In English

Vehicle data analysis using cloud-based stream processing

Morhaf Alaraj ; Philip Bogdanffy
Göteborg : Chalmers tekniska högskola, 2018. 68 s.
[Examensarbete på avancerad nivå]

In the automotive industry, maintenance planning can be seen as a set of strategies that aims to increase the uptime of vehicles. Unexpected events such as if a vehicle suddenly stops due to internal component failures or due to a flat tire are in the context of maintenance planning referred to as unplanned stops. For transport vehicles, unplanned stops are critical and will in most cases lead to late deliveries and possibly damaged goods and can in the worst case imply dangerous consequences for the driver of the vehicle. As data become more and more available and the connectivity in vehicles get better, more advanced techniques can be applied when it comes to maintenance planning. The demand of creating self-learning or automated systems that can predict unplanned stops is increasing due to the big amounts of data that are generated by today’s systems. Manually analyzing vehicle data is becoming an unsustainable approach and it is hard for humans to keep up with the digitised systems due to increased complexity and big data amounts in the vehicles. In this thesis, we propose a concept that enables stream processing on vehicle data on a remote machine. The implementation of the proposed concept is built using state-of-the-art streaming components, namely Apache Spark Streaming and Apache Kafka. This thesis focuses more on the design of the system architecture and the components of the concept. The results show that it is possible to create machine learning models that continuously evolves and learns from data streams. Implementations of the proposed concept can for example be used to detect anomalies in vehicle components remotely without re-configuring any software inside the vehicles. The machine learning models that were trained with a Volvo data set did not deliver the desired prediction accuracy for the area of maintenance planning. Future work in this area would require further research in which online machine learning algorithms that best fits this vehicle data and also how features should be chosen to be able to predict anomalies in vehicle data.

Nyckelord: Anomaly detection, Stream processing, Machine learning, Maintenance planning



Publikationen registrerades 2018-09-18. Den ändrades senast 2018-09-18

CPL ID: 255952

Detta är en tjänst från Chalmers bibliotek