Correlation Recommendations

The 'Correlation Recommendation' tool is one of two recommendation features that TrendMiner's 'Recommendation Engine' provides (the other being the 'Fingerprint Deviations'), and is implemented under the light bulb icon, situated top right of the TrendMiner user interface.

The Recommendation Engine - correlation feature enables you to identify the most significant linear relationships between an active tag and one or more potential influence factors for your current view, taking into consideration all layers present in the focus chart. Suggestions include influence factors in upstream process sections with potentially significant time delays.

Note: For those familiar with the influence factors tool. The recommendation engine can be used in much the same way as the influence factors tool, to generate a hypothesis. The difference between the two is that the recommendation engine takes all tags and layers present in the focus chart into consideration; while the analysis for influence factors only applies to one specific tag in your base layer.

Why use the correlation recommendations tool?

In practice, the correlation recommendation feature can be used to quickly and conveniently generate a hypothesis about possible contributors to an abnormal situation of interest.

For example:

  • You find an anomaly in your process.
  • You search for similar occurrences with the Search tools.
  • You add those occurrences as layers on your view and then request recommendations.
  • The Recommendation Engine will then provide suggestions for possible causes of this behaviour for all selected occurrences in the form of other tags. Creating further insight into other process parameters that misbehave.

Early indicators

The recommendation engine will automatically take time shifts of up to 1 hour into consideration. The best result in the time window within the current view and the hour before, will be displayed as the result. Results with a significant time shift will be indicated with a label 'early indicator'.

The “Early indicator”-badge is only shown if the time shift is bigger than 10% of the time range selected in the Focus Chart.

How to use the recommendation engine

Before launching the recommendation you have to be aware that:

  • Results are being computed based on the time period selected in the focus chart; Maximum time span is dependant on resolution.
  • Suggested results are sourced from tags that are available in the TrendMiner Store Index. Unindexed tags will not pop up as recommendations.

In order to launch the recommendation engine:

  • Ensure that the correct period of interest is selected in the focus chart.
  • Click on the light bulb icon  located top right of the screen. The recommendation engine will now start retrieving results, taking into consideration each tag and layer visible on the screen.

  • Results are retrieved and returned per visible tag. 

 

Screen_Shot_2018-10-15_at_11.13.00.png

 

  1. Click on the tag of interest to check potential correlations with other tags.
  2. Click on any result to add it the view.

 

Screen_Shot_2018-10-15_at_11.16.17.png

Note: tags that are already in your active tag list (hidden) will be shown with an empty eye icon. In order to visualise the tag go to the tag menu click the eye so that the iris appears. Alternatively you can delete the tag then reintroduce it to the tag list.

Recommendation Results

  • The recommendation engine will display results as soon as one result is calculated. Subsequent candidate evaluation is prioritised based on:
    • the length of focus-time-ranges (small time-ranges get preference over longer time-ranges to prevent congestion)
    • the total number of candidates to be processed
  • Multiple layers present in the Focus Chart are taken into account:
    • An optimal time shift is determined for each layer individually.
    • Each layer is considered evenly important. The significance shown is the average of all single-layers.
  • Using the correlation calculation it is possible to see the difference between positively and negatively correlated results.

Note: closing the recommendation engine will cause a reset of results

Recommendation settings

The recommendation engine settings can be adjusted to search for correlations with a minimum significance only, tags containing specific characters and to return results for early indicators. In order to change the settings:

  1. Click on the light bulb icon  located top right of the screen to open the recommendation engine.
  2. Click on the settings icon  and adjust your settings as required:
    • Min significance: returns results above the value entered
    • Filter expression: returns tags containing what is entered in this field
    • Early indicator checkbox: returns results with time shifts.

Screen_Shot_2018-10-15_at_12.13.00.png

  1. Click the "save settings"-button.
  2. Close the recommendation engine.
  3. Reopen the recommendation engine to start searching for results with the adjusted settings.

Limitations

  • The Recommendation Engine will only return indexed tags.
  • The early indicator only returns results between the current time and one hour.
  • Recommendations are restricted to time-ranges of 90 days.