Train the Component Baseline

Click

Start Training & Monitoring

on the Training Page. The user will see the progress bar begin to populate as data is acquired from the variable frequency drive.

The application will create a new model at each half-hertz increment of the operation speed of the component. The model is automatically switched by reading the command frequency of the drive. Once the number of training iterations is acquired, the given frequency bucket will automatically switch from training to monitoring. Multiple frequencies can be trained in parallel depending on the variability in speed of the application. FactoryTalk Analytics GuardianAI is designed to switch automatically across frequencies between training and monitoring.

While the component is training, the user cannot edit its configuration. To make any edits, the user must click

Stop Training & Monitoring

.

Once the training is completed, an indicator with name Quality is tagged with a badge as

Good

or

Uncertain

based on the data it captured. This tag is available for each completed bin frequency. When the user clicks on the badge, the

Baseline Quality Metrics

pop-up window is displayed. It provides the following details:

State
Output Frequency
Quality Score
Normal Cluster Score
Quality over Time plot

The plot shows the good and uncertain data captured over time. Good data is indicated with green dot and uncertain data is indicated with yellow dot. The user can retrain the baseline based on the preference. The training will restart progress from 0%.

If the quality score is less than 90%, FactoryTalk Analytics GuardianAI considers that it is uncertain quality data and provides a recommendation to retrain the baseline. It is users choice to choose

Continue monitorting

or

Retrain

the baseline.

Click

Stop Training & Monitoring

to stop training and monitoring the component. Once the user stops training, any frequency bucket in progress will be reset back to 0%. Frequencies that were already fully trained are preserved and will resume monitoring once the user starts training again.

A dialogue box will prompt the user to accept the reset of the in-progress buckets and confirm the action.

The user must stop training to make any edits to the component configuration.

Re-training a component is essential if a major physical change is made to the system. An example might be replacing the motor or full re-alignment of a coupling. FactoryTalk Analytics GuardianAI starts the training process under normal conditions, but it is possible there was already an existing degradation at the time when the baseline was acquired. In this scenario, the application will monitor further degradation from baseline. If a user makes a major physical change without re-training, there is a possibility that FactoryTalk Analytics GuardianAI will view the normal operation as anomalous behavior. If too many false positives are detected after a major maintenance event, then FactoryTalk Analytics GuardianAI should be re-trained to acquire a new baseline of the component's behavior.

Click

Retrain

to retrain the component and aquire a new baseline.