| Summary: | Diagnostic health monitoring without prior knowledge is still a hard problem in the prognostic and health management field. A multivariate diagnostic health monitoring strategy is proposed based on telemetry data for in-orbit spacecrafts with component degradation. Compared with the existing univariate or direct diagnostic health monitoring methods, multivariate diagnostic health monitoring methods can avoid constructing one-dimensional synthesized health index and setting empirical thresholds for different health states. In our developed strategy, a deep forest algorithm combined with an effective feature extraction approach and fuzzy C -means clustering algorithm is proposed to achieve more accurate assessment of the current health state. First, a partitioning window is utilized to deal with the raw telemetry data and then features which have high monotonicity and trends are extracted for diagnostic health monitoring. Then, a fuzzy C -means algorithm is used to handle unlabeled telemetry data and determine states of degrading component. Finally, a deep forest classifier is adopted to obtain the prognostic model for online probabilistic diagnostic health monitoring. Verification results on a simulated spacecraft attitude control system can demonstrate the effectiveness and feasibility of the proposed multivariate diagnostic health monitoring strategy.
|
|---|
