Model based fault detection neural technique for electromechanical servomechanisms

Incipient failures of electromechanical actuators (EMA) of primary flight command, provoked by progressive wear, can be identified with the employment of several different approaches. A strong asset is expected by the development of a prognostic algorithm capable of identifying the precursors of an electromechanical actuator failure. If the degradation pattern is well understood, it is possible to trig an early alert, leading to proper maintenance and servomechanism replacement. Prognostic, though, is as it is. As such algorithms are strictly technology-oriented and based on accurate analysis of the cause and effect relationships, they may show great effectiveness for some specific applications, while mostly failing for different applications and technologies. This work proposes an approach with a demonstrated benefit from a prognostics point of view. Friction, backlash, coil short circuit and rotor static eccentricity failures are considered. A model-based fault detection neural technique is defined and used for the assessment of the data obtained through Fast Fourier Transform (FFT) analysis of the components under normal stress conditions. A simulation test bench has been developed for the purpose, demonstrating that the method is robust and is able to early identify incoming failures, reducing the possibility of false alarms or non-predicted problems.