Multimodal, data-driven control for quality improvement in manufacturing processes

The manufacturing industry is faced with ever more stringent quality requirements caused by a trend towards stricter safety standards and higher customer requirements. At the same time, guaranteeing quality is more difficult by the increasing complexity of the produced products and the limits of manual quality inspection. Due to the high processing speed and repeatability, these manual checks are increasingly being replaced by computer vision for online quality control, resulting in a higher efficiency. Main quality traits are related to color, size, shape and the absence of defects. Current applications of computer vision for quality control purposes are mainly targeted to detection, for example to detect defects, the presence or absence of a particular component or characteristic, or the correct assembly. Products with identified issues are then either manually fixed, brought back into the manufacturing process in an earlier processing step, or simply discarded. At the same time, due to the increasing digitalization of manufacturing processes, an increasing amount of additional information is available, including the parameter settings of the machines as well as inline sensor measurements on the process. The objective of this thesis is to investigate if the vision-based quality inspection can be complemented with these additional information sources. Studying the correlation between this multimodal set of in-process sensor data, machine settings and image-based analysis should allow to gain better insights in the root causes for particular quality defects. Subsequently, research will be performed on how to incorporate these insights into a novel methodology for multimodal, data-driven control towards an improved parameterization of particular processing steps, in turn leading to a higher product quality. This will contribute to a more efficient, cost-effective and sustainable manufacturing practice, due to an improved operational performance and a reduction of the amount of scrap material resulting from quality defects.