Condition monitoring: turning readings into work somebody actually does

Deciding coverage before deciding technique

Start from the assets whose unexpected failure genuinely hurts, and for each, from the failure modes that have actually occurred or are plausible given the duty. That determines what is worth measuring and how often. Coverage decided the other way round, by applying a favoured technique across everything accessible, produces a large route, a technician permanently occupied collecting data and very little analysis. Keep the covered list short enough that every point is reviewed by a person, and record why each asset is on it, so the list can be pruned when the equipment's role in the plant changes.

Baselines and alarm levels have to come from your machine

A reading means nothing without a reference. Establish a baseline when the asset is known to be running acceptably, and set alert and action levels relative to it, adjusted as evidence accumulates about how that particular machine behaves under its actual duty. Generic thresholds taken from a table produce alarms on machines that have always run at that level and silence on machines deteriorating within nominal range. Where duty varies with product or season, record the operating context with the reading, otherwise a normal difference in load will be interpreted as a developing fault.

Somebody has to look, and it has to be the same somebody

The review is where the value is created, and it is the step most often unresourced. Name the person, give them protected time, and make the review a scheduled activity with an output: a list of assets with a condition judgement and a recommended action. Rotating this among whoever is available destroys the continuity that makes trend interpretation possible, since recognising that a reading has drifted requires familiarity with how that machine normally behaves. If nobody can be given the time, reduce the coverage until the remaining points can be properly reviewed.

The route from an abnormal reading to a scheduled job

Define what each severity level triggers: increased monitoring frequency, a physical inspection at the next opportunity, a planned intervention within a stated window, or an immediate stop. Each level needs a named decision owner and a way of entering the maintenance work queue with the finding attached, so the technician who eventually does the work knows what was seen. Findings communicated verbally or by informal message disappear. Where the plant already runs a work request process, condition findings should enter through it rather than through a parallel channel that only the monitoring team can see.

Closing the loop by recording what was found on stripdown

When an asset is opened, whether after an alert or after a failure, the condition of the components should be recorded and compared with what the monitoring indicated. This is the only way to learn whether the thresholds are set sensibly, and it is routinely skipped because the pressure is to get the machine running. Make it a short, required step with photographs, and review the accumulated comparisons periodically. Over time the record tells you which techniques earn their place on which assets, which is a far better basis for programme decisions than a technique's general reputation.

Frequently asked questions

Should operators be involved in condition monitoring?

Yes, for the simple checks. Operators are with the equipment continuously and notice changes in noise, vibration, smell, leakage and temperature long before a scheduled route does. Give them a short structured check with clear criteria and an easy way to raise a finding, and make sure raised findings visibly produce a response. The specialist techniques requiring interpretation stay with trained staff, but the operator report is often the earliest signal available and costs almost nothing to obtain.

How often should readings be taken?

Frequently enough that the interval is comfortably shorter than the period between a detectable change appearing and the failure occurring. That period differs by failure mode: some deteriorate over months, others over days. Setting one interval for the whole route is convenient and wrong, because it over-samples slow modes and misses fast ones. Where a mode develops faster than any practical route frequency, continuous measurement or a different maintenance strategy is the answer.

What do we do with historic data nobody has analysed?

Treat it as a one-off diagnostic exercise rather than a backlog to clear. Pick the assets that have failed unexpectedly, look back at their readings, and see whether a signal was present and missed. That tells you something useful about thresholds and review discipline. Then decide the ongoing coverage you can genuinely support and stop collecting the rest, because continuing to gather data nobody reviews consumes technician time and creates a false sense of protection.