Automatic Discovery of Precursors in Time Series Data (ADOPT) code release(ARC-18198-1)

ADOPT (abbr. Automatic Discovery Of Precursors in Time series) is a data mining/ machine learning algorithm that processes large volumes of time series data and identifies precursors to adverse events. An adverse event may refer to any unsafe event ranging from a negligible safety hazard to a catastrophic accident, depending on the scope of the analysis. A precursor is an early indicator of an increasing likelihood of the adverse event. Identifying precursors is important in the context of a proactive safety management because precursors detect the increasing severity of the underlying hazard much earlier, giving sufficient time to identify, analyze and implement corrective actions. ADOPT analyzes large volumes of historical data to find complex trends among several sensory variables simultaneously to find precursors. ADOPT's data mining approach captures real-world effects such as human factors, weather, geographic constraints, operating procedures, airline strategies etc. that are difficult to capture using first-principle models. The following are ADOPT's capabilities - new functionality to discover precursors to events of interest by mining time series data, reduces the time required by subject matter experts to discover and analyze precursors from large volumes of data, detects the likelihood of the adverse event earlier to alert the operator on an impending failure, assists in explaining the event of interest with identified precursors.The ADOPT algorithm is a general methodology that does not make any special assumptions about the system and does not need specialized knowledge on the system states enabling it to operate directly on the observed time series data. Further, it can scale well to multivariate time series and can analyze large number of flights, which may enable a faster turnaround time for subsequent tasks such as hazard identification and safety risk analysis. A subject matter expert may require a few hours to analyze a flight for precursors which is not scalable considering the thousands of flights operated on a daily basis. Also, human experts may not be able to visualize hundreds of time series variables to notice complex variations and trends in the data. ADOPT may be used to speed this process by analyzing the thousands of flights that operate every day to short-list only the significant precursors which may then be analyzed by a subject matter expert, reducing the turnaround time for safety analysis.While the motivation to find precursors came from analyzing safety (negative events), the method is general to be applied to performance issues as well as to understand positive events such as finding precursors to a high throughput operation at an airport.