Skip to content

Generation of model predictions (Inference)

To generate model predictions, start by selecting the appropriate asset and aspect where the prediction results are available. The result is added to the chosen aspect, which should already exist in the IoT asset model.

The inference engine generates predictions for the quality variables used during model training. The destination aspect must contain the following variables, named exactly as they were in the training dataset:

  • Identifier: A unique variable representing serial numbers of parts, matching identifiers in the selected process data.
  • Target: The quality feature used as the target variable during model training.
  • Tolerance Low: The lower tolerance limit for quality acceptance checks.
  • Tolerance High: The upper tolerance limit for quality acceptance checks.

The two modes of inference are:

  • Single Point: Generates a prediction at each data point for the selected asset that meets the filtering criteria defined during the model configuration and training. Aggregated data for input is calculated starting from the first data point in the manufacturing phase.
  • Aggregation: Aggregated data is calculated once after the completion of the manufacturing phase.

To start generating predictions, click "Start Prediction" in the model versions overview, which displays the prediction configuration window.

Model Predictions


Last update: January 21, 2025