Joint mining of fluid knowledge and multi-sensor data for gas–water two-phase flow status monitoring and evolution analysis

Abstract

Gas-water two-phase flow process widely exists in petroleum, chemical and other industrial processes, which is intricate, manifested by dynamic evolution of fluid structures and real-time fluctuation of flow parameters. Detailed description and real-time monitoring of flow status are of great significance to safe operation of industry processes. In this paper, a monitoring framework jointly mining the fluid knowledge and latent property behind multi-sensor data is proposed to address the issues of uncertainty and sample scarcity in transitional flow statuses, which enables real-time monitoring on structural evolution and parameter change of flow process. Firstly, the shared iterative algorithm with multi-task weight discovery is proposed to establish the relative semantic description system to characterize flow process knowledge by mining the relationship between multi-sensor data and different fluid semantic properties, which is helpful for interpretable monitoring of flow status structure and parameter information. Secondly, latent properties mined from multi-sensor data by sparse exponential discriminant analysis contribute to flow characterization completeness and discrimination. Thirdly, the process knowledge and latent information of flow status is transformed into different degrees of describable semantic and latent properties, which obtain relative representations of flow status in the joint description space realized by rank support vector machine. The results demonstrate that the method enables precise monitoring and detailed description of typical and uncertain flow statuses in the absence of transitional flow status modeling samples.