Scalable time-series feature engineering framework to understand multiphase flow using acoustic signals
Maruti Kumar Mudunuru, Vamshi Krishna Chillara, Satish Karra, Dipen N. Sinha · Proceedings of meetings on acoustics · 2017
Time-series signals are central to understand and identify the state of a dynamical system. They are ubiquitous in many areas related to geosciences, energy, and climate. As a result, the theory and techniques for analyzing and modeling time-series have vast applications in many different scientific disciplines. One of the key challenges that the time-series data analysts face is that of data overload. The sheer volume of the data generated at the sensor makes it difficult to transport it to centralized databases. These aspects pose an obstacle in detecting changes in the system response as early as possible. Instead, a workflow for an efficient and automatic reduction of collected data at sensors can enable timely analyses and decrease event detection latency. Such a workflow can be useful for many sensing applications. An attractive way to construct a computationally efficient workflow for automated analysis of sensor data is through machine learning. Herein, we present a novel framework for feature construction, feature selection, and feature filtering. As a first step, we construct comprehensive time-series signal features. In the second step, we perform feature selection using machine learning algorithms. The proposed framework is tested and validated against datasets obtained from multiphase flow loop experiments.