Skip to content
Beaconby Winda
FactOperations, Maintenance and Process Safety · Maintenance and reliability

What is predictive maintenance in oil and gas?

·10 sources

Short answer

Predictive maintenance times work by the measured condition of equipment, such as vibration, oil condition or wall thickness, not by a fixed calendar. Work is planned when the data show a failure developing. It is also called condition-based maintenance; ISO 17359 gives general guidelines for setting up condition monitoring programs.

By , Editor·Researched and drafted by ·Checked by , Editor·Last checked ·Next review

Why does it matter?

A test record that says "pass" does not show when the equipment will fail. In its 2011 to 2013 inspections of ageing offshore installations, Key Programme 4 (KP4), the UK Health and Safety Executive (HSE) found too little use of data trending to forecast when equipment would fail its performance criteria 1. It called for more use of trend data on riser emergency shutdown valves, such as closure times and leakage rates, so that maintenance or replacement could be done before the valves failed, and for trending on gas detection and shutdown equipment to avoid fixing it only after a fault 1.

How does predictive maintenance work?

Most equipment loses its resistance to failure gradually. For many failure modes there is a point where the loss becomes detectable, the potential failure (P), before the point where the equipment can no longer do its job, the functional failure (F) 2. The time between them is the P-F interval, and the US Department of Defense (DoD) reliability-centered maintenance manual states that a condition check must come round at least once within it to catch the problem in time 2.

The same manual treats a condition-based task as applicable only when four things hold: the warning sign can be detected, the P-F interval is reasonably consistent, checks can practically be made within it, and it leaves enough time to act 2. The task must also bring failure risk down to an acceptable level, or pay for itself where only cost is at stake 2.

NASA's guide describes the data side: readings start from a baseline and are judged by trend analysis, pattern recognition, comparison with limits and correlation between techniques 3. Its example sets an alert level about two standard deviations from the mean and an alarm at three, and once the alert is passed the monitoring interval is cut to one-third to one-quarter of its previous length 3. ISO 17359:2018 gives general guidelines for setting up such a program for machines 5.

Which techniques are used in oil and gas?

TechniqueWhat it detectsTypical equipment
Vibration monitoring and analysisBearing and gear defects and other changes in rotating parts; NASA calls it the most common predictive technique 3Pumps, compressors, turbines, motors, fans, gearboxes 3
Lubricant and wear particle analysisMachine wear, lubricant condition and contamination 3Rotating machinery and engines 3,4
Infrared thermographyHot and cold spots, for example in energized electrical equipment 3Electrical systems, transformers, motors, valves, heat exchangers 4
Airborne ultrasonic detectionHigh-frequency noise from gas leaks and some lubrication problems 3Pressure and vacuum systems, heat exchangers, valves 3,4
Nondestructive testing (NDT) for wall thicknessWall loss from erosion and corrosion 3Heat exchangers, condensers, tanks and piping 4
Performance and process dataDrift in flow, pressure, temperature, efficiency or valve closure time 1,3Pumps, heat exchangers, shutdown valves 1,4

Sources: NASA RCM guide, 2008 3; US Department of Energy (DOE) O&M Best Practices Guide, Release 3.0, 2010 4; HSE KP4, 2014 1.

Where does machine learning fit?

Machine learning is one way to turn condition data into a warning, not a separate type of maintenance. A review by researchers at the US National Institute of Standards and Technology (NIST), published in 2019, separates diagnostics, which detects and isolates a fault, from prognostics, which predicts the future state or remaining useful life 6. It groups prognostic methods into physics-based models, data-driven methods such as neural networks and support vector machines, and hybrids of the two 6.

The review lists limits that apply to plant data: data-driven models must be trained on data that represent the faults expected, which can be hard to validate; a remaining-life estimate without a measure of its uncertainty has little value; and collecting clean, consistent data is a major bottleneck 6. The DoD manual places health monitoring and prognostics within condition-based maintenance plus (CBM+), and treats reliability-centered maintenance as the analysis that shows whether such technology is needed at all 2. See How is machine learning used to predict corrosion rates?.

A NIST survey of US discrete manufacturers for 2016, which excluded petroleum and chemical plants, found that on average 17.3% of maintenance was predictive, 31.8% preventive and 45.7% reactive 7.

What can predictive maintenance not do?

  • Find failures that give no warning in normal operation. Protective devices such as relief valves act only on demand, so the DoD manual covers them with failure-finding tests rather than condition checks 2. The US Chemical Safety and Hazard Investigation Board (CSB) found that routine inspection and weekly function testing of the Macondo well's blowout preventer in 2010 covered the parts used in daily drilling but missed latent failures in its emergency systems 9.
  • Replace all interval work. NASA notes that predictive testing does not suit every type of equipment or failure mode, and that interval-based tasks remain appropriate where abrasive, erosive or corrosive wear occurs or where failure is clearly linked to age 3.
  • Pay off without investment. The 2010 DOE guide quotes study estimates of 8% to 12% savings over preventive maintenance, but notes that much diagnostic equipment costs more than $50,000, that training needs considerable funding, and that managers may not readily see the savings 4.

Where integrity problems show up

Element or stepIntegrity linkIn the Library
Measuring wall thickness as a condition checkNASA lists ultrasonic thickness readings among predictive checks for erosion and corrosion of pipe walls, and DOE lists thickness testing for heat exchangers, tanks and piping 3,4; for pressure-containing equipment, inspection can instead be prioritized by risk, the subject of API RP 580 (4th edition, August 2023) 10How does ultrasonic thickness measurement work?
Trending shutdown valve closure times and leak ratesHSE called for this on offshore riser emergency shutdown valves, to act before they fail their performance criteria 1Performance standard (for safety-critical elements)
Watching erosion of fluid catalytic cracking (FCC) slide valves between turnaroundsAfter a 2018 explosion at a Superior, Wisconsin, refinery, the CSB recommended monitoring slide valves between turnarounds with process data and mechanical checks, reviewed quarterly, to inform decisions such as a shorter turnaround cycle 8What is a turnaround (planned plant shutdown)?
Testing devices that act only on demandCondition data from normal operation can miss latent faults; routine tests at Macondo in 2010 did not reveal failures in the blowout preventer's emergency systems 2,9What is a safety instrumented system (SIS)?

Sources: HSE KP4, 2014 1; DoD Manual 4151.22-M 2; NASA, 2008 3; DOE, 2010 4; CSB Superior report, 2022 8; CSB Macondo report Volume 2, 2014 9.

Common questions

Is predictive maintenance the same as condition-based maintenance?
Usually, yes. NASA treats predictive testing and inspection as another name for predictive maintenance or condition monitoring, and the DOE guide labels predictive maintenance as condition-based. NIST places condition-based and reliability-centered maintenance under a wider predictive maintenance umbrella.
Is predictive maintenance the same as risk-based inspection (RBI)?
No, though they can share data such as thickness readings. Predictive maintenance acts when measured condition shows a failure developing, while API Recommended Practice (RP) 580 (4th edition, August 2023) sets out programs that prioritize inspection of pressure-containing equipment by risk.
Does predictive maintenance need machine learning?
No. NASA's methods are trend analysis, pattern recognition, comparison with limits and correlation of techniques. Machine learning is one data-driven route to diagnostics and prognostics, with its own data needs. For how predictive work compares with preventive and corrective work, see [Preventive vs predictive vs corrective maintenance](/library/maintenance-strategies/).

Beacon explains these concepts in its own words for education. It does not reproduce or replace the standards cited, and Winda is not affiliated with or endorsed by the organizations that publish them. Engineering decisions need the current edition of the standard and a qualified engineer.

Sources

  1. 1.Key Programme 4 (KP4): Ageing and life extension programme · UK Health and Safety Executive (HSE), Energy Division
  2. 2.DoD Manual 4151.22-M, Reliability Centered Maintenance (RCM), incorporating Change 1 · US Department of Defense (via Whole Building Design Guide)
  3. 3.Reliability-Centered Maintenance Guide for Facilities and Collateral Equipment · US National Aeronautics and Space Administration (NASA)
  4. 4.Operations & Maintenance Best Practices: A Guide to Achieving Operational Efficiency, Release 3.0 (PNNL-19634) · Pacific Northwest National Laboratory for the US Department of Energy, Federal Energy Management Program
  5. 5.ISO 17359:2018, Condition monitoring and diagnostics of machines: General guidelines (standard page) (ISO 17359:2018, edition 3, confirmed 2023) · International Organization for Standardization (ISO)
  6. 6.A review of diagnostic and prognostic capabilities and best practices for manufacturing (G. W. Vogl, B. A. Weiss, M. Helu), Journal of Intelligent Manufacturing 30(1): 79 to 95 · US National Institute of Standards and Technology (NIST) authors, via PubMed Central
  7. 7.Economics of Manufacturing Machinery Maintenance: A Survey and Analysis of U.S. Costs and Benefits (NIST AMS 100-34, D. S. Thomas and B. A. Weiss) · US National Institute of Standards and Technology (NIST)
  8. 8.Investigation Report: FCC Unit Explosion and Asphalt Fire at Husky Superior Refinery (No. 2018-02-I-WI) · US Chemical Safety and Hazard Investigation Board (CSB)
  9. 9.Investigation Report Volume 2: Explosion and Fire at the Macondo Well (No. 2010-10-I-OS) · US Chemical Safety and Hazard Investigation Board (CSB) ·
  10. 10.2025 Publications Catalog (entry for RP 580, Elements of a Risk-Based Inspection Program) (RP 580, 4th edition, August 2023) · American Petroleum Institute (API)