Article scientifique rédigé par Thibaut Le Magueresse, data-scientist chez Amiral Technologies, Jérémie Derré et Florent Mercat, Airbus Operations – Acoustics Testing Team
Cette publication scientifique est rédigée en anglais.
Introduction
This paper presents a study of sensor fault detection from an acoustic test bench, performed by machine learning. The concerned rig is based on the modal generation and detection principle, aiming at characterizing the acoustic properties of engine liners. Such a test mean is instrumented with a large number of sensors (i.e. more than a hundred and seventy sensors), and used in the frame of heavy experimental campaign with significant test matrices. In that context, it is of prime interest to ensure that all the sensors are behaving as expected, and to validate the current test point, before moving forward and change the configuration. Data validation becomes therefore fundamental, as the causes of failure are numerous in such a complex environment, and automatic processing avoids time losses, especially regarding the huge data quantity. The proposed approach lies in machine learning software, whose inputs are the temporal raw data, which therefore do not need to be pre-proceeded. First comparisons between the AI and human validations are presented in this paper.
