I created this page as a way to share PDF copies of my publications on the rotating machine fault detection research I conducted during my graduate degree. The PDFs are released here as per the publication licenses; it's pretty cool that Elsevier allows you to share your publications on your personal blogs!
I graduated with a BSc in Electrical Engineering from the University of Calgary, Canada. I then pursued a MSc in Electrical Engineering under Dr. Qing Zhao at the Advanced Control Systems laboratory at the University of Alberta, Canada. During my research I became fascinated with deconvolution techniques to extract faults from vibration, and was inspired by Hiro Endo's application of Minimum Entropy Deconvolution (MED). I graduated in 2011, but have since had some hobby interest in the field on the occasional weekend.
This interest in MED led me to begin experimenting with different filter design optimization cost functions that were better designed for rotating machine faults. If anyone else is exploring new deconvolution techniques, I recommend using the non-linear optimization Matlab toolkit to solve any cost functions iteratively. Using this method, you can try out any cost function you can dream up without solving any of the math. Save the math until you have good results on Matlab first. Anyways, this led to the papers below based on Maximum Correlated Kurtosis Deconvolution (MCKD) and Multipoint Optimal Minimum Entropy Deconvolution (MOMEDA), as well as the convolution region fix for MED (MEDA). Generally, I feel the deconvolution methods make more sense than most other vibration fault detection methods. For example, using the wavelet transform and selecting a band that has maximum kurtosis can be thought of generally as just applying a bandpass filter; it is taking a frequency range of the signal and discarding the rest of the information + spectrum. A lot of information is lost here.
Although the deconvolution techniques have come a long way in rotating machine fault detection, there is a lot more work to be done and discovered. Here are a few brainstorming bullets for inspiration:
In retrospect, I still think this is the best deconvolution solution for our rotating machine fault extraction - MCKD and MED aren't as good IMO. It solves directly for the solution (no iteration), and uses an infinite impulse train as the goal. What I figured out afterwards is that I derived a known equation from a different angle, look here at page 441 for more context on what the MOMEDA solution is: Least squares and psuedo inverses
I was never a huge fan of this research direction for rotating machine fault detection. It's interesting, but I don't think it will surpass or even equate to other approaches. Gang Li did some cool research down this line.
The main content of the thesis is the MCKD and adaptive sinusoidal modelling content. There is an extra chapter on applying the data to long-term monitoring of industrial steam turbines.