Vibration analysis is a maintenance technique that measures the vibrations produced by rotating machinery to detect developing faults before equipment breaks down. Every machine vibrates, and the specific pattern of that vibration changes as components wear, loosen, or fall out of alignment. By reading those patterns, technicians can identify problems like bearing damage, shaft imbalance, or structural looseness weeks or months before a failure occurs.
How Vibration Tells You What’s Wrong
A vibrating machine produces a signal that can be broken down into two core measurements. Amplitude tells you how severe the problem is: bigger vibrations mean more energy and usually a worse condition. Frequency tells you the source of the problem, because different faults vibrate at different speeds. A third measurement, phase, reveals the direction and timing of the vibration, which helps distinguish between faults that might otherwise look similar on a chart.
Think of it like diagnosing a strange noise in your car. How loud the noise is tells you how urgent the problem might be. Whether it’s a low rumble or a high-pitched squeal points you toward the specific part that’s failing. Vibration analysis applies the same logic, but with precise sensors and mathematics instead of a mechanic’s ear.
From Raw Signal to Useful Spectrum
Machines don’t vibrate at a single, clean frequency. The raw vibration signal is a messy waveform that combines dozens of overlapping patterns all happening at once. To make sense of it, analysts use a mathematical tool called the Fast Fourier Transform (FFT). The FFT takes that complex waveform and separates it into its individual frequency components, much like a prism splits white light into individual colors.
The output is a frequency spectrum: a graph showing vibration amplitude at each frequency. Instead of staring at a squiggly line that’s nearly impossible to interpret, an analyst can now see distinct peaks. Each peak corresponds to a specific vibration source, whether that’s the shaft spinning once per revolution, a gear meshing, or a bearing defect. This is what makes vibration analysis diagnostic rather than just descriptive. The FFT reveals which frequencies are present and how strong they are, pointing directly toward the component causing trouble.
Common Faults and Their Signatures
Different mechanical problems produce predictable vibration patterns. Learning to recognize these signatures is the core skill of vibration analysis.
Imbalance
When mass is unevenly distributed on a rotating part, the heavy side pulls outward with every revolution. This shows up as a strong vibration peak at exactly one times the shaft’s rotational speed (often written as 1X RPM). The vibration amplitude increases with the square of speed, so a machine running twice as fast produces four times the vibration from the same imbalance. A telltale sign is a roughly 90-degree phase difference between horizontal and vertical measurements at each bearing.
Misalignment
When two connected shafts aren’t lined up properly, the vibration spectrum typically shows elevated peaks at both 1X and 2X RPM, sometimes with a strong axial (lengthwise) component. Couple unbalance, a related condition, produces equal vibration amplitude at both bearings but with phase readings 180 degrees apart, meaning the two ends of the shaft are vibrating in opposite directions.
Mechanical Looseness
Loose bolts, cracked frames, or a poor foundation let the machine move in ways it shouldn’t. The vibration spectrum for looseness often contains multiple harmonics of the running speed (2X, 3X, 4X, and beyond) rather than a single dominant peak. Looseness caused by cracked structures or loose mounting bolts tends to produce two pulses per revolution in the raw waveform, while a loose connection between the machine and its base may show just one pulse per revolution along with a 180-degree phase shift across the joint.
Bearing Defects
Rolling element bearings generate very specific vibration frequencies depending on which part is damaged. A defect on the outer race, the inner race, or the rolling elements themselves each produces its own characteristic frequency based on the bearing’s geometry: the number of rolling elements, their diameter, the pitch diameter of the bearing, and the contact angle. These frequencies are mathematically predictable, so an analyst can calculate them in advance and watch for those exact peaks to appear in the spectrum. Early bearing damage often shows up at high frequencies before it becomes visible at the fundamental bearing defect frequencies, making it one of the earliest detectable faults.
Route-Based vs. Continuous Monitoring
There are two main ways to collect vibration data, and most facilities use one or a combination of both.
Route-based monitoring is the more affordable option. A technician visits the facility on a regular schedule, attaches portable sensors to each machine, and collects readings with a handheld analyzer. Hardware costs are minimal since the same sensors and analyzer move from machine to machine. The downside is that you only get snapshots. Anything that happens between visits goes undetected. Some measurement points may also be physically dangerous or impossible to reach while equipment is running, creating blind spots in your data.
Continuous monitoring uses permanently installed sensors that feed data to a central system in real time. This catches problems the moment they begin developing, provides long-term trend data, and eliminates the need for someone to physically access every measurement point. Sensors can be installed on hot, elevated, or otherwise hazardous equipment during a planned shutdown, then monitored remotely from that point forward. The tradeoff is cost: permanent sensor networks, cabling, and the software to manage continuous data streams require significantly more upfront investment.
Many facilities take a tiered approach, putting continuous monitoring on their most critical or hard-to-reach equipment and using route-based collection for everything else.
How Machine Learning Is Changing the Field
Traditionally, interpreting vibration data required a trained analyst who could look at a spectrum and recognize fault patterns. Machine learning is beginning to automate that process. Algorithms trained on large datasets of known fault conditions can classify new vibration readings with remarkable accuracy.
A 2025 study tested three different machine learning approaches on a standardized database of motor faults. Support vector machines correctly identified fault conditions 95.4% of the time, and a nearest-neighbor algorithm reached 92.8%. The best-performing model combined a deep neural network with FFT-based features and achieved 99.7% accuracy. These systems don’t replace human analysts entirely, but they can flag anomalies in continuous monitoring streams far faster than a person reviewing charts manually, making them especially valuable for facilities with hundreds or thousands of monitored assets.
What Vibration Analysis Actually Prevents
The practical value comes down to timing. An undetected bearing failure can seize a shaft, destroy surrounding components, and shut down a production line for days. Vibration analysis typically detects that same bearing defect weeks to months before catastrophic failure, giving maintenance teams time to order parts, schedule a repair during planned downtime, and avoid the cascading damage that comes with an unexpected breakdown. The cost difference between a planned bearing replacement and an emergency teardown of an entire machine assembly is often an order of magnitude or more.
Beyond preventing breakdowns, regular vibration monitoring builds a baseline picture of each machine’s normal behavior. When a new vibration pattern appears, it stands out against that baseline immediately, even if the absolute vibration levels are still within acceptable limits. This trend-based approach catches slow-developing problems that a single measurement might miss entirely.