GAN Data Augmentation for CAD/PLC Classification

Background & context

For an elevator-control PLC auto-generation system, CAD drawings had to be classified reliably — but the labeled CAD dataset was far too small to train a robust classifier.

This project used a GAN to synthesize additional CAD-drawing data, then trained a classifier that keeps improving as new data arrives through incremental learning.

Key responsibilities & role

• Data augmentationDesigned and trained a GAN to synthesize CAD-drawing samples, expanding a scarce dataset.
• ModelingBuilt a PyTorch deep-learning classifier for the CAD/PLC drawings.
• Incremental learningImplemented incremental learning so the model improves as new data is added, without full retraining.

Tools & knowledge

Results & achievements

→ Expanded the training dataset 10× — from 700K to 7M records — via GAN augmentation.

→ Improved classifier robustness for the elevator-control PLC generation task.