
The work is implemented as part of the project “Thermal Twin 4.0″ funded by the Vienna Business Agency – A Fund of the City of Vienna.
CO2 and resource savings through physical- and data-driven approaches
In the Thermal Twin 4.0 project an overall model of the incineration process of a thermal waste treatment plant is developed through a combination of physical (control and process engineering) and data-driven (machine learning) approaches. Physical correlations are used here to develop a model of the processes within the treatment plant, which given the input streams may simulate its combustion behavior.
Data-driven approaches – statistical as well as AI technologies – come into play to accommodate for incomplete data about the input streams.
Together, they form a digital twin of the incineration as a whole, which can then be used for an optimization of the energy efficiency and reduction of the CO2 emissions of the plant.

Challenge
Due to the impossibility of sampling from the inputs of the thermal treatment plant, a general coarse classification regarding thermal characteristics was done. However, even within such a class, the combustion characteristics may vary strongly.
The challenge of Cloudflight is to determine more fine-grained classes of the input streams so that the happenings inside the thermal treatment plant become more predictable.
The difficulty is that several input streams are processed in parallel, of which each has different temporal effects on the combustion process (for example liquids are burned faster than solids).
Idea
Use the physical model of the thermal plant and past historical data of the combustion characteristics of the input streams to predict the fine-grained classes of the used inputs.
Solution
Since the solution is still in progress, Cloudflight plans to implement the following using the physical model:
- During the process of incineration, constantly compute probabilities of how likely a certain input class currently is.
- Assume certain input streams to be known and try to find the remaining most likely inputs by minimizing the error between the actual and modeled sensor values.
- Link the available supplier information with the identified input streams to determine the combustion characteristics of the input streams.

The TU Wien is an institution for research & education for over 200 years. In the Thermal Twin 4.0 project, the general coarse classification of the input streams regarding thermal characteristics and the sub-models of the treatment plant are developed by an expert team of the Institute for Chemical, Environmental and Bioscience Engineering.

ENRAG offers tailor-made digital twin software development for technical applications. In this project, ENRAG develops the solver and integrates all sub-models to generate an equation based digital twin of the whole plant.

Wien Energie supplies energy to two million people and 230,000 commercial and industrial facilities. Wien Energie brings the domain know-how into this project and provides a large amount of historical sensor data, which is measured throughout the whole combustion process.

Cloudflight is a provider of customized software solutions. Its team of experts uses data-driven (statistical and AI) approaches to characterize the input fuels based on historical data and using the combustion process model.