Large language models (LLMs) for flexible interface to industrial decision-support tool

Semester project or master thesis Fall 2024 or Spring 2025

Context

The design and operation of the next generation of waste management and energy supply systems should ensure industrial Global Warming Potential (GWP) reduction targets agreed internationally and enable transition towards a circular economy of materials whilst being financially viable. Related challenges stem from the range and complexity of involved technologies and engineering domains, variety of possibly conflicting performance criteria, and interaction potential within the system as well as with external interfaces, in the context of uncertain energy supply markets.

This project is part of the development of a simulation and optimization platform for decision-support in industrial decarbonization and value recovery pathways, focusing primarily on the development of the platform interface to industrial stakeholders (decision-makers and engineers). This work is a collaboration of IPESE and R&D at Kanadevia Inova AG (KVI), an Engineering, Procurement and Construction company active in the waste treatment sector (>600 plants built), based in Zürich (Switzerland).

Objective

This explorative project supports innovatively the interaction of humans (engineers, managers, directors in the industry) with the information from the platform environment (developed in OpenModelica/Python). The following aspects will be covered:

  1. Understanding and inventory of the available documentation of the decision-support tool (tutorials, workflows, databases of models and economic, process and life-cycle information)

  2. Screening of available language processors for machine-learning supported interface

  3. Exploration of possibilities (very creative step!): use the trained interface to build workflows tailored to the need of the user, automatically build OpenModelica models from scratch, etc.

  4. Protoype of possible interface(s)

This work will increase the interaction potential of humans to the analyzed waste treatment technologies by providing more flexible and understandable interface than fixed/established workflows, thereby increasing industrial stakeholders’ trust and use of the decision-support tool.

Skills and Background

  • Coding skills in Python
  • Willingness to work independently and strong curiosity for the exploration of interface possibilities
  • Previous experience with LLMs is of advantage
  • Understanding of energy systems (the following courses may be of advantage, but not regarded as strict requirements): ME-454 (Modelling and Optimisation of Energy systems), ME-451 (Advanced Energetics), ME-409 (Energy Conversion and Renewable Energy)

Location & Organization

The project is done remotely, either on EPFL campus, IPESE laboratory (located in Sion - train fare Lausanne-Sion is reimbursed by EPFL) or as home office. Regular meetings with the supervisor are ensured (exact schedule to be determined, but typically 1-2x/week), and discussions with Professor Maréchal, Fachhochschule OST or industrial research colleagues may be organized in the current of the semester.

Contact

The work is supervised by Julie Dutoit, PhD candidate and R&D process engineer at KVI. If interested, please send your CV with a short description of your specific interest to julie.dutoit@hz-inova.com.

References