A data-driven AI workflow for build-time estimation (BTE)

• Build-times of 3D models are mainly predicted employing a global AI modeling approach using all parts in the heterogenous dataset
• As an alternative, the dataset can be divided into subsets of homogenous parts whose characteristics and building times are comparable
• This helps a data-driven algorithm to better learn the mapping between the 3D objects’ characteristics and their printing time
• Allows to gradually construct and maintain a reference repository composed of 3D objects, their characteristic features and the associated AI models for BTE estimation

Input(s):

• 3D Objects (STL, Native CAD)
• Correct BTE

Main feature(s):

• Extraction of features characterizing the 3D objects and automatic selection of the most important features for estimating the objects’ build-time
• Use data-driven methods to divide the heterogenous set of 3D objects into homogeneous subsets
• Train independent AI models on the identified subsets for estimating build-times of the 3D objects
• Incremental learning and performance improvement as more data becomes available
• Capturing any validated BTE estimation into a reference repository

Output(s):

• Reference repository composed of 3D objects - features - AI BTE models
• Estimation of the build-time of 3D objects
• Interactive notebook implementing the validated AI workflow to be used for research experimentation

• Access to a large dataset of 3D objects to build independent AI models on the identified subsets of the data. This might not be readily available
• The AI workflow should be trained on a dataset in which the 3D objects were printed in the correct/optimal orientation and their build-times were computed accurately to make a reliable ground-truth

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

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