This is very application specific question. In rotating machines discussed in webinar, simpler RMS level of vibration can already indicate you the failure. Then you can move to FFT based analysis or other more effective filtering methods, such as enveloping. Anomaly detection using AI algorithms is interesting new way emerging, please follow ST press releases for pre-made solutions we are planning to launch on this shortly. In some other applications you might want to use other sensors such as ultrasound microphones, temperature, current sensing and related algorithms.
read here
https://newsroom.st.com/media-center/press-item.html/t4259.html
