Artificial Neural Network Modelling for Biodiesel Production Using Waste Camel Bone based Heterogenous Catalyst

Abstract

In recent scientific findings, utilizing solid wastes, especially biowastes from animal bones, as heterogeneous catalysts for biofuel production has gained attention in shifting from first- to third-generation feedstocks for a more environmentally friendly and sustainable transition from petroleum diesel to biodiesel. However, commercialization is still far off, as challenges and limitations demand further technological advancements to achieve a holistic production process with bearable costs to support the economy. Modeling studies have always been essential to understanding the behaviour of process and reduce costs as they can predict experimental results for untested conditions. This study addresses the element of process modeling in biofuel production by targeting critical parameters such as catalyst loading rate, alcohol-to-oil molar ratio, reaction time, and temperature to maximize biodiesel yield from non-edible seed oil. Experimental study was carried out in which waste camel bone was utilized as a heterogeneous base catalyst for the conversion of non-edible Jatropha curcas seed oil to biodiesel. Three types of artificial neural networks were developed namely: feedforward (FFNN), cascade forward (CFNN), and patternet (PNN), applying various training algorithms. A comparison among them was made, and it is demonstrated that the FFNN performed better overall than the other networks in predicting oil yield. CFNN with Levenberg–Marquardt (LM) achieved perfect accuracy (R = 1.0) and outperformed FFNN under scaled conjugate gradient (0.9789 vs 0.9347), though its performance collapsed under Bayesian regularization (BR) (R = 0.04). FFNN demonstrated more stable performance across algorithms (R = 0.9961–1.0), while PNN also achieved near-perfect fits with LM (0.9999) and BR (1.0) but showed much weaker performance under gradient-based algorithms (R = 0.5664–0.6923). The results highlight the potential of integrating machine learning in biodiesel production to capture process patterns and improve yield prediction, offering a pathway to overcome current limitations and move closer to commercialization.