Residence time distribution (RTD) study of solids in a three-phase pilot-scale bubble column photoreactor has
been carried out in order to provide data for the development of an artificial neural network model usable for process optimisation.
The experimental data indicated that the RTD of solids was a complex nonlinear function of gas and liquid velocities
as well as the contacting pattern (co-current and countercurrent flow of gas and liquid). In this study, the solid particle
RTD data were modeled using feed forward artificial neural networks (ANN). The networks were trained with 250-
sets of input-output patterns using back-propagation algorithm. The trained networks were tested using 50-sets of RTD
data previously unknown to the networks. Out of several configurations, a 3-layered network with 6-neurons in its hidden
layer yielded optimal results with respect to the validation data. The optimal model and empirical data exhibited good
agreement with a correlation coefficient of 0.995.