Date of Award

6-2026

Document Type

Thesis

Publisher

Santa Clara : Santa Clara University, 2026

Department

Bioengineering

First Advisor

Prashanth Asuri

Second Advisor

Hamed Akbari

Abstract

Membrane bioreactors (MBRs) are increasingly employed in biomanufacturing due to their ability to produce high-quality effluent and support complex biological processes. However, membrane fouling remains a critical operational challenge, driving increased energy consumption, frequent cleaning cycles, elevated transmembrane pressure (TMP), and accelerated membrane degradation. Conventional operator-driven cleaning strategies are reactive by nature and insufficient for managing the dynamic, nonlinear fouling behavior characteristic of full-scale MBR systems.

This study presents a data-driven machine learning pipeline for predicting TMP and fouling behavior in an industry-scale MBR system, with the goal of optimizing cleaning schedules and extending membrane lifespan. A Random Forest regression model was trained and validated using k-fold cross-validation on a dataset of over four million data points collected across four distinct operational phases. A structured preprocessing pipeline was developed to address missing values, normalize sensor readings, and extract statistically meaningful features including pre- and post-backwash TMP and standard deviation of backwash cycles. Model performance was evaluated using root mean square error (RMSE) and coefficient of determination (R!). Model Predictive Control (MPC) was subsequently integrated to translate predictive outputs into actionable flux optimization strategies, increasing filtration duration from 474.5 seconds at baseline to approximately 4,160 seconds while reducing fouling rate and cleaning frequency. Visualization outputs were designed to prioritize interpretability for facility technicians, centering on commanded flux and TMP trend graphs over more complex dimensionality reduction techniques.

Results confirm that the reconstructed TMP signal aligns closely with expected trends from literature, validating data integrity prior to model training. The proposed pipeline demonstrates that machine learning can meaningfully support predictive maintenance, reduce operational costs, and extend membrane service life in both wastewater treatment and biomanufacturing contexts.

Download

Share

COinS