Forecasting the multicomponent adsorption of nimesulide and paracetamol through artificial neural network

Abstract

The multicomponent adsorption modeling is a challenge since the prediction of the interactions between the adsorbates is hard. An alternative is the use of models based on artificial intelligence. This study evaluated the single and binary adsorption of nimesulide and paracetamol on activated carbon (AC) using an artificial neural network (ANN). The characterization revealed a typical AC with a specific surface area reaching 866.12 m2 g−1 and point of zero charge (pHPZC) of 6.50. The adsorption capacity for each pharmaceutical compound was investigated as a function of adsorbent particle size, adsorbent dosage, contact time, and initial concentration. Experimental results indicated that nimesulide presented a higher affinity for the active sites of the AC than paracetamol, which can be attributed to hydrophobic and π−π dispersion interactions between nimesulide and AC. During the binary adsorption, nimesulide molecules competed with paracetamol molecules, suppressing the adsorption of paracetamol. The highest adsorption capacities were found for the adsorbent dosage of 0.5 g L−1 and particle size of 150 μm, in which the AC removed 98% of nimesulide and 76% of paracetamol in 300 min. An optimal ANN trained by the Bayesian regularization backpropagation algorithm and structured with three hidden layers with five neurons was successfully developed to simultaneously predict the single and binary adsorption of nimesulide (R = 0.9989) and paracetamol (R = 0.9985).