Machine Learning based Data Stream Merging in Additive Manufacturing

Artificial intelligence, especially in the form of machine learning methods, has the potential to stabilize and optimize manufacturing processes in terms of productivity, quality and resource efficiency as numerous publications in the recent past verify. To make this possible, information from various sources (machine controller, external sensory, manufacturing execution system etc.) must be tapped and aligned as a prerequisite for its utilization. Optimally, a suitable primary key (e.g. timestamp or position information) is available and can be used to merge the raw data streams. If such a key does not exist, however, merging data must be carried out manually by domain experts at great expense. To support the envisaged process stabilization and optimization, we present a machine learning based method to generate a valid data stream from various, heterogeneous sources. The herein detailed approach aims to mitigate the problem of data merging using the process knowledge of domain experts and transfer it to machine learning models. The paper demonstrates the applicability of this approach in the context of a Laser Metal Deposition case study. Therein, conducted experiments show unstable deposition heights, which lead to unwanted process changes that affect the geometric and metallurgical component quality. The eventual aim of this use case is to stabilize and optimize the deposition process. The presented approach shows promising results to pave the way for achieving this goal by merging information from various sources in such a way that meaningful conclusions can be drawn.