Neural network modeling of organics removal by activated carbon cloths

The adsorption of organic compounds onto an activated carbon cloth is studied in a dynamic reactor. An experimental design is carried out to investigate the influence of operating conditions (initial concentration C-o, flow velocity U-o, and bed thickness Z) and adsorbate's characteristics. A slow intraparticular diffusion is shown by flattened breakthrough curves, and adsorption capacities are high and range between 50 and 250 mg g(-1). The transfer zone Z(o), assessed by the Adams and Bohart equation, is low (3 mm). All experimental results are modeled by a neural network to take into account the specific diffusion of cloths. Parameters related to the adsorbate-adsorbent affinity in a batch reactor are consequently introduced in the input layer of the neural network (intraparticular coefficient K-w and Freundlich parameters K-f and Un), added to operating conditions whose influence was shown (C-o, U-o, and Z) and time t. The statistical quality of the neural network modeling is high (r(2) = 0.956). Furthermore, the Garson connection weight method allows the relative influence of input neurons to be determined. This analysis confirms the influence of parameters relative to adsorbant-adsorbate affinity.