Similarity search

By using absolute values or signal shape, you can apply the Similarity Search feature to search for events in your data that present similar profiles to an event you find interesting.

You can quickly find and analyze peaks related to unit shutdowns based on variables that are characterized by a signal shape. By doing this you can easily determine how frequently the signal shape appears in the data stream, and hence how frequently a certain event occurs in the operation of your process.

Note: The similarity search will search through the time period highlighted by the context bar (range selected by using the double slider). Time periods that are filtered out or not yet indexed, are excluded from the search.

How to perform a similarity search

Select a set of tags and a time window that contains the behavior of interest, then visualize the event in the focus chart. When activated, the similarity search tool will search for the complete pattern highlighted within the focus chart. 

Note: The index resolution setting determines the minimum length of a search pattern. The length of your query period needs to be a minimum of 4x the index resolution for similarity searches. 

  1. Within the search menu, select the similarity search icon .
  2. Change search options if required (see search options below)
  3. Press the button 'Find similar situations' to start the search.
  4. Wait for the result list to appear. By default, the results are sorted and grouped by match score (see working with the search result list to change the order of the results). Results are also indicated in blue under the context bar.

You can navigate through the results list to identify results of interest, and visualize them in the focus chart by clicking on a result. You can also visualize events by adding additional layers, just click on the add layer icon.  .

 

Search options

  • Tags can be included or excluded from the search by selecting the checkbox next to the corresponding tag color. A tooltip with the tag name will appear when hovering over the colors.
  • For each tag, the similarity search can be performed on absolute values or on signal shape. 
    • Absolute values: Select this option when the absolute level and evolution over time needs to be matched. 
    • Signal shape: Select this option when the absolute level of the measurement is less important, but the signal shape or measurement evolution (up / down) needs to be matched. The signal shape option will not rescale the measurement, but will allow vertical displacement of the pattern.
  • The user can change the minimum match score (default = 60% match quality). Increasing this threshold value speeds up the search, but fewer results will be found. Decreasing the threshold has the opposite effect.

Check how should I interpret similarity search scores to learn more about this score.

Weights

Occasionally, the best overall search results may not match the queried feature of interest particularly well, a peak for example. In these cases, weights may be used to increase the importance of a selected sub-window of the query.
  1. Highlight a selected area.  
  2. Use the weight buttons to set the weights. The line thickness of the trend will change accordingly.  
  3. Press the button 'Find similar situations' to start the search.

Hint

To give one tag more weight over the others, you can alter the tag scales. The similarity score is determined based on the relative variation of a tag to its total range in the focus chart. By increasing the tag scale the weight of a corresponding tag will decrease in your similarity search.   
 
In certain cases no results are found. This may occur when:
  • No matches with a sufficiently high score are found in the selected time range.
  • A user tries to search in a period that is not yet (fully) indexed.
  • The window of the focus covers a substantial part of the context range.
  • Filters are active that exclude significant parts of the context bar.

Similarity search deep dive video

Cross-asset similarity search

Open 'Advanced options' to search for the occurrence of a pattern that occurs in one asset into another asset (e.g., check if a pressure dip on reactor A also occurs on reactor B).

To perform a cross-asset search (i.e. verify whether a pattern that was observed in asset A also occurs in asset B), follow these steps:

  1. Add tags from asset A and asset B to your list of active tags
  2. Hide all tags from asset B by clicking the eye icon  .

    Similarity_Search_-_advanced_-_1.png

  3. Visualize the pattern of interest in asset A.
  4. Open the 'Advanced options' for your similarity search. In the mapping dropdown on the right, map the tags of asset B to those of asset A.

    Similarity_Search_-_advanced_-_2.png

  5. Close the 'Advanced settings' and press the 'Find similar situations' button.
  6. Click the results or add layer(s)  from the results list. Results are now defined on the tags from asset B, and will visualize the corresponding tags.

Cross-asset search deep dive video