A pump score of 71. Down from 85 ten days ago. No alarm has fired yet. That single number tells you more about what's coming than any threshold-based alert your SCADA system will generate this week. Here's why degradation scoring works the way it does, and how we've built it into Midstreamly's monitoring layer for midstream rotating equipment.
What a Degradation Score Actually Is
Most operators are familiar with alarm-based monitoring: a vibration sensor trips a high-high threshold, the DCS fires an alert, someone investigates. Binary. On or off. The problem is that the threshold was likely set at commissioning, based on manufacturer tolerances, not on what that specific unit looks like when it's running well.
A degradation score is a continuous health index, typically scaled 0 to 100. One hundred means the asset is behaving exactly as it did during its healthy baseline period. Scores in the 70s indicate developing fault signatures. Scores below 50 are serious. The number updates every 15 minutes in our system, pulling from live sensor feeds, comparing them against a unit-specific baseline built over 18 months of operating history.
Not a fleet average. Not a manufacturer curve. This unit, on this service, at these operating conditions.
The Four Input Signals
We track four primary signal groups for rotating equipment degradation scoring. Each captures a different failure mode pathway.
Vibration Amplitude
Overall vibration level in velocity (in/s or mm/s RMS) is the most direct indicator of mechanical imbalance, misalignment, and looseness. Rising amplitude often precedes audible changes by weeks. In our data, units showing a score decline from the 80s to the low 60s typically show amplitude increases of 0.3 to 0.8 in/s RMS before any single-point alarm trips.
Frequency Spectrum Drift
Raw amplitude tells you something changed. Frequency spectrum tells you what. We track 1x (running speed), 2x, sub-synchronous, and bearing defect frequencies. Drift in these spectral components maps directly to fault classes: rotor unbalance, misalignment, bearing race defects, looseness. A score decline driven primarily by spectral drift in the bearing defect frequency range points toward bearing degradation, not the seal or impeller.
Bearing Temperature Deltas
Absolute bearing temperature matters less than delta from baseline. We've seen units with bearing temperatures sitting at 185°F that are perfectly healthy, and units at 140°F trending toward failure. What matters is the rate of rise relative to the asset's own history at comparable load conditions. The scoring model weights temperature delta against load to avoid false positives during summer ambient peaks.
Lube Oil Differential Pressure
For compressors and larger pump trains, lube oil differential pressure is an early-warning signal for bearing film degradation. A narrowing differential often precedes mechanical bearing distress by 7 to 14 days. This signal is underutilized in most alarm-only setups because the threshold is set wide enough to avoid nuisance trips. In the scoring model, small sustained shifts in this differential register as score reductions before any alarm would fire.
Why Unit-Specific Baselines Matter
This is the part we push hardest in every conversation with operations teams. Fleet-average baselines don't work for rotating equipment scoring. Here's why.
Two nominally identical centrifugal pumps on the same gathering system will have different baseline vibration signatures. Different piping configurations, different foundation stiffness, different upstream flow conditioning, different impeller wear patterns from months of operation. A baseline built from fleet averages introduces systematic bias that either masks real degradation or generates constant noise from normal unit-to-unit variation.
We build each asset's baseline from its own 18-month operating history. That window is long enough to capture seasonal ambient variations, typical maintenance intervals, and normal load swings. Short baselines (30 or 90 days) miss seasonal effects and post-maintenance settling. Eighteen months is the minimum that produces stable, low-drift baselines in our experience across midstream service conditions.
Unit-specific matters for another reason: it makes the score interpretable to the operator who actually runs that unit. When a score drops, the maintenance team can pull the contributing signals and immediately recognize whether the pattern is consistent with past fault history on that machine. That context speeds diagnosis significantly.
Score Trend vs. Single Alarm: A Concrete Comparison
Consider a natural gas compressor on a gathering header. Two scenarios:
Scenario A: Alarm trip. The vibration high-high threshold fires at 11:42 PM. The unit is shut down per procedure. Inspection finds advanced bearing wear on the drive-end bearing. Repair time: 6 days. Production impact: significant.
Scenario B: Score trend. The degradation score begins declining from 85 on a Tuesday. By the following Friday it's at 62, a 23-point drop over 10 days. The 30-day failure probability ranking flags this unit as the highest-risk asset in the fleet. A work order is generated in SAP PM. An oil sample is pulled. Bearing wear metals are elevated but not yet at alarm levels. The unit is scheduled for a planned bearing replacement the following Tuesday during a scheduled maintenance window. Repair time: 1 day. Production impact: minimal.
Same failure mode. The difference is the 7 to 21-day early detection window the scoring model provides. In our tracking, the median detection lead time before a fault-forced shutdown is 14 days. That's 14 days to plan a maintenance window, source parts, and schedule crew.
How Multivariate Scoring Catches What Single-Signal Alarms Miss
This is where it gets technically interesting. Many fault modes in rotating equipment produce signals that individually stay below alarm thresholds but collectively indicate a developing failure. Fact: bearing outer race defects in their early stages often show frequency components that are 30 to 40% of the alarm threshold while simultaneously showing a 4 to 6°F bearing temperature rise and a small but consistent lube oil differential pressure drop.
Any one of those three signals, monitored in isolation, looks like noise. The alarm doesn't fire. The operator sees nothing. The multivariate score integrates all three, weights them by their historical correlation with fault progression on that asset, and reduces the score accordingly. The decline is detectable. The alarm-only system is blind to it.
We've found this multivariate sensitivity particularly valuable for suction-side pump bearing degradation, where vibration signatures are partially masked by hydraulic noise. The temperature and differential pressure signals carry the detection weight in those cases.
Fleet-Wide Risk Ranking and Work Order Integration
Individual asset scores feed a fleet-wide 30-day failure probability ranking dashboard. Every unit in the monitored fleet gets a rank, updated daily, based on its current score trajectory, the rate of score change, and the historical fault patterns associated with that type of score decline. The maintenance team can sort by risk rank every morning and immediately see which assets need attention in what order.
The ranking directly drives work order generation. Midstreamly integrates with both SAP PM and Maximo. When an asset's score crosses a configurable threshold or its rank moves into the top tier, the integration can auto-generate a predictive maintenance work order with the relevant signal data pre-populated. The maintenance planner reviews and approves, but the triggering and data collection are automated. No manual export, no copy-paste from the monitoring screen into the CMMS.
For operations teams running 40 to 80 monitored assets across multiple compressor stations, that automation difference is significant. Manual triage of 80 assets daily isn't realistic. Reviewing a ranked list of 5 to 10 assets with elevated risk scores is.
What This Looks Like in Practice
In our experience deploying at midstream operators, the first 90 days post-commissioning are the most valuable for calibration. The scoring model is running, baselines are accumulating, and the operations team starts to develop intuition for what score levels mean on their specific equipment. Score of 90+ on their gas lift compressors means normal operations. Score of 75 on one of their multistage centrifugals is a flag. Score of 60 anywhere is a work order.
Those thresholds aren't universal. They're specific to each operator's risk tolerance, their spare parts availability, their maintenance crew capacity. We configure them collaboratively. The scoring model provides the signal; the operations team provides the operational context. Together they produce a maintenance prioritization workflow that is, in practice, substantially more reliable than any threshold-based alarm system running on the same equipment.
Rotating equipment doesn't fail on a schedule. But it does degrade in patterns. Tracking those patterns with a continuous, multivariate, unit-specific score is how you get ahead of failures instead of reacting to them.
Want to see how degradation scoring works on your specific rotating equipment fleet? Request a demo and we'll walk through the signal inputs and baseline methodology for your asset types.
