Application of CHMMs to two-phase flow pattern identification

Abstract

In this paper, the application of continuous hidden Markov models (CHMMs) in identifying two-phase flow patterns is investigated. Air–water two-phase flows were realized in a transparent vertical tube with a 2 m length and a 19 mm inside diameter. Local void fraction signals were collected using a single-step index multimode optical fiber probe located at the center and mid-length of the tube. Walsh–Hadamard transform, an autoregressive model and an innovative method based on the passage length of the phases were used to extract signal features required in the CHMM implementation. CHMMs were trained for nine reference flow conditions, and were used to identify the flow patterns of 60 different flow conditions. Two different approaches were compared to treat log-likelihood results: maximum total likelihood and maximum likelihood. The results from the passage length-based method, in combination with the maximum total likelihood approach, were in relatively good agreement with a theoretical flow pattern map and photographs of the flow captured during the experiments. In sum, the results showed that a CHMM has a good potential in identifying two-phase flow patterns.