Bearing condition monitoring (BCM) and damage identification are usually performed by vibration-based signals and supervised learning algorithms. However, this approach is impractical in many industrial facilities because some industrial motors are unable to provide access to vibration-based signals or they are prevented from performing their functions under damaged conditions. In this context, this work employs a density-based and fractal approach to extract features from current-based signals. These features create an unlabelled database that feeds a fuzzy c-means algorithm to perform BCM and the support vector machines to classify bearing damages in an unsupervised learning approach. Tests with several bearing damages under various load and speed conditions are reported, presenting promising results.