Genetic Programming-Based Feature Construction for System Setting Recognition and Component-Level Prognostics

Extracting representative feature sets from raw signals is crucial in Prognostics and Health Management (PHM) for components’ behavior understanding. The literature proposes various methods, including signal processing the time, frequency, time–frequency domains, selection, unsupervised learning. An emerging task data science Feature Construction (FC), which has advantages of both selection In particular, constructed features address a specific objective function without requiring label during construction process. Genetic Programming (GP) powerful tool to perform FC PHM context, as it allows obtain distinct depending on analysis goal, i.e., diagnostics prognostics. This paper adopts GP extract system-level machinery setting recognition component-level Three fitness functions are considered training, requires set statistical time-domain input. methodology applied vibration extracted test rig run-to-failure tests under different settings. performances evaluated through classification accuracy Remaining Useful Life (RUL) prediction error. Results demonstrate that GP-based classify known novel operating conditions better than learning methods.