Towards the discovery of boiler physics in a Kraft recovery boiler using sparse identification of non-linear dynamics

Abstract

This dissertation seeks to discover the underlying physics, that govern a soot blowing procedure for the removal of ash deposits, in a Kraft recovery boiler. The methodology consists of the design of a thermodynamic model, aimed at calculating fouling levels in an online capacity, the discovery of the strengths and shortcomings of the SINDy algorithm, and the implementation of this algorithm to discover the soot blowing physics. Fouling levels are determined by means of heat transfer coefficients, as local fouling measurements are not available. The methodology is applied to the secondary superheater in the Kraft recovery boiler at the SAPPI Ngodwana mill. Implementation of the SINDy algorithm shows, that it is possible to discover simple, interpretable physics models, that can be used for generalisation outside of the training data domain, however, significant data pre-processing is required before consistent general models can be established.