AI models for boiler heat exchangers


Boiler heat exchanger design is one of the most demanding engineering tasks one can think of. Not only the formulas are quite complicated and their correct application is often confusing. Also much experience is required to take all the design restrictions, safety standards and findings from already done projects into account.

When we speak about modelling (or re-engineering) of already existing boilers, the standard procedure is:

  1. Gather boiler geometry (bundle and tube dimensions, pitches, tube wall thickness and so on)
  2. Develop a design model based on fundamental heat transfer equations
  3. Collect measured data for different load points (if we deal with solid fuels also for different fouling conditions)
  4. Validate and adjust the model to hit the measured values
  5. Use the model to check how new designed equipment fits into the whole system

As simple it looks as complicate is it. Especially step 4 requires lots of efforts caused by the high number of possible adjustments (used heat transfer formulas, fouling assumptions, construction specifics like hoppers and cleaning equipment and so on).

An interesting alternative or supporting method is an AI (artifical intelligence) driven procedure:

  1. Collect measured data for different load points
  2. Develop an AI model based on these measured values
  3. Use the model to check how new designed equipment fits into the whole system

Compared to the first method the AI model does not provide "inner" heat

transfer values but is much faster to develop, no validation and adjustment is necessary and – depending on the boundary conditions – it is also more accurate.

In the example the outlet HP steam temperature was calculated a) with a design model with all the heat

transfer equations and parameters required and b) with an AI model as an artifical neuronal network (ANN):

The ANN was fed only with flue gas and steam flow, the fluid inlet temperatures and the dry oxygen content (= values that are normally measured).

No other geometry or fluid parameters were used.

The chart shows that both outlet steam temperatures are equal within a wide range of 60 to 100% load.

Of course ANNs are NOT FOR DESIGN new plants. But for creating models of running systems they are amazing!


Dr.-Ing. Martin Horeni, 11/2018