Nowadays, industries typically monitor the health of its machinery based on sensors, collecting data (e.g. vibration signals) with high frequency in order to provide real-time information, and thus avoid any delay in detection of an abnormal behaviour. Machine learning algorithms are often applied to classify degradation condition (e.g. normal, damaged, critical) and infer about the Remaining Useful Life (RUL), automating the process of fault detection and/or of prognostics to make preventive decisions. In this context, the definition of the essential features to predict important measures such as RUL can be challenging and highly application-dependent. Moreover, the machine learning performance is inherently limited if incomplete or erroneous features are defined. Deep learning is a data-driven approach that emerges as an alternative for human-based feature description, and it has presented good performance in the prediction of reliability-related metrics such as RUL and system health indicators. Therefore, this work proposes the use of a Convolutional Neural Networks (CNN) to predict RUL of bearings under accelerated degradation. Real data provided by IEEE PHM 2012 Data Challenge was used in which vibration time-series was monitored. Although specific results do not present excellent performance, the model may adopt other ways for improvement with also possibility of using other deep learning approaches.