Automatic failure diagnostics techniques and health features investigation for electro-hydraulic actuators.

In passenger aircraft, the Electro-Hydraulic Servo Actuators (EHSA) are still the leading technology used for primary and secondary flight controls. Their Maintenance, Repairing and Overall procedures (MRO) are predominantly manual and high-priced. Therefore, Lufthansa Technik has developed an innovative and automatic testing procedure, with the purpose to reduce the MRO cost and improve the diagnostic accuracy. The servo actuator is commanded by means of a modular excitation signal in order to extract significant features from the unit response. Then, the analysis of the feature points out the faulty components inside the unit. The present thesis introduces new troubleshooting techniques by the extraction of additional health features concerning the elevator servo actuator. The first part of the thesis covers the correlation study between the features extracted by the manual and the automatic test. High values of correlation coefficient validate the automatic testing procedure, which is suitable for investigating deeper the servo actuator behaviour. The second part includes the investigation of further improvements concerning the diagnostic procedure. The work developed consists in analysing the LHT shop database through non-parametric statistical methods, such as Kernel density estimator, evaluating the sub-components durability and the health features goodness. Subsequently, the same database has been used to identify possible correlations between the sub-components. The algorithm, developed in Python, calculates the conditional probability between different shop events. Then, the relationships highlighted are evaluated in the feature extraction section, which aims to isolate the behaviour of the component. Given a set of time series recorded in the LHT test field, the study of two typical degradations has been addressed: the mode switching valve spool contamination and the servo valve feedback spring degradation. By using post-processing methods, a new set of features has been selected in order to study the degradations. The health feature validation has been driven by the shop load information. Finally, the k-means clustering method has been used to highlight the strong correlation between the features and the degradations.