Predictive modeling of the length of prepared CNT by CVD through ANN-MPSO and GEP

Floating catalyst chemical vapor deposition (FC-CVD) is considered as one of the most appropriate techniques for the preparation of carbon nanotubes (CNTs) on the industrial scale. This paper tried to model the length of CNTs prepared by FC-CVD using two approaches, i.e. gene expression programs and hybrid artificial neural networks. In this regard, the effect of various FC-CVD parameters, viz. temperature, time, preheat temperature, Ar gas flow, methane gas flow, ethylene gas flow, Al2O3 catalyst, and Fe catalyst, on the length of CNTs, were investigated. At first, a hybrid artificial neural network-modified particle swarm optimization strategy (ANN-MPSO) has been used to model the CNTs length as a function of practical variables. In the next step, the same modeling of the problem was done using gene expression programming (GEP) instead of ANN-MPSO. The accuracy of the developed hybrid ANN-MPSO and GEP models was compared with regard to the linear combination of mean absolute percentage error and correlation coefficient as criteria. The results confirmed that the ANN model upgraded by the meta-heuristics strategy could be effectively applied for an accurate predictive model in the estimation of the length of CNTs as a function of the most important practical FC-CVD parameters. Also, the sensitivity analysis confirmed that the precursor type of carbon (including CH4 and C2H4) and the preheat temperature have the highest and the least effect on the length of CNTs, respectively.