I'm new to machine health/IoT analytics, but I'm trying to compare the frequencies produced by accelerometer sensors at four different locations of a machine. Each sensor produces an X and a Y output CSV so in total there are 8 CSVs. Measurements are collected at semi-routine intervals (~2 hours) but there are some bigger and smaller gaps.

Each CSV is the same where each row has a timestamp, sampling rate, sample length, and then data in the subsequent columns. The sample rate is 8192Hz with sample length of 4096.

My thought is to use FFT to identify the dominant frequencies for each timestamp and then compare the frequencies over time to evaluate the performance.

I'm not sure if I can FFT because the time intervals aren't 100% consistent. I also don't know what numbers to use for the variables. Any help is appreciated.

I've tried averaging the data for each timestamp and then comparing the averages over time, but this didn't reveal any insightful information. I was hoping to see peaks of high activity and static points of no activity. Instead, my averages were basically a wave.

I'm hoping to see peaks which reveal high vibration, thus activity, and low points (no/minimal activity).