Development of New Algorithms for the Data Analysis of MEMS Production Processes

The manufacturing industry is currently in the middle of a data-driven revolution, which requires the ability to manage large amounts of data. Bosch Sensortec, a leading global supplier of MEMS sensors for consumer electronics, exploits big data technologies to store massive amounts of data from the whole manufacturing process. Despite the employed cutting-edge technologies, the process of storing, retrieving, preprocessing and analyzing the manufacturing data can be challenging and time consuming. The solutions for the data analysis are often commercial and require extensive trainings and/or programming language knowledge. This thesis aims at increasing the overall efficiency of the manufacturing data analysis using two approaches: developing an internal tool for data analysis and investigating the relevance of the manufacturing data stored in the big data cluster. In the first approach, a full-stack reactive web-based dashboard with access to the big data cluster was developed. The dashboard design was focused on being tailored to the ongoing reliability test (ORT) engineers needs, providing abstraction to collect, preprocess, analyze, export data and generate reports. Addressing maintainability, malleability and low cost, the front and back-end of the dashboard was completely developed using opensource software and Python language. The second approach focused on the data currently being used for identifying a single failure mode of a sensor. According to field experts, 90 features were relevant for assessing the failure mode. Exploiting data mining and machine learning techniques, specifically data reduction, from the initial 90 features, 44 were considered the most relevant, suggesting a potential of data reduction of approximately 50\%. This result could lead to the increase of the overall efficiency of the manufacturing data analysis, data storing, and/or reducing the scope of manufacturing tests. This thesis recommends to gather feedback from the ORT engineers regarding the dashboard, and to further investigate through data analyses, and field experts opinion, the true necessity to collect the potentially not relevant features.