Vibration monitoring is one of the oldest tools in rotating equipment maintenance. It's also one of the most misapplied. In our experience deploying predictive analytics across midstream pump and compressor fleets, the gap isn't usually in sensor coverage. It's in understanding what each measurement type is actually telling you, and what it can't.
This guide covers the four main vibration measurement methods used in midstream, where each belongs, and how to combine vibration data with process historian context to catch fault modes that vibration alone will miss.
The Four Measurement Types and Where They Fit
Overall Vibration (RMS/Peak)
Overall vibration measures total energy across the full frequency spectrum, typically reported in inches per second (IPS) or mm/s. It's the broadest signal you can pull from a machine, and it's useful for exactly that reason: a clear upward trend in overall level tells you something is changing, even if you don't yet know what.
For midstream applications, overall vibration works well as a first-tier health indicator on centrifugal pumps and pipeline compressors where access is limited. A single accelerometer on the bearing housing gives you enough signal to trigger an alert. What it won't do is tell you whether the energy shift comes from an impeller imbalance, a worn bearing race, a soft foot issue, or a structural resonance. That discrimination requires spectrum analysis.
Spectrum Analysis (FFT)
Spectrum analysis decomposes vibration into its frequency components. This is where the diagnostic value lives. Running speed harmonics, sub-synchronous vibration, bearing defect frequencies, gear mesh frequencies: each fault mode has a characteristic frequency fingerprint, and spectrum analysis is how you read it.
In gas gathering and transmission, spectrum analysis is the standard tool for centrifugal compressors. It's good at catching rotor imbalance, misalignment, and rolling element bearing degradation early, often 7 to 21 days before a fault becomes operationally significant. The trade-off is complexity. Reading spectra requires trained analysts or automated pattern-matching software. Raw FFT data is not an alarm system. It's a diagnostic tool, and treating it like the former is a common source of both missed faults and false positives.
Proximity Probes (Eddy Current)
Proximity probes measure shaft displacement directly, expressed in mils (thousandths of an inch) peak-to-peak. Unlike accelerometers mounted on the bearing housing, proximity probes track the shaft itself, which matters for fluid-film bearing machines where rotor dynamic behavior is the primary failure mode.
API 670 installations on large centrifugal compressors in gas transmission typically include X-Y proximity probe pairs at each journal bearing. These give you shaft centerline position (static) and orbital motion (dynamic). A shaft orbit drifting from its design operating position is often an early indicator of bearing wear, oil film instability, or changes in alignment under operating temperature. This is not a technology to retrofit casually. It requires precision installation, qualified hardware, and a system like a Bently Nevada 3500 to process the signals correctly. But on high-consequence compressors, it's the right tool.
Accelerometers for High-Frequency Analysis
High-frequency accelerometers, operating in the 10 kHz to 100 kHz range, are used primarily for early rolling element bearing diagnostics via techniques like Envelope Analysis (also called High-Frequency Resonance Technique, or HFRT). Bearing defect frequencies, which appear well below the noise floor in standard FFT, become detectable in the envelope spectrum long before they register as elevated overall vibration levels.
In midstream, this technique is most valuable on pipeline booster pump bearings and gas gathering compressor ancillary equipment where bearing replacement cost is manageable but unplanned downtime is not. We've seen bearing defects detected 4 to 6 weeks in advance using envelope analysis on assets that showed no concerning trend in overall vibration or standard FFT. That lead time is real. It changes the maintenance conversation from reactive to scheduled.
Centrifugal vs. Reciprocating: Different Physics, Different Strategies
The measurement approach that works for a centrifugal compressor will under-serve a reciprocating compressor. The failure physics are different enough that they require separate monitoring frameworks.
Centrifugal Compressors
Centrifugal compressors are dominated by rotordynamic behavior. The primary fault modes, rotor imbalance, aerodynamic instabilities, surge, and bearing wear, all produce vibration signatures that spectral analysis and proximity probes can detect. Standard API 670 protection systems handle trip and alarm logic. The challenge is early-stage fault detection before the signal is obvious enough to trigger a protection system.
For this, spectrum trending with automated baseline comparison is effective. Most vibration analysis platforms can track amplitude changes at specific frequencies (1X, 2X, bearing defect frequencies) and flag statistically significant shifts from a trained baseline. That's where the 7-to-21-day detection window comes from in practice.
Reciprocating Compressors
Reciprocating compressors are fundamentally different. They are high-vibration, low-speed machines where the mechanical signature is dominated by crankshaft rotation and valve events. Vibration levels that would indicate a fault on a centrifugal machine are normal operating states on a reciprocating unit. This is important. Out-of-the-box vibration alarm thresholds from a centrifugal program will produce constant false positives on a reciprocating fleet.
The useful measurements on reciprocating compressors are cylinder pressure analysis (rod load, P-V diagrams), rod drop monitoring, and time-waveform analysis of valve events. These require specialized instrumentation beyond standard accelerometers. Vibration analysis remains useful for crankshaft bearing monitoring and structural assessment, but it's a supporting diagnostic, not the primary health indicator.
What Vibration Monitoring Misses
Here's the thing most vibration-centric programs don't account for: roughly 30 to 40% of midstream pump and compressor failures originate in process conditions, not mechanical degradation. And vibration sensors don't see process conditions.
Consider valve wear on a reciprocating compressor. A degraded suction valve lets gas backflow into the cylinder on the compression stroke, reducing volumetric efficiency before any mechanical fault is detectable in vibration. The first signal is performance degradation: lower discharge pressure, higher suction temperature, reduced throughput. None of that shows up in an accelerometer.
The same is true for fouling on centrifugal pump impellers. Fouling shifts the pump's operating point on its curve, increasing power draw and reducing flow, but the vibration signature may remain within normal bounds until the fouling is severe enough to cause mechanical imbalance. By that point, performance has been degraded for weeks or months.
Cavitation is another process-induced fault. While it does produce a characteristic broadband noise in the vibration spectrum, the signal can be subtle and easy to miss without a matched baseline. A suction pressure drop that puts the pump into cavitation is visible in the process historian immediately.
Combining Vibration with Process Historian Data
In our data, combining vibration signals with process historian variables, suction and discharge pressure, bearing temperatures, lube oil pressure and temperature, interstage temperatures, and flow rates, increases detectable fault coverage from roughly 60 to 70% of failure modes compared to vibration alone. That's not a theoretical number. It reflects what the combined signal space can see versus what vibration alone captures.
Practical note from Nadia Okafor, Head of Data Science: During my time deploying PI Asset Framework at midstream clients, the pattern I saw most often was rich vibration data sitting siloed from the process historian. The analysts running the vibration program had no visibility into suction pressure trends. The control room operators watching process data had no visibility into vibration. Neither team could see the full fault picture. The integration step, even just pulling contextual process variables into the same time-aligned view as vibration, consistently improved diagnosis accuracy.
The specific combinations that have proved most useful in midstream:
- Centrifugal compressor surge detection: Vibration sub-synchronous components + discharge/suction pressure ratio + flow rate. Surge onset produces a characteristic vibration spike, but the process variables give you early warning before the mechanical event.
- Reciprocating compressor valve health: Cylinder temperature asymmetry across banks + suction temperature + discharge flow. Valve degradation shows here before it shows anywhere in vibration.
- Pump bearing thermal runaway: Bearing temperature rate of change + lube oil supply temperature + bearing housing vibration. The thermal signal often leads the vibration signal by 30 to 60 minutes.
- Fouling detection: Power consumption vs. flow (efficiency curve deviation) + differential pressure. Efficiency loss is detectable weeks before vibration responds.
Setting Baselines and Alarm Thresholds
The most common mistake in vibration monitoring programs is setting alarm thresholds from standards documents rather than from operational baseline data. ISO 10816 and API 670 vibration severity criteria are protection limits, not early warning thresholds. Running an alarm at 80% of the trip level is not predictive maintenance. It's a slightly earlier warning that you're about to have a problem.
Effective early warning requires a statistical baseline built from normal operating data. The minimum viable baseline for midstream rotating equipment is 18 months of continuous operation across seasonal temperature variation. Compressors and pumps behave differently in winter operating conditions, and a baseline that doesn't capture that seasonal range will generate spurious alarms every time the ambient temperature changes. Seriously. We've seen programs with 3-month baselines that spent more time managing false positives than detecting real faults.
Once a baseline is established, alarm thresholds set at 2 to 3 standard deviations from the baseline mean give you a statistically defensible trigger for anomalous behavior without excessive false positive rates. The specific threshold depends on how much process variability your asset sees under normal operation. High-variability assets need wider bands. Steady-state baseload machines can use tighter tolerances.
Baseline Invalidation Triggers
A baseline isn't static. It needs to be reset after any of the following events:
- Major maintenance (bearing replacement, seal replacement, impeller changeout)
- Process operating point changes (different throughput contracts, changed inlet conditions)
- Driver changes (motor rewinding, turbine overhaul)
- Structural changes (foundation repairs, piping modifications)
Running against a stale baseline after a major maintenance event is how programs generate alarm floods that train operators to ignore alerts. That's the worst outcome: not missing a fault, but training your team not to respond when a real fault appears.
Where to Start
If you're evaluating or improving a vibration monitoring program for a midstream pump and compressor fleet, the practical starting point is an honest audit of existing data coverage: what variables are being collected, how they're being stored, and whether the vibration historian and the process historian can be time-aligned.
The monitoring technology is not the limiting factor for most operators. The instrumentation is already in place on most assets that matter. The gap is almost always in the analytics layer and in the integration of signal types that have historically been managed by separate teams with separate tools.
Closing that gap is where the detection improvement is. Not in adding more sensors.
