Difference between revisions of "Automatic Generation of Descriptive Features for Predicting Vehicle Faults"

From ISLAB/CAISR
 
Line 4: Line 4:
 
|TimeFrame=Fall 2019
 
|TimeFrame=Fall 2019
 
|Prerequisites=Artificial Intelligence, Learning Systems, Data Mining
 
|Prerequisites=Artificial Intelligence, Learning Systems, Data Mining
|Supervisor=Sepideh Pashami, Sławomir Nowaczyk,
+
|Supervisor=Sepideh Pashami, Sławomir Nowaczyk, Mahmoud Rahat
 
|Level=Master
 
|Level=Master
 
|Status=Open
 
|Status=Open

Latest revision as of 20:56, 27 October 2019

Title Automatic Generation of Descriptive Features for Predicting Vehicle Faults
Summary Automatic Generation of Descriptive Features for Predicting Vehicle Faults
Keywords
TimeFrame Fall 2019
References
Prerequisites Artificial Intelligence, Learning Systems, Data Mining
Author
Supervisor Sepideh Pashami, Sławomir Nowaczyk, Mahmoud Rahat
Level Master
Status Open

Generate PDF template

Whenever a Volvo truck visits an authorised workshop, aggregated vehicle statistics (parameters) such as average vehicle speed, total fuel consumption, etc. are stored in a database called Logged Vehicle Data (LVD). This information has great potential for understanding how the vehicle is used, what is its current condition, and therefore what are the most likely faults that can occur in the near future.

However, the data is very noisy, often inaccurate and with a lot of duplication. The aggregated statistics in their raw form, at any given point in time, are not the correct input for diagnostics and predictive maintenance. Instead, they should be combined across the whole lifetime of the vehicle, to capture the main relevant usage patterns.

The goal of this project is to generate descriptive features that can be used for predicting vehicle faults. To this end students will evaluate a number of models, including regression models, random forests, and deep neural networks.

Preliminary steps for this project are as follows:

  1. create models for each LVD parameter based on a single data readout
  2. create models for each LVD parameter based on the complete vehicle history
  3. combine those models to find commonalities between different parameters
  4. extract descriptive features from the combined models