Towards objective flow pattern mapping with the K-means clustering algorithm

Abstract

Paper presented at the 6th International Conference on Heat Transfer, Fluid Mechanics and Thermodynamics, South Africa, 30 June - 2 July, 2008.

Two-phase flow regime prediction is of great importance for designing evaporators and condensers because the influence of the heat transfer coefficients is strongly related to the flow regimes. These flow regimes are often presented using flow pattern maps. As most flow pattern maps are based on visual observation, they lack objectively defined flow regime transition criteria. In order to add flow characteristics to the transitions boundaries, a sensor was developed which measures the capacitance of the two-phase flow. Due to the difference in dielectric constant of liquid and vapour and the dependency on the capacitance to the internal distribution of liquid and vapour in the cross-section of the tube, the sensor is able to characterize two-phase flow regimes. A large number of experiments was done with air-water flow. The setup was able to cover three main flow regimes for horizontal flow in a 9 mm tube, namely stratified, annular and intermittent flow. A multivariate analysis was performed to find characteristic signal parameters. The average signal value, together with the variance and a high frequency contribution factor were found suitable. These parameters were used as input features for the k-means clustering method, which groups the sensor data into a given number of classes. The influence of the weight parameters of the features was mapped, as well as the influence of the distance function. A comparison between the visual classification based on high speed camera images and the cluster classification shows a remarkable agreement.