TH NextGen - Cross-correlations

The cross-correlation diagnosis tool enables you to identify the most significant linear correlation between the visible tags of your current view and other process parameters present in TrendMiner, taking into consideration all layers available in the focus chart.

Why use the cross-correlation diagnosis tool?

In practice, the cross-correlations diagnosis tool can be used to generate a hypothesis quickly and conveniently about possible root causes of an abnormal dynamic behaviour of interest.

A typical workflow would take the following form:

  1. You find an anomaly in your process.
  2. You search for similar occurrences with the search tools.
  3. You add those occurrences as layers on your view and then perform the cross-correlation analysis.
  4. The cross-correlations feature will then provide suggestions for possible causes of this behaviour for all selected occurrences in the form of other tags.

This helps you create further insight into other process parameters that misbehave and generate new hypotheses to investigate further.

How to use the Cross-correlations diagnosis tool

As a first step, it is important to know that the diagnosis is linked to the current visual view. For the diagnosis:

  • The complete period in the focus chart is used. A maximum time span is in place which depends on your index-resolution (e.g. 90 days for an index resolution of 1 minute). The time span of the analysis is the sum of the durations of all layers.
  • All layers present in your view are taken into consideration.
    • An optimal time shift will be determined for each layer individually. The minimum and maximum shift will be shown for each candidate in the results.
    • Each layer is considered equally important. The cross-correlation value shown in the results is the average of all single layers.

After setting up your view, you can proceed with the actual diagnosis:

  1. In TrendHub, open the diagnose menu.
  2. Choose the cross-correlations option.
  3. Fill out the settings (see cross-correlation settings section)

  4. Click the "Diagnose" button to show the results.
    • Suggested results are sourced from tags that have been indexed. Unindexed tags are not included in the diagnosis.
    • A result is shown as soon as it is computed, and subsequent results are added to the list as a stream until the diagnosis is completed.
    • Only the top 300 candidates are shown for each tag of interest (sorted on absolute value).
    • A result will be marked with an early indicator label if the time shift is larger than 10% of the time range selected in the Focus Chart.


The results show a cross-correlation value. The closer the value is to 100% the higher the correlation of that result with the tag of origin.

From the result list you can add or remove the candidate from your focus chart. You can then choose to redo the diagnosis, including the newly added tag.

Note: Adding a candidate tag will automatically apply the maximum shift to the focus chart time range so you can quickly see what is happening for that candidate.

Results will persist when changing menus or making changes in the focus chart. A message will be displayed to notify that the listed results may not be up to date with the current view.screenshot4.png

Cross-correlation settings

Upstream shift (maximum): A correlation between process parameters can be instantaneous or there can be a lag between the cause and effect. The upstream shift indicates how far back in time TrendMiner is looking to find correlations. The default value is 1h but can be changed to any value up to 24h.

To determine the ideal time shift, TrendMiner automatically checks different time shifts between the current start time of the focus chart and the "maximum shift". For each time shift TrendMiner will then check the correlation, and the highest correlation and corresponding time shift will then be returned as result.

Note: The specific shifts for which the correlation analysis is performed depends on the specified maximum shift. The specified maximum shift is divided into 100 equidistant intervals and the analysis is performed for each interval. The smallest possible shift is equal to the index resolution.
For an index resolution of 1 minute and a maximum shift of 100 minutes, the correlation will be performed in 1-minute steps. For the same resolution, and a maximum shift of 300 minutes, the correlation will be performed in 3 minute steps.

Tag filter expression: The diagnosis is done over all indexed tags present in TrendMiner if this field is left blank. Using this field will only return results of tags that match the entered filter-criteria. Using the tag-filter, results in faster computation time and more specific candidates to choose from.

Early indicators: Choose to only return candidates that are an early indicator as result. A candidate is labelled as early indicator if the time shift is larger than 10% of the time range selected in the Focus Chart.

Period for Analysis: The diagnosis is done for the period selected in the Focus Chart, including all visual tags and layers. The first and most important step of a cross-correlation analysis is to first set the correct periods of interest.



  • Only indexed tags can be returned as a result.
  • Only the top 300 candidates are listed as result for each visual tag (sorted on absolute values).
  • The maximum time shift that can be set is 24 hours.
  • The maximum time period that can be diagnosed is dynamic with your index resolution (90days for an index resolution of 1 minute). The time period of the analysis is the sum of the durations of all layers.
Enjoying TrendMiner? Share your experience to help the community.
Rate TrendMiner