Vibration analysis is one of the oldest tools in rotating equipment reliability. It is also one of the most misunderstood. Engineers run route-based surveys, collect data monthly, generate PDF reports no one reads, and then act surprised when a compressor trips offline with no warning. The problem is rarely the sensor. The problem is how the data gets used.
This article covers the mechanical and signal-processing fundamentals that matter for oil and gas compressor monitoring, with particular attention to the differences between centrifugal and reciprocating machines. We will also cover how integrating proximity probe data with a process historian changes what you can actually diagnose.
Sensor Selection: Accelerometers vs. Proximity Probes
Not all vibration measurements are equivalent. Accelerometers and proximity probes capture fundamentally different phenomena, and choosing the wrong one for a given machine type produces data that looks reasonable but misses the fault you care about.
Accelerometers measure casing vibration in units of acceleration (g) or velocity (in/s, mm/s). They capture high-frequency content well. Bearing defect frequencies for rolling element bearings, gear mesh frequencies, and aerodynamic excitation in centrifugal stages all live above 1 kHz. An accelerometer mounted on a bearing housing in good contact with the casing will see these events clearly. For a centrifugal compressor with rolling element bearings, accelerometers are the right starting point.
Proximity probes are non-contact eddy current sensors mounted in the bearing housing, pointed at the shaft. They measure shaft position directly, in microns or mils of displacement. For machines running on fluid film (hydrodynamic) bearings, the shaft can move significantly within the bearing clearance, and that motion is exactly what you want to see. Accelerometers mounted on the casing will barely register what is happening inside the bearing when clearances are still within tolerance. Proximity probes will catch it early. Large centrifugal compressors, screw compressors, and most reciprocating units use fluid film bearings. These machines require proximity probes for meaningful shaft dynamics data.
The Bently Nevada 3500 Series rack is the field standard for proximity probe signal conditioning. Our Field Integration Lead, Lars Eriksson, spent seven years commissioning 3500 Series systems across Permian Basin and Eagle Ford sites before joining Midstreamly. His experience is direct: the 3500 rack's integration quality matters as much as sensor placement. A poorly conditioned signal gives you noise, not data.
Frequency Domain Analysis: Reading the FFT Spectrum
Time domain waveforms show you that something is wrong. The frequency domain tells you what is wrong. Fast Fourier Transform (FFT) analysis decomposes the vibration signal into its constituent frequencies, revealing the signature of each fault mechanism as a discrete peak or pattern in the spectrum.
Three frequency zones matter most for compressor diagnosis:
1x running speed (synchronous): A dominant 1x peak with a consistent phase angle is the classic signature of rotor imbalance. Mass asymmetry distributes centrifugal force at shaft speed. When you see 1x amplitude climb over time with a corresponding phase shift, that is an imbalance trend. The phase change distinguishes developing imbalance from a stable 1x that simply reflects the machine's normal residual unbalance. One data point: in our baseline dataset across 18 months of Permian Basin centrifugal compressor monitoring, more than 60% of detected faults showed a 1x or 2x harmonic as the primary indicator before any other symptom appeared.
2x running speed (twice synchronous): A 2x peak elevated relative to 1x, combined with elevated axial vibration, is the standard coupling misalignment signature. Angular misalignment drives a strong 2x radial component and bending stress in the coupling. Parallel misalignment produces a strong 2x as well, but the axial component is less pronounced. When 2x exceeds 1x in amplitude and axial readings are 30 to 50% of radial, start with coupling inspection before replacing bearings.
Sub-synchronous: Any spectral energy below 1x is worth immediate attention. Fluid film bearings can develop oil whirl instability at approximately 0.42 to 0.48x running speed. If whirl locks onto a rotor natural frequency, it becomes oil whip, which is destructive. Sub-synchronous content can also indicate surge in centrifugal compressors. Either way, sub-synchronous peaks in a continuous monitoring system trigger an alert on the same shift, not at the next route survey.
Nyquist sampling applies directly here. Your sampling rate must be at least twice the highest frequency you want to resolve. For rolling element bearing defect frequencies on a high-speed centrifugal stage running at 10,000 RPM, you may need to resolve frequencies above 10 kHz. Under-sampling aliases high-frequency faults into the wrong spectral bins. This is a configuration issue that route-based surveys frequently get wrong.
Shaft Orbit Plots and Bearing Condition
Two orthogonal proximity probes at 90 degrees in the same axial plane produce a shaft orbit. The orbit traces the actual path of the shaft centerline within the bearing clearance over one revolution.
A healthy fluid film bearing on a well-balanced rotor produces a nearly circular orbit, centered near the bearing clearance center, with slight eccentricity in the direction of the load. Deviations from this pattern are diagnostic. A figure-eight or elliptical orbit with an inner loop indicates 2x vibration from misalignment or a rub contact. A precessing orbit that changes shape between measurements suggests developing bearing wear or changing process conditions shifting the rotor's equilibrium position within the clearance.
Bode plots complement orbit analysis by showing amplitude and phase as a function of speed during startup or shutdown. The rotor's critical speed, where it passes through its first bending natural frequency, shows as an amplitude peak and a 180-degree phase shift. How sharply the amplitude peaks and how quickly the phase shifts gives you the effective damping. A Bode plot taken when the machine was commissioned establishes the baseline. Subsequent startups that show the critical speed migrating or the peak amplitude increasing indicate a change in bearing stiffness or rotor mass, both worth investigating before they become unplanned downtime.
Bearing Spall and Broadband Noise Floor
Rolling element bearing spall produces a different spectral signature than the synchronous fault frequencies. When a spall develops on the outer race, each rolling element strikes the spall at the Ball Pass Frequency Outer race (BPFO). This generates an impulsive event that excites the bearing's structural resonance frequency, which is typically in the 2 to 20 kHz range depending on bearing geometry.
In the spectrum, early spall often appears as a raised noise floor in the high-frequency region before discrete BPFO sidebands become visible. By the time you see a clean, discrete BPFO peak, the spall is already well developed. We have found that tracking the RMS noise floor in the 5 to 20 kHz band with an 18-month baseline is more sensitive to early bearing degradation than waiting for a recognizable BPFO peak. In controlled test data, noise floor rise precedes visible BPFO sidebands by 7 to 21 days on average under steady operating conditions.
That 7-to-21-day window is the maintenance planning horizon. Not a crisis. A window.
Overall Vibration Trending vs. Spectrum Analysis
Overall vibration, typically expressed as velocity RMS in mm/s or in/s, is useful for alarm thresholds and ISO 10816 compliance assessments. It summarizes everything in the spectrum into a single number. Fast to read. Easy to alarm on. Also easy to miss a developing fault that is real but not yet large enough to move the overall number.
Spectrum analysis is more sensitive but requires interpretation. A fault that shows up as a discrete frequency peak may contribute only a few percent to the overall vibration. The overall number looks fine. The spectrum tells a different story. This is why overall vibration trending and spectral analysis are complementary, not interchangeable. Use overall for fleet-level health scoring. Use spectra for fault identification once a machine is flagged.
Route-Based Surveys vs. Continuous Monitoring
Monthly route-based surveys are the industry standard at most midstream facilities. They are also fundamentally limited. A route survey gives you a snapshot. The machine was running normally at 10:00 a.m. on the 14th. What happened between the 14th and the 15th is invisible.
Bearing failures, fluid film instabilities, and process-induced vibration excitation can develop over hours to days. A monthly survey has no chance of catching a fault that initiates and propagates to failure inside a 30-day window. In our experience monitoring compressor fleets across Eagle Ford gathering systems, approximately 40% of bearing-related unplanned shutdowns occurred within two weeks of a clean route survey result.
Continuous monitoring does not eliminate maintenance personnel. It changes what they do. Instead of walking routes and collecting data, technicians spend time investigating specific flagged machines with actionable spectral data already in hand. The diagnostic work is higher value. The surprises are fewer.
Process Context: Why Vibration Alone Is Not Enough
Vibration amplitude is process-dependent. A centrifugal compressor running near surge will show elevated sub-synchronous content and increased overall vibration regardless of mechanical condition. A compressor with a fouled suction filter running at reduced flow will have a different operating point and different bearing loads than the same machine running clean. Interpreting a vibration trend without process context produces false positives and missed diagnoses.
This is where integrating proximity probe data with a process historian like AVEVA PI changes the analysis. When Midstreamly ingests 3500 Series proximity probe channels alongside PI tags for suction pressure, discharge pressure, inlet temperature, and recycle valve position, the vibration analyst has the full picture. A sub-synchronous event that correlates with a recycle valve opening event is likely process-driven. The same sub-synchronous signature with no corresponding process change is a bearing or stability issue that needs mechanical investigation. Context eliminates guesswork.
This integration also enables fault detectability at scale. Across the fault signatures we track, the combined vibration-plus-process dataset improves fault detectability by 60 to 70% compared to vibration data alone, based on our internal benchmark against route survey outcomes on the same equipment population.
Putting It Together
The fundamentals are not complicated. Use the right sensor for the machine type. Analyze in the frequency domain, not just overall amplitude. Read shaft orbits for fluid film bearing machines. Track broadband noise floor, not just discrete fault frequencies. Combine vibration data with process historian tags to separate mechanical faults from process-driven events. And monitor continuously, not periodically, if the machines are critical enough to matter when they fail.
Vibration analysis done well gives you a 7-to-21-day window to plan maintenance before a fault becomes a trip. That is the difference between a scheduled replacement during a planned outage and an emergency compressor pull at 2 a.m. with a gas gathering system offline. We have seen both. The first one is much better.