Model based analysis of precursors of electromechanical servomechanisms failures using an artificial neural network

Several approaches can be employed in prognostics, to detect incipient failures of primary flight command electromechanical actuators (EMA), caused by progressive wear. The development of a prognostic algorithm capable of identifying the precursors of an electromechanical actuator failure is beneficial for the anticipation of the incoming failure: a correct interpretation of the failure degradation pattern, in fact, can trig an early alert of the maintenance crew, who can properly schedule the servomechanism replacement. Prognostic, though, is strictly technology-oriented as it is based on accurate analysis of the cause and effect relationships. As a consequence, it is possible that prognostics algorithms that demonstrate great efficacy for certain applications (electrohydraulic actuators, for examples) fail in other circumstances, just because the actuator is based on a different technology. The research presented in this paper proposes a prognostic technique able to identify symptoms of an EMA degradation before the actual exhibition of the anomalous behavior; to this purpose friction, backlash, coil short circuit and rotor static eccentricity failures are considered. An innovative model-based fault detection neural technique is proposed to analyze information gathered through FFT analysis of the components under normal stress conditions. A proper simulation test bench was developed: results show that the method exhibit adequate robustness and a high degree of confidence in the ability to early identify an eventual malfunctioning, minimizing the risk of false alarms or unannunciated failures.