Skip to content
Service-King

Knowledge base · Service-King

Predictive maintenance: data, sensors and history

The replacement date written on the parts card helps you plan your work. Prediction requires more than that: condition data, a method to evaluate and verify that the indications actually help make decisions.

Service-King team Published: Updated:

A vibration sensor on an industrial motor.
Topic illustration · Service-King

Three ways to organise maintenance

Disaster recovery, planned and predictive
ApproachBasics of operation
After crashResponse to a problem.
ScheduledTime, mileage or other established criteria.
Condition-based/predictiveCondition data and assessment indicating need for action.

The planned life of a part is not a measurement of its actual wear. Similarly, the percentage entered by the user does not automatically become a failure forecast.

Schedules, condition and forecasts have different foundations

Replacement after the agreed time is the plan. The response to the measured condition uses the current information. The forecast further aims to assess future behavior. The end-of-life date of a part does not automatically measure wear or predict failure.

It is worth naming the approach used according to the actual source. This allows the user to understand whether they are seeing a human-entered parameter, a sensor value, or a model result. Each piece of information requires a different method of evaluation and responsibility.

History should link the symptom, action and result

If the entry ends with "fixed", it is difficult to judge whether the problem has returned. Record the significant symptom, conditions, items listed, and test result. Similar fault names should have consistent meanings.

The absence of an entry is not proof of the absence of a failure. When changing the recording method, the number of events may increase even though the technical situation has not worsened. The person analyzing the data should know this context. They have common ground machine cards and reports.

Put maintenance history in order first

Repeatable fault names, replacement dates and inspection results help you see recurring problems. Without this data, it is difficult to evaluate even a simple rule. No entry does not mean no failure. Can only mean that the team didn't register it.

In Service-King, the machine card combines parts, log and documentation. This is the foundation of service organization. Automatic vibration measurements, signal analyzes and predictive models require a validated range of devices and integrations.

How to assess a predictive maintenance project

Select a specific type of damage and check if the data allows you to recognize it in advance. Compare the result with your current planning method. Account for missed issues and false positives. More alerts do not mean better service.

The decision should take into account the cost of inspection, parts availability, and the impact of unnecessary downtime. This is an assessment of the process, not just the model.

Start with a limited group of machines

Select a device for which you know typical problems and can check the data. Write down how much notice time will be useful and who will take action when indicated. An alarm that appears without any response may not improve the result. Likewise, a very cautious model can cause costly, unnecessary downtime. Therefore, the assessment must be combined with the organization of service, the availability of parts and the method of confirming the work performed.

Related topic: maintenance activity workflow.

How to assess model predictions or rule-based alerts

Prediction utility assessment
ResultQuestionCost to be included
Good warningDid it show up in time to respond?Inspection and planned action.
False alarmDid it trigger unnecessary downtime?Verification and loss of availability.
OmissionDid the problem occur without a signal?Effects of failure.
No data availableDoes the user see the vulnerability?Risk of false trust.
Process changeDoes the model still fit the conditions?Reassessment and maintenance.

Start with one clearly identifiable problem

Select a group of similar devices and the type of damage for which you have data and the ability to assess. Determine the amount of notice needed and the action to follow. An alarm with no response option does not necessarily improve the score.

Compare the design with the current handling method and simple rule. Record the cost of sensors, analysis, control and maintenance. Don't assume that a more complex model will be more useful if the data is incomplete or there are no parts to perform the action.

What can be organized in Service-King

Records, parts, journal and documentation help build the history. Handling requests and deadlines organize the team's work. Automatic measurements and predictive models require a validated range of devices, data and integrations.

Start with quality registration and a useful report. When you know which problems reoccur and what their consequences are, it is easier to assess the next stage. The practical scope is described maintenance and history and listings.

Scenario: an alert cannot be acted on immediately

The analysis tool indicates a possible problem, but a replacement part is not available. It is worth determining in advance what action should result from the warning and what warning time is useful. An accurate signal without the ability to react does not guarantee an improvement in the result. The design should include the service organization, not just the model.

Prepare priorities and responsibilities for assessment. Not every alert has to lead to downtime, and the decision depends on proper technical knowledge and procedures. Record what was checked and why a particular course of action was chosen. This will ensure that subsequent analysis evaluates actual decisions rather than counting just closing the message as a success.

Check the cost of false positives. If each triggers expensive controls, an assessment of their frequency and impact is needed. On the other hand, a model that ignores important issues may also be unusable, despite a nice statistical average. It is worth comparing it with a simple rule and previous planning in similar conditions.

Data quality may change when a sensor is replaced, a machine is upgraded, or the way of working changes. The result should have a source and a known scope. The lack of current data does not constitute confirmation of good condition. In maintenance, there needs to be responsibility for assessing the continued suitability of the model and re-examining it after significant changes.

Service-King can organize records, parts, tickets and activity history. Automatic analytics and measurements require a confirmed design. This foundation helps to first distinguish between a data problem and a decision problem. Only then is it worth expanding the scope to additional devices, maintaining the actual possibility of reaction and assessment of the effect.

Arrangements to record after this scenario
AreaPractical arrangement
TargetNotice time needed and possible action.
RatingAppropriate technical responsibility.
CostInspections, downtime and missed issues.
DataSource, freshness and changes in conditions.
MaintenanceReassessment after upgrade or measurement change.
HistoryDecision and actual result of action.

Getting started: a checklist

  • One machine or group of similar machines.
  • Reliable failure and service history.
  • Known source of status data.
  • Comparison with simple schedule.
  • Person who evaluates the indication and performs the action.

See machine card and inspection planning.

Questions worth clarifying before implementation

Does percentage of fitness mean automatic prediction?

No. You need to know the source and method of calculation. The value entered by the user has a different meaning than the measurement or model result.

Can service be improved without sensors?

Yes. Reliable history, timelines and documents can improve planning and resolving recurring issues. This should not automatically be called a prediction.

How to check the benefit?

Compare decision effects, false positives, omissions, and costs to a baseline. The number of warnings generated alone is not enough.

Go to solutions related to this topic

Free consultation

Let's start with failure history and the data available.

Describe the devices, symptoms, and current maintenance planning. We will determine what can be sorted out in the records and what scope of analysis requires further assessment.

Contact Service-King

Describe your case.

Fields marked * are required.

1. How can we help? *Your message will be sent to: Sales.
2. How can we contact you?
I prefer contact:
3. Case details

Go to the contact form

Website language

Choose a language.