Predictive maintenance on air compressors: indicators, alerts, gains
Predictive maintenance eliminates 70 to 75% of breakdowns on average, according to the US Department of Energy, and extends component life. Here are the indicators to monitor and the alert thresholds.
Three maintenance strategies: which one to adopt
Corrective maintenance (repair it when it breaks) is the cheapest to set up and the most expensive to live with. Its real cost is not the repair but the production lost during the stoppage: that figure is specific to your site, and it is the first one to establish before any decision.
Calendar-based preventive maintenance (a service every X months) reduces failures and, according to the US Department of Energy's O&M best practices guide (2010), saves 12 to 18% compared with reactive maintenance. But it over-consumes parts: components that are still good get replaced "just in case".
Predictive maintenance (monitoring indicators, intervening when a threshold is crossed) is the optimum: you intervene just before the failure and replace only what needs replacing. According to the same guide, a functioning predictive programme saves a further 8 to 12% compared with preventive maintenance, eliminates 70 to 75% of breakdowns on average, and cuts downtime by 35 to 45% and maintenance costs by 25 to 30%.
Between preventive and predictive maintenance, what decides is less the number of running hours than the cost of a stoppage: on a compressor that is critical to production and has no standby machine, predictive maintenance is justified first. The reasoning is set out at the end of this article.
The seven indicators to monitor continuously
| Air temperature at compression discharge | Normal range, alert and shutdown: the manufacturer's values; track drift at comparable load and ambient temperature |
|---|---|
| Oil temperature | Thresholds: the manufacturer's values; follow the trend rather than the instantaneous value |
| Intake filter differential pressure | Replacement threshold: the manufacturer's; a dirty filter reduces capacity and efficiency (US Department of Energy) |
| Air/oil separator differential pressure | New: 2 to 3 psid (≈ 140 to 210 mbar) at full load; replacement usually suggested at 10-12 psid (≈ 690 to 830 mbar), according to the US Department of Energy's Sourcebook |
| Motor bearing vibration | ISO 20816-3 zones A to D (the standard that replaced ISO 10816-3), depending on power and support; the screw airend is outside the standard's scope |
| Specific energy consumption, kWh/m³ compressed | Measured vs rated: alert if drift > 5% |
| Cumulative hours vs scheduled service hours | Alert at 90% of the manufacturer's threshold |
None of these indicators can be managed "by hand" with periodic readings. All of them require regular automatic acquisition, with several months of history to detect slow drifts.
Which sensors to add, and which are already there
A compressor run by an electronic controller already measures some of these quantities internally: discharge pressure, compression discharge temperature, running hours, load status. When the controller exposes them, for example via Modbus RTU or Profinet, the job is to open up access to that data through a gateway, then to add what is missing.
On older compressors, a few external sensors need adding: a piezoelectric accelerometer on the bearing, a differential pressure sensor, a three-phase power meter. These are standard electrical catalogue components, trivial next to the price of the machine they monitor.
- Temperature sensor: most machines already have one; check Modbus access
- Differential pressure sensor: to add on the intake filter and the air/oil separator
- 3-axis accelerometer: to mount on the motor bearing and the screw bearing (one per critical bearing)
- Three-phase power meter: to install upstream of the compressor to track specific energy consumption
- Hour counter and status signal (on-load/off-load/stopped): usually already built in
Before investing in sensors, list the measurement points already fitted and accessible. Every measurement already available on the machine's controller is one less sensor to buy: all that is then missing is the gateway that brings it out.
Data architecture: from sensor to alert
Once the sensors are in place, the data has to be centralised before alerts can be triggered. Three levels of architecture are possible, from the simplest to the most sophisticated.
- Level 1, local monitoring: a PLC or an industrial PC reads the sensors, displays the values and triggers alarms on site. One-off investment, but no long history and no trend analysis.
- Level 2, remote monitoring: a gateway pushes the data to a server, with email alerts and historical curves. A monthly subscription per machine rather than an investment, and it is what lets you see a slow drift.
- Level 3, predictive analytics with ML: the data feeds a machine learning model that detects anomalies before a simple threshold is crossed. Reserved for large fleets or critical processes.
For a single-compressor site below 100 kW, level 2 (cloud monitoring with static thresholds) covers most use cases and is easy to justify. Level 3 only becomes relevant on a large fleet or for critical 24/7 processes.
Six failures that instrumentation can predict
- Saturated air/oil separator → ΔP rises gradually towards the replacement threshold (10-12 psid according to the DOE, see above). A planned intervention instead of a breakdown with oil carried into the network.
- Clogged intake filter → intake ΔP increases. Consequence if left untreated: overheating and loss of flow.
- Screw or motor bearing at end of life → the vibration signature changes (BPFO/BPFI harmonics appear). Detectable several weeks or even months before failure (NASA reliability-centred maintenance guide, 2008).
- Fouled cooler → the discharge temperature rises gradually, for lack of exchange on the air or water side.
- Sticking inlet valve → specific energy consumption degrades sharply. A source of hidden energy cost.
- Drifting pressure sensor → the machine may keep running, but in degraded mode without anyone knowing. Detected by cross-checking several sensors.
These failures can be spotted with modest instrumentation and a monitoring system that keeps history: slow drifts before the breakdown, sudden faults as soon as they occur. The calculation is simple: compare that cost with the cost of a single unplanned stoppage on your site.
When instrumentation is justified
The case for predictive instrumentation depends on three parameters: the cost of an unplanned stoppage, how critical the compressor is to the process, and the hourly cost of corrective maintenance.
On a single-compressor site with no standby machine, a stoppage costs the entire production. On a site with redundancy, the cost is limited to the on-call callout and the part. The more exposed the site, the more the instrumentation is justified.
| Critical site with no standby machine | A single unplanned stoppage costs more than the instrumentation |
|---|---|
| Site with N+1 redundancy | The gain lies in parts and planning, no longer in production |
| Lightly used site | Calendar-based preventive maintenance is usually enough |
Start by costing a complete stoppage of your site, lost production included. It is the figure that weighs most, and it is specific to your business: until it is established, no investment comparison makes sense.
Recommended models for this topic
Selected by our technical team
A compliance question?
Our technical team answers within one working day with a free preliminary study, sizing and a quotation.
How to calculate and reduce pressure drop in a compressed air network
A 1 bar pressure drop costs around 7% in energy across the whole compressed air installation. Here is how to calculate it, measure it and reduce it: pipe diameter, network layout, filters, dryer, fittings.
Costs and energyPermanent magnet VSD: what the drive really saves
What a permanent magnet drive changes on a varying load: idle hours removed, and the annual gain to be worked out from your own load profile, across 6 typical power ratings.
GuideHow to size an air compressor: flow, pressure and storage
Sizing a compressor properly is not a matter of picking a power rating in kW: it means calculating the real flow, the right pressure and suitable storage. A 4-step method, with a worked example.