Physics-Informed Machine Learning: the Next Big Trend in Food Process Modelling?

Abstract

The goal of this short review is to introduce a newhybrid modelling approach, i.e. physics-informed machine learning (PIML), todeal with transport phenomena-based problems and related applications in foodengineering. To evaluate its potential, we investigate the fundamentals of themethod and most relevant contributions.Overall, PIML is in a development phase but has alreadyshown interesting capabilities to find solutions of partial differentialequations. This approach integrates powerful machine learning tools like neuralnetworks with knowledge-guided learning to find physically consistentsolutions. Both forward and inverse problems can be tackled without the need ofa large data set for training.Considering the features of PIML, including cost ofimplementation and computing speed, we conclude that this new approach willplay a key role in the virtualisation of food products and processes, and thedevelopment of digital twins. We can expect more contributions of PIML in foodengineering in the next few years.