Enhancing Operations Planning and Scheduling in Dynamic Production Systems by Using CLIP

Julia Markert, Matthias Kerzel, Michael Variola, Dominik Saubke, Stephanie von Riegen, Lothar Hotz, Pascal Krenz

Abstract: Individualization and sustainability are current trends that lead to new challenges for the production of goods, such as producing efficiently, on-demand and in small batches. One result of this are efforts towards more local production in micro, small and medium-sized enterprises (MSME) in order to make value creation cycles smaller. This enables environmental benefits such as shorter transport distances and closed-loop product life cycles, in addition to economic advantages like more independence from global trade and support of the local economy. But more individualized production and smaller scales increase product variety. Thus, to meet changing demands, local MSMEs will have to collaborate more, especially for complex products requiring specialized knowledge. As a result, MSMEs face a multitude of challenges, such as the planning of their production and especially of value chains across the network companies. Recent developments in computer science have opened up new possibilities to support such planning processes in networks. This paper explores how one of these technologies, CLIP, which was introduced in 2021 by OpenAI, can be utilized to support Operations Planning and Scheduling (OPS) tasks. The acronym stands for “Contrastive Language–Image Pre-training” and it is a neural network that uses text-image pairs. Additionally, the use case of local furniture production in a network is presented. CLIP is tested using data from a repository as well as real-world data and the results are analyzed.