Comparison of artificial neural networks with empirical correlations for estimating the average cycle time in conical spouted beds

Abstract

Conventional spouted beds have been extensively used in many real-life applications but are not suited for all types of materials, especially fine particles, which require internal devices to improve their motion in the spouted bed. However, unlike conventional spouted beds, there are almost no mechanistic or empirical models available for the design of spouted beds with internals. Given the availability of an extensive but not experimentally designed database, the main purpose of this study is to present an analysis of neural networks and empirical models in terms of their suitability to fit and predict average cycle times in conical spouted beds with and without draft tubes. The parameters investigated are particle size, density, contactor angle, gas inlet diameter, static bed height, and draft tube features. Although the amount of information is always a key factor when fitting models, the size of the database used in this study strongly affects the fitting performance of empirical models, whereas artificial neural networks are more influenced by how the data are scaled. Results of model verification show that both techniques are suitable for predicting average cycle times for data outside the range covered by the database.