Is predictive maintenance worth it? What the evidence supports, and how to test it in your plant

What the research supports on savings

The U.S. Department of Energy's operations and maintenance guide states that earlier studies have estimated a properly functioning predictive maintenance programme can save 8% to 12% over a programme that uses preventive maintenance alone. It presents that as an estimate drawn from earlier studies, not a measurement of its own. It compares with preventive-only maintenance, assumes the programme works as intended, and does not net out the cost of starting it.

Many pages quote larger savings. None of them is backed by a source in the research library behind this guide, so this guide does not repeat them.

Proactive maintenance and performance

In a survey of plant managers and maintenance managers, proactive strategies (preventive and predictive maintenance) and aggressive strategies (total productive maintenance) had strong positive relationships with operational performance. The same study notes that these strategies need greater commitments of training, resources and integration than reactive, fire-fighting maintenance. A survey shows association, not cause, so it supports taking the idea seriously, not assuming the payoff.

Tie maintenance to production outcomes

A case study of a paper mill proposes a framework that links maintenance to the quality, efficiency and effectiveness of operations, and shows how an effective maintenance policy can influence productivity and profitability. It is one process in one industry. The lesson that transfers is the method: judge a maintenance decision against the production outcome it is meant to protect.

What condition monitoring has to do to earn its place

Condition-based approaches depend on data and models good enough to detect and anticipate the failures that matter. A review of the field maps those methods and covers current practice, but it is a review of methods, not evidence that monitoring pays back in any operation.

Two cautions from the playbook sources apply. Where no preventive task is justified, reliability-centred maintenance uses a deliberate default (run to failure, redesign, or duplicate) rather than an arbitrary interval, so monitoring is not the only alternative to a calendar. And a monitored signal is only useful if the measurement system is trustworthy and the chart separates ordinary variation from a real signal; a signal says when something changed, not why.

A test you can run

This test is Portolan's, built from the playbook steps. First, rank assets by what their failure costs and pick the few where a failure is expensive. Second, keep only those whose failure gives a detectable warning early enough to act. Third, record maintenance cost and unplanned downtime for those assets before the programme starts. Fourth, compare the programme with the preventive-only baseline, including sensors, data and skills. Fifth, decide at a set review date against the checkpoint, and stop or narrow the programme if it has not earned its cost.

The path, step by step

These steps come from the Maintenance and asset reliability playbook. Each has a checkpoint you can observe.

1. 01

   Classify assets by function and criticality

   List the assets that carry operational, safety or environmental consequence. For each, state the function it must perform and what happens when it fails.

   CheckpointA criticality ranking tied to operational and safety consequences, with the top assets named.

   Common failureTreating every asset as equally critical and spreading maintenance thin.

2. 03

   Choose tasks by failure consequence

   For each significant failure mode, decide whether a scheduled task, condition-based check, redesign or run-to-failure is justified. Delete tasks that cannot address a credible mode.

   CheckpointA task decision per major failure mode, including the explicit run-to-failure decisions.

   Common failureAdding more preventive tasks to every asset and calling it reliability.

3. 05

   Justify condition monitoring with data

   Use condition-based or predictive monitoring only where the failure mode gives a usable warning and the consequence justifies the cost. Pilot on one asset class and compare with the existing task.

   CheckpointA pilot showing earlier detection or lower cost than the existing task, with false alarms counted.

   Common failureBuying sensors for assets whose failures give no warning or carry little consequence.

4. 06

   Review reliability and update the strategy

   Review downtime, repair time and task yield on a fixed cadence. Move assets between strategies as the failure history changes.

   CheckpointA quarterly reliability review that changed at least one task or asset strategy.

   Common failureSetting the maintenance plan once and never revisiting it.

What this guide does not settle

The research library has no controlled trial or measured return-on-investment study of predictive maintenance. The one quantified figure is an estimate quoted from earlier studies, the survey shows association, and the case study is one paper mill. Competing pages quote larger figures; this guide does not, because no source here supports them.