Injection molding is a widely used process for manufacturing a large range of polymer parts, including fiber-reinforced polymer composites. The complex nature of this process, combined with the non-Newtonian behavior of molten polymers, implies significant challenges in terms of control, prediction, and optimization of the processing parameters that influence manufacturing quality. While various computational and numerical models have been proposed to simulate the injection molding process and ensure the desired part quality (see, for instance, industrial software such as Moldex3D or Moldflow, they bring their own challenges, including high computational costs, the need for solver tuning, frequent parametric studies, and the economic burden associated with licensing.

In this study, machine learning surrogates are explored as an alternative to conventional modeling software. They are based on Graph Neural Networks (GNNs) that have recently demonstrated remarkable potential for simulating physical systems. In a graph, physical systems are encoded as nodes end edges, capturing their organization. Then, state-of-the-art techniques such as the message passing framework are tested for their potential to learn physics. In the long term, this project aims to establish a novel graph-based framework capable of efficiently simulating complex physics across various material models and processing conditions.

A numerical model is developed for 2D mold filling and quantitively validated against a benchmark. The prediction results demonstrate that the learned simulator is able to capture the physics of melt front behavior. Further, the prediction runtime is significantly faster than the ground truth, completing in under 1 second compared to \textasciitilde hours, representing a substantial improvement in efficiency. However, the accuracy of the predictions can vary, with errors accumulating especially during testing. While the scalar loss metric provides an estimate of the model's performance, it lacks expressiveness and can be unreliable on its own.

We perform a study on acoustic metasurfaces, aiming to achieve simultaneously low resonance frequencies (below 400 Hz), high attenuation bandwidth (greater than 200 Hz), and high attenuation coefficient magnitudes (above 0.8), while maintaining a surface-like structure.
We propose the implementation of geometrical optimization through genetic algorithms, as well as the incorporation of a chamber to induce resonator coupling in a supercell hexagonal Helmholtz resonator metasurface, to achieve the stated objectives simultaneously.
Results show that genetic algorithms can effectively increase the attenuation bandwidth while maintaining a moderate attenuation coefficient magnitude. Incorporating a chamber induces resonator coupling, causing frequency locking and pulling phenomena. A narrow chamber can effectively lower the resonance frequencies and enhance the attenuation coefficients at those frequencies, while maintaining a surface-like structure. However, incorporating a chamber may lead to a reduction in bandwidth. By combining the genetic algorithm optimization with chamber integration, we observe a significant reduction in bandwidth narrowness, while the benefits of frequency locking and pulling are maintained.
In conclusion, genetic algorithms have the potential to achieve wide attenuation bandwidths, while chamber incorporation holds promise for attaining low resonance frequencies with high attenuation coefficients. Using both methods simultaneously may enable the achievement of all objectives.

The detrimental effects of Cu contamination during steel recycling and production are primarily due to the segregation of Cu at Fe grain boundaries (GBs). A promising approach to mitigate these effects is the introduction of alloying elements that inhibit Cu segregation at Fe GBs, which can be investigated through atomistic simulations. Currently, popular simulation methods include density functional theory (DFT)-based simulations and empirical interatomic potentials (EIPs)-based simulations. DFT calculations can provide simulations with high accuracy, while the high computational cost limits their application to simpler GB structures. Moreover, simulations utilizing EIPs may lack reliable potentials, especially when predicting GB segregation tendencies across a broader range of alloy systems. To address these challenges, universal machine learning interatomic potentials (uMLIPs), which are trained on DFT data and applicable for most of elements, have emerged as a promising alternative. Although uMLIPs have shown potential in various materials simulation tasks, their reliability for out-of-distribution tasks, such as simulating GB segregation behavior, remains unproven.

This thesis project evaluates the performance of the best available uMLIPs, specifically MACE-MP-0, CHGNet, M3GNet, and SevenNet-0, in predicting single-solute GB segregation energies, GB energies for both body-centered cubic (BCC) Fe and face-centered cubic (FCC) Fe systems, and solution enthalpies for BCC Fe and cementite. The results were compared against existing studies conducted via DFT calculations to assess the accuracy and applicability of each uMLIP. Additionally, some EIPs were tested for comparison, serving as an extra reference. The findings reveal that MACE-MP-0 generally outperforms the other uMLIPs in both accuracy and stability of convergence. While all tested uMLIPs perform well in BCC Fe systems, CHGNet(v0.2.0) and SevenNet-0 show reduced accuracy in FCC Fe simulations. Although most of the simulations using uMLIPs converged well in BCC Fe GBs, many unconverged cases were reported in FCC Fe systems, particularly for uMLIPs other than MACE-MP-0 and CHGNet(v0.3.0). Furthermore, a consistent underprediction of segregation tendencies for highly segregating solute elements, such as Cu, is observed in the results of MACE-MP-0 and CHGNet. This suggests that while uMLIPs hold significant potential for atomistic simulations, fine-tuning pre-trained uMLIP models for out-of-distribution tasks, such as calculating GB segregation energy, is recommended to improve accuracy and convergence behavior. This work offers a valuable benchmark for using uMLIPs in future GB segregation studies.

For the use of screening potential glassmaking recipes the seeding method has been applied to a Molecular Dynamics simulated CaO-SiO₂ system in order to attain the parameters for the Classical Nucleation Theory and construct a Time-Temperature-Transformation (TTT) diagram. Using this TTT diagram, the non-crystallisation temperature and the critical cooling rate of the material was determined, two quantities important for the prevention of crystallisation during the glassmaking process. The implementation of the seeding method on the CaO-SiO₂ system involved creating a novel local bond order based detection method for distinguishing the crystalline structure of Wollastonite-1A from the glass melt. Regrettably this method had less than desirable precision which resulted in results for the nucleation rate that can only be used qualitatively. In contrast the results for the crystalline growth rate can be used quantitatively, while the resulting TTT diagram again can only be used qualitatively.

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.

Quenching and partitioning (Q&P) is a novel processing route that can create microstructures that combine high strength, ductility and easy formability without the excess use of alloying elements by also keeping the production cost low. Therefore, this method is ideal to be applied to steels that are used in the automotive industry. The material is initially annealed at a high temperature to form austenite. Then it is quenched to a temperature below the martensite start temperature to form a controlled fraction of martensite. Second annealing follows, at a lower temperature than the first one, that stabilises the austenitic phase at room temperature. It has been found that the austenite grain size at the end of the first annealing (prior austenite grain size) has a significant impact on the processing method, the final microstructure and consequently on the final properties of the material. In this thesis, phase-field modelling was employed to investigate the role of prior austenite grain size on the quenching and partitioning process. Simulations of the processing were performed for three different prior austenite grain sizes (6, 25 and 67 μm). Microstructures consisting of about 10.5% austenite and 89.5% martensite were created by simulating the first quenching of the Q&P process. Then, the evolution of the microstructure and carbon distribution were investigated for partitioning at 400 ◦C for times up to 4000 s, and quenching to room temperature. The results gave insights into the effect of the prior austenite grain size on the morphology of the microstructure and the kinetics of carbon during partitioning. It was found that a finer microstructure was derived with decreasing PAGS. Moreover, partitioning becomes more efficient because of the shorter distance that carbon has to diffuse, and the even spatial distribution of austenite. Finally, wide grain size distributions of the austenite that is present during partitioning can lead to significant variations in the carbon content among the austenite grains of the final microstructure, especially for short partitioning times. This affects the efficiency of the partitioning process and has an impact on the mechanical stability of the austenite in the final microstructure and consequently on the TRIP phenomenon.

Laser surface treatments offer interesting prospects for the creation of architectured microstructures formed by distinct phases in metastable austenitic steels. In the present study, a laser-based localised heat treatment was developed to locally create an austenitic region in a quenched Fe25Ni0.2C martensitic microstructure. The highly localised laser heat flux gives rise to high spatial gradients in peak temperature and heating rate. This results in strong variations in the microstructures observed over short distances, which are related to local changes in the martensite to austenite phase transformation temperatures and formation mechanisms, the occurrence of grain growth and recrystallization in the newly formed austenite, and the tempering of the initial martensite. Moreover, thermal stresses and surface effects influence the final microstructure. In this work, effects of heating rates and peak temperatures are studied by dilatometry whereas Electron Backscatter Diffraction (EBSD) and optical microscopy are used to assess austenite grain size and morphology. This information is linked to a Finite Element Model of the local thermal history to investigate the evolution of the local microstructure throughout the zone affected by the laser heat source. This research provides insight into localised microstructural control of steels with laser surface treatment and provides a thermal model for detailed understanding of the mechanisms controlling the microstructural changes taking place during these treatments.

Metamaterials are a new class of materials where the properties crucially depend on the design of the unit cell that is periodically repeated in space. In this study a new metamaterial unit cell concept has been proposed, inspired by a class of space structures known as deployable masts. The ability of these structures to contract to a fraction of their size made them suitable candidates for energy absorbing applications. One of the main design targets of energy absorbers is the ability to tune the material response to specific applications. Tunability of the mechanical response of the metamaterial concept requires deep understanding of the influence of design parameters. The prime focus of this study was to gain this understanding via data-driven insights. Conventionally, the design of a new material is carried out by making educated guesses about the design parameters and subsequently performing expensive and time-consuming experiments. In this work computational simulations were utilized to create databases of mechanical responses. These databases are later used to model the relationship between the inputs and the output response. This generated the issue of how and what method to use to effectively determine this relationship. This work explored state-of-the-art machine learning methods to enhance a recently proposed data-driven framework with thegoal of designing a new super-compressible metamaterial with large energy absorption. Importantly, the data-driven design process included the influence of manufacturing imperfections on the mechanical response of the metamaterial.The study revealed that by tuning the design parameters, significantly different mechanical response of the structure was achievable. The proposed learning model has enabled mapping of the influence of design parameters in the design space, moreover the sensitivity to those parameters varied across the design space. The increased energy absorption has been attributed to the resistance to bending of the main load carrying components of the design. It was demonstrated that the number of those components and the elastic modulus were scaling factors for the quantities of interest. Based on the insightsgained, a unit cell metamaterial design with significantly improved energy absorbing capability was proposed.

Piezoelectric energy harvesting technology is an alternative source for powering low power electronics. The ability of piezoelectric materials to convert ambient vibrational energy into usable electrical energy is seen as a promising battery-free solution to be used in inhospitable areas to self-power electronics for a longer time with little maintenance. The goal of this research is to maximise the mechanical input: elastic strain energy experienced by a lead zirconium titanate (PZT) bilayer piezoelectric buzzer by introducing four different boundary conditions during loading. Investigation of how mechanical input from each of these boundary condition influences the stored electrical output is carried out. The effect of three different static loads on each of the stored electrical output for four boundary conditions are studied. In order to accurately compare the data, the mechanical input (elastic strain energy) is calculated while the electrical output (stored energy) is measured experimentally. Given the brittle nature of ceramics, the maximum load bearing capacity for the PZT ceramic disc is determined by mechanical tests such as the ball-on-the-ring and uniaxial compression tests. This is to ensure that the experiments do not fracture the sample. Results show that, by inducing a bending mode in the buzzer, mechanical input values almost four orders of magnitude higher can be reached when compared to boundary conditions without bending modes. A similar result was found for the stored electrical energy with values for the bending mode of almost three orders of magnitude higher than when no bending is involved. The comparison between the calculated mechanical input and the measured electrical output shows good agreement in the boundary conditions involving bending. The energy conversion is highly efficient for the full range of applied loads for these boundary conditions. In the case of non-bending boundary conditions, the stored electrical energy is one order of magnitude high than predicted for the mechanical input and thus the model is in poor agreement with experiment.

The Pd-based metal membranes have been focused on many studies, because they are the most promising candidates for hydrogen separation. If Pd-based membranes are designed in nanoscale, then the introduced high volume fraction of grain boundaries can act as fast diffusion paths for hydrogen atoms that leading to an increase of hydrogen separation rate. However, due to the substantial amounts of grain boundaries, the nanocrystalline material tends to show a rapid grain growth at operating temperature (that leading to the coarse grains), which may lead to the failure of this membrane in further applications. However, grain growth could be effectively inhibited by thermodynamic stabilization method-that is solute segregation to grain boundary region. In this thesis, a thermodynamic grain boundary segregation model is developed and is applied to many types of metallic binary alloys, including the Pd-Cu, Pd-Zr, Pd-Y, Fe-Zr, Y-Fe and so on. From this model, the segregation tendency of alloying constituents is determined, and the solute concentration in grain boundary is obtained. Moreover, the grain boundary energy (GB) of investigated alloys can be calculated via this model, by comparing the calculated GB energy with that of pure Pd, we are able to determine whether these alloys can be stabilized by the segregation-introduced thermodynamic stabilization. These calculated results have been compared with those reported in the literature, verifying the predictability of this model. In addition, X-ray diffraction (XRD) analysis, including the XRD residual stress measurement and X-ray peak broadening analysis have been performed on both Pd_0.7Cu_0.3 and pure Pd thin films, the experimental results will help us determine whether the solute segregation occurs and whether this alloy can be stabilized by the solute segregation.