Predictive maintenance: acting on a warning before the machine stops

The decision the warning is supposed to trigger

A useful prediction answers a scheduling question: roughly how much running time remains before the fault becomes a stoppage or a quality problem. That window is what allows the work to be moved into a slot production can afford. Where the warning arrives with no sense of urgency attached, it either gets acted on immediately, wasting remaining component life, or filed and forgotten. Build the response into the alert itself: what the finding means, how long the asset can reasonably continue, and which decision level owns the call. Without those three, monitoring generates information the organisation cannot use.

Who is authorised to stop a running machine

This is the argument the approach provokes, because the machine is still producing and the case for stopping it rests on an engineering interpretation. Decide the authority in advance and by consequence: a technician may raise the finding, a reliability or maintenance engineer confirms it, and a named person weighs it against production commitments. Where safety or product integrity is involved, the maintenance judgement should prevail without negotiation. Where it is purely economic, production and maintenance decide together with the cost of both options on the table, and the decision is recorded so it can be reviewed against what actually happened.

Replanning production around an early intervention

The benefit is only realised if the work is scheduled into a window the plant can spare, so the warning has to reach the planner in time to move it. In practice this means the maintenance and planning routines have to be connected: a standing item where developing findings are reviewed against the production schedule for the coming weeks, and a mechanism to reserve a slot before the situation becomes urgent. Plants that keep the two conversations separate find the warning arrives in engineering, sits until it becomes critical, and is then executed as a breakdown anyway.

Buying the part while there is still time

An early warning is also a purchasing opportunity: it converts an emergency order at premium price and expedited freight into a normal one. That only works if the finding triggers a parts check, if the plant knows which components a given fault indication implicates, and if someone raises the requisition immediately rather than waiting for the intervention to be approved. For long-lead items on critical assets the warning may still not be early enough, which is an argument for stocking those particular parts rather than for abandoning the approach.

False alarms and the credibility they consume

Every incorrect prediction spends organisational goodwill, and after a few unnecessary interventions production stops releasing equipment on an engineering opinion. Track the outcomes: for each intervention, record what was found and whether the condition justified the action, and for each unexpected failure, record whether monitoring gave a signal that was missed. That record is the only honest basis for adjusting thresholds. It also protects the function, because a documented history of accurate calls is a far stronger argument for equipment access than any assertion about the value of the technique.

Frequently asked questions

Which assets justify a predictive approach?

Those where failure is expensive, where deterioration gives a detectable warning over a usable period, and where the plant can actually act within that period. If any one is missing, the effort is wasted: a critical asset that fails without warning needs redundancy or scheduled replacement instead, and an asset whose failure costs little should simply be repaired when it breaks. Work through the asset list against those three tests before committing to any monitoring programme.

Does predictive maintenance replace scheduled routines?

It replaces some of them and leaves others in place. Scheduled replacement of components whose failure is genuinely age-related still makes sense, as do statutory inspections and tasks required by insurers or regulators regardless of condition. What predictive methods usually displace is the periodic strip-down performed in the hope of finding something, which is expensive, disruptive and carries its own risk of introducing faults during reassembly.

How do we justify the investment to a finance director?

Not with a general claim about downtime reduction. Build the case on named assets: what an hour of stoppage on each one costs in lost output and recovery, how often it has failed unexpectedly, and what proportion of those failures would plausibly have given warning. Compare that against the full cost of monitoring, including the analyst time to review findings, which is the element most often omitted and the reason many programmes quietly lapse.