The anticipated high volume of gravitational-wave observations in the near future will require the development of reliable, unsupervised techniques for data quality assessment and signal detection and interpretation. We present a simple noise monitoring pipeline for gravitational-wave detectors that uses self-similarity analysis and an unsupervised machine learning anomaly detection algorithm. The approach may be used in real time to detect non-astrophysical noise transients at different time scales, as well as to identify periods of noise non-stationarity. We demonstrate how it works with two examples of data collected by one of the LIGO interferometers during the third observation run of the LIGO, Virgo, and KAGRA collaborations.