The manufacturing industry is faced with the challenge of how to produce parts with the targeted quality in the face of constantly changing environmental factors and raw materials using the least amount of resources. This challenge is currently met by trained domain experts that adjust manifold process parameters to try to achieve the targeted quality. Since machines produce data as well as products there is a natural interest to address this challenge using data-driven approaches. In practice, however, data quantity is secondary to data quality since machine learning algorithms perform best on evenly distributed amounts of well-structured data, which in the case of supervised learning systems have to be labelled as well. Several techniques such as data augmentation and simulation exist to circumvent this problem. However, data augmentation reduces sample quality by introducing noise and simulating manufacturing processes with their multitude of physical interactions to the required accuracy proves challenging. As such, efficiency in regards to data usage is considered paramount in manufacturing scenarios. To achieve this, this work sets out to combine the knowledge of available domain experts with learning systems. The following two research questions will be addressed during this venture towards a PhD: (1) How can we reduce the required effort to extract knowledge to create ontologies? (2) How can we guide the learning process with unaggregated knowledge to increase its data efficiency?