---
title: The P-F Curve and the Problem With Ranking Predictive Technologies
slug: p-f-curve-predictive-technologies
url: /topics/articles/p-f-curve-predictive-technologies/index.md
publishedAt: "2026-10-26T18:45:49"
updatedAt: "2026-10-26T18:45:49"
excerpt: "In most illustrations of the P–F curve, predictive technologies appear in the same order: data analysis, ultrasound, vibrations, oil analysis, audible noise, and heat to the touch. Although the curve represents a pedagogical image, no technique has a fixed position. We must also add the commercial bias each technologist brings to their PdM tool and […]"
authors:
  - Jorge Alarcon
---

# The P-F Curve and the Problem With Ranking Predictive Technologies

By [Jorge Alarcon](/authors/jorge-alarcon/index.md)

In most illustrations of the P–F curve, predictive technologies appear in the same order: data analysis, ultrasound, vibrations, oil analysis, audible noise, and heat to the touch. Although the curve represents a pedagogical image, no technique has a fixed position. We must also add the commercial bias each technologist brings to their PdM tool and the values they prioritize.

![P-F Curve The race to detect first](https://cms.reliamag.com/wp-content/uploads/2026/10/P-F-Race-JAB.jpeg)

The failure mode, the component, the operating conditions, and, above all, how it is measured decide which one will detect a problem first. Lubricant analysis can anticipate vibrations when wear particles or chemical changes appear; vibrations can signal an imbalance before it generates wear; and ultrasound can detect increased friction before the defect shows up in the vibrational spectrum. Many well-documented cases show all three situations, but none allow universal classification.

The problem with the famous curve illustration is that it distributes techniques along degradation rather than presenting a specific sequence that would suit any machine and any breakdown. This article reviews published cases, details which PdM tools were used in each, and proposes a P–F curve **by failure mode** (Figure 1).

## Before Ordering Technologies, You Need to Define What You Are Comparing

"Oil analysis", "vibrations," or "ultrasound" are families of tools, not a single measurement. Each technique looks at a different physical quantity and has its own detection window:

To read the cases rigorously, it is useful to distinguish **three levels of evidence**, which are also used in Figure 1:

* **Documented time sequence:** the same component was followed with two or more techniques, and which alarm was published first.
* **Confirmed diagnosis:** one technique identified the anomaly and the inspection confirmed it, but without temporal comparison with the others.
* **Physical expectation:** The mechanism explains why a tool *should* detect earlier, even if the case does not measure it.

## When Oil Analysis Gets Ahead of Vibrations

### Low-Speed Gearboxes: Bearing Wear

In a seed crushing plant in Latin America, the reducer of a soybean elevator showed iron increasing from 65 to 527 ppm and chromium from 1 to 8 ppm, while phosphorus decreased. That month's vibrations were still within acceptable limits, although the trend was already increasing; the first vibratory alarm came later, and the vibratory analysis helped us better locate the affected element. 

In a second lifting reducer in the same plant, the oil gave the first signal again: the vibrations did not indicate input-shaft bearing failure until **more than a month later**. By replacing the oil and bearing, the deterioration was confirmed, and the vibratory indications disappeared.

![Slow gearbox / thrust bearing - wear with particles](https://cms.reliamag.com/wp-content/uploads/2026/10/Case-a.jpeg)

**PdM tools in this case**

**Documented:** elemental spectrometry (Fe, Cr, P), particle count, viscosity, water and AN in the oil; severity trend and vibration spectrum. Cr is consistent with bearing steel (AISI 52100), and P drops as anti-wear additives are consumed. The plant also installed desiccant vents due to external contamination (silicon and high organic flour dust).

**Recommended in slow gearboxes:** combine spectrometry with **PQ index** (detects large particles that spectrometry underestimates) and launch **analytical ferrography** when both rise; in vibration, low-frequency accelerometers (≈500 mV/g), long waveforms covering several turns of the output shaft, and **envelope** for bearing defects.

The correct reading is not "in gearboxes the oil always wins"; with that regime and sampling/vibration settings, the oil alarm criterion was reached earlier. And an "acceptable" vibration is not a vibration without information – the trend was already growing.

### Turbine-Reducer: Thrust Bearing Deterioration

At the ethylene plant in Nevada, the oil from a turbine-reducer assembly of the rapid cooling water system showed an increase in ferrous particles in January. The turbine and gearbox shared a tank, so the signal did not locate the source. Analytical **ferrography and SEM analysis** identified mainly **Babbitt** particles, discolored by temperature and with cut shapes. The axial bearing was suspected and the intervention was prepared (work order, balanced spare gears, checked bearings). Meanwhile, the vibration, which had been low until then, began to grow in the **axial direction**. The inspection confirmed the thrust bearing was failing and would have lasted little longer; the early shutdown prevented damage to the gear train and a possible trip of the charging gas compressor.

**PdM tools in this case**

**Documented:** routine analysis of ferrous particles → analytical ferrography and SEM (material type, temperature, mechanism) → increasing axial vibration → inspection.

**Recommended:** with a shared tank, ferrography and SEM tell *you what* wears, but not *where*. **Axial position probes** (API 670) and **metal-temperature RTDs in thrust pads **locate Babbitt wear in a thrust bearing and usually move before axial vibration.

## When Ultrasound Gets Ahead of Vibrations

### Centrifuge at 5,250 rpm: Incipient Bearing Defect

An ethanol plant in Europe identified, by ultrasound, a compressor bearing in an anomalous early phase, when the defect was not yet visible with the vibratory tools used. It was decided to replace it at the next scheduled shutdown and, in the meantime, permanent monitoring was installed. Over the following months, deterioration progressed slowly in the ultrasonic spectrum until it became clearly visible in the vibration and demodulated spectra, with 1× sidebands.

**PdM tools in this case**

**Documented:** portable ultrasonic collector for detection; permanent system and software for tracking; vibration and envelope spectra for confirmation.

**Recommended:** set static alarms in dB-μV on a reference specific to each bearing (the rules of practice of +8 dB / +16 dB are a starting point, not a standard) and, before concluding that ultrasound "wins", check whether a well-configured high-band enclosure would have seen the defect.

![Bearing in a high-speed machine - friction and impacts](https://cms.reliamag.com/wp-content/uploads/2026/10/CASE-B.jpeg)

The advantage appeared on a fast machine, not a slow one, and only with a specific vibration configuration. The publication does not provide an exact interval between detections or enough data to compare instrument sensitivities.

### Recirculating Fan: One Month Difference Between Alarms

A water treatment plant controlled the bearing on the non-driven side (NDE) of a recirculating fan. From start-up, friction was higher than in similar bearings, and from the sixth month, it increased until the **second alarm**. Despite re-greasing several times, the signal did not return to baseline. The vibrational acceleration did not rise until **about a month later**, when the alarm went off; the margin allowed for a controlled shutdown to be planned.

**PdM tools in this case**

**Documented:** wireless system (continuous ultrasonic friction and level alarms) versus conventional vibration sensors.

**Diagnostic key:** If friction goes down when greasing, the problem was lubrication; if **it doesn't return to reference**, there's a mechanical defect. Ultrasound-guided lubrication (dosing while observing the dBμV drop and recording grams) turns that task into a diagnostic assay.

## When Vibrations Are the First Relevant Tool

### Seven-Stage Centrifugal Compressor: Unbalance After Overhaul

A seven-stage centrifugal compressor at about 8,000 rpm after a revision with rotor destacking exceeded the commissioning acceptance criterion of 1 thousand peak-to-peak; in one probe, component 1× exceeded 2 mils. An eccentric mass generates a force synchronous with rotation, so **amplitude and phase at 1×** directly describe the problem, although other defects also produce 1×.

**PdM tools in this case**

**Documented:** four **proximity probes** (X-Y pairs at ±45°); in the absence of a keyphasor, a **temporary laser tachometer** with reflective tape was used as a phase reference; low-speed *runout* compensation  (0.21–0.44 mils). The phase difference between extremes indicated a torque imbalance and was balanced in **two planes** by coefficients of influence (test weights of 6.8 g, least squares adjustment) with corrections of 18.6 g at 233° and 28.2 g at 0°.

**Case Lessons:** Erroneous wiring of the one-ended probes and balancing rings numbered the holes in opposite directions.

If the imbalance has not yet caused friction or bearing damage, there is no reason to expect particles or chemical changes in the oil, nor for a friction-oriented ultrasound to detect a low-frequency excitation first. That is why we prioritize vibrations.

![Centrifugal compressor at 8,610 rpm - unbalance after overhaul](https://cms.reliamag.com/wp-content/uploads/2026/10/CASE-C.jpeg)

## Turbines and FTIR: Where There's Evidence and Where It's Missing

A study of gas turbine oils in three natural gas plants in Europe linked laboratory tests to deposits. One of the plants had about 7,500 liters of R&O ISO VG 32 oil with an amino antioxidant; an increase in oxidation in the oxidation FTIR signal was evidenced by around **1,713 cm⁻¹ **at the same time as a slight increase in the acid number; in a shutdown prior to further maintenance, varnish was observed on the main bearing.

**PdM tools in this case**

**Documented:** Oxidation FTIR (ASTM D7414, band 1,800–1,660 cm⁻¹), AN (ASTM D974), insoluble MPC (ASTM D7843), RULER (ASTM D6971), RPVOT (ASTM D2272), and counting reported as ISO 4406. The insoluble MPC reached **ΔE 24, **yet varnish had already adhered to the component surface. In another plant, the phenolic antioxidant was depleted before the amino acid, without the RPVOT falling below 50%.

**Practical conclusion:** the generic limits are indicative; alarms should be set by combination of oil and equipment, and trends should be monitored. The study did not measure vibration or bearing temperature in parallel.

![Gas turbine - oil oxidation and varnish](https://cms.reliamag.com/wp-content/uploads/2026/10/CASE-D.jpeg)

The study supports FTIR for tracking deposit-bound chemical degradation, but it does not show that it consistently outperformed vibration-based methods. Nor should FTIR always be placed first: antioxidant consumption by RULER or FTIR can occur before measurable oxidation products accumulate.

## The P–F Curve Applied to Cases

Figure 1 reconstructs the curve for each case family. The time axis is qualitative: only the intervals that the sources publish are indicated. The marker shape indicates the level of evidence, so the figure promises no more than the cases show.

Each of the figures shows the P-F curve from a different perspective. In **A, **the oil opens the P–F window, and vibration closes it by localizing the issue. In **B**,  ultrasound detects friction, and failed relubrication separates lubrication from damage. In **C**, vibration 1× is the only relevant signal at the beginning. In **D**, the fluid degrades before the asset loses condition, and vibration, if it reacts, does so only at the end.

## How to Draw a Technically Defensible P–F Curve

It is more useful to work with a **machine-component-failure mode tab** than with a generic curve. Each tab should include:

* The function that is to be preserved and the functional failure that is intended to be avoided.
* The degradation mechanism: oxidation, wear, imbalance, friction, or instability.
* The measured magnitude and the specific tool: FTIR, PQ and ferrography absorbance, envelope defect frequency, amplitude and phase.
* The acquisition conditions: load, speed, measurement point, periodicity, method and alarm criteria.
* The available evidence: first anomaly, first alarm, diagnosis and confirmation by inspection.

These cases show why these dates don't mean the same thing: a gearbox may show increasing vibration before raising an alarm; a shared tank can warn of wear without revealing where; an FTIR can signal oxidation without a dynamic change in the rotor.

Before deciding which technology to place furthest to the left, ask another question: which signal appears first in the mechanism we want to control, can we capture it with our predictive maintenance program, and how much time does it leave us to act? The order of the tools on the curve should be the consequence of that answer, not the starting point.