Published:
July, 2019
Category:
Development
Client:
Airbus
Digital Mock Ups for Artificial Intelligence
How it Works
For this manufacturing AI research, we created digital mock-ups from CAD data. The digital mockups were then used to create images to train deep neural networks.
The first step is to convert CAD drawings into discrete 3D models. Then, soft items, such as sealant, are added to the model. Textures and physics surfaces are then added to model realistic surfaces such as painting variations. Finally, the movement of the robotic arm is modelled to capture realistic lifelike images from a camera emulator.
Millions of images were created by varying foreign bodies, paint/sealant failures, physical damage, and more.
The output included the renders themselves along with failure and component masks. These were used to train the neural networks using an adversarial training paradigm.
