SPAVer - Slimme Procescontrole van Anaerobe VERgisting

Organic waste streams (for example manure, food waste or sewage sludge) are valuable resources for depleting nutrients and energy by means of anaerobic digestion. Stable operation of anaerobic digesters of organic waste streams is however a big challenge. Therefore, a lot of digesters are running sub-optimal or are not running at all. Next to economic damage for the farmer and the suppliers of digesters, there is a waste of valuable resources due to lower conversion of circular raw materials to fatty acids or methane. Mechanistic models are used for automatic process control, but the underlying microbiological and physical chemical processes in digesters are so complicated that the control is not robust.

Machine learning algorithm-based solutions have shown the potential to learn from data collected from complex systems, such as digester bioreactors. and accurately identify and predict their behaviours, even without actual knowledge of the mechanism(s) in the reactor itself. For example, Artificial Neural Networks, a type of deep machine-learning algorithms have already Artificial Intelligence, for instance Machine Learning or a Recurrent Neural Network, can predict the behaviour of a complex system without actual knowledge of the mechanism in the reactor itself. Neural networks have already successfully predicted the production of biogas and were able to optimize it in relation to input- and output parameters.

Combining a Neural Network as input for the automated process control could optimize the efficiency of the bioreactor. This project is to test the possibility for process control and optimization on a 6 liter laboratory scale digester.

Problem statement:

Can an Artificial Neural Network-based solution be developed to predict or optimize process conditions of a 6L bioreactor at Saxion?

Project of research goal:

  • Creating a databank with selected parameters using (old) data of the bioreactor.
  • Developing and comparing for performance Neural Network-based models for digester optimization using selected (and derived) parameters and data from the bioreactor databank.
  • Validating the best performing Neural Network-based models using validation data sets from the databank.

Looptijd:

September 2021 – September 2022

Partners:

HoSt, ToPerform, Methaplanet, Universiteit Twente

Financiering:

RVO-SIA

Betrokken onderzoekers:

Hans Gelten, Gerco Pijffers, Simon Hageman, Richard van Leeuwen