Ash fouling characteristic analysis and prediction for pillow plate heat exchanger in waste heat recovery based on attentive-feature decision algorithm

Ash fouling on heat exchanger surfaces in waste heat recovery and utilization is an inevitable result of solid fuel combustion and particle deposition, affecting the efficiency, availability and operating cost of an energy utilization system. Fouling prediction plays a critical role in reasonable design and efficient operation of heat exchangers. However, credible monitoring and accurate prognostic towards fouling condition always remain a challenge, largely due to the complex flue gas component and abominable service environment of heat exchangers. In this paper, a novel ash fouling prediction method combined with Effective Particle Size Filtering and Multistep Attentive-Feature Decision algorithm (EF-MAFD) for Pillow Plate Heat Exchanger (PPHE) is proposed. First, a numerical fouling model with Discrete Phase Model (DPM) combined particle deposition and removal model is developed. Validation work is conducted on Nusselt number, pressure drop coefficient and normal restitution coefficient with existing experimental data. The effect of geometrical configuration of PPHE and condition on the fouling characteristic is then evaluated with this model. Next, the EF-MAFD ash fouling prediction model is established, in which the initial particle size distribution is transformed into an effective diameter innovatively and two indicators evaluating the fouling state, namely critical fouling resistance and critical fouling time are defined and predicted. The results indicate that fouling mainly occurs downstream of the welding spots and shows an asymptotic growing trend over time. PPHE with smaller diameter of welding spot, channel height and pitch ratio possess anti-fouling potential. In comparison to existing methods, the newly developed EF-MAFD method has the capacity to deal effectively with the complex particle distribution and provides the most satisfying prediction ability with coefficient of determination (R2) of 0.93 and mean absolute prediction error (MAPE) of 7.8 %. The proposed method could be utilized as a reliable tool to support fouling condition-based maintenance and design for various types of heat exchangers operating in fly ash conditions.