Machine learning-based prediction of missing parts for assembly

Manufacturing companies are faced with the challenge of managing increasing process complexity, while at the same time having to meet ever higher demands in terms of on-time delivery and product costs. Especially at points in the value chain such as assembly, where different material flows converge, it is often not possible to provide the components required for an order in a timely and synchronized manner. Early identification of missing parts at the beginning of assembly can help to take countermeasures to meet the required delivery dates. To achieve this, this thesis develops machine learning based prediction models that can predict potential missing parts at the start of assembly at an early stage in the value chain. The development of the models was carried out as case studies at manufacturing companies in the machine industry. As a basis for the development, an extensive systematic literature search was conducted on existing approaches for the prediction of lead times of production orders. The result was that no approach exists that takes into account the full complexity of manufacturing companies. In particular, with regard to the data used, it became clear that information about the product to be manufactured—so-called material data—has not been used up to now. Based on the systemic review, a model for predicting missing parts from inhouse production was implemented. It was shown that classification approaches achieve the best possible model quality for components from in-house production. With the defined modeling approach—classification—it was then verified that material data has a significant influence on the model quality and is therefore relevant for the prediction of missing parts at the start of assembly. Finally, a model for predicting delivery delays in the purchasing process was implemented, which makes it possible to predict potential missing parts from suppliers at the time of ordering. The case studies show that the use of machine learning for the prediction of missing parts in both in-house production and the purchasing process can identify delays in the start of assembly at an early stage. The developed models are therefore suitable as a support system for production planners and controllers as well as purchasing departments to improve material availability at the start of assembly.