Data-driven equipment feature analysis and anomaly detection

With the continuous applications expansion in the aerospace domain， the health management of complex equipment is oriented from scheduled maintenance towards data-driven predictive maintenance. A data-driven anomaly detection framework known as TS-ADF is proposed， which achieves effective identification of potential anomalies through the establishment of normal patterns， reconstructive analysis and feature fusion of multidimensional operational data. Specifically， preliminary screening is involved by using density peak clustering， deep features of time series are captured through LSTM-AE， and anomaly points are validated via time-frequency analysis and parameter variation analysis. Experimental results demonstrate the method's effectiveness in anomaly detection， which can serve intelligent health management and predictive maintenance of equipment.