AI/ML Models

You are here:

The AI/ML Models section are where multivariate models are defined and trained.

Add a new AI/ML model

  1. Expand the “Configuration” menu on the sidebar navigation. 
  2. Expand the “Lists” menu. 
  3. Select “AI/ML Models” to view the list of models.
  4. Select “Add” to add a new AI/ML model.
  5. Populate the fields.  Data fields with an asterisk (*) are required.
    • Name*: Enter the name of the AI/ML model.   
    • Comments: Any comments you would like to store for this model.
    • External Identifier: Enter a unique identifier that represents this record in an external system.
  6. Select “Save” upon completion.

Develop an AI/ML model

In this section you will define the paramaters and training data for the model.

Details

  1. Locate the AI/ML model record and select the View/Edit button for the desired record to update the name, comments, and external identifier.
  2. The following fields are available for viewing but not editing:
    • Valid: If the model is trained and ready to use. 
    • Status: The current status of the model. The status will self-adjust as the model training progresses.
  3. The Train the Model button will appear only after an asset and telemetry types have been selected and sample training data has been generated.
    Clicking this button will automatically create a Scheduled Job that will run immediately. The Scheduled Job will self-delete after it completes running.  Click this button to schedule the model for training. 

Training Information

The Training Information section lists the parameters used for training the model.

  1. Select “Edit” to edit the training information.
  2. Populate the fields.  Data fields with an asterisk (*) are required.
    • AI Model Name*: The name of the model as defined by the AI algorithm.
    • Asset*: The asset that contains the telemetry to use for the training of the model.
    • Start Date*: The UTC date and time of the start of the telemetry training range. Assetas will recommend a start date based on the training telemetry data included in the following section. When you edit the Training Information, a button to ‘Use this Value’ will appear. Click the button to automatically set the suggested start date.
    • End Date*: The UTC date and time of the end of the telemetry training range. Assetas will recommend an end date based on the training telemetry data included in the following section. When you edit the Training Information, a button to ‘Use this Value’ will appear. Click the button to automatically set the suggested end date.
    • Date Grouping*: The date grouping to use for the telemetry data. Assetas will recommend grouping based on the training telemetry data included in the following section. When you edit the Training Information, a button to ‘Use this Value’ will appear. Click the button to automatically set the suggested grouping.
    • Sliding Window*: The number of data points (between 28 and 2,880) that are used to compute the anomaly score of the subsequent point. The default value is 300. The maximum number is the number of data points available within your start date/end date range (for example, 48 for four years of monthly data).
    • Align Mode*: How to align multiple variables (time series) on time stamps. The default is Outer. Inner means the model will report detection results only on timestamps on which every variable has a value, i.e. the intersection of all variables. Outer means the model will report detection results on timestamps on which any variable has a value, i.e. the union of all variables.
    • Fill Method*: How to fill missing data in the merged dataset.
      • Linear: the average of the previous and subsequent value.
      • Previous: propagates the last value to fill gaps.
      • Subsequent: Uses the next valid value to fill gaps.
      • Zero: fills missing data with 0.
  3. Select “Save” upon completion.
  4. Re-Train the Model button will appear in the ‘Last Trained On’ field only after the model has been trained at least once. Click this button to schedule the model for re-training immediately. 

Training Telemetry

The Training Telemetry section lists all available data for model training based on the asset and date range selected in the Training Information section.

  1. Prior to analyzing, edit each training telemetry to mark it for inclusion in the analysis. To analyze the available telemetry for the training asset and refresh the data in this list, click the Analyze button. 
  2. View the available telemetry for the training asset by clicking the first button in the Actions menu.
  3. Edit the training telemetry. On the next page, populate the following fields:
    • Include: Toggle on to include this type of telemetry’s data points in the training model.
    •  Aggregate: The aggregate function to apply to this data if there are multiple values in the date grouping.
  4. Select “Save” upon completion.

Sample Training Data

The Sample Training Data section displays the generated training data based on the parameters set in the Training Information section and the telemetry selected in the Training Telemetry section. Once training data are generated, the model may then be trained.

Click Generate to generate the first 20 rows of the training data for review.

Export the full set of training data. Data are available for export in one of the following export types. The exported file will be generated after selection. Please refresh the data by clicking the Generate button prior to export.

  • One-Table Excel: creates a single CSV file, viewable in Excel, that contains all training telemetry types.
  • Compressed CSV Package: creates a folder of compressed CSV files, one for each training telemetry type.

Manage AI/ML models

  1. Expand the “Configuration” menu on the sidebar navigation. 
  2. Expand the “Lists” menu. 
  3. Select “AI/ML Models” to view the list of AI/ML models.
  4. To the right of each record, under the Actions menu, you may:
    • View detailed information or edit the AI/ML model.
    • Copy this AI/ML model as a new model.
    • Delete this model.
Table of Contents