Deep neural networks in chemical engineering classrooms to accurately model adsorption equilibrium data

The latest industrial revolution, Industry 4.0, is progressing exponentially and targets to integrate artificial intelligence and machine learning algorithms with existing technology to digitalise chemical processes across the industry, especially in the area of online monitoring, predictive analysis and modelling. Machine learning algorithms are being constantly implemented in both academic laboratories and industry to uncover the underlying correlations that exist in the high-dimensional and complex experimental and synthetic data that describes a chemical process. Indeed soon, proficiency in artificial intelligence methodology will become a required skill of a chemical engineer. It is therefore becoming essential to train chemical engineers with these methods to help them to adapt to this new era of digitised industries. Keeping these issues in mind, we introduced deep neural networks to the final-year chemical engineering students through a computer laboratory exercise. The exercise was delivered in fast-track mode: the students were asked to develop deep neural networks to model and predict the equilibrium adsorption of uptake of three different acids by activated carbon at four different temperatures. In this manuscript, we discuss in detail this laboratory exercise from delivery and design to the results obtained and the students’ feedback. In the classroom, the students compared the adsorption equilibrium data obtained using the established theoretical adsorption isotherms and empirical correlations with the neural networks developed in the classroom. The experience obtained from the classroom confirmed that this exercise gave the students the essential knowledge on the AI and awareness on the jargons in the world of machine language and obtained the required level of coding skills to develop a simple neural net with one layer or a sophisticated deep networks to model an important unit operation in chemical engineering and to accurately predict the experimental outcomes