Three leaved yam starch physical / engineering properties evaluation using Response Surface Methodology and Artificial Neural Network network

Abstract

This study evaluated some physical properties of three-leaved yam starch (TLYS). The angle of repose (AOR) (wood and glass), bulk density, true density, bulk volume, and surface area were analyzed with varying temperatures (60 °C, 67.5 °C, and 75 °C) at a constant air velocity (1.75 m/s), as temperature increased, the AOR glass decreased significantly. In contrast, the AOR metal increased, the highest bulk density (0.61 kg/m3) was observed at 75 °C, bulk volume and bulk density decreased significantly with an increase in temperature, and surface area increased with an increase in temperature. The effect of the operating parameters (time, temperature, and air velocity) on the responses (bulk density, true density, bulk volume, and surface area) was investigated, modeled, and optimized via Response Surface Methodology (RSM). The Analysis of Variance (ANOVA) showed a second-order polynomial model with bulk density (R2- 0.999, Adj R2-0.9997, Pred R2-0.9979), true density (R2- 0.999, Adj R2-0.9997, Pred R2-0.9977), bulk volume (R2- 0.9970, Adj R2-0.9932, Pred R2-0.9527) and surface area (R2- 0.9953, Adj R2-0.9892, Pred R2-0.9247) indicating a close relationship between the experimental and predicted responses. The 3D graphs showed a significant impact of the process factors on the response. The optimal bulk density (0.81 kg/m3), true density (0.55 kg/m3), bulk volume (25 m3), and surface area (684 m2) were obtained at a temperature (57.5 °C), time (3 h), and air velocity (2.25 m/s). Artificial Neural Network (ANN) technique with 3 backpropagation algorithm (B·P.) algorithm was employed to analyze TLYS engineering properties; each algorithm was evaluated with 3 neurons in the input layer, 10 neurons in the hidden layer, and an output layer with four neurons. Coefficient of determination (R2) and mean square error (M.S.E.) have been implemented and correlated to test the adequacy of the model. Results showed that the Bayesian regularization had the best prediction for all the algorithms with an MSE (2.5465E-9) and R2 (9.999E-1) for the responses. Scanning Electron Micrograph (SEM) and proximate analysis indicate that the TLYS contains starch. This information from this study can be effectively utilized in the design parameters of the TLYS post-harvest process/machinery. Hence the post-harvest process evaluates the nutritional quality, protects food safety, and reduces losses between harvest and consumption.