APPLICATION TYPES - DIAGNOSIS

The diagnosis of faults in avionics systems was the first complex application area in which the EFAR technology was suceesfully tested. The data used in diagnosis was collected by several hundreds of sensors monitoring the operation of multiple on-board aircraft systems over extended periods of time. The data logs showed occasional anomalies in the behaviour of avionics components. However, the parametric singularities by themselves were not sufficient to indicate whether the anomaly causes were due to transient flight conditions or to component malfunctioning. Furthermore, in the latter case it was next to impossible to determine what other conditions, reflected in the recorded data, were associated with the malfunctioning, in order to develop a more general description of that type of incident.

The use of EFAR technology allowed the separation of the two categories of incidents by showing that the parametric singularities associated with component malfunctioning were correlated with other events present in the flight data. More specifically, EFAR tools allowed analysts to reliably relate flight parameter anomalies with consistent behaviours of other components which contextualised the anomalies. By developing, asessing, and revising hypotheses, engineers were able to develop and test diagnostic cases and anomaly types, and to decide on the specific measures that needed to be taken.

The input of the expert's domain knowledge working as a data analyst was critical in the success of the EFAR technology. The volume of data would had previously defeated uninformed analysis technologies, which were not guided by a knowledge-based, domain-anchored strategy. Equally important was the rendering of the individual avionics system parameters in a format that was aligned with the specific features for which avionics engineers were looking for in flight data. Ultimately, the EFAR technology proved itself capable to assist domain experts to develop causal models, based on real-world data, which helped identify causes for problems beyond what was known without using sophisticated data analysis.