in the aforementioned context. Moreover, the computational burden of the training procedure can be decreased by extending the conventional GPR technique into a co-kriging regression with multi-fidelity information (multiGPR). This surrogate modeling strategy reduces the need to collect high-fidelity information by collecting information from a low-fidelity model thoroughly. Thus, enabling accurate training datasets to be created from a less representative, but computationally less taxing models. In this work, the multiGPR approach is used to construct accurate and efficient surrogates for the behavior of fiber-reinforced composite materials. Training data is obtained from selected RVE configurations that consist of linear-elastic fibers randomly embedded in a matrix that has pressure-dependent plasticity and used to train single and multi-fidelity GPR constitutive models. Both approaches are trained with various combinations of loading scenarios and their prediction capabilities are investigated to represent the training cases in addition to their prediction capabilities under unseen load cases.

Metamaterials are a relatively new group of materials whose behaviour strongly depends on the design of their internal structure. They can be employed in a wide range of applications, one of which is presented in this thesis. As cardiovascular diseases account for around 30% of deaths worldwide the research done in the field of Materials Science may find a real life use in the form of a magnetically activated heart assisting device. Such a structure was designed on the basis of a newly developed magnetostrictive material with the use of finite element simulations and machine learning based analyses. The computational approach enabled the investigation of the structure’s deformation and determination of the influence of parameters, which define the metamaterial’s geometry, before commencing prototyping and experiments. Geometry, which resulted frommultiple iterations necessary to match the heart’s shape and deformation patterns, was parametrized and simulated in each configuration to create a database. Regression was performed on it with Artificial Neural Networks and Sparse Gaussian Process Regression in order to predict possible bounds and importance of specific parameters. At the subsequent stage the structure was optimized, with the aim of matching the deformation of a healthy myocardium, which concluded the project. Obtained sections of the design space allowed for qualitative as well as quantitative description of the device’s capabilities. It was established that most of the behaviour in all directions, longitudinal, radial and rotational, is determined by position of the main active element within the structure as well as its size and the size of vertical elements encompassing the myocardium. Optimization process confirmed the predictions and led to the first magnetically activated, metamaterial based design of a heart assisting device.

Mechanical metamaterials are a new emerging class of materials which achieve properties outside the bounds of conventional materials. A metamaterial consists of a unit cell which is periodically repeated in space. In this study, a new metamaterial unit cell is proposed, derived from a class of space structures known as deployable masts. What makes these masts particularly interesting is their ability to contract to a fraction of their original length. In order to use such a structure as a unit cell, requires a deep understanding of the design parameters impact on material response. To guide this project, a novel data driven approach to design will be implemented. Here, computational simulations are used to create a database of mechanical responses, which in turn is used to model the relationship between input and output responses. This approach essentially flips the conventional approach of mechanical design on its head by using computational simulations to define the design space before manufacturing and testing. This replaces designer intuition with predictive charts, becoming increasingly useful for non-intuitive problems. This study validates the data driven approach through mechanical testing of a metamaterial unit cell. This testing is done at the macroscopic scale, utilizing a hobbyist 3D-printer (Ultimaker 2) to manufacture the structure. This study demonstrates that the material model is capable of accurately predicting the unit cell response. The limitation and possibilities for fused deposition modelling printed parts to be used as functional components is also investigated. Based on the insights gained from the data driven design process and experimental validation, design parameters are proposed for which a metamaterial unit cell exhibits both extreme compressibility and a high compressive strength.