Hi Guys,

Basically i'm got some engine sound data in MATLAB form that needs analysis and diagnostics. I was able to source out a couple of methods of plotting the data. This was done via FFT, power spectrogram and a probability density plot.

But here's the problem, i don't know any techniques for extracting features from these analysis results. Because i need to use these features rather than the analysis results (shown in my diagrams) for fault detection and diagnosis.

I'd deeply appreciate it if anyone can push me in the right direction.

 

Just think about whats happening. Main bearings should thump once per rev. Rod bearings twice that. The camshaft turns at twice the speed of the crankshaft. The moments that create noise are when force reverses. Higher frequencies can be created by ball and roller bearings that rotate at many tines the shaft they are on, Then figure in how the individual cylinders fall into the crank rotation. These things are all frequency related and can be predicted. looking at the spectra with this knowledge should let you divide and conquer. I have used knowledge of the engine and a timing light to isolate noises to a cylinder. So looking at the sampled dataset with timing references may shed some insight as well. I envy you, this sort of analysis has always intrigued me. I would find the project most interesting.

Regards

cww (a motorhead in my youth)