Part quality is crucial in the Additive Manufacturing (AM) sector. Detection of anomalies and defects in the process with the use of recorded sensor signals is becoming increasingly important in maintaining quality. The objective is to reduce off-line quality controls and their associated costs by utilizing this data. Therefore, many monitoring systems are included in Laser Powder Bed Fusion (LPBF) systems in various. Purpose of this work is to detect pores with a high accuracy by using Artificial Intelligence (AI), to overcome the limits of manual evaluation due to the large amount of recorded data.

Therefore, different heterogeneous data have to be homogenized and merged with respect to their scope, structure and speed. 316L stainless steel was processed on a LPBF-system “Aconity MINI”, which is equipped with different monitoring systems. This are pyrometer, high-speed camera and the embedded sensors (e.g. oxygen sensor), whereby their individual output data differs strongly The raw data of the pyrometers are already available in terms of their local coordinates regarding their position on the built plate. These data are further reduced to the data located along the scan path. The high-speed camera provides videos of the individual parts, and its layers are available, that need to be first converted into images for the evaluation process. Thereby different image features such as brightness, number of pixels, and size of the detected melt pool are included into the analysis. Furthermore, slice data is used to determine the position of each image with respect to the original CAD data. This approach is shown in Figure 1 and allows to combine all data sources for further detection of salience, which shows significant deviation from stable process conditions. Labelling of the monitoring data has been implemented by means of computer tomography.

This is followed by AI training, which is carried out by means of supervised learning. To do this, multi-view learning (MVL) and automated machine learning (AutoML) will be combined. This is intended to increase the added value of the simultaneous use of a wide variety of ML methods and heterogeneous data sources in a cost-efficient manner and thus offer innovative, data driven solutions, especially in LPBF-production.