Unequal time series clustering applied on flight data

Multiple signals are measured by sensors during a flight or a test bench and their analysis represent a big interest for engineers. These signals are actually multivariate time series created by the sensors present on the aircraft engines. Each of them can be decomposed into series of stabilized phases, well known by the experts, and transient phases. Transient phases are merely explored but they reveal a lot of information when the engine is running. The aim of our project is converting these time series into a succession of labels, designing transient and stabilized phases. This transformation of the data will allow to derive several perspectives: on one hand, tracking similar behaviours or patterns seen during a flight; on the other, discovering hidden structures. Labelling signals coming from the engines of the aircraft also helps in the detection of frequent or rare sequences during a flight. Statistical analysis and scoring are more convenient with this new representation. This manuscript proposes a methodology for automatically indexing all engine transient phases. First, the algorithm computes the start and the end points of each phase and builds a new database of transient patterns. Second, the transient patterns are clustered into a small number of typologies, which will provide the labels. The clustering is implemented with Self-Organizing Maps [SOM]. All algorithms are applied on real flight measurements with a validation of the results from expert knowledge.