Defined in Task

5.1 / 5.4

Short description

SubFUSE is a state estimation (soft-sensory) tool which fuses subspace methods for system identification with Kalman filtering. The tool is solely data-driven, it requires IO data (measurements) as input.

In particular, the tool is developed for the estimation of input fuel characteristics of combustion-thermal power plants. Relying on regular process data including flue gas composition measurements, an estimate of the chemical structure of the fuel fed to the furnace of a fluidized bed boiler is provided. The tool has been tested in a simulated environment which aims to replicate the dynamics of the pilot problem.

The approach can be applied for alternative state estimation purposes as well, given that a suitable plant model can be generated and proper measurements are provided. The tool has been implemented in MATLAB, and is available in script format.

Example of usage

The tool targets of solving the WP3 pilot problem on fuel characterization, as a part of the heat exchanger fouling monitoring problem. Using IO data pairs of the process of interest, the tool proceeds in two steps: first, a sufficient Linear Time-Invariant (LTI) approximation of the governing dynamics is conducted utilizing subspace identification. Once the approximate dynamics are available, a standard Kalman filter is used for state estimation.

For example, Figures 1 and 2 illustrate the performance of the tool in soft-sensing the chemical composition of the fuel fed to a Circulating Fluidized Bed (CFB) boiler. In the learning phase (Figure 1) the tool learns to mimic the dynamics of combustion.

Figure 1: The identified LTI model (red) of a nonlinear combustion dynamics using IO measurement pairs (blue). Training data (left) are separated from the validation data (right) by a black line located at timestep 600.

Based on the IO relationship identified from data, the tool estimates (soft-senses) the chemical composition of the fuel using standard flue gas measurement data available at the power plant of interest (figure 2).

Figure 2: Actual (blue) and estimated (red) nitrogen content of the fuel used for heat generation in the combustion-thermal power plant of interest.


The tool is implemented in MATLAB script (.m file). Input data (measurements) are provided as numerical vectors (matrices). Estimation outcomes are provided as numerical vectors (matrices).

A link with StreamPipes is enabled by an OPC-UA client/server component (see FUSE OPC-UA tool).

Subordinates and platform dependencies

The tool is implemented in MATLAB script (.m file). MATLAB, a product Mathworks is required to run the application. (SubFUSE has been tested on MATLAB version 2020b).

MATLAB (2020b) is available on all major operating systems, including Windows, Unix/Linux and MacOS. The tool uses the MATLAB core, additional toolboxes are not required. Open software such as Octave is known to be able to interpret m-files, but FUSE-codes have not been tested with Octave.


The SubFUSE code tool is available for download (contact or ).


Current state is TRL 4 (validated in lab) currently being raised to TRL 5 (validated in a relevant environment).


Neuvonen, M., I. Selek and E. Ikonen (2021) Estimating Fuel Characteristics from Simulated Circulating Fluidized Bed Furnace Data. Int. Conf. on Systems and Control (ICSC’21), 24–26 Nov 2021, Caen, France, 2021, pp. 107–112.

To be considered in particular for the following COGNITWIN pilots

  • Sumitomo SHI FW Energia Oy
  • WP3