Understand Patterns in Volvo Truck Lifetime Repair History
|Title||Understand Patterns in Volvo Truck Lifetime Repair History|
|Summary||A project in collaboration with Volvo AB on finding good representations for data-driven description of Volvo truck's repair and maintenance history|
|Keywords||Data Mining, Knowledge Representation|
|Prerequisites|| Programming competence
Artificial Intelligence, Learning Systems and Data Mining courses
|Supervisor||Sławomir Nowaczyk, Sepideh Pashami, Fredrik Moeschlin (at Volvo)|
This project aims at analysing historical records of workshop visits, detailing repair and maintenance work, for 70.000 Volvo trucks over a period of more than 10 years. The idea is to use data mining and machine learning techniques in order to find structure and regularities in this data. From both scientific and practical point of view, it is interesting to provide classification (either as a taxonomy or an ontology) of different service events, and how are they related to vehicle's usage.
We have access to the data containing different vehicles, different faults and different components. However, this data is represented in a complex, structured form. In order to make it easier to use, it is necessary to analyse this data and extract useful information from it, both relatively simple one (e.g. what are the most common faults and/or long term driving cost factors), as well as more complex and abstract one (e.g. it is important to distinguish between repair and maintenance actions, as well as between operations affecting different subsystems of the truck).
Moreover, by analysing the frequency and cost of different repairs, together with finding relations and dependencies between various classes of them, one can expect to significantly decrease cost and increase effectiveness of maintenance and diagnosis.
In particular, we are interested in modelling the interdependencies between different faults and repairs. The goal is to create a model of truck's lifetime maintenance needs, which would at any point in time give accurate estimates of what are the most likely problems this vehicle may encounter. Possible approaches to investigate include Partially Observable Markov Decision Process for modelling component state of health/wear, or Bayesian Networks for capturing relationships between various faults.