Recent advances in high-throughput experimental and computational techniques have led to an accelerated discovery of new materials. On the contrary, development of processing routes required for sustainable manufacturing of advanced materials is lagging behind.
At 4D ICE, we aim to fill this gap by developing computational tools for rapid and systematic discovery of time-, energy- and cost-efficient processing routes for the synthesis of materials with tailored functionality. Our work, which lies at the triple-point of microstructure modeling, machine learning, and high-performance computing aims at enabling controllability in advanced materials processing.
Optimal control of microstructural evolution during materials processing is an unexplored area. Key fundamental questions within this field need to be answered, for example: How do processing parameters influence microstructural evolution? How does one optimally control processing parameters to tailor advanced materials? What are the limits to types of materials that may be fabricated given a processing method? What are the minimal requirements for processing that facilitates the synthesis of a material with specific properties and behavior? Our research group will address these questions by coupling phase-field method (PFM) with machine learning algorithms (MLA) for determining optimal materials processing trajectories. Both these techniques will be seamlessly integrated into a fully-parallel computer framework, for instance, via closed-loop control strategy, as shown above.
Computational approaches to design optimal materials processing routes involve two components: (i) comprehending the behavior of control parameters that modulate microstructural evolution by developing predictive multicomponent, multiphase-field modeling approaches, and (ii) coupling PFM with computer-based MLA for detecting time-dependent process trajectories. The group will lead the research in both these areas by developing model-agnostic control tools that are transferable between different models.
I. Routes to controlling microstructural evolution
(a) Rapid solidification processing
Rapid solidification (RS) processing is one of the most common mechanisms that governs the microstructural evolution, and consequently, the mechanical and functional properties of components fabricated via additive manufacturing (AM). The state-of-the-art PFM are known to be reasonably accurate in terms of predicting microstructural evolution close to equilibrium. However, lack of microstructure models that can account for the non-equilibrium multiphysics accompanying dynamic melt pool viscosity, high solidification rates, complex thermal cycling, and convection, considerably limits our capabilities of establishing optimal process control in AM.
To enable controllability in AM, the group will formulate predictive PF models for microstructural evolution during RS of multicomponent alloys. This would allow us to deduce the process parameters that can be optimized to achieve desired material behavior. In due course, we will expand our computational framework to optimize process control down to atomic length scale and diffusion time scales (beyond what Molecular Dynamics or Monte-Carlo algorithms can achieve), by developing phase-field-crystal (PFC) models for RS.
(b) Self-assembly in soft-matter
Nanoscale self-assembly of polymers makes them highly promising candidates for numerous functional applications in soft matter nanotechnology, including optics, electronics, and acoustics. However, the absence of robust control strategies has considerably limited their commercial viability.
A key prerequisite for establishing optimal process control in polymer processing is a knowledge of how external agents, such as temperature gradients, solvent annealing, directional epitaxy, shear, and electromagnetic fields [Phys. Rev. E (2015), Phys. Chem. Chem. Phys. (2016)], modulate polymer nanostructures. The group will address this challenge by adopting a hierarchical modeling approach based on PF and PFC methods. In future, we will devise analogous frameworks to optimally control visible light-mediated self-organization in biological cells and organelles for biomedical applications.
(c) Proliferation of material flaws
Improper materials processing commonly lead to unwanted flaws such as porosity, cracking, poor surface finish and deformation. While the limited efforts on enhancing processing control have mostly focused on accelerated, and cost-efficient routes of materials manufacturing, strategies to simultaneously inhibit the proliferation of undesirable, deleterious flaws have rarely been explored. The group will develop predictive PF and PFC models to study the influence of various process parameters on microstructural flaws propagation. In future, we will devise optimal control strategies to minimize flaws in processing.
II. Coupling PFM/MLA for optimal process control
Materials processing implies reaching the desired end-state, N, by starting from a known initial state via a series of intermediate states, k. It involves the control of isolated or combined external stimuli, f, such as, electromagnetic, thermal, mechanical, and convective fields, or selective removal/aggregation of matter, all of which may vary in space and time. Therefore, development of tailored processing routes for obtaining desired material properties involves exploring an infinite-dimensional space, s, of time-dependent trajectories through a suitably defined space of material states and processing parameters.
The research team at 4D ICE will develop specialized analytical and computational tools to generate such optimal processing trajectories. Recently, we designed a feedback control system based on supervised learning to optimally guide microstructural evolution as illustrated in the adjacent figure.Thus, our computational approach allows us to compute trajectory-dependent functionals, such as, how does a finite change in thermodynamic variables influence the cost function, which, in turn, will guide the materials processing.
In future, we will formulate novel control algorithms based on the concept of unsupervised learning and couple them with PFM/PFC to determine optimal processing trajectories in materials processing. Since every step of PFM/MLA coupling is amenable to parallelization, we will resort to high-performance computing to achieve optimal control under realistic processing conditions.