Configuring the PCA Model block

Load the XML file containing your PCA model. These files are deployed directly from the Discrete & Batch Troubleshooter, where you are asked if you want to export your PCA model to an XML file.
Specify a Calculation method:
- Continuous: Fields in the output ports are updated on every execute when new data is available in the input port. The execution time is based on the sampling period in the data source.
- Moving Window: This is a window period that is user defined and specified in seconds. Calculations are performed on the data over the defined window period, up to the current execute. The values are averaged over the window period, and the fields in the output ports updated on every execute. The block must execute for at least as long as the specified Moving Window period before good quality outputs are produced by the PCA Model block; the block will produce bad quality outputs until such a time.
- Batch: Calculations are performed every time a new batch is started, and a trigger is thus needed to indicate the start of a new batch. This triggering field is selected from the list of all the input fields, and must be a Discrete field of string or integer values. (If an input field with double values is used, and these values change with every execute, the results from the model will be similar to using Continuous field data.) Fields in the output ports will only be updated when the value of the selected trigger field in the parameter port changes.

NOTE: Bad quality samples in the trigger field are ignored. Only changes in good quality sample values are considered as triggers for executing the model.

NOTE: Batches can only be calculated when the parameter port is linked to the data source.

An average value over the last batch is calculated.

It is important to remember that the results of the model are calculated only at the trigger signalling the start of the new batch. This does not necessarily coincide with the last data point of the previous batch, and this might result in a delay between the end of one batch, and the calculation of the model at the start of the new batch. This delay needs to be considered when using the data, or when sinking the results of the model block.

Cutoff threshold: When comparing batches, a cutoff threshold value can be set, determined from values from the Distance to model plot for a series of batches which includes well performing batches. An ideal batch will have a very low Distance to model value, as the model predicts the process accurately, and the model outputs are thus very similar to the process outputs. Greater Distance to model values indicate poor process outputs, and an under performing batch. Distance to model values above the user defined threshold will generate an alarm in the blueprint, signalling the poor batch. The cutoff threshold can be set in one of two ways:

from the Distance to model plot from the PCA model generated in the Discrete and Batch Troubleshooter: see how to determine the cutoff threshold in the Discrete & Batch Troubleshooter.
in the Architect: after visualizing the model outputs for a few runs, decide on an optimal threshold value, either from your process knowledge or through trial and error.

Map all the required fields in the PCA model to the available double fields in the input port. Automatic mapping occurs when [<<] is selected. The input port must contain all fields used in constructing the loaded PCA model, and these input fields must be mapped to the required fields of the PCA model.
Enable / disable the variable contribution baseline: This baseline is a constant factored into the model output calculations, calculated from the difference between the variable contributions of the new model data less the variable contributions from the original data set. If this baseline value is near to 0, i.e. the modeled variable contributions are very similar to the variable contributions of the original data set, this contribution baseline can be disabled.