Project Improvement 3 min readORG-F38

Plan predictive maintenance around one detectable failure

Share

You want earlier warning of a fan bearing failure. The first question is not which sensor to buy: it is what observable sign could precede the failure and what someone would do when that sign appears.

Choose an asset and a failure mode

Pick an asset with a meaningful failure consequence and a failure mode that can be investigated. 'Fan condition' is broad; bearing damage with a known operating context and maintenance history is more specific. Existing measurements or periodic inspection may be enough to begin.

Design the response before the alert

  • Choose measurements for the failure mechanism with the appropriate specialist; more data is not automatically better.
  • Keep units, time and operating state with values so unlike conditions are not compared blindly.
  • Define who reviews an alert, the response time and how findings will be recorded.
  • Evaluate unnecessary inspections as well as failures that were not detected.
Measurements should lead to a checkable finding and an owned response.
Predictive maintenance pilot decision path

Measurements should lead to a checkable finding and an owned response.

Run a trial that can be judged

  1. 01

    Record failure history and current maintenance practice as the comparison basis.

  2. 02

    Collect suitable data under authorized conditions with known quality.

  3. 03

    Define the detection rule and decision owner before evaluating results.

  4. 04

    Follow alerts through to actual findings and decide how the method should continue or change.

A report store is not a prediction system

AutoNexit can retain the plan, responsibilities and reports. It does not acquire live measurements, analyze signals or perform predictive diagnosis. A failure-free trial period alone does not establish the ability to detect every failure.

Primary sources

Technical claims and product boundaries were checked against these primary sources. Fact-checked

  1. ISO 17359:2018 Condition monitoring and diagnostics of machines — General guidelinesInternational Organization for Standardization
  2. ISO/TC 108/SC 5 — Condition monitoring and diagnostics of machine systemsInternational Organization for Standardization

Frequently asked questions

Is machine learning required?

No. Match the method to the problem and available data; a reviewable rule may be an appropriate starting point.

Can we copy a threshold from an article?

Installation, operating conditions and measurement method matter. Set practical limits with appropriate technical sources and specialists.

What about data from before a repair?

Record the repair and changed conditions; mixing before/after data without context can mislead the assessment.

Should false alerts be removed?

Keep them in the evaluation with their cause and investigation cost; they are part of judging usefulness.

Related articles

ORG-F37

Preventive vs Predictive Maintenance: Data, Cost, and When Each Approach Fits

Preventive maintenance acts on a time or usage interval; predictive maintenance acts on evidence that indicates changing asset condition. Neither is automatically superior. The right policy depends on failure consequence, failure pattern, detectability, data quality, intervention cost, and the organization’s ability to act on a warning.