Laura Marcano
Department of Electronic Engineering, OsloMet – Oslo Metropolitan University, Norway / Department of Electrical Engineering, University College of Southeast Norway (USN), Norway
Anis Yazidi
Department of Computer Science, OsloMet – Oslo Metropolitan University, Norway
Davide Manca
Department of Chemical Engineering, Politecnico di Milano, Italy
Tiina Komulainen
Department of Electronic Engineering, OsloMet – Oslo Metropolitan University, Norway
Download article
http://dx.doi.org/10.3384/ecp18153132Published in: Proceedings of The 59th Conference on Simulation and Modelling (SIMS 59), 26-28 September 2018, Oslo Metropolitan University, Norway
Linköping Electronic Conference Proceedings 153:19, p. 132-139
Published: 2018-11-19
ISBN: 978-91-7685-494-5
ISSN: 1650-3686 (print), 1650-3740 (online)
Feedback is one of the key factors that makes industrial simulator training an effective learning tool. Usually, the trainees receive feedback from the instructor, who guides them through the simulation tasks. However, nowadays the availability of expert instructors is scarce while the training demand is increasing. Therefore, there is a need for new simulator training practices that could allow the trainees to be more independent and decrease the need to rely so often on the instructor. This could be achieved by offering the trainees online automated feedback. This article presents a method for developing a tool meeting those requirements is presented. Simulation data were gathered representing different execution paths of the same scenario. Data were then analyzed and clustered using different clustering techniques. Interestingly, “good” and “bad” performances are shown to be separable using different techniques for clustering multivariate time series. Furthermore, we introduce the concept of enclosing data tunnel representing the trajectory of well-behaving execution paths in a reduced dimensional space. By conditioning the mal-behaving performances to be less than 20 % of the total simulation time inside the tunnel, an accuracy on 68 % was obtained. Being more flexible and allowing the mal-behaving performances to be inside the tunnel for a maximum of 35 % of the total simulation time, the accuracy of the enclosing tunnel was increased to 84 %.
simulator training, online feedback, data clustering, enclosing tunnel, execution path