The AI model is created through the use of Teachable Machine and Cloud Vision API.

First check is done to check for the presence or absence of the label on the machine, and this binary classification is done through the use of Teachable Machine. If a label is detected to be present, then it would go to the second check. If not, the user will be notified that the label is absent.

Second check is a multi-label classification model that would check if the label has passed or failed the QC check. Below are the labels that could be returned to the user:

1. Correct: The label on the machine is placed in the correct orientation and position on the control box.

2. Crumpled: The label on the machine is crumpled.

3. Handwritten: There are handwritten markings on the label on the machine.

4. Missing-field: There are missing fields in the label on the machine.

5. Torn: The label on the machine is visibly torn or folded.

6. Wrong-position: The label on the machine is upside-down or placed at an angle.

After the check is done through our custom API mode, the user will see that the label on the machine has passed or failed, and if failed, what is the main reason for failing.

Overall, the accuracy achieved was 0.9769332 from the Vision API training. While the dataset was recreated by us, the accuracy scores prove that it is possible to train a machine learning model to do AI inspection for QC.