Session Synopsis:

The discussion on using machine learning (ML) and artificial intelligence (AI) in process systems engineering applications began at least as early as the 1990s. However, results were not readily translated into real-life applications in the subsequent two decades. Venkatasubramanian (2019) lists a few reasons, including that some of the problems to be tackled were inherently difficult, other problems had already found acceptable solutions using alternative approaches, and a lack of sufficient computing power. The latter has evolved substantially since then. Our ability to acquire and store massive amounts of data and to use this data to train very large-scale ML models has led to productivity boosts in areas such as e-commerce, computer vision, natural language processing, and, more recently, generative AI, among others. Science and engineering examples of using ML/AI tools include the development of new materials, such as catalysts, and the estimation of physical properties of organic molecules, among others. The successful application of ML models and AI tools in process systems engineering, including modeling dynamical systems where models are used for optimization and control, must consider some particular needs. First, prediction accuracy is key. For example, overpredicting the purity of a plant’s product leads to the production of an off-spec product, which can incur penalties. Underpredicting the same variable will cause the plant to deliver too pure a product, effectively giving away revenue. Second, the data used to train ML models represent a physical system that obeys material and energy conservation laws. ML models and their predictions must be consistent with these conservation laws. Third, and further to the previous point, the models should be interpretable from the perspective of the governing equations of the underlying physical system. Finally, model maintenance should be prominently considered. Process plants inherently change over time. Some changes are known and obvious (e.g., a piece of equipment is replaced), while others (such as catalyst deactivation, corrosion, and fouling) occur gradually over extended periods. Owing to these changes, the predictions of an ML model will diverge from the outputs of the physical system, and the model should be regularly updated to correct this gap. In this session, we highlight open problems related to the above needs in using AI/ML models for process systems engineering applications and delineate promising directions for research.

Session Synopsis:

Autonomous or self-driving labs (SDLs) that bring together experimental platforms, advanced measurement systems, robotics, automation, and decision-making are ushering a revolution in materials discovery. The integration of these many elements into SDLs that can be used to design sustainable materials and products to be manufactured at scale poses many process systems engineering challenges. These are driven in part by the need to explore poorly understood materials/formulations for which experimental failure rates and uncertainty are high. Beyond product design, questions arise on what roles SDLs, which typically operate at a very small scale, can play in process development and scale-up and how they can enhance existing approaches to process synthesis and design.

Session Synopsis:

The integration of Artificial Intelligence (AI) into chemical sciences is revolutionizing research, development, and industrial processes, significantly enhancing our capabilities in various areas, from material discovery and reaction optimization to process operations and business decision-making. Recently, advances in generative AI have shown great potential in driving substantial breakthroughs in innovation and productivity, transforming workflows and solution strategies for complex problems. However, developing generative AI solutions in chemical engineering and related fields presents unique challenges, including the complexity of modeling chemical systems and the necessity for high-quality, domain-specific data. This session will explore the challenges and opportunities of generative AI, providing insights into how to effectively and responsibly harness its transformative power. In particular, we consider the use of generative modeling, large language models, and AI assistants for property prediction, reaction optimization, molecule/material discovery, process design and optimization, supply chain management, scientific discovery in general (e.g., an AI co-scientist), and enhancing educational outcomes in core chemical engineering courses.