Generative Artificial Intelligence for Chemical Process Engineering
Speaker: Artur M. Schweidtmann
Abstract: Generative AI has gained immense traction across diverse sectors, exemplified by ChatGPT’s language generation and GitHub Copilot’s code generation. Generative AI also holds immense potential to reshape chemical process engineering by offering advanced data handling, modelling, and decision-support capabilities, ultimately driving innovation and efficiency in the industry. However, there are only limited applications in chemical engineering so far.
In my talk, I will give an overview of opportunities for GenAI for process engineering. I propose promising applications for generative AI in process engineering including autocompletion of flowsheets, autocorrection of engineering documents, P&ID generation, and AI-assisted HAZOPs. My talk will discuss three main areas of development: (1) data, (2) information representation, (3) model architectures including mechanistic information. I will further demonstrate our web application for the digitization of Piping and Instrumentation Diagrams that uses computer vision algorithms transforming PDF documents in smart P&IDs.
References
Schweidtmann, A. M. (2024). Generative artificial intelligence in chemical engineering. Nature Chemical Engineering, 1(3), 193-193.
Generative Artificial Intelligence for Control Structure Prediction
Speaker: Lukas Schulze Balhorn
Abstract: Control structure design is an important but tedious step in chemical process development. Generative artificial intelligence (AI) promises to reduce development time by supporting engineers. Previous research on generative AI in chemical process design mainly represented processes by sequences. However, graphs offer a promising alternative because of their permutation invariance. We propose two generative AI models to predict control structures from flowsheet topologies: (i) A sequence-to-sequence model and (ii) a sequence-to-sequence model. For the graph-to-sequence model, we compare four different graph encoder architectures, one of them being a graph neural network (GNN) proposed in this work. Compared to a purely sequence-based approach, the graph-to-sequence model improves the top-5 accuracy for a relatively small training dataset of 1,000 flowsheets from 0.9% to 28.4%. However, the sequence-based approach performs better on a large-scale dataset of 100,000 flowsheets, reaching a top-5 accuracy of 89,2%. These results highlight the great potential of generative AI models to accelerate chemical process development but its effectiveness on industry relevant case studies should be tested.
Process Design through Deep Reinforcement Learning and Graph Neural Networks
Speaker: Qinghe Gao
Abstract: Process synthesis experiences a disruptive transformation accelerated by artificial intelligence. We propose a reinforcement learning algorithm for chemical process design based on a state-of-the-art actor-critic logic. Our proposed algorithm represents chemical processes as graphs and uses graph convolutional neural networks to learn from process graphs. In particular, the graph neural networks are implemented within the agent architecture to process the states and make decisions. We implement a hierarchical and hybrid decision-making process to generate flowsheets, where unit operations are placed iteratively as discrete decisions and corresponding design variables are selected as continuous decisions. We demonstrate the potential of our method to design economically viable flowsheets in an illustrative case study. The results show quick learning in discrete, continuous, and hybrid action spaces. The method is predestined to include large action-state spaces and an interface to process simulators in future research.
[ Slides ]