Adsorption of <scp>5G</scp> blue reactive dye using passion fruit pomace: Kinetics, <scp>ANN</scp> modelling, and process optimization

Abstract The present work reports on the use of artificial neural networks to predict the adsorption of 5G blue reactive dye (5GBRD) on yellow passion fruit pomace in a fixed‐bed process and the % dye removal optimization. The samples were characterized using a thermogravimetric analyzer and scanning electron microscopy. Batch adsorption experiments were conducted to analyze the impact of the initial concentration of 5GBRD, contact time, and solution pH and temperature. For the fixed‐bed adsorption experiments, the processing time (0–55 h), inlet flow rate (1–4 mL min −1 ), initial dye concentration (35–70 mg L −1 ), and bed height (15–23 cm) were evaluated. The predictive model was built using a multilayer perceptron machine learning (artificial neural network [ANN]) model, and the process optimization used the dividing rectangles (DIRECT) algorithm. The best ANN model architecture was 4–4–1 and the accuracy of testing data were as follows: coefficient of determination ~0.97, mean squared error ~0.004, mean average error ~0.04, and root mean square error ~0.06. The DIRECT optimization algorithm indicated that the maximum % dye removal is achieved at 43.8 h, 3.7 mL min −1 , 66 mg L −1 , and 19.3 cm. The ANN model and DIRECT optimization algorithm are valuable tools for practical applications in adsorption process modelling and optimization.