SUMMARY

With the significant increase in the number of physical objects connected to the Internet, various data are generated from all over the world. In manufacturing era especially, many parameters have impacts on speed, quality, and performance of the production. Finding the relationships between these parameters and gaining insights from them not only can result in an increase in performance and efficiency, it can also help us prevent unexpected behaviors in the product lifecycle. Finding these correlations could be challenging and time-consuming, however, with the traditional methods that are not optimized for the Internet of Things (IoT) infrastructure. In the recent decades, machine learning is being used as a primary method of analyzing the data that involves multiple parameters and unknown behaviors. However, there is a limited work performed in the area of machine learning for IoT to allow automated analytics for the embedded systems. This thesis presents an efficient and scalable architecture for sense-making, analysis, and prediction using machine-learning algorithms for IoT systems. The proposed framework focuses on the architectural elements from embedded system data models on the edge level to the fog and cloud-level elements, which results in the production of centralized decision-making for IoT systems. An application of this architecture as a case study will be presented that predicts the quality of the produced parts using FDM 3D printers with considering the history of the humidity and temperature of the PLA filaments.