Clean technology for sequestering Rhodamine B dye on modified mango pod using artificial intelligence techniques

Abstract

In this study, an acid-modified mango pod (AMMP) was prepared as adsorbent for removal of Rhodamine dye and the adsorption uptake was also determined using machine learning algorithms namely, Artificial Neural network (ANN) and Adaptive neuro-fuzzy inference systems (ANFIS) respectively. The prepared adsorbent was characterized with SEM, FTIR, EDX, and PXRD techniques. Surface chemistry revealed the presence of O–H stretching of free hydroxyls and alcohols and phenols, C–H stretch of alkanes, C≡C stretch of alkynes, CO stretch of ketones, lactones, and carboxylic anhydrides, –CC- stretching of alkenes, C–N stretch of aliphatic amines and O–H bend of carboxylic acids. Surface morphologies of the AMMP revealed well-developed and open porous surfaces needed for an efficient adsorption of the dye molecule. EDX analysis showed an increase in the carbon contents from 79.94% (raw) to 80.12% (AMMP) by weight and 86.89% (raw) to 89.11% (AMMP) by atom. The crystallinity structure revealed the new and intense peak formations and the presence of ordered (organized) crystalline structures on AMMP. Optimal ANN architectures, activation functions and training algorithms were selected after several stimulations with different network parameters while the ANFIS model was tested with three clustering approaches namely, Grid partitioning, fuzzy c-means, and subtractive clustering to carry out the extensive studies to optimally predict the adsorption efficiency/capacity of Rhodamine dye onto AMMP. The performance of the developed models was evaluated using the following statistical metrics; the optimal ANN model gave RMSE ​= ​10.422, MAD ​= ​3.673, MAPE ​= ​5.409, R2 ​= ​0.977 while with optimal ANFIS model gave a RMSE ​= ​9.246, MAD ​= ​4.938, MAPE ​= ​4.672, R2 ​= ​0.984 ​at the testing phase. The lower value of the statistical parameters indicates better performance in both models. Batch adsorption studies gave qmax of 456.67 ​mg/g, 389.51 ​mg/g, and 410.56 ​mg/g for the experimental data, ANN and ANFIS models respectively thus, suggesting their good correlations. Technoeconomic aspects of the study present that AMMP is approximately 8 times cheaper, translating to saving cost of 225.2 USD/kg when compared with 259.5 USD/kg for commercial activated carbon.