Three-dimensional reconstruction of bubble distribution in two-phase bubbly flows with the dynamic programming method

Abstract

A three-dimensional bubble reconstruction method is proposed in this thesis to analyze two-phase bubbly flows. Gas/liquid two-phase flows have important roles in the nuclear and chemical industries and other engineering fields, but they are not completely understood yet. This indicates more experimental investigation is required. Bubble distribution is very important for the accurate description of bubbly flows and many methods have been proposed. However, conventional methods assume a fixed shape or size of bubbles, and this is not a realistic assumption for the flows with bubbles of complex shapes. The fundamental strategy of the proposed method is similar to tomography, which is an iteration process of guessing and evaluating. The advantages of the proposed method are that it does not need to separate a phantom of a single bubble from another in the shadow image and it does not need to assume fixed shapes or sizes of the bubbles. In the proposed method, a bubble distribution is represented as a metaball object with few parameters needed to describe three-dimensional complex bubble shape. Thus, the proposed method is able to analyze limited data acquired from digital cameras. In contrast, ordinary tomography requires large amount of data. The method is an application of dynamic programming. In this case, the problem of searching the optimal metaball object that represents the bubble distribution the best is divided into smaller problems of searching a semi-optimal object. This approach reduces the hypothetical solution space, and thus the number of match testing processes. Knowledge base is employed here for efficiency. The search algorithm consults the knowledge base to determine whether a solution candidate is worth performing a match testing. This approach isolates the search algorithm from the knowledge base so that a developer can easily add new rules. The method has been applied to synthetically generated shadow image sets in various conditions to evaluate the performance and the images acquired from the real experiment. The result indicates that the reconstructed bubble distribution can be very accurate.