Comparative study between EMD, VMD, SAGE, CLEAN, and ESPRIT-TLS algorithms for real-time fault bearing recognizing in induction machines.

Real-time monitoring of electromechanical faults is today at the heart of scientific research in the industrial sector. Indeed, the idea of this work is therefore to show, through a comparison based on performance indicators (precision, calculation time, occupied memory size, and convergence), which the processing algorithm of the signal is more robust in detecting the spectral characteristics of a fault using the MCSA method (Machine Current Signature Analysis). To do this, we used indicators such as NMSE to calculate the estimation errors on the estimated frequencies of the faults by these algorithms. The MATLAB simulations of the detection of a bearing fault, known by its spectral characteristics (two, four and six harmonics) and injected into a stator current signal with a variable level according to SNR, showed that the ESPRIT-TLS method exceeds the VMD, EMD, SAGE, and CLEAN algorithms and occupies the first place in the precise discrimination of this type of fault, especially when its amplitudes are low and even when close frequencies are used.