Abstract
This study presents a broad perspective of hybrid process modeling andoptimization combining the scientific knowledge and data analytics inbioprocessing and chemical engineering with a science-guided machine learning(SGML) approach. We divide the approach into two major categories. The firstrefers to the case where a data-based ML model compliments and makes thefirst-principle science-based model more accurate in prediction, and the secondcorresponds to the case where scientific knowledge helps make the ML model morescientifically consistent. We present a detailed review of scientific andengineering literature relating to the hybrid SGML approach, and propose asystematic classification of hybrid SGML models. For applying ML to improvescience-based models, we present expositions of the sub-categories of directserial and parallel hybrid modeling and their combinations, inverse modeling,reduced-order modeling, quantifying uncertainty in the process and evendiscovering governing equations of the process model. For applying scientificprinciples to improve ML models, we discuss the sub-categories ofscience-guided design, learning and refinement. For each sub-category, weidentify its requirements, advantages and limitations, together with theirpublished and potential areas of applications in bioprocessing and chemicalengineering.