[home] [Personal Program] [Help]

---

Session: Poster pitches day 1
Session starts: Monday 26 June, 09:50
Presentation starts: 09:50

---

Michael J. Scott (RMIT University)
Michael J. Candon (RMIT University)
Wim J.C. Verhagen (RMIT University)
Oleg Levinski (Defence Science and Technology Group)
Pier Marzocca (RMIT University)

Abstract:
The proposed paper describes the use of a machine learning approach for fault diagnostics and prognostics of aircraft control surface free-play using on-board sensors. The significance of this research is in detecting and tracking the aircraft control surface free-play on an all-movable horizontal tail. This will progress proactive condition-based maintenance of free-play through the development of a data-driven Prognostics and Health Management (PHM) framework. Free-play requires labour-intensive maintenance, limits aircraft performance, and can induce aeroelastic asymmetries (i.e., limit cycle oscillations), reducing component life and is therefore costly to operators. The aim of this work is to predict free-play using aircraft on-board sensors; the detailed research questions are outlined by the future work in the authors’ recent paper [1]. The research method utilises time- and frequency-domain signal processing of control surface actuator loads, generating 17 features (e.g., peak frequency, peak amplitude, standard deviation, skewness, kurtosis, etc.), which are labelled “high” or “low” in severity based on measured free-play values. Subsequently, a supervised machine learning model is used to diagnose free-play severity, while prognostics estimate Remaining Useful Life (RUL) for free-play using an optimised exponential degradation model. Preliminary results show good agreement with exact free-play measurements, with a K-Nearest Neighbours (KNN) binary classification model generating an accuracy of 88.6%. The prognostics degradation model produces good early to mid-life prediction for the piece-wise linear RUL target. However, the overall RUL prediction has a root-mean-square-error value of 37.47 samples, which is impacted by over-estimation near the nominal failure threshold later in the degradation life. Nonetheless, this is an important step towards predicting free-play on an aircraft stabilator using on-board sensors. The final manuscript will incorporate more representative flight data and maintenance interventions, as well as increased complexity in diagnostic and prognostic models to handle the non-linearities and behaviour of control surface free-play, further fine-tuning prognostics model parameters. [1] M. J. Candon, M. Scott, S. Koschel, O. Levinski, and P. Marzocca, ‘A Data-Driven Signal Processing Framework for Enhanced Freeplay Diagnostics in NextGen Structural Health Monitoring Systems’, in AIAA SCITECH 2022 Forum, American Institute of Aeronautics and Astronautics, 2021. doi: 10.2514/6.2022-2131.