Managing Input Parameter Uncertainty in Digital Twins

Discrete event simulation models can be used in digital twins to support the design and decision-making process of manufacturing systems. In many industrial contexts, the collection of real-time data from the system is a costly task and the performance predicted by digital twins may be affected by input uncertainty, due to the scarcity of data used to input simulation parameters, thus leading stakeholders to biased decision-making. Literature approaches treat this problem mainly from a theoretical point of view and are applied on very simple systems that do not adequately represent real factories. The aim of this paper is to explore the advantages and drawbacks of an input uncertainty simulation technique, namely the metamodel-assisted bootstrapping procedure. This technique is applied and extended to evaluate the production rate of a lab-scale manufacturing system. We show it is possible, despite the scarcity of data, to build a reliable confidence interval on the production rate and to identify those parameters whose effect on the performance is most relevant. Moreover, the marginal contribution of each input parameter to the performance can be quantitatively assessed, thus enabling stakeholders to identify which parameters to focus on in data collection activity. © 2022 Elsevier B.V.. All rights reserved.