Consider this scenario: Your pump fails at 2 AM on a Sunday. An emergency repair will cost $50,000. Catching it three weeks earlier with vibration analysis? $500.
Here's the 7-step workflow maintenance teams should use daily to prevent failures.
Get these basics in place first:
Once that foundation exists, here's the workflow.
Identify what category of equipment you're analyzing:
Each machine type has statistically-derived alarm levels based on what's normal for that equipment category. A direct-driven overhung fan might have an alarm threshold of 0.325 inches per second, while a precision pump would be much lower.
Then note the configuration details:
Example: You're analyzing a direct-coupled centrifugal pump with rolling element bearings running at 1,784 RPM. Your database sets the alarm at 0.30 inches per second for this machine type. Knowing it's direct-coupled with rolling bearings means you can immediately rule out belt problems, gear problems, and journal bearing issues. You went from 35 possible faults down to 23 before collecting any data.
Now you attach a sensor to the machine and take measurements.
The sensor is typically an accelerometer: a device that measures how fast the vibration is changing (acceleration). It's calibrated to output a specific voltage for a specific amount of vibration, usually 100 millivolts per G.
Measure at every bearing location in three directions:
Take readings under normal operating conditions. Record speed, load, and temperature while you're at it.
Different problems show up in different directions. Imbalance shows strongest in radial directions (horizontal and vertical). Misalignment shows strongest in axial. If you only measure one direction, you'll miss half the problems.
A typical motor-pump setup has four bearings - two on the motor, two on the pump. Measuring all three directions at each bearing gives you 12 data points.
The sensor generates an analog electrical signal - voltage that changes over time as the machine vibrates.
This raw signal looks like a complicated squiggly line with multiple wiggles happening at once. It's capturing everything vibrating in the machine:
The signal is measured in volts, but you can convert it to acceleration (G's) using the sensor's calibration. From there you can mathematically convert to velocity (inches per second) or displacement (mils), which are often more useful.
This raw signal contains all the information you need. The challenge is extracting the useful parts from the noise.
The analog signal is continuous and infinite. You can't process infinity, so you need to turn it into something manageable.
The analyzer samples the signal at a very high rate - taking thousands of data points per second - and converts each sample into a digital number. This creates a digital version of the analog signal that looks virtually identical but can be processed by a computer.
Proper sampling captures all the information without losing anything important. Sample too slowly and you miss high-frequency problems. Sample correctly and you get an accurate digital representation of what's happening.
The analyzer runs Fast Fourier Transform (FFT) and turns it into a clean graph with individual peaks. Each peak represents something vibrating - shaft rotation, bearing cage, blade pass, or a defect.
What you see:
The raw signal is impossible to interpret, but the graph shows clear peaks where problems hide. This is where 99% of vibration analysis happens.
Modern analyzers do this automatically; you just need to know how to read what comes out.
Now you're looking at a graph with several peaks. Figure out what each one represents.
First: Find the peak that matches shaft speed.
If your machine runs at 1,784 RPM, look for the peak at 1,784 cycles per minute. That peak is your 1X RPM reference point. Everything else gets compared to this.
Then classify other peaks:
The equation to use is peak frequency ÷ shaft speed = order of vibration
Example: Your shaft spins at 1,784 RPM. You see a peak at 10,704 cycles per minute. Divide: 10,704 ÷ 1,784 = 6.0. That means something vibrates exactly 6 times per shaft rotation. Got a six-blade fan? That's normal—each blade creates a pulse as it passes. But if that peak is three times higher than your alarm level, look for blade damage or imbalance.
Result: This step narrows possibilities from 35 problems to about 10.
Now that you know what frequencies are present and how strong they are, you compare them to your alarm thresholds.
Your database has alarm bands set up for different frequency ranges:
What happens: If any peak exceeds its threshold, an alarm triggers. The database automatically generates a report telling you which bearing, which direction, and which frequency band exceeded the limit.
Analysis is worthless if nobody acts on it. Create a crystal-clear work order.
Don't write: "High 1X radial vibration detected at position 3V exceeding 0.45 in/sec RMS velocity."
Write this:
Be clear about urgency. Consider:
Save everything to your CMMS: spectrums, waveforms, complete analysis. Set a reminder to check again after repair to verify the fix worked.
Seven steps from sensor to repair. Start with 10-20 critical assets, run weekly or monthly, document everything.
The payoff: Catch problems 4-8 weeks early, cut emergency repairs 70-80%, plan maintenance during shutdowns.
Ready to implement this? Grace Technologies' GraceSense automates this workflow with continuous monitoring and expert support. Contact us to prevent your next 2 AM breakdown.
Zero harm, Zero downtime,