SPP 2331:  Machine Learning in Chemical Engineering. Knowledge Meets Data: Interpretability, Extrapolation, Reliability, Trust

• Subject Area: Thermal Engineering/Process Engineering ; Biology ; Chemistry ; Computer Science, Systems and Electrical Engineering ; Mathematics ; Physics
• Term: since 2021
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 441958259

Chemical Engineering (CE) is at a crossroad. Worldwide, the chemical industry has a 10% share of the total energy utilization and relies almost entirely on fossil sources. A transformation of the chemical industry to renewable energy and feedstock supply is of the utmost importance in Germany as the world's third largest chemical supplier. Renewable resources fluctuate in time and space, requiring dynamic operation and new paradigms for the design of flexible plants. Simultaneously, the chemical industry needs to continuously optimize their plant operation, increase plant availability, and shorten time-to-market to ensure competitiveness. CE is currently ill-equipped to facilitate this fundamental change by itself. CE is deeply rooted in physics and chemistry and combines models and simulation with experiments. Models cover scales from molecular to enterprise and the environment. Experiments play a major role to identify, calibrate, or validate models for process design and operation. However, developing models and suitable mathematical methods is expensive and many phenomena cannot be fully described by tractable models. To tackle the transformation of chemical production, we envision a close collaboration between Machine Learning (ML) as emerging field and CE with its wide set of methods based in physics and chemistry. ML has a great track record in working on heterogeneous and large data sets and performing creative tasks. Applications like AlphaGo and autonomous driving show impressive results highlighting ML's potential. So far, ML applications within CE focus mostly on data analytics and replacing existing physicochemical models by surrogates. Joint interdisciplinary research between ML and CE has the potential for breakthrough results. CE has a track record of working with applied mathematics and computer science and co-developing methods with applicability well beyond CE, e.g., in partial differential equations (PDE), differential algebraic equations (DAE), and deterministic (global) optimization.We identified six areas of collaborative research for this Priority Programme (PP), which open up new methods for CE, formulate new types of problems for ML, and jointly generate advances for methods in both ML and CE. These areas are #1 optimal decision making, #2 introducing / enforcing physical laws in ML models, #3 heterogeneity of data, #4 information and knowledge representation, #5 safety and trust in ML applications, and #6 creativity. Under the umbrella of these areas / topics, the PP will have collaborative projects between groups from chemical engineering and ML, which promise progress regarding process synthesis (especially regarding feedstock transformation), process flexibility, material selection, generation of alternatives, and uncovering hidden information. This PP can hence make a large contribution towards readying Germany's chemical industry for a sustainable future.

• DFG Programme: Priority Programmes
• International Connection: United Kingdom

Projects

• Coordination Funds (Applicant Mitsos, Ph.D., Alexander )
• Graph-Based Generative Machine Learning for Optimal Molecular Design (Applicants Grohe, Martin ;  Mitsos, Ph.D., Alexander )
• Hybrid Physics-Neural Network Soft Sensors for Dynamic Operation of Liquid-Liquid Separation Processes (Applicants Dahmen, Manuel ;  Jupke, Andreas )
• Improving simulations of large-scale dense particle laden flows with machine learning: a genetic programming approach (Applicants Mostaghim, Sanaz ;  van Wachem, Berend )
• Kernel Methods for Confidence Regions in Optimal Experimental Design and Parameter Estimation (Applicants Bortz, Michael ;  Esche, Erik )
• Machine Learning and Optimal Experimental Design for Thermodynamic Property Modeling (Applicants Herzog, Roland ;  Richter, Markus )
• Machine learning for design of chemical engineering unit operations - a microevaporator, leading to a 3D structured multiphase absorber (Applicants Dittmeyer, Roland ;  Friederich, Pascal ;  Stroh, Alexander )
• Machine Learning for Explainable Roundtrip Polymer Reaction Engineering (Applicants Beuermann, Sabine ;  Fiosina, Jelena )
• Machine Learning for the Design and Control of Power2X Processes with Application to Methanol Synthesis (Applicants Kienle, Achim ;  Sager, Sebastian ;  Seidel-Morgenstern, Andreas )
• Reinforcement Learning for Automated Flowsheet Synthesis of Steady-State Processes (Applicants Burger, Jakob ;  Grimm, Dominik )
• Safe Reinforcement Learning for Start-up and Operation of Chemical Processes (Applicants Lucia Gil, Sergio ;  Repke, Jens-Uwe )
• Translating thermodynamic knowledge to computers (Applicants Fellenz, Sophie ;  Hasse, Hans ;  Leitte, Heike )

• Spokesperson: Professor Alexander Mitsos, Ph.D.