Modeling of acetosolv pulping of oil palm fronds using response surface methodology and wavelet neural networks

Abstract

Mathematical models based on response surface methodology (RSM) and wavelet neural networks (WNNs) in conjunction with a central composite design were developed in order to study the influence of pulping variables viz. acetic acid, temperature, time, and hydrochloric acid (catalyst) on the resulting pulp and paper properties (screened yield, kappa number, tensile and tear indices) during the acetosolv pulping of oil palm fronds. The performance analysis demonstrated the superiority of WNNs over RSM, in that the former reproduced the experimental results with percentage errors and mean squared errors between 3 and 8% and 0.0054 0.4514 respectively, which were much lower than those obtained by the RSM models with corresponding values of 12 40% and 0.0809 9.3044, further corroborating the goodness of fit of the WNNs models for simulating the acetosolv pulping of oil palm fronds. Based on this assessment, it validates the exceptional predictive ability of the WNNs in comparison to the RSM polynomial model.