Résumé:
This thesis proposes an innovative approach for diagnosing faults in rotating machinery by utilizing Variational Mode Decomposition (VMD) to enhance detection accuracy and robustness in complex environments. It begins with a review of condition-based maintenance strategies, highlighting the limitations of traditional methods and the potential of advanced signal processing techniques.
VMD is compared to Empirical Mode Decomposition (EMD) to demonstrate its superiority in isolating fault signatures from noisy signals. A criterion based on Shannon entropy is proposed to optimize the number of Intrinsic Mode Functions (IMFs). Additionally, advanced methods such as Wavelet Multi-Resolution Analysis (WMRA) is integrated with VMD to refine the detection of subtle faults.
A major contribution of this research lies in developing a hybrid framework combining VMD and Long Short-Term Memory (LSTM) networks for fault classification and severity assessment. This combination effectively leverages signal decomposition and sequential learning, achieving high accuracy even with limited data and demonstrating strong potential for industrial applications.
In conclusion, the VMD method proves particularly effective for diagnosing faults in rotating machinery under challenging operating conditions, opening new avenues for condition-based maintenance
