oa South African Journal of Chemical Engineering - Neural network modelling of minimum bubbling and slugging velocities
|Article Title||Neural network modelling of minimum bubbling and slugging velocities|
|© Publisher:||South African Institution of Chemical Engineers (SAIChE)|
|Journal||South African Journal of Chemical Engineering|
|Affiliations||1 University of KwaZulu-Natal and 2 University of KwaZulu-Natal|
|Publication Date||Jan 2012|
|Pages||1 - 12|
|Keyword(s)||Bubbling bed, Fluidization, Neural network and Slugging|
Depending on the rate of coalescence above a gas distributor in gas fluidized beds, bubbles of diameter approaching the column diameter can be formed. This phenomenon often occurs in columns with a large bed height to bed diameter ratio. In this event, "slugging" is said to occur. The transition from a bubbling regime to a slugging regime is accompanied by a drop in bubble/slug rising velocity. Slug formation is in most cases an undesirable phenomenon: slugs lead to a drop in efficiency of mass transfer processes due to a poor gas-solid contact and in the case of laboratory and pilot plant operation may lead to serious scaling problems. Experiments were performed in a 150 mm diameter column, with three closely sized fractions of river sand particles (diameter of 0.6-0.8 mm, 0.8-1 mm, and 1-1.4 mm). All these experiments were initially conducted using air at ambient temperature and pressure. Air velocities were chosen such that a transition from a bubbling to slugging regime could take place. Different experiments were then performed using a mixture of sand particles (29% of a fraction 0.6-0.8 mm, 23% of a fraction 0.8-1.mm, and 48% of a fraction 1-1.4 mm) at higher fluidization velocities with fully developed slugging under ambient and increased temperatures of 25, 200, 300, and 400°C. The detection system consisted of two pressure-sensing probes spaced 5 cm vertically apart. The shape of the ∆P-time trace from such a pair of pressure probes not only gives a good indication of the presence of a bubble, but also can be used to determine its velocity and size. Initially, correlations taken from the literature for bubble and slug velocities were compared with the experimental data. In turn, the experimental data was then used to train a neural network. A feed-forward back-propagation neural network was employed to model the relations among particle size, temperature, air fluidization velocity, bubble/slug size and rising velocity. Results indicate the neural network model was capable of predicting the velocities of bubbles and/or slugs and identifying the transition from a bubbling fluidized bed to a slugging bed. An improvement was evident if compared with a previously used criterion for the onset of slugging.
Article metrics loading...