This paper presents an industrial application of machine learning techniques in fluid mechanics to the two-phase flow identification problem, under the context of forced high-speed oscillation with reference to the cooling gallery inside automobile engine pistons. We propose a video classification method to characterize two-phase flow patterns within a simplified cooling gallery model. Different experimental evaluations show that our approach is effective to identify two-phase flow patterns. To our knowledge, this is the first paper to introduce visualized study of two-phase flows with machine learning techniques in related engineering domains.