Data-Driven Identification of Two-Phase Microfluidic Flows

A data driven approach for the identification of fast two-phase flow nonlinear dynamics was investigated with the characteristics to be easily adopted in different experimental conditions and embedded in a portable device for real-time applications. Two experimental campaigns were conducted using different inputs flow rates and the optical signals acquired were processed and used for the identification step. Starting from the optical process monitoring, an analysis procedure in the frequency domain for the flow characterization was developed and a classification based on bubble frequency obtained. The nonlinear autoregressive with exogene input (NLARX) technique based on the wavelet network was used as identification method for the bubble flow modeling. To evaluate the models performance both in time and frequency domain, two errors were used. Two types of data patterns were considered. In the single patterns a model per experiment was identified and in the collective pattern the same model was used for more experiments.