TutORial: Change Detection & Prognostics for Transient Real-World Processes Using Streaming Data

By Satish Bukkapatnam and Ashif Sikandar Iquebal

Recent advances in sensor arrays and imaging systems have spurred interest in analyzing high-dimensional, streaming time series data from real-world complex systems. These time series data capture the dynamic behaviors and causalities of the underlying processes and provide a computationally efficient means to predict and monitor system state evolution. More pertinently, they can provide the ability to detect incipient and critical changes in a process, which is essential for real-time system integrity assurance. However, effective harnessing of information from these data sources is currently impeded by the mismatch between the key assumption of stationarity underlying most change detection methods and by the nonlinear and nonstationary (transient) dynamics of most real-world processes. The current approaches are slow or simply unable to detect qualitative changes in the behaviors that lead to anomalies. For most real-world systems, the vector field of state dynamics is a nonlinear function of the state variables, i.e., the relationship connecting intrinsic state variables with their autoregressive terms and exogenous variables is nonlinear. Time series emerging from such complex systems exhibit aperiodic (chaotic) patterns even under steady state. Also, since real-world systems often evolve under transient conditions, the signals so obtained tend to exhibit myriad forms of nonstationarity. This tutorial presents a delineation of these diverse transient behaviors, and a review of advancements in change detection and prognostication methods for nonlinear and nonstationary time series. We also provide a comparison of their performances in certain real-world manufacturing and health informatics applications.