Description
The webinar focused on how surrogate models and generative AI are transforming process simulation in chemical engineering, presented by Ryan Muir and Mihaela Hahne from AVEVA. The speakers discussed different types of simulation models, including white box, black box, and surrogate models, and demonstrated how machine learning can be integrated into process simulation tools using ONNX models. They showcased examples of using AI to create gray box models and automate flow sheet generation from PFDs. The presentation also covered AVEVA's academic partnerships and resources available to students and professors. The session concluded with a Q&A where participants asked questions about data handling, uncertainty quantification, and the integration of AI agents with simulation software.
Webinar on Surrogate Modeling and AI
The meeting began with John introducing the webinar format and requesting participants to share their names, affiliations, and LinkedIn profiles in the chat window. Alberto Siccardo, representing the IChemE CAPE group, expressed gratitude for the collaboration with AICHE. John then introduced the speakers:
Ryan Muir is the Senior Product Manager for AVEVA Process Simulation and oversees product strategy, development roadmaps, and marketing for AVEVA’s next-generation simulation platform. Prior to his current role, Ryan worked as a principal enterprise architect at AVEVA, where he specialized in process simulation applications for enterprise-scale digital twin solutions. He was also previously a developer for AVEVA Process Simulation and has process engineering experience in the specialty chemicals market. He received his bachelor’s degree in chemical engineering and master’s degree in environmental engineering from the University of Cincinnati.
Mihaela Hahne is the Global Director of Business Development at AVEVA, where she has played a key role in developing and strengthening the company’s academic partnerships since 2016. With more than a decade of experience at AVEVA, she has previously held roles including Program Manager for Portfolio and Manager of Licensing and Operations.
With over 15 years of experience in the software industry, Mihaela has worked across both operational and commercial functions, building strong connections between technology providers, universities, and industry. She is passionate about advancing educational initiatives and fostering collaboration between academia and industry through the integration of innovative technologies and joint research programs.
AVEVA and AI Simulation Overview
Mihaela presented an overview of AVEVA's history and its role in providing industrial software solutions, highlighting its evolution from a Cambridge spin-off to a global leader in digital transformation. She emphasized AVEVA's commitment to innovation and its partnerships with various technology and alliance partners. Ryan then discussed the use of AI in simulation, explaining the differences between white box, black box, surrogate, and gray box models. He highlighted the advantages of using machine learning models in process simulation, such as addressing lack of understanding, performance issues, model unreliability, and IP concerns.
ML Integration in Process Simulation
Ryan discussed the integration of machine learning models into process simulation environments, focusing on the use of ONNX (Open Neural Network Exchange) as a bridge between different machine learning pipelines and simulation tools. He explained how ONNX allows for the interoperability of ML models across various platforms without exposing sensitive data to the simulation tool. Ryan demonstrated the process of developing a gray box model for a pump using historical plant data, training an ML model, and integrating it into AVEVA process simulation. He also touched on the use of surrogate models for complex simulations and the potential of automated flow sheet generation using AI agents. Ryan emphasized the flexibility and security offered by this approach, allowing users to leverage ML models without compromising sensitive data or requiring extensive data science expertise.
Ryan presented on the integration of machine learning and process simulation, highlighting the use of surrogate models and their advantages over empirical models in capturing detailed simulations while maintaining speed. They discussed data handling, the role of uncertainty quantification in safety-critical processes, and the potential for autonomous modeling pipelines. Questions from attendees addressed the use of ONNX models, the interaction of AI agents with simulations, and the application of machine learning to physical properties.