Tema da tese: Real-Time Simulation of Supply Chain Processes Supported by Big Data in an Industry 4.0 Context
Autor: António Amaro Costa Vieira
Programa Doutoral: Programa Doutoral em Sistemas Avançados de Engenharia para a Indústria
Orientadores: Luis Miguel da Silva Dias, Maribel Yasmina Campos Alves Santos
Data de Defesa: 02/10/2019
The dynamic and complex nature of Supply Chains exposes these networks to disruptive events, which have consequences that are hard to quantify. To avoid such problems, simulation may be used, as it allows the uncertain and dynamic nature of systems to be considered. Furthermore, such systems comprise several processes, with the respective sources generating data with increasingly high volumes and velocities. This data can add the required level of detail to the simulation models, hence paving the way for the development of simulation models in Big Data contexts.
This thesis analyzed the real case of an automotive electronics Supply Chain and proposed a Decision-Support System, supported by a Big Data Warehouse and a simulation model. The first stores and integrates data from multiple sources and the second reproduces movements of goods and information exchanges from such data, incorporates risks and allows their impact to be analyzed.
Albeit the organization, where this thesis was inserted, has technological conditions and reference business processes, several data problems were observed. Simulation was used to detect and bypass some of these issues, since those incoherencies were only identified by the needs inherent to the use of simulation. The identification of such issues in this excellence environment suggests the novelty of this research and that similar projects may experience equivalent problems. Hence, a classification of the identified data issues was proposed, serving as a milestone for future similar projects. Having bypassed such data issues, simulation was used to support the decision-making, allowing to: (1) run simulations using the stored data; (2) incorporate risks either by allowing users to fire disruptions in runtime, or by applying statistical distributions; (3) adapt the simulation model in real-time to eventual data changes; (4) load the system’s state from the data to analyze specific time slots without having to wait for the simulation to reach the start of the analysis; (5) and allow the simulation to run beyond the stored data, adding, thus, predictive features to the simulation models.