Publication
Towards an Artificial Intelligence based Approach for Manufacturing Resilience
Nurten Öksüz-Köster; Sebastian Bouschery; Martin Schlappa; Martin Unterberg; Jan Sporkmann
In: 22. VDI-Kongress AUTOMATION 2021 - Navigating towards resilient production. VDI Automatisierungskongress (AUTOMATION-2021), June 29-30, Virtual, Germany, VDI, 2021.
Abstract
Resilience describes a system’s recovery from disruptions or the capability to deal with future shocks. Not only the current COVID-19 pandemic has highlighted the importance of resilience: floods, earthquakes, bankruptcies, politics and similar events regularly disrupt the manufacturing industry. Yet, only few companies work on managing their resilience, partially since it (or the lack thereof) only becomes evident after catastrophic events. A better understanding of a firm’s resilience provides the basis for an active resilience management, which is necessary in order to survive as a manufacturing company in the future. Both qualitative and quantitative approaches for determining manufacturing resilience exist, especially in the context ofsupply chain management. Quantitative approaches can be further classified into deterministic and probabilistic approaches. Whilst deterministic approaches to manufacturing resilience lack complexity to describe manufacturing systems with several participants, quantitative approaches to manufacturing resilience fail to leverage advances in AI technologies up to this date. However, advances in AI technologies coupled with an ever-increasing amount of available data from manufacturing networks create an opportunity to approach the measurement of manufacturing resilience in a fundamentally new way. Such a new data-driven approach will become more and more valuable as manufacturing companies operate under increasingly volatile, uncertain, complex, and ambiguous conditions. The SPAICER project addresses this problem by focusing on a quantitative and data-driven approach to manufacturing resilience. Previous measurement of resilience come with several shortfalls: (1) They are usually based on qualitative inputs that need to be manually gathered via e.g. surveys, interviews or questionnaires with huge amounts of questions which potentially leads to (2) a subjective assessment of the resilience level; (3) Traditional approaches are neither data-driven nor automated, which hinders scalability and constant measurement; (4) They usually focus on certain parts of the company (e.g. production resilience, logistics resilience, etc.) and only incorporate input from few individuals within area of focus, not providing a holistic view and neglecting e internal dependencies; (5) Most research focuses on event studies after certain disruptions occured, which implies that resilience can only be measured ex-post focusing on reactive resilience and mostly ignoring anticipative resilience; (6) In terms of data, only single data sources are used. Moreover, most approaches only consider internal data sources such as ERP systems and neglect external data sources (weather, news, traffic, etc.) to better understand the whole environment impacting the processes. With our approach, we intend to overcome these shortfalls by utilizing an AIbased approach for manufacturing resilience on a micro- (local) meso- (company-wide) and macrolevel (global, network-oriented). The goal is to address manufacturing resilience through an objective and data-driven lens by using AI technologies and taking a holistic and cross-company view and thus enabling a reactive as well as anticipative resilience analysis.