Setup and Usage¶
To setup and use "Insights Hub Quality Prediction" application, follow these steps:
Open Insights Hub Quality Prediction application in the Launchpad.
Open overview of the prediction models.
Click “Create Model” to add new prediction model, either for new version or to clone to an existing model in actions menu with setup wizard. To setup a prediction model, follow these steps:
- In "Select model information", enter "Model name", "Description of model version", select asset and aspect with process data used for model training and click “Next”.
- Select input features, enable additional features considered in training and click “Next”.
- In “Select filtering features”, add process variables with their values and enable Zscore filter to remove outlier values.
- In “Aggregation”, select calculation methods to build aggregated variables of the manufacturing phase.
- In “Select quality data”, select the CSV format quality data file to train the model from IDL.
- In “Select quality variables”, select the quality variables from the data file.
- In “Configure model”, select model type and parameters in the user dialog or automatically by optimization routine.
- In “Review training and save model”, click “Submit” to finalize and store the model.
Display and evaluate the model training results
These are the following results displayed and evaluated:
- Evaluate the model accuracy with selected metrics with Regression score(R2), Root Mean Square Error(RMSE) and Mean Absolute Error(MAE) for the training and testing data sets.
- Visualization of prediction and true results as runtime chart, scatter chart and deviations histogram.
- Analysis of the feature importance to define the process parameters having the major impact on the predicted quality result.
- Visualization of process and quality data used for the model training as runtime charts.
- Analysis of the process parameter distributions and correlation between parameters.
Managing the model versions
- Open View all versions from the action panel
- Navigate to the most actual model version
Setup for generation of predictions (inference)
Select asset and aspect to write prediction results. The inference engine will generate predictions of the quality variable used for the training of the model. The destination aspect should contain variables for the quality result, high tolerance and low tolerance exactly with the same name as in the quality file used to train the model.
Select one of the two modes of inference:
- Single point – Prediction result will be generated at each data point of the selected asset, which meets filtering criteria defining the manufacturing phase used to configure and train the model. The aggregated data used as input for the model to generate prediction will be calculated starting from the first data point of the manufacturing phase.
- Aggregation – the aggregated data will be calculated only one time after completion of the manufacturing phase.
Display Operator dashboard
Evaluate current production in the predicted quality results.
Select the process and quality prediction variables defined in the model configuration to display automatically on the dashboard. Further variables can be selected from the corresponding process and quality aspects.
Current prediction result and current values of process parameters which have the highest importance for the model are displayed.
The notification service provided by the rules engine in Insights Hub Monitor can be used to inform users if the prediction value exceeds tolerance limits.