The increasing availability of data and computational resources has led to a revival of machine learning (ML) applications in the chemical process industry. Successful applications of ML include real-time optimization and planning, fault detection and diagnosis, and high-fidelity predictive models of complex processes. While ongoing research has focused on weaknesses of ML, such as the lack of interpretability of its black-box models, little light has been shed on the environmental and societal impact of applying ML on a large industrial scale. Artificial neural networks of complex processes consist of millions of parameters, which makes them resource and time-intensive to train and optimize. Researchers in the computational intelligence community have estimated that training a neural network on a popular image recognition dataset (ImageNet) emits 100,000 lb CO<sub>2</sub>e. The primary aim of this work is to answer the following question: given the anticipated growth of ML applications in the chemical process industry, what is the environmental impact of applying ML at a large scale? Contrary to studies in the computational intelligence community that only consider the environmental impact in terms of CO<sub>2</sub> emissions, this study addresses other impact categories, such as raw material depletion, land use, and water consumption. To answer the central question, the environmental impact of ML per computational unit (floating-point operations) is analyzed. Then, the computational demand of some of the most promising ML applications in the chemical process industry is estimated. The extent of the environmental impact depends on two major variables: the source of energy that is used to power data centers, where computation at a large scale takes place, and the efficiency of ML algorithms. Knowing this relationship, sensitivity analysis is conducted, and different future scenarios are presented. Answering this question will allow engineers to understand the impact of the growth of ML and set realistic goals for its implementation. In addition, it will inform scientists about the extent of progress in computational efficiency that needs to occur for ML to be applied at a large scale with minimal environmental impact.

Evaluating the sustainability of machine learning applications in the chemical process industry

Fall 2022 abstracts, posters, and presentations will launch in mid-September.

ACS Fall 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.