Adaptive Learning for Stall Pre-cursor Identification and General Impending Failure Prediction
Frontier Technology, Inc (FTI) received a contract award to assist the Navy by developing innovative computational tools for the analysis of performance and usage data to predict aircraft engine stall. In the Phase I, FTI will provide a proof of concept of a computational tool which can extract and analyze engine and field data to predict engine stall and identify the engine data indicators which led to that stall prediction.
Modern propulsion systems for Naval Aviation, full authority digital engine control [FADEC] systems, and supporting health management systems have the capability to collect and analyze large datasets of engine performance and usage information. This data offers the potential to detect impending failures and component performance degradation in flight and post-flight and flag these for compensation by the engine control system. This could also help to provide more timely maintenance attention which would positively impact operational availability, reliability, and safety. An example of a primary concern is intermittent gas turbine engine compressor stall (due to gradual component wear and fouling in service) that causes mission aborts and results in extensive troubleshooting at the flight line.
Innovative computational tools are needed for the analysis of performance and usage data to predict aircraft engine stall. These tools should allow time for control system software to take preemptive corrective action in order to avoid engine stall events, as well as identify and assess the state of the engine components causing this behavior and estimate the remaining useful life (RUL) of these critical engine components. Machine and adaptive learning techniques, effective searching algorithms of engine large data sets, statistical analysis methods, and adaptive neural networks are some of the tools seen to have promise for attacking this problem.
The proposed computational tools should be able to provide diagnostics and prognostics of aircraft engines, modules, subsystems, and components. The tools should also be able to: conduct failure mode and effects analysis to identify the failure modes and root causes and assess their impact; use machine learning techniques and neural networks to extract rules and knowledge underlying the available engine large data sets; predict impending engine stall events and other types of performance degradation, and estimate remaining useful life of the critical components driving the degrades engine performance; verify the performance and accuracy of the results against the available engine data sets; adaptively learn from newly generated engine data; inform engine maintenance staff and field engineers of imminent problems; interface seamlessly with the current engine FADEC and other hardware systems; and provide engine designers with field-data based feedback to enable them to improve the design of current FADEC systems and future propulsion systems.
KEYWORDS: Data Mining; Data Fusion; Adaptive Learning; Neural Networks; Engine Stall; Stall Precursor