Dissertations and Theses (Open Access)

Date of Award

Fall 8-2021

Degree Name

Master of Science (MS)

Advisor(s)

LUIS G. LEON-NOVELO, PhD

Second Advisor

MICHAEL PHILLIPS, PhD

Third Advisor

GOO JUN, PHD

Abstract

For this thesis, we utilized random forest (RF) algorithm in combination with Quality by Design (QbD) in simulated commercial-scale penicillin production dataset. The RF algorithm was primarily used in the selection of critical process parameters (CPPs). In this analysis, the primary critical quality attributes (CQA) was set to be the final penicillin yield. After the CPPs were selected by the RF algorithm, we would propose design of experiment (DoE) by using fractional factorial design to establish mathematical relationships between CPPs and CQAs. Among the five RF models, we chose two models before proceeding to the DoE step. One using the bioreactor state at the process time of 60 hours, and the other leveraging all of time points through 230 hours. Of these two models,  the 60h time point was selected due to having the most qualitatively similar result as to the reference paper. The other one model used all time points, and we chose it because of its low RMSE value. The conclusion that we derived from this analysis was that the utilization of random forest algorithm is not inferior to the combination of FMEA with traditional multivariate statistical method and could aid to eliminate the determination of criticality step in traditional QbD approach.

Download

Share

COinS