In-situ selective epoxidation of Colocynthis Vulgaris shrad seed oil for the synthesis of a methacrylated biobased resin; An artificial neural network (ANN) modelling approach

Abstract

Colocynthis Vulgaris shrad seed oil (CVSSO) contains unsaturated fatty triglycerides predominantly (oleic, linoleic and linolenic acid), indicating its suitability for epoxide development. The challenges associated with petroleum-based epoxides, such as price escalation, non-renewability and environmental pollution, have led the research for its substitution with vegetable-based epoxide; artificial intelligence techniques have been adopted to model and predict complex processes. This study predicted CVSSO epoxidation parameters using an Artificial Neural Network (ANN). First, pure CVSSO was epoxidized via a conventional in-situ method with peroxyacetic acid. Then, methacrylated biobased resin was synthesized by further modifying the epoxidized oil with methacrylic acid in the presence of hydroquinone. The input parameters of the neural networks are time, temperature, composition and stirring speed, while the output parameters are iodine value, oxirane value, conversion, yield and selectivity. ANN was evaluated with 11 backpropagation (BP) algorithms; each algorithm was evaluated with four neurons in the input layer, a hidden layer with ten neurons and an output layer with five neurons. Coefficient of determination (R 2 ) and mean square error (MSE) have been implemented and correlated to test the adequacy and predictive ability of the model. Results showed that the Levenberg Marquardt algorithm gave a good prediction for iodine and oxirane value with mean square error (MSE) of 2.2397E-11 and 1.67718E-12; at the same time, Bayesian regularization predicted conversion, yield and selectivity with MSE of 4.3925E-12, 1.86759E-12 and 8.53013E-13 respectively; nevertheless, Levenberg Marquardt and Bayesian regularization gave the best prediction than other different BP algorithms. Other statistical indicators such as mean absolute error (MAE) showed 0.0232,0.0781,8.14,12.13 and 19.73), while mean absolute percentage error (MAPE) gave (0.00028%, 0.000864%, 0.000376%, 0.00019% and 0.00017%) for iodine value, oxirane value, conversion, yield and selectivity respectively; the results indicate a high forecast on the ANN simulated process parameters and suggest a good agreement between the experimental ad predicted values. The characterization results from physicochemical and FTIR analysis indicated that CVSSO was suitably epoxidized, and further modification to Methacrylated epoxidized Colocynthis Vulgaris shrad oil (MECVSSO) was actualized. These results show that ANN can potentially estimate epoxidation parameters of CVSSO. This development may have significant potential to improve the product quality of epoxides and reduce time and cost by minimizing epoxidation experiments.