Automatic Detection of Vibration Patterns During Production Test of Aircraft Engines

Abstract

Every day, several aircraft engines exit Safran plant in the south of Paris. Each engine is assembled and sent for a last bench test before shipment to the aircraft. Among all operations implemented during this hour length phase and after the first run-in, we realize a slow acceleration and a slow deceleration. During those two steps the engine is almost stabilized and a Fourier spectrogram may be evaluated for each rotating speeds. The engine has two shafts and two accelerometers. Hence for each operation we draw four spectrograms with a rotating speed on the x-axis and frequency ordinate. These images are transferred to a team of specialized engineers whose work is to look at the spectrograms and annotates zones they found unusual or patterns they already saw on past observations. Then if something may be of concern, a decision is taken to send the engine back for inspection. This still is today’s process, but several years ago we tried image analysis algorithms on the spectrograms and discover a way to help our team in the detection of abnormal patterns. We finally implement two algorithms which employed together give a pretty good detection of interesting zones. Subsequent sections in this paper will present the original data, show some abnormal patterns and describe our set of algorithms. Now, Safran develops a big-data environment with a datalake that capitalizes all measurements and our production test vibrations analysis should be operational in the coming years among lots of other algorithms.