Contribution au diagnostic des défauts des arbres tournants

Abstract

The main objective of this thesis is the numerical integration of acceleration signals to identify the most suitable parameter for detecting and specifying faults in rotating shafts. The proposed approach has been validated on both simulated signals and real signals obtained from the Machine Fault Database (MaFaulDa) related to rotating shaft faults. In this context, there are four operational categories that are taken into account, namely normal state, unbalance, vertical misalignment, and horizontal misalignment. The dataset under study contains 880 acceleration signals, which are then integrated into velocity signals and displacement signals. The feature vectors used consist of scalar indicators as well as frequency-based indicators, totaling 36 indicators for each parameter. The classification method employed is the artificial neural networks Multi-Layer Perceptron (MLP).