Conference Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 148 - Design and Analysis of Experiments
Type: Contributed
Date/Time: Monday, August 8, 2022 : 10:30 AM to 12:20 PM
Sponsor: Section on Physical and Engineering Sciences
Abstract #322476
Title: Active Learning for the Construction of Digital Twins from Design of Experiment’s Data
Author(s): Rosa Arboretti and Elena Barzizza and Nicolò Biasetton and Riccardo Ceccato and Marta Disegna and Luca Pegoraro* and Luigi Salmaso
Companies: University of Padova and Università degli Studi di Padova and University of Padova and University of Padova and University of Padova and University of Padova and University of Padova
Keywords: Machine Learning; Experimental design; Adaptive sampling; Physical experiments
Abstract:

In Industry 4.0 great interest lies on the development of digital twins that can be used as virtual representations of a physical system. Active learning (AL) is a supervised learning technique based on machine learning (ML) devoted to the sequential selection of data points that maximize information acquisition, in the aim of developing models that can reliably predict some responses of interest. In this context, we present a novel AL methodology that is based on a quantification of prediction uncertainty and model’s variable importance. The objectives of the procedure consist in the maximization of global prediction accuracy while minimizing the number of experimental trials for model training. A case study is also presented that employs the proposed methodology to predict the physical properties of innovative metal alloys for application in sustainable power generation systems.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2022 program