Mr Ramy,

depends on signal strength u can go for thresholding technique to segment particle signal at output side otherwise if it in frequency domain prefer respective filtering techniques

Hi Ramy,

in terms of segmentation, Keogh's paper is a classic:

http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=989531

For the detection issue, try searching for: anomaly detection, state change detection, and event detection.

Regards,

Philip

Ordinal patterns could be useful for  you:

http://link.springer.com/article/10.1140%2Fepjst%2Fe2013-01852-9

http://digital.csic.es/bitstream/10261/21608/1/GetPDFServlet.pdf

http://www.sciencedirect.com/science/article/pii/S0016003213001543

Once you have converted your time series into a symbolic sequence, you can apply a clustering procedure to get the different segments.