i am newbee. after some research and interviews with the expert, they are sure that it is possible thing to do with AI. and i also cant find any post related to this.
There is a sketch of end geometry and mold (geometry). the end geometry is similar to mold geometry, but not the same. to create mold geometry there are alot of factors to be considered (shrinkage, etc.). and we have several thausend pairs of them in CAD-Data (dxf, dwg, step etc.)
so the idea is to create AI-modell which can generate mold geometry by looking at end geometry.
my first question, could FT do such a tasks?
which kind of input data can FT handle? would 2D-CAD will work?
should the 2D-CAD first convert to picture, teach them in AI-FT-Modell, generate a mold picture, and finaly converst them in 2D-CAD?
Yes, FT, specifically a type of GAN called a Conditional Generative Adversarial Network (CGAN), can potentially be a good fit for this task. CGANs excel at learning relationships between input data (end geometry) and desired output (mold geometry).
FT typically works best with numerical data. While FT itself might not directly handle 2D CAD file formats (DXF, DWG, STEP), there are two approaches you can consider:
Preprocessing to Vectors:
Extract relevant geometric features (e.g., dimensions, angles) from your 2D CAD files and convert them into numerical vectors. These vectors will be the input for your CGAN model.
Tools like libraries for these CAD formats or custom scripts can help automate feature extraction.
Image-based Training:
Convert your 2D CAD data into images (e.g., PNG). While this approach loses some geometric precision, it leverages FT’s ability to handle image data directly.
Ensure consistent image sizes and preprocessing steps for all images in your dataset.