Predictive Maintenance

Predictive Maintenance

Predictive Maintenance involves using data to track machinery health status and predict when maintenance should be performed. Effective use of predictive maintenance can prevent breakdowns and unplanned downtime and the associated costs.

Common maintenance strategies

Two commonly used maintenance strategies are:

  • Run to Failure (R2F)
  • Preventive Maintenance (PvM).

R2F involves repairing or replacing equipment only after it breaks down, which has high costs associated with it, due to  unplanned downtime and possibly even the price paid for a new machine. Preventive Maintenance, on the other hand, is an approach by which we repair or replace equipment well in advance of a possible breakdown, which means that we might replace a perfectly working piece of equipment. As you can see, both these strategies have their own downsides: with the R2F strategy we incur costs from being too late, while with the PvM strategy we spend too much on maintaining healthy machines.

Why predictive maintenance?

A third (and better) maintenance strategy is:

  • Predictive Maintenance (PdM)

PdM tries to find the sweet spot between the two aforementioned maintenance strategies. Leaning on various statistical methods, PdM predicts the moment at which a machine might fail. This allows for appropriate maintenance scheduling in advance of this predicted breakdown. This way, the pricey and unexpected downtime caused by the breakdown will decrease. At the same time, it will prevent unnecessarily frequent scheduling of maintenance checks, which also cost something.


A trucking company currently employs the R2F maintenance method and repairs their trucks when they fail. These breakdowns are costly, because the truck often breaks down while in operation – while on the road. However, this company also collects and stores sensor data from within the truck, such as engine temperature, vibration intensity, and fluid levels. By using PdM techniques on this data we predict when a truck is about to fail and schedule a maintenance event before the expected breakdown.

At this point you might ask yourself: what if the company doesn’t have sensors to collect data? Don’t worry – predictive maintenance can still be implemented. Data can be collected, from other sources, for example, by extracting data from the maintenance logs, even historical maintenance invoices, or elsewhere.

Anomaly detection

Anomaly detection (or outlier detection) is, again, a set of statistical methods used to detect rare or abnormal events. While this method is widely popular in tracking of financial transactions, it can also be used in an industrial setting, when applied to equipment or machinery. The value this technique offers is clear: based on “normal” operational data when machinery is “healthy”, we can identify when new data points deviate from the “normal” behavior. When we notice these deviations, we can take action.

Want to know more?

If you have any questions or thoughts, feel free to contact Alexander by sending him an email.