Has anyone tried to model rolling element bearing vibration signal with HMM or HSMM? I have placed a lot of effort and time in trying and I haven't come to any significant result. Could it be that HMM are not suited for rolling element type of signals? Has anyone tried it?
You can do an easy test. Train the HMM. Then flatten all probabilities in the transition matrix to a constant (depending on the topology, ergodic, Bakis, whatever transitions were modeled). If the classification rate does not deteriorate significantly, there is no need for a Markov assumption and the marginal probabilities are all you need to estimate.
Also see: http://jpvanoosten.nl/docs/AI-MAI-2010-J-P.van.Oosten.pdf where we (i.e, my MSc. student Jean-Paul van Oosten) showed that for a number of problems, flattening the transition matrix had NO detrimental effect on the classification rate.
Alternatively, given the cyclic nature of the processes involved you may want to look into detailed spectral features as in:
https://www.researchgate.net/publication/5326393_Robust_harmonic_grouping_by_octave_error_correction
Article Robust harmonic grouping by octave error correction
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering