(340ad) Multiscale Systems Engineering for the Development of Sustainable Technologies

Multiscale systems engineering across molecular to infrastructure scales is essential to realize sustainable technologies. Membranes enabled by advances in nano engineering devices offer the advantages of solute-targeting (tailorability), modularity, and low energy separations for several applications such as water treatment for trace contaminants, resource recovery such as uranium and lithium from sea water, hydrocarbons from crude oil, and alkenes from alkanes. Yet the promise of nanoengineering has been slow to manifest because novel membranes are rarely evaluated in the context of a process at scale to assess sustainability, feasibility and economics of the application in realistic feed conditions for example, the inclusion of trace contaminants. The proposed multiscale-systems engineering framework could address this challenge.

Current work. My research develops molecular-to-systems engineering frameworks to enable novel membrane separations for heavy metal (e.g., Pb) remediation from water and Li recovery for LIB recycling. The key pillars of our framework (attached figure, described below) are general and widely applicable to virtually all facets of sustainable technologies.

Materials-process targeting: Process targeting frameworks make use of successive analyses with restrictive assumptions to quantify the feasibility limits of integrated processes. Process targeting has been successfully applied in coal-to-liquids processes, gas separation networks, and crude production. We recently published an easy-to-use multiscale materials-process targeting framework to screen candidate sorbents for specific applications. We analyze recently published sorbents and show Pb remediation and Li recovery separations are systems, not materials, limited and identify the device-scale research challenges

Superstructure optimization: Superstructure optimization provides a mathematical framework to search through several alternative flowsheet configurations simultaneously to find the optimum configuration. We propose a superstructure optimization framework to design a novel membrane-based diafiltration cascade, in which a dialysate buffer is used to offset concentration effects, to recover metal rich solutions in the fractionation step of a lithium-ion battery recycling process. Treating the membrane areas, flows and concentrations of all streams as decision variables, we search through multiple, complex recycle and feed injection strategies using an epsilon-constrained method to identify Pareto-optimal configurations. The framework quantifies trade-offs in binary solute recovery, staging complexity vs membrane area, and defines design heuristics for continuous diafiltration cascades enabled by the analysis of thousands of optimal designs.

Data informed design of experiments: The knowledge of processes that govern solute selectivity and membrane performance in multicomponent feed systems is necessary to advance membrane processes. We use nonlinear regression to fit time-series data from membrane characterization experiments run in diafiltration mode to fit a family of differential-algebraic equation models for the system. The synergy between data analysis and diafiltration experiments, including an upgraded experimental setup demonstrated five-times faster characterization of membranes which in turn enables the efficient building of predictive models. The current work lays the foundation for a collaborative (experimental-theoretical) model-based design of experiments to improve parameter precision and enhance model discrimination.

Bayesian uncertainty quantification: Model simplifications are necessary for multiscale analysis but add bias and over/under confidence due to epistemic uncertainty. We propose a Bayesian hybrid model which contain a simplified mechanistic model, a data driven Gaussian process (GP) model, and random noise. The outputs of the hybrid model encode uncertainty information as probability distributions which may be used to make future decisions under uncertainty. Using the Bayesian hybrid model based decision-making framework, we demonstrate accurate decision-making despite epistemic uncertainty in a ballistic firing and reaction engineering studies and optimal training strategies for the hybrid model.

Future directions. Data-science and new applications provides exciting opportunities to augment multiscale systems engineering frameworks by enabling top-down design:

Materials informatics: I plan to develop materials informatics methods to automatically learn structure-property relations by leveraging massive online materials databases. The relationships provide a bridge to connect materials-process targeting and superstructure optimization with (macro)molecular engineering to realize inverse design of novel materials and systems optimized for specific separations. For example, the optimizing the self-assembly of nanostructured polymers into monolithic columns with microchannels can improve the capacity of adsorbents thereby decreasing material requirement for future sustainable separations leading to potential economic benefits.

Bayesian optimization: Black-box optimization facilitates the solution of otherwise intractable problems, such as the navigation of vast materials databases for candidate materials. Bayesian optimization uses self-learning and uncertainty information to balance exploitation (best prediction) versus exploration (highest uncertainty) to efficiently navigate design spaces. I intend to use Bayesian optimization to find candidate materials for new device-scale integration techniques such as the discovery of such as ideal compositions of ternary polymer-solvent-nonsolvent inks which will enable additive manufacturing of nanostructured polymers.

Other applications. Multiscale system engineering is essential for all aspects of sustainable technology; as an independent researcher, I am eager to adapt these system engineering paradigms to exciting new application areas. For example, in reaction engineering, the discernment of reaction pathways (data driven design of experiments) along with accurate uncertainty information (Bayesian uncertainty quantification) can inform process flow sheet development (superstructure optimization) and the synthesis of efficient heat and mass exchange networks (materials-process targeting framework). In the field of electrochemistry, knowledge about the degradation mechanism of electrodes can inform material selection, cell design, operating strategies, and scale-up feasibility.

Career goals and teaching interests:

As a 5^th year Ph.D. student at the University of Notre Dame (advisor: Prof. Alexander Dowling), I am actively seeking an Industry position starting Summer or Fall 2022. Specifically, I am looking to diversify my research portfolio by developing skills in machine learning, computational optimization, data-driven materials discovery, learning about new technologies. My research has been recognized by the Chemical and Biomolecular Engineering Exemplary Candidacy Award and the Patrick and Jana Eiler’s Graduate Fellowship, both awarded by the University of Notre Dame. My work as a Teaching Assistant for the Chemical Engineering Laboratory, Thermodynamics, and Numerical and Statistical Analysis courses was recognized with a Notre Dame Learning Outstanding Graduate Student Teacher Award. I enjoy working in diverse teams and sharing my work with non-experts and have taught introductory coding workshops to new graduate students.