Budapesti Műszaki és Gazdaságtudományi Egyetem - Villamosmérnöki és Informatikai Kar

Automatizálási és Alkalmazott Informatikai Tanszék

The topic is available also in Hungarian

Petrochemistry is a well-known industry. It operates with huge quantities of products, like t/h (ton/hour). It means that a slight change in the production gets multiplied and scaled up immediately. The goal of this project is to optimise the production line for profit. To achieve this, the production is first improved on a digital twin, and when it is tested and evaluated before putting it into production. In the concrete case, digital twin means a software based simulation that mimics the behaviour of the production plant based on mathematical formulas and iterations.

In the frames of this project, in silico optimization is conducted to find better production parameters. We do not do chemical process modelling as it is already done by external software systems (e.g. Hysys, Pro2, DWSim). The task is to develop an interface to such modelling systems and apply optimization techniques. Our primary goal is to conduct optimization for business goals. It means that the costs and revenues are also embedded in the models. This is how we can optimize the profit.

Lately, the outcome of the simulation software is estimated with artificial neural networks (ANN). One of the goals of this project is to develop such ANNs, so called surrogate models.

The student can work on the following topics.

- Advanced local search algorithms
- Global optimization techniques
- Develop feed forward neural networks
- Deep reinforcement learning techniques

Maximális létszám: 10 fő