Humpday | Grace Technologies Blog

The Maintenance Engineer's 7-Step Workflow for Catching Equipment Failures Early

Written by Adarsh Iyengar | Mar 11, 2026 6:30:00 PM

The Cost of Waiting Too Long

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.

 

Before You Start: Quick Setup

Get these basics in place first:

  • Identify critical assets: Focus on equipment where failure means serious downtime or safety issues
  • Install sensors: Tri-axial accelerometers at each bearing (sensors that measure in all three directions at once)
  • Take baselines: Measure during normal operation when equipment is healthy
  • Set alarm levels: Based on machine type (a cooling tower can vibrate more than a precision pump)
  • Connect to CMMS: So alarms automatically generate work orders

Once that foundation exists, here's the workflow.

Step 1: Define the Machine Type 

Identify what category of equipment you're analyzing:

  • Cooling towers
  • Reciprocating compressors
  • Direct-driven overhung fans
  • Belt-driven fans or blowers
  • Motor generator sets
  • Chillers
  • Turbine generators
  • Horizontal pumps
  • Vertical pumps

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:

  • Motor type (AC induction, DC, VFD)
  • Connection (direct coupled, belt-driven, gearbox)
  • Bearing type (rolling element or journal bearings)
  • Running speed (RPM)

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.

 

Step 2: Measure with a Calibrated Sensor 

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:

  • Horizontal (H) - side to side
  • Vertical (V) - up and down
  • Axial (A) - along the shaft, front to back

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.

 

Step 3: Capture the Voltage vs Time Signal 

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 shaft rotating
  • The bearing components spinning
  • Any defects or problems
  • Background noise

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.

Step 4: Sample and Digitize the Data

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.

 

Step 5: Convert to Frequency Domain with FFT

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:

  • A peak at shaft speed (how fast the shaft is rotating)
  • Peaks at bearing frequencies (how fast the bearing components are spinning)
  • Peaks at blade pass frequency (if it's a fan or pump)
  • Peaks at any defect frequencies (if something's wrong)

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.

 

Step 6: Identify Peak Frequencies

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:

  • At 1X shaft speed = Something on the shaft itself (imbalance, eccentricity, bent shaft, loose mounting)
  • Exact multiples (2X, 3X, 4X) = Usually coupling problems or misalignment. Could also be blade pass frequency on fans or pumps.
  • Below shaft speed (0.4X, 0.38X) = Bearing cage frequencies or certain types of looseness
  • Not exact multiples (3.049X, 4.56X) = Probably bearing defects. Bearing frequencies are almost never exact multiples.
  • Groups of equally-spaced peaks = Something's beating or pulsing. Advanced bearing damage or electrical problems.

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.

 

Step 7: Compare to Acceptable Ranges

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:

  • Overall level (0.30 in/sec for this machine type)
  • 1X running speed band (90% of overall = 0.27 in/sec)
  • 2X running speed band (40% of overall = 0.12 in/sec)
  • Bearing frequency bands (15-25% of overall)

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. 

Final Step: Write It Up and Fix It

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:

  • Where: Motor bearing 3, vertical direction
  • Problem: Loose mounting foot (soft foot condition)
  • Fix: Tighten foot bolts to spec, inspect all four feet, adjust shims as needed, verify alignment after repair
  • When: Schedule during next planned shutdown (not critical yet)

Be clear about urgency. Consider:

  • How much over threshold? (10% over vs 3X over makes a difference)
  • How fast is it getting worse? (doubled in a week = urgent)
  • What happens if it fails? (critical process = fix now, backup available = less urgent)

Save everything to your CMMS: spectrums, waveforms, complete analysis. Set a reminder to check again after repair to verify the fix worked.

 

The Bottom Line

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,


 


connect with us