Artificial intelligence and data mining methods applied in process engineering

Artificial intelligence (AI) and big data processing methods have proven their value in many fields thanks to their ability to solve complex problems that are often difficult to formulate explicitly (multilingual translation, image generation, autonomous driving, etc.). However, process and product engineering is an interdisciplinary science of complexity: (i) the complexity of physical and chemical phenomena (chemistry, mass transfer, hydrodynamics, etc.), (ii) complexity of functional properties (reactivity of an intermediate, energy content of a fuel, viscosity of a polymer, etc.), (iii) complexity of media (mixtures, formulated products, biological media, etc.), (iv) the complexity of application sectors (semiconductors, food, pharmaceuticals, etc.), (v) the complexity of operating facilities (reactors, separators, processes, etc.), and (vi) the complexity of management strategies (regulation, supply chain, planning, etc.). Furthermore, process and product engineering operates in a volatile and highly competitive market, where industries must differentiate themselves by developing innovative products quickly and cost-effectively, while ensuring quality, performance, and sustainable production.

It therefore seems natural that AI methods have a role to play in process engineering. The recent explosion in these methods might suggest that this is a new development, but connections have existed for a long time: data-driven trend models dating back to the early 20th century, expert systems in the 1980s and 1990s, and so on. Since understanding couplings and scaling-up are constant concerns in process engineering, many data processing and modeling approaches are in common use: multiple linear regression (MLR) and principal component analysis (PCA) are part of the curriculum for many undergraduate programs in process engineering. A literature review highlights the use of supervised learning methods and hybrid, unsupervised, and combinatorial methods. While semi-supervised and reinforcement learning methods remain marginal, they are growing and opening up new avenues that the community is beginning to explore. However, new methods emerge regularly, sometimes based on entirely novel approaches. This article therefore aims to provide an overview of the AI methods used in process engineering, their underlying principles, their specific characteristics, and their relevance to practical applications, and then to expand this perspective to include the most promising methods.

Whether a process engineer is focused on the optimal conversion of raw materials into products or on developing new products, a significant portion of their time is spent working with data. This data is compared, clustered, classified, filtered, regressed, and so on. These basic functions can be handled...