A global bandwidth smoothing approach in the multi-response optimization for oil-in-water emulsion
Response surface methodology (RSM) data was examined for the ultrafiltration of oil-in-water emulsion procedure via the use of three-phase experimental design methods namely: design of experiment (DOE), modeling phase of the estimated regression model, and the optimization phase. A quadratic regression model (QM) was employed to fit the data, which performed well in areas where the data exhibited a quadratic trend or curvature. Nevertheless, the QM showed limitations near the boundaries with linear trend of the data due to a boundary bias problem. To address this issue, ordinary least squares (OLS) regression and a global or fixed bandwidth (FB) applied to local linear regression ( model was considered via an experimental design technique such as the Central Composite Design (CCD). The model aimed to improve goodness-of-fit statistics, the zero-residual line, and optimization results. The comparative analysis, including 3D surface plots, the outcome showed that ( performed better than both OLS and QM in terms of the design technique. The ( model demonstrated superior goodness-of-fit statistics, minimized residual errors, and better optimization results, achieving an overall desirability of 61.31%, as compared to 56% for OLS and 55% for QM respectively.Top of Form
Keywords: Global local linear regression model, Chemical oxygen demand (COD), Permeate flux (PF), Trans-membrane pressure, Turbidity removal, Velocity.
*Correspondence: oeguasa@biu.edu.ng, +2348028631580
Download Paper