GOURI, Nesrine2026-02-042026-02-042025https://dspace.univ-annaba.dz//handle/123456789/4555In modern industrial applications, the reliability of production facilities is strongly linked to the performance of rotating machines driven by static power converters. This configuration offers high energy efficiency but demands rigorous maintenance to prevent costly failures. This thesis addresses the problem of detecting bearing faults, which are among the most common and critical failures in such systems. The goal is to extract useful information that characterizes the machine's health state by analyzing vibration signals. To achieve this, we develop a methodology based on mathematical modeling and numerical analysis of the measured vibration signal. A novel feature extraction approach is proposed, combining Minimum Entropy Deconvolution (MED) and the Van Cittert Algorithm (VC). The MED technique is used to construct an inverse filter that reduces the masking effects of the transmission path. Then, the VC, regularized using the Tikhonov-Miller method, enhances the reconstruction of impulsive components related to faults. This combined strategy improves the clarity of fault indicators and supports more accurate and timely fault detection. The results contribute to more efficient condition monitoring and predictive maintenance in converter-fed rotating machinery.PDFenfault diagnosis; rotating machine; vibration signal; bearing; deconvolution; iterative algorithmThesis