• 02
  • Aug

6 tips on choosing the correct data for successful predictive maintenance

Senseye Blog6 tips on choosing the correct data for successful predictive maintenance

Time and again we encounter excited maintenance managers who possess terabytes of data; eager to kick-off their predictive maintenance project and avoid machine downtime. It never stops being painful when we take a look at the data and find that much of it us unusable for predictive maintenance purposes. 

We’ve put together six crucial tips to help you get on the right track with collecting relevant machine condition data and start to detect machine failure: 

 

Stick to UTC time

Non-UTC timestamped data is a pain. It’s trivial to change the timecode on incoming data but what causes issues is when data sources from multiple sites are brought together and someone needs to keep track of what data is in which time zones and the correct correlation of events. It saves frustration and potential confusion for everyone involved in the project if Coordinated Universal Time really is! 

 

It should be analogue

Digital data from SCADA and PLC units can be interesting to capture machine state information but it rarely contains any data useful for condition monitoring analysis (manual or automated) to understand the mechanical health of the asset. Good examples of analogue data sources are:

  • Vibration
  • Current
  • Temperature

Ideally you should capture more than one analogue data source per asset (depending upon complexity) as the richer your ‘data picture’, the better to understand long-term mechanical degradation and calculate Remaining Useful Life (prognostics). 

 

It rarely needs to be high frequency

In most cases, the frequency of data captured can be far lower than you’d think. Useful information can be gathered if data is captured every hour; more useful if it’s an aggregated value captured every 5 minutes (for example the mid, min and max of a data source for the preceding time period). As always there are exceptions; most notably vibration where whilst a Root Mean Square (RMS) value can be taken as a snapshot and provides an indication of the amplitude (vibration energy), specialist condition monitoring hardware needs many thousands of snapshots per second in order to calculate condition indicators. Whether you store data onsite or on the cloud, we suggest storing aggregated data – it’s easier on your network and storage policies whilst still able to show you the things that you really care about.

 

 

It should be captured in a known state

Capturing data at a regular interval without paying attention to machine state can result in data which is difficult to work with – as the machine is likely to be doing a different thing each time. Ideally data should be captured when the machine is performing a known and consistent operation. This allows for rapid trending of specific data.

 

Ensure it is accessible

It’s no good capturing any data if getting access to it becomes a point of major frustration to the project. A huge folder of .csv files isn’t useless but it’s much more difficult to work with than a well-structured database or even better a modern factory historian or IoT-middleware platform with a RESTful web API. Security is extremely important here and the larger factory historian and IoT-middlware providers will be able to make penetration test results available which should help satisfy IT security requirements.

 

Capture it from machines with a low MTBF

When a predictive maintenance project is getting started, it’s crucial to demonstrate a rapid ROI as the start-up costs are not insignificant. It’s an often-overlooked point that the machinery being monitored should have a failure mode that is expected to be seen in the data being captured, within a timeframe suitable to demonstrate ROI for further scaling.

 

In summary

Follow these rules and you’ll exponentially increase the likelihood of success of any pilot projects and a smooth transition to scalability. We’ve put together a little cheat-sheet to help you remember:

ideal data for predictive maintenance.pngClick the image to download a PDF copy

The Uptime as a Service Tool – Senseye

Our automated condition monitoring and prognostics software helps you to enable predictive maintenance and reduce unplanned downtime without needing to be an expert. If you’d like to learn more about our easy to use software, download your free brochure here:

 Free Senseye flyer

Senseye prognostics for predictive maintenance

Keep me notified