Build time estimation (BTE) for (high-end) FDM

• The current (before the project) accuracy of BTE for FDM had 65% predictions within tolerance. Early tests show that 88% is feasible with a trained AI model
• Two variants can be provided: a pretrained model on a large database present at Materialise (the basic user model) and tools to train models on the customers database (the advanced user model)
• The system can improve over time as more data becomes available

Input(s):

• 3D Models (STL, STEP, Native CAD)
• Production data or database plugin

Main feature(s):

• Train AI model on a set of data to predict the FDM build time of a 3D model

Output(s):

• The build time

• Organized, quality AM production data: this is only limited available
• Legacy data is not always useable
• There is currently not a (commercial) system to use customer specific data
• For the training of the AI-model at the user, training algorithms need to be provided. At this moment, free to use packages are used (Keras …) but this might change in the future

• AM users and service bureaus
• Existing (software) customers

• Different business models can be applied: license, pay-per-use …
• OEM contract