This work investigates the application of empirical, statistical and machine learning methods to appraise the prediction of transmembrane pressure (TMP) by oscillating slotted pore membrane for the treatment of two kinds of deformable oil drops. Here, we utilized the previous experimental runs with permeate flux, shear rate and filtration time as features, while TMP of crude oil and Tween-20 were two distinct targets. For 87 experimental runs, Response surface methodology (RSM) and Artificial Neural network (ANN) modelling were opted as statistical and machine learning tools, respectively, which were comprehensively compared with empirical slot-pore blocking model (SBM) considering accuracy and generalization. ANN with 10 neurons in the hidden layer could approximate the TMP of both oils better than RSM and SBM, which is reflected by computed performance metrics. Under the given conditions, almost similar analysis were predicted for TMP of both oils except changes in magnitude which were interpreted by (1) line plots, which showed that TMP of crude oil and Tween-20 were linearly related to flux rate and filtration time, and there was an inverse relationship between TMP and shear rate, (2) contour plots, which illustrated the strong interaction effect of flux rate and time on TMP, and (3)- sensitivity analysis, which revealed the influential sequence of variables on TMP as; flux rate > filtration time > shear rate, for both cases. The optimisation of the process showed that minimum TMP can be attained by maintaining higher shear rate and lower flux rate and time. Conclusively, the current findings indicate the utilization of ANN for the accurate assessment of TMP and can be helpful for the process designing and scale up.
Read full abstract