(415b) Machine Learning Assisted Anti-Biofouling Modification of Commercial Reverse Osmosis (RO) Membranes

Reverse Osmosis(RO) is a vital technology to our water purification industry. In USA, most of the regional water purification plants uses Reverse Osmosis (RO) to desalinate and purify water for regional pure water supply. However, the RO membranes used to produce pure water is prone to fouling, and leads to higher operating pressure, flux decline. Frequent chemical cleaning is required to maintain the optimum production performance but it severely shortens the membrane life cycle. Studies on RO membrane autopsies of regional plants reveal a severe decline in total flux with significant decline in rejection performance before their intended lifecycle. Our autopsy on Grand Forks Regional Water Treatment Plant reveal a membrane surface coated with a gelatinous foulant. It is mostly organic (microorganism shells, Gram-negative bacteria and Gram-positive bacteria) in nature (98% of dry mass).

Different process conditions can instigate fouling, which can occur through different mechanisms. However, for most cases, biofouling occurs as the microorganism deposits, adhere, propagate on membrane surface causing an external barrier to separation. The initial microorganisms can be classified as simple chain of amino acids. In this research, a ML assisted Anti-biofouling modification of commercial RO is initiated. The commercial polyamide RO (Dupont XLE) membrane surface is activated by “carbodiimide” chemistry. The search for the ML assisted anti-biofouling additive is initiated by mining of reported hydrophilic and lipophobic polymers to build enough database for a supervised ML model. Some polymeric membranes are naturally antibacterial and have low fouling properties, however the search for an ideal additive can be a herculean task. Our tree based regression model (RF) is interpreted by applying SHapley Additive exPlanations (SHAP). This lead to identification of features or group of atoms with positive and negative contributions towards anti-biofouling properties. A reference point is created to initiate the screening algorithm to screen 19,233 polymer motifs from PolyInfo database for a potential for anti-biofouling properties. Two of the polymer additives were chosen based on their possible participation in carbodiimide chemistry for the surface modification. The modified membranes were characterized using FTIR, XRD, SEM and tested for water permeation and salt rejection. Anti-biofouling properties were tested with static protein adsorption and fluorescent staining.