Data-Driven development of layer-by-layer nanofiltration membranes and processes

• Datengetriebene Entwicklung von Layer-by-Layer Nanofiltrationsmembranen und -prozessen

Rall, Deniz Patrick; Wessling, Matthias (Thesis advisor); Szymczyk, Anthony (Thesis advisor)

Aachen (2020, 2021)
Book, Dissertation / PhD Thesis

In: Aachener Verfahrenstechnik series - AVT.CVT - chemical process engineering 8 (2020)
Page(s)/Article-Nr.: 1 Online-Ressource (xi, 223 Seiten) : Illustrationen, Diagramme

Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2020

Abstract

Synthetic membranes provide an essential technological basis for modern drinking water treatment and disposal. A quest for ion selectivity becomes increasingly important, as opposed to the conventional task of high ion retention. This quest requires a new versatile membrane fabrication approach and screening pathway. By now, exploring the design space of synthetic membranes is generally based on an evidence-based experimental effort relying on a large number of trails through an educated guess and experimental design context. This thesis offers a novel data-driven methodology using artificial neural networks (ANNs) for the development of layer-by-layer nanofiltration membranes in contrast to established screening pathways. Using statistically substantiated predictions through machine learning techniques provides the possibility to skip a large number of these experimental steps. Furthermore, machine learning enables a well-founded physical analysis of the performance spectrum, creating an interface between the membrane in the laboratory and its model-based optimization. Starting from an extensive data-set of layer-by-layer nanofiltration membranes, this thesis presents for the first time that an ANN can predict and improve ion retention of salts and water flux values based on given membrane synthesis protocols. An innovative deterministic global multi-objective optimization identifies the upper bound (Pareto front) of the delicate retention and permeability trade-off. Next, an extension by a mechanistic ion transport model offers hybrid modeling systems that are embedded in a state-of-the-art membrane process model to simultaneously design the membrane synthesis protocols along with the process layout yielding favorable results immediately. Ultimately, hybrid modeling bridges the gap between knowledge creation at the small-scale (local transport models) and decision making at large-scale (process optimization). These steps pioneer solving complex multi-scale process optimizations in membrane science and enable decision making across multiple scales with accurate transport models. The presented data-driven development methodology unleashes material scientists and engineers to overcome limitations in membrane material development - enabling fast, non-intuitive solutions that may remain hidden when only considering conventional screening pathways.