Open Access

Gerhard Gudergan, Denis Krechting, Jonathan Reinartz

• The acquisition, processing and analysis of internal and external data is one of the key competitive factors for corporate innovation and competitive advantage. Many firms invest a significant amount of resources to take advantage of advanced analytics methods. Machine learning methods are used to identify patterns in structured and unstructured data and increase predictive capabilities. The related methods are of particular interest when previously undiscovered and unknown structures are discovered in comprehensive data sets in order to more accurately predict the outcome of manufacturing or production processes based on a multitude of parameter settings. So far, this knowledge is often part of the individual or collective knowledge of experts and expert teams, but rarely explicit and therefore not replicable for future applications. On the one hand, it is demonstrated in this paper how different machine learning algorithms have been applied to better predict the output quality in the process industry. On the other hand, it is explained how the application of machine learning methods could contribute to making previously not accessible process knowledge explicit. In order to increase the prognostic accuracy of the model diferrent methods were combined, later on compared and evaluated within an industrial case. In this paper a comprehensive approach to knowledge-based process engineering is being presented.