| Summary: | Rolling element bearings are one of the most critical mechanical components of various rotary machinery. Their natural operation is necessary for the smooth operation of the whole system. Any deviation from this balance indicates the existence of a fault that urgently needs to be detected and dealt with, in order to avoid its dispersion to the rest of the system and breakdown of the machinery. This task is managed by Prognostics & Health Management (PHM), which aims at early diagnosis and prognosis of the imminent failure. Artificial Neural Networks (ANNs) and Convolutional Neural Networks (CNNs) have shown great advantages in automated fault diagnosis and prognosis without depending on the human factor. In this paper a study of Neural Networks, their supervised learning and their usefulness on diagnostics and prognostics of rolling element bearings is conducted. Their effectiveness is shown after their implementation in vibration data, which have been acquired from the experimental platform PRONOSTIA of FEMTO – ST institute. This paper proved the superiority of 1D CNNs over ANNs in fault diagnostics and their capability in estimation of Remaining Useful Life (RUL) on rolling bearings under the same operating conditions.
|
|---|
