This thesis was written as a result of the cooperation with the Budapest branch of Continental
Automotive Systems Ltd., one of the leading automotive manufacturing companies in the world.
The plant produces a large volume of measurement data during the production of various
automotive products on dozens of production lines. Currently there exists only a limited scope of
predictive data analysis solutions that operate on the production data. In order to better understand
possible connections between production quality metrics, the Vehicle Dynamics (VED) business
unit proposed several case studies about currently existing production quality issues observed
during the ER100 electronic brake system product family.
In this thesis, I first analyze the existing traceability workflow based on the central
Manufacturing Execution System (MES) and also investigate the data analysis capabilities of the
current standard Business Intelligence (BI) and reporting tool. Afterwards, I design and implement a
complete machine learning data pipeline starting from the integration of heterogeneous data sources
using SQL Server Integration Services (SSIS) to provide input data for several different machine
learning models that can eventually provide predictions after successful training.
After the theoretical overview of the used software tools and machine learning techniques, two
real-life case studies provided by the VED engineering team are analyzed by deploying multiple
machine learning models using popular machine learning and deep learning frameworks, such as
Scikit-learn and TensorFlow. In order for an effective understanding of the underlying data during
the data exploration phase, I created an interactive web dashboard using Bokeh. In addition to the
on-premise implementation, I carried out the training and deployment of various prediction models
in the cloud, as well using Amazon SageMaker, which is a recent addition to Amazon Web Services
(AWS) machine learning product portfolio.
Finally, I summarize the data analysis results of the case studies along with the potential future
work to further improve results and facilitate a production-ready rollout