Abstract: The interface detector with reduction boundary samples(RI-detector) is presented based on introduce the term boundary samples of self-space, which can detect the abnormal degree of equipment rapidly without fault sample. Take the Iris data set as examples for analysis, and then find that RI-detector is an abnormal detection method with high detection rate, high false alarm rate and strong compression capability for data. RI-detector shows a better detection performance by comparison with other commonly used anomaly detection methods; especially where the data sets has a clear boundary. It not only reflects the various states of bearing, but also reflects the fault degree pass the abnormal degree of the same equipment failure when analyzed the bearing state data used RI-detector. The RI-detector can detect the faults of equipment by learning normal data without fault data, which is built with reduction boundary samples, their boundary location information and the self-radius. It can efficiently detect the faults of the equipment that lacks fault data.

Key words: abnormal degree, anomaly detection, bearing, interface detector, negative selection algorithm