New prognostic neural method by discrete wavelet transforms for electromechanical flight controls affected by progressive faults

Matteo D L Dalla Vedova,Paolo Maggiore,Nicola Lampariello,K Kontis,S Pantelakis

doi:10.1051/matecconf/201823300017

Matteo D L Dalla Vedova, Paolo Maggiore + Show 3 more

Open Access

https://doi.org/10.1051/matecconf/201823300017

Copy DOI

Abstract

Progressive failures affecting onboard electromechanical actuators (EMA), especially if related to primary flight commands, could be a critical issue for the aircraft reliability and, in the worst cases, could compromise its safety. In the last years strong interest is expected by the development of prognostic algorithms able to provide an early identification of the precursors of EMA progressive failures. In this work authors proposes a new prognostic method based on two artificial neural networks (ANN), a basic and an enhanced feedforward neural network, performing the fault detection and identification of two critical progressive faults often affecting the EMA brushless motor (i.e. turn-to-turn short circuit of a stator coil and rotor static eccentricity); in order to identify a suitable data set able to guarantee an affordable ANN classification, the said failures precursors are properly pre-processed by a Discrete Wavelet Transform, extracting several features used as input of the proposed prognostic algorithm.

Highlights

In the recent decades, the electromechanical actuator (EMA) is playing an increasingly important role as an augmented flight control system in fly-by-wire architectures
It must be noticed that, with respect to other EM models available in literature, the numerical model shown in the previous sections is able to calculate the instantaneous value of each current phase (Ia, Ib, Ic) in case of unbalanced electromagnetic system; it is possible to correlate the progressive faults with the dynamic response of these signals by means of an algorithm, based on the Wavelet analysis, that evaluates the filtered phase currents; for this purpose, each phase current is filtered by three low pass signal filter, in order to attenuate noise and disturbances
The principal issue, for a neural network and more generally for prognostics, is to preprocess output signals from sensors, since the network needs a separable set of data to accomplish an affordable classification

Summary

Introduction

The electromechanical actuator (EMA) is playing an increasingly important role as an augmented flight control system in fly-by-wire architectures In this scenario, studies in prognostics will be necessary to reduce maintenance costs and preserve safety because, otherwise from mechanical fatigue, which can be estimated with a reliable confidence level, EMA electrical failures, like a partial stator phase short-circuit or rotor eccentricity, are difficult to predict using an external analysis. Proper PHM strategies could handle progressive failures in a more effective way leading to a substantial reduction of system redundancies, operating costs, maintenance interventions and, on the other hand, improving the aircraft safety and reliability plus simplifying logistics (as reported in [2]) To these purposes, in this paper is proposed be the authors a new Fault Detection and Identification (FDI) technique [3], based on Artificial Neural Networks (ANNs), able to identify the failure precursors and predict the corresponding damage entity. The algorithm effectiveness has been evaluated by a dedicated MATLAB-Simulink® numerical model, able to analyse the EMA performance and the responses of two progressive faults; the so obtained results demonstrate the robustness of the method and the ability to identify the incoming failures, reducing the possibility of false alarms rather than non-predicted problems

Primary flight control EMA

EMA numerical model

Considered EMA failures

Wavelet Neural Network

Proposed ANN fault detection algorithm

Conclusions