https://www.mdu.se/

At a CFB boiler a system has been tested based on a Modelica model together with a decision support system. The model is a physical model including energy and material balances, chemical reactions like combustion and gasification reactions. For the combustion system we primarily consider equilibrium conditions while for gasification the kinetics is important and thus PLS-models built on experimental data in a pilot plant are combined with literature data and a physical model. The simulation model is first developed in Modelica, but then placed as an object in Simulink/Matlab, from which data is communicated to and from the data base through OPC-server. Measured data are collected from the process data base and inserted as initial data into the simulation model, including the boiler, separator, heat exchangers and steam system. A simulation during 300 seconds is performed and the data after this is compared to the initial data. If we have steady state conditions, the values after the simulation will be the same as the initial data, while if the data are not balanced, the difference will correspond to a balanced state between all measured data and the physical correlations in the boiler. This procedure is repeated on a regular basis and the trend of the difference between the measured and the balanced data is plotted and analyzed with respect to slope respectively variance. These data are combined with other type of information like standard deviation of sensors, which corresponds to noise; is the data value changing at all? Input of manual information like lab-data, unexpected events like noise; maintenance actions; activities like how many times a valve has been opening and closing; combination of data like Energy and Mass balances combined with conductivity in blow down from steam drum to detect possible leakages in piping or boiler systems;

All this information is introduced into a BN, Bayesian Net, which has been built from known relations, but where the quantitative data is built from experience and statistics. In this way we can then detect possible faults or probable faults coming up. This information is used by both the operators and maintenance staff. The mathematical simulation model over the CFB boiler and results from the utilization is presented in this paper.