Automatic threshold setting for anomaly detection

The paper presents automatic method for calculation of threshold value and its use for detection of anomalies in vibration signals for the purpose of condition monitoring. Anomalies are understood as time series (trends) calculated from consecutive spectra from individual vibration signals with significantly different characteristics than remaining trends. In this sense, the paper presents a classification method within anomaly (or novelty) detection; yet, without any black-box techniques. The method is based on the detection of specific gap of histogram of selected statistical indicator. Thus, the method relies on hypothesis that any deterioration of technical condition of rotary machinery results is increase of vibrations. The method is validated on real data. The performance of the method is compared with other typical methods, both wideband, and narrowband.