Recent developments and prospects of machine learning in chemical engineering

We are delighted to present this special issue on recent developments and prospects of machine learning (ML) in chemical engineering. A term coined in the late 1950s, ML has produced some impressive results as of late. Based on continuous advances in various fields, from software engineering, to algorithm theory and practice, to computer hardware, to big or small data and statistics, ML now offers some exciting tools that can be well used to solve problems in various domains of interest to chemical engineers. Application of such tools in standard or customized form has already elicited interest among chemical engineers in many areas such as materials discovery or property prediction, process monitoring and control, optimization, numerical simulation, and many others. In fact, application and further development of ML in each of these areas would easily qualify as an entire research field in its own right. Therefore, we consider it quite timely that this special issue focuses on the interplay between ML and the broader field of chemical engineering. All publications in this special issue emphasize that ML methods have long been evolving and used by scientists and engineers. However, they also point out that through this long development, many recent capabilities afforded by ML have crossed thresholds to the point that problems until recently thought to be out of reach are now “miraculously” tractable. It is telling that in fall 2024, during the time manuscripts for this special issue were under review, Nobel Prizes were announced in Physics and in Chemistry for work in ML and in the broader artificial intelligence (AI) area. Just as telling is the corollary that as broadly applicable and powerful as these ML/AI tools are, their prudently managed use and further development for generic or specialized use requires significant and sophisticated human input. In this context, the publications in this special issue examine areas where combination of ML with domain expertise bears promise. Specifically:

Recent advances in continuous nanomanufacturing: focus on machine learning-driven process control (Venkatesan et al.) demonstrates how ML enables real-time control and monitoring in roll-to-roll (R2R) nanomanufacturing. The challenges of high-throughput fabrication, such as balancing precision and scalability, are addressed through data-driven techniques that fuse multimodal data and optimize process parameters in real-time. This work exemplifies how ML can significantly improve manufacturing competitiveness by enhancing efficiency and precision while driving down costs.

In Uncertainty quantification and propagation in atomistic machine learning (Dai et al.), the authors explore the critical role of uncertainty quantification (UQ) in ensuring the reliability of ML predictions. As ML is increasingly used in chemical and materials sciences, understanding and managing uncertainty becomes paramount, particularly for out-of-distribution data. By categorizing and benchmarking existing UQ and uncertainty propagation (UP) methods, this article provides a roadmap for improving confidence in ML-driven simulations and modeling. The work underscores the need for robust probabilistic frameworks as ML models are applied to high-stakes scenarios.

A tutorial review of machine learning-based model predictive control methods (Wu et al.) addresses the intersection of ML and control theory. Integration of ML with model predictive control (MPC), the de facto standard of advanced control in process industries, offers new possibilities for handling complexity, nonlinearity, and uncertainty, ever present in chemical processes. This article balances theory and practice, providing guidance on challenges such as data quality, generalization, computational efficiency, and safety. The inclusion of an open-source example ensures accessibility and encourages broader adoption of ML-based MPC methods.

The review Accelerating process control and optimization via machine learning (Mitrai and Daoutidis) focuses on how ML can streamline and enhance decision-making in process optimization. By automating algorithm selection and configuration, ML can enable faster and more efficient problem-solving. The authors delve into the representation of decision-making problems, the integration of ML into traditional optimization workflows, and the potential for ML to accelerate progress in chemical engineering and beyond.

Finally, Machine learning applications for thermochemical and kinetic property prediction (Tomme et al.) highlights the role of ML in addressing a long-standing challenge in kinetic modeling: predicting thermochemical and kinetic properties. The article emphasizes the interplay between data quality, representation, and model performance, stressing the critical need for high-quality datasets. This work demonstrates the potential for ML to fill gaps left by experimental and quantum chemical methods, advancing the accuracy and efficiency of chemical process design.

These articles not only illustrate the current state of the art but also point to exciting future directions. As ML continues to advance, its role in scientific discovery and technological innovation is poised to grow. Collectively, the following issues appear to be important in future expectations:

1. Enhanced interpretability and reliability: Future ML models will likely place greater emphasis on explainability and uncertainty quantification, ensuring that model predictions are both reliable and actionable in high-stakes applications.

2. Integration with domain knowledge: Hybrid approaches that combine ML with traditional scientific and engineering principles will continue to gain prominence, leveraging the strengths of both data-driven and analytical methods.

3. Data-efficient learning: Advances in techniques like transfer learning, self-supervised learning, and active learning will reduce the dependency on massive datasets, making ML accessible to a wider range of applications.

4. Real-time adaptation: As ML becomes more embedded in real-time systems, the ability to adapt to changing conditions and optimize processes on the fly will transform fields like manufacturing, logistics, and energy.

5. Sustainable AI: Maintaining a balance between replacing humans by machines and empowering humans with the use of machines, has been an important and persistent challenge since the early days of the industrial revolution and appears to remain such in the future.

We hope that this special issue serves as a valuable resource for researchers, practitioners, and students, inspiring them to explore the vast potential of ML. The articles collectively underscore the importance of collaboration across disciplines, as the convergence of data science, engineering, and physical sciences holds the key to addressing some of the most pressing challenges of our time.

We thank the authors, reviewers, and contributors for their efforts in bringing this special issue to fruition and look forward to the exciting innovations that lie ahead.