DateTuesday December 14, 2021 from 12:00 PM to 1:00 PM
LocationOnine - MS Teams
Many TU/e researchers are advancing or using Artificial Intelligence in their research projects. To support cross disciplinary learning and to strengthen the TU/e AI network, EAISI organizes a series of (internal) meetings during lunch time, where various researchers talk about their projects, followed by Q&A.
Login details for this online meeting will be shared in Outlook.
Please send an email to eaisi@ if you are interested to present or join. tue.nl
Johan Lukkien | Dean of department Mathematics & Computer Science
Welcome & introduction
|12:30||Charly Bastiaansen | Program manager at EAISI||Introduction to moonshot 'NextGen Industry'|
Development of a physics based, multi-scale numerical framework to model Additive Manufacturing processes
Alvaro Chaim Correia
Decision tree-based models are popular discriminative learners that have been widely studied by the statistics and machine learning communities. Yet, their relationship to generative models remains largely unexplored. We show that decision trees fit in the formalism of tractable probabilistic graphical models and are readily represented as Probabilistic Circuits, a class of deep probabilistic models with tractable inference routines. This reinterpretation equips decision trees with a full joint distribution over the feature space and leads to Generative Decision Trees (GeDTs) and Generative Forests (GeFs), a family of novel hybrid generative-discriminative models. This family of models retains the overall characteristics and predictive power of DTs and RFs while additionally being able to handle missing features and detect outliers.
Additive Manufacturing, or 3D printing, allows for the manufacturing of unprecedented designs and functionality with a minimum of waste material. In addition, as products can be made directly from digital drawings, it omits the necessity to have specific tooling, which allows for an economic production of unique, tailored products and the on demand manufacturing in small batches. In the past decades, a variety of printing processes have been developed in order to print a wide range of materials varying from soft polymers to steels and technical ceramics.
The quality of the printed products such as residual deformations and the presence of cracks and voids depends on many factors. These factors comprise the properties of the raw materials, printer process settings and the conditions during printing. In order to predict the quality of the printed product a fundamental understanding of the mutual relations between the process conditions, the evolution of the material properties and the resulting geometric and physical properties is key. Material characterisation by testing provides insufficient information as it reveals the material state after printing. Additional in-situ measurements only reveal global parameters such as temperature distributions and deformations. The actual material evolution during the transformation processes (e.g. curing or melting) and the resulting residual stresses and deformations in relation to the actual print process can only be studied in detail by advanced physics based numerical models.
In this presentation, various novel multiscale modelling techniques are presented for different print technologies and material systems, including polymers, metals and ceramics. Special attention is given to the integration of model order reduction techniques and artificial intelligence in the framework to increase the computational efficiency of the analysis to enable the integration in closed loop control systems and process parameter optimization tools.