Induction Motors are very important machines for most industries, and an unexpected interruption of them may cause an emergency plant breakdown. There are different spectral analysis techniques that can be used in induction motor signals of voltage, current or vibration, to identify different motor failures, but all of them have limitations. Specifically, there is not a technique capable to detect the broken rotor bar (BRB) failure in all induction motors cases. The Nonlinear Mode Decomposition (NMD) is an adaptive tool based on time-frequency spectral analysis techniques, that identifies and decompose a signal in interdependent oscillations. The method is based on the Windowed Fourier Transform or the Wavelet Transform, in an adaptive form, and it can decompose a signal in different meaningful oscillations, for a reliable analysis. In this work, the NMD have been used, in a combination with the multiple signal classification (MUSIC) and a digital resampling, to detect the BRB fault, analyzing current induction motor signals. The methodology is tested with both simulated and real signals from induction motors with different levels of BRB failure, obtaining satisfying results.