The complexity of the current installations requires advanced and effective monitoring techniques. The most commonly used technique is the vibratory analysis. Despite the large number of existing methods for detection, diagnosis and monitoring of bearing defects, the scientific community is widely interested in learning methods. These methods allow automatic detection and reliable diagnosis. This paper presents anew real-time unsupervised pattern recognition approach for the detection and diagnosis of bearings defects: RT-OPTICS. This approach focuses on two steps of damage evolution: defect detection by classification and monitoring of the new cluster representing the degraded state of the bearing. These two steps are performed by a two-dimensional method implementing scalar indicators: Kurtosis and Root Mean Square values. These two indicators provide additional information about the presence of defects in the bearing. The first step deploys RT-OPTICS based on the real-time unsupervised ordering points to identify clustering structure (OPTICS) classification to detect defects on inner and/or outer bearing races. The next step is to monitor the state of degradation using three parameters of the new cluster: the center jump, density and contour of this cluster. After a validation on simulated signals which variations of parameters were tested, this approach was tested under experimental conditions on a test bench made up of N.206.E.G15bearings, with varying load and angular velocity. A comparative study is carried out between the suggested approach and (i) a classical approach: monitoring of scalar indicators over time and (ii) a dynamic classification method (DBSCAN).