Knowledge Graphs (KGs) have proven to be a key factor to foster the success of Smart Manufacturing. The data generated in these contexts need to be semantically harmonized and made available for a myriad of applications. However, due to the huge level of heterogeneity, as well as the underlying data quality, there exist the requirement to continuously update and improve these KGs by means of knowledge graph completion (KGC) methods like Link Prediction or Node Classification. Besides, state-of-the-art KGC approaches fail to consider the quality constraints while generating predictions, resulting in the completion of KGs with erroneous relationships. To successfully adopt KGs and its enhancements made by KGC methods, a Neuro-Symbolic approach is required. The approach will enable manufacturing experts to understand and apply the output obtained by the methods and properly evaluate whether the results make sense according to their knowledge. This thesis will research innovative Neuro-Symbolic methods to improve KGC methods in smart manufacturing scenarios. Emphasis will be placed on the representation of heterogenous and relation-based knowledge and the combination with learning-based approaches to enable explainability and transparency of the predictions. The thesis will validate the methods on one or more real use cases at Bosch. Evaluations will be performed on public benchmarking challenges in smart manufacturing recognized by the community and internal Bosch datasets.