Gas-Liquid Flow Regime Identification Via a Non-Intrusive Optical Sensor Combined with Polynomial Regression and Linear Discriminant Analysis

The identification of gas-liquid flow regimes for co-current upward flow in a vertical pipe is investigated using Polynomial regression (PR) and Linear Discriminant analysis (LDA) on the response from a non-intrusive optical sensor. The average voltage and associated standard deviation can be extracted as features from the sensor and used as inputs to establish the coefficients of an optimal regression model. It is interesting to note that the feature relationship presents a clear 4th order polynomial trend. Further to this, due to the structural similarity between the slug and churn flow regimes, further discrimination was required in the form of a Linear discriminant analysis to give accurate classification between these two flow regimes. According to the offline test and real-time flow conditions considered, a classification accuracy of 100% is achieved. The results of this paper show the capability of applying a relatively simple supervised regression model and LDA to the response from a non-intrusive optical sensor to identify gas-liquid flow regimes efficiently and accurately. Applications to detection of flow regimes in process industries is a key application of this research