Artificial Intelligence

AI has arrived in the process industries giving statistics a new name. Many of the currently applied solutions are rooted in well-known concepts but the tool chain is novel. New tools open the door to a step-change in the understanding of processes. Today’s process experts rightly claim that AI must explain its results and recommendations. Integrating rigorous first-principle simulation and measured data from plants are the current challenges. The session will present results from our industries and cover the state-of-the-art as well as experimental approaches.

Session Chair:

• Armin Fricke, Capital-Gain Consultants

Schedule:

Abstracts:

Pre-reformer reactor: Explaining and integrating ML models with rigorous simulation

Jan C. Schöneberger, Chemstations Europe

The availability of sufficient plant data, both in quality and quantity, is a barrier to widespread usage of Machine Learning (ML) models in the process industry. Flowsheet simulators can solve this problem effectively by providing reliable data that considers the underlying physical models. They can also act as evaluation and validation tools for ML models. The resulting hybrid models unleash a true potential for ML applications by combining data-driven predictions with mass and energy balances. The ML model then becomes an attractive alternative to rigorous simulation, which can solve more complex process engineering problems in lower calculation times.

In this presentation, we introduce our 5-step workflow for building and evaluating ML models based on rigorous simulation models. We build an ML model for a pre-reformer reactor and evaluate it by implementing the model as a unit operation.

Divide and conquer: Benefits from modular surrogate modelling of flowsheets

Michael Bortz, Fraunhofer Institute for Industrial Mathematics

Flowsheet simulation based on the equilibrium stage model consists in solving large nonlinear systems in order to obtain values of KPIs for the chosen values of specifications. That is, the KPIs are given implicitly by the specifications via the solution of the nonlinear system. AI-based surrogates, trained with a sufficiently large set of previously simulated solutions, offer the option to identify an explicit functional dependence. However, an open issue so far is the limited transferability of the surrogates, e.g., with respect to different flowsheet topologies.

In this contribution, we make one step into this direction by training surrogates for single flowsheet units which are then interconnected. We illustrate that in this approach, not only the training effort is reduced, but also the convergence properties of the rigorous simulation in the presence of recycle streams can be improved.

Leveraging AI for the Future of Sustainability Innovation

Tomas van der Heijden, Merantix

Global sustainability regulations are vast, growing, and increasingly difficult to assess, resulting in many companies focusing available resources on risk management and compliance, rather than forward-looking strategies to identify new opportunities and accelerate the transition to net zero. We believe leveraging machine learning technology is crucial to change this, especially in the chemicals industry where the sustainability regulatory environments are incredibly dynamic, variable and complex. Our mission is to harness natural language processing to guide companies along the path to positive environmental impact and shine a light on the rapidly evolving sustainability landscape. To do this we are building a state-of-the-art sustainability intelligence and management platform to unlock organisations' full potential at every step of the sustainability management journey. Our first goal is to help organisations find, understand and act on environmental regulatory data specific to their business, which will be the subject of this talk.