Analyzing motor vibration data

[sab_sab]

I am collecting data into two folders. One of these folders contains 60 excel files with vibration test data from the motor when new bearings were installed. These 60 files were collected over a period of one day and have time stamps. The other of these two folders has ~1000 excel files for two months worth of data from the motor with different loads applied to it. The folder with data collected using new bearings will be used to form a baseline. The other folder's contents will be used as test data. I will conduct data analysis that includes signal processing and prediction for fault detection. I am in the first phase now and need to extract features from the vibration data in order to establish trends. For that, I am doing FFT analysis and my question is this: When I plot power vs. frequency for the FFT analysis of the first file in the first folder, I see the peaks. I then plotted the power vs. frequency plots for the remaining 59 excel files on the same graph (overwriting them). Is this a correct way of looking at/presenting the data? Does anyone have an suggestions on how to identify trends or how to extract features? I am a bit lost here and any help would be greatly appreciated.

 

[klystron]

Please give more detail on:
What is your sampling speed?
How many samples to you take in a block?
What is the time delay between blocks of samples data?
How many bits is the AtoD?
What type of vibration sensor do you use?
What is the motor rotation speed?
Is the motor rotation speed constant?
Do you measure and store the motor rotation speed?
What type of Windowing do you do before spectral analysis?
What is the bearing details?
How do you "induce" fault conditions in the bearing?

 

[sab_sab]

Fs is 25600Hz
There is no consistency for the time delay. Recordings I have are randomly recorded samples.
AtoD is 8 bits
Type of sensor is accelerometer
Motor vib speed is 9200 rpm and it slightly changes over time. lets say btw 9180-9280 rpm
I didnt apply any windowing data.
I have given only the data and I am supposed to apply fft and time domain functions to fins trends in the data I have.
Do you have any idea?

 

[klystron]

What is the block size ? The FFT line resolution is equal to sampling frequency/block size
The 8 bit A2D is not good. Try to get at least a 12 bit A2D(the quant. noise floor is much lower)
You must use windowing prior to taking the FFT . Without windowing there will be spectral artifcats related to the non zero start andd stop sample values
You can do time domain averaging if you have a tacho sync signal. This will improve the output of the FFT
Study the bearing fault frequncy formulas and look for engery at the bearing fault frequencies in the frequency domain