Create a Process Anomaly Detection Guided Model

Perform the following steps to define a TwinThread Process Anomaly Detection model.

To define a Process Anomaly Detection Guided Model

1. Navigate to the asset page for the asset you wish to add a model for, select the Asset Actions menu, then Create Analytics Model, and then select Add Process Anomaly Detection Guided Model to access the Process Anomaly Detection Guided Model Details side sheet.

   

2. Provide a name for the model in the Model Name field.

3. Confirm the asset that is populated in the Asset field.

   To choose a different asset, clear the current select with the delete icon . Then, select a new asset from the Asset field.

4. Specify tags to include in the model.

   By default, the tag selection drop-down will be filtered by the current asset.

5. Optionally, configure a Filter.

   A filter allows you to define a condition that trains the model on when the asset or process is not running. The filter tag can be a string, discrete, or analog tag.

   Note: If the selected filter tag filters out a large portion of the sample data set, the model may fail due to not having enough sample data.

   1. Select a Filter Tag.

      By default, the Filter Tag search field will be filtered and only search for tags under the selected asset. To search for tags across all assets, clear the filter.

   2. Choose a Filter Operator. Available options are: Equal, NotEqual, GreaterThan, GreaterThanOrEqual, LessThan, LessThanOrEqual.

   3. Set a Filter Value representing a condition that indicates when the target asset or process is stopped.

   4. Select Next to proceed.

6. Optionally, configure an Operational Mode Indicator.

   An Operational Mode Indicator identifies the operational mode or batch. Integer and String tags are supported. Note that:

   • Only one tag can be selected as an Operational Mode Indicator.

   • The selected tag must be excluded from the set of tags provided in step 1.

7. Set a value for the Training Window.

   This defines a period of time used to train the algorithm. The selected time period should include only normal operating values. Available range options are:

   • Last 3 Months

   • Last 1 Month

   • Last 1 Week

   • Custom

     A minimum of 7 days must be provided in custom ranges. Selecting Custom allows you to define the following fields:

     Start Time - Defines the start date for algorithm training. The algorithm will sample values from after this date to learn normal operating parameters for the selected tags.

     End Time - Defines the end date for algorithm training. The algorithm will sample values from before this date to learn normal operating parameters for the selected tags.

     Note: The effectiveness of the model is dependent on the quality of the training data provided. Ensure that the provided range includes only in-range values for the selected tags.

     Note 2: Provided training data is sampled every 30 minutes. Training data will be impacted if there is missing data in the sample set, which can happen, for example, if a data source was offline for a period of time.

8. Select Save to finish configuring the model.

   Note: If the parent asset is associated with an asset type template, you'll have the option of saving this model to only the asset, or to the asset type for replication to other asset instances. For more information, see Asset Types.