Classify

Classify parts by predicting their label and creating part sets.

Perform this task on parts that do not have manually assigned labels.

From the Assembly ribbon, click the Classify > Classify tool.

Figure 1.
From the guide bar, select the Classifier drop-down menu to select an existing classifier.
From the guide bar, select the Files drop-down menu to select a trained machine learning model file from the selected classifier’s folder.
The Parts selector activates.
With the Parts selector active, select the parts to classify under the labels of the selected classifier/trained model.
From the guide bar, click to set a value for the Certainty Ratio (see Certainty Ratio, Unclear Predictions, and Unrecognized Parts).
Click Classify to classify the selected parts.
The parts are color-coded based on the predicted labels. The legend that is displayed after classification is complete is temporary and only remains while you are using the Classify tool.
Part sets are created for review and traceability after exiting the Classify tool.

Labels are assigned to all parts selected for classification even if some of them have no obvious similarities to the original manually-assigned label contents used for training.

For example, if a machine learning model trained to distinguish springs from dampers is used to classify wheel rims and suspension control arms, the spring and damper labels are also going to be assigned to all wheel rims and control arms selected for classification.

Certainty Ratio, Unclear Predictions, and Unrecognized Parts

When a part is processed by a trained machine learning model, the likelihood of the part matching each label is calculated. Sometimes, predictions are returned that the machine learning model knows are uncertain.

The Certainty ratio is the ratio of the probability for a part to classify under the second most-likely label over the probability for the same part to classify under the most-likely label. The default value is 0.8. While using the Classify > Classify tool, from the guide bar, click

to modify the Certainty ratio value.

If the certainty ratio calculated for a part is greater than the threshold value, then the machine learning model prediction is considered uncertain, and the part is assigned to a part set named:
<label_name>Unclear(_<certainty_ratio>)_<iteration>
If the certainty ratio calculated for a part is lower than the defined threshold, then the prediction is considered certain enough and the part is assigned to a part set named:
<label_name>(_<certainty_ratio>)_<iteration>
If the part does not appear to be shaped like any parts that have been labeled, then the machine learning model marks the part as unrecognized and assigns it to a part set named:
Unrecognized(_<certainty_ratio>)_<iteration>

For example, if a trained model has three labels: wheel, panel, and bolt, and the prediction is 0.52 for wheel, 0.45 for panel, and 0.03 for bolt, then the classification ratio is 0.45/0.52 = 0.86. This is above the default 0.8 certainty ratio. As a result, the part is predicted as "wheel", but given the part set “unclear”.

Note: These prediction values are internal.

Add Trained Model File

Add a third-party trained machine learning model file.

The resultant binary (.aic) files from model training can be shared and used on different machines.

From the Assembly ribbon, click the Classify > Classify tool.

Figure 2.
From the guide bar, click to select a file for Add Existing .aic File.
A new classifier folder is created in the project directory and a duplicate of the trained model file is saved in it, which you can use later for classification.
Note: You can use these models for classification, but you cannot label additional parts in this classifier or retrain it.