The fault detection, diagnostics, and prognostics of bearings play a key role in increasing the reliability, availability, and efficiency of rotating machinery. Signal processing techniques are useful for the health condition monitoring of rotating machinery. Fast-Fourier transformation, Wavelet transformation, and Hilbert-Huang transformation are three main time-frequency techniques to analyze vibration signals and extract representative features to monitor the health status of bearings. Hilbert-Huang transformation has attracted lots of attention due to its advantage in handling non-linear and non-stationary signals. However, identifying an appropriate feature, fault diagnosis, and remaining useful life prediction of roller bearings has remained an open challenge and highly dependent on the applied applications and methods. In this paper, two accelerated failure tests have been conducted on real bearings at NTNU to collect vibration data of bearings from a healthy state to the failed state. The degradation mechanism to degrade the bearings is contamination which is mixing solid particles, i.e., “Silisum Carbide” and a lubricant to pour into the bearing continuously. Empirical Mode Decomposition (EMD) has been applied to decompose the signals into different Intrinsic Mode Functions (IMFs) and the most informative IMF has been selected to extract Hilbert energy as a time-frequency feature for degradation modeling. RMS as an energy representative feature in time-domain has also been employed and compared with Hilbert energy to predict the failure time of the experimental bearings.