TU Darmstadt / ULB / TUbiblio

On the Challenges of Real World Data in Predictive Maintenance Scenarios: A Railway Application

Kauschke, Sebastian and Janssen, Frederik and Schweizer, Immanuel (2015):
On the Challenges of Real World Data in Predictive Maintenance Scenarios: A Railway Application.
In: KDML: Workshop on Knowledge Discovery, Data Mining and Machine Learning, Trier, [Conference or Workshop Item]

Abstract

Predictive maintenance is a challenging task, which aims at forecasting failure of a machine or one of its components. It allows companies to utilize just-in-time maintenance procedures instead of corrective or fixed-schedule ones. In order to achieve this goal, a complex and potentially error-prone process has to be completed successfully. Based on a real-world failure prediction example originated in the railway domain, we discuss a summary of the required processing steps in order to create a sound prediction process. Starting with the initial data acquisition and data fusion of three heterogeneous sources, the train diagnostic data, the workshop records and the failure report data, we identify and elaborate on the difficulties of finding a valid ground truth for the prediction of a compressor failure, caused by the integration of manually entered and potentially erroneous data. In further steps we point out the challenges of processing event-based diagnostic data to create useful features in order to train a classifier for the prediction task. Finally, we give an outlook on the tasks we yet have to accomplish and summarize the work we have done.

Item Type: Conference or Workshop Item
Erschienen: 2015
Creators: Kauschke, Sebastian and Janssen, Frederik and Schweizer, Immanuel
Title: On the Challenges of Real World Data in Predictive Maintenance Scenarios: A Railway Application
Language: English
Abstract:

Predictive maintenance is a challenging task, which aims at forecasting failure of a machine or one of its components. It allows companies to utilize just-in-time maintenance procedures instead of corrective or fixed-schedule ones. In order to achieve this goal, a complex and potentially error-prone process has to be completed successfully. Based on a real-world failure prediction example originated in the railway domain, we discuss a summary of the required processing steps in order to create a sound prediction process. Starting with the initial data acquisition and data fusion of three heterogeneous sources, the train diagnostic data, the workshop records and the failure report data, we identify and elaborate on the difficulties of finding a valid ground truth for the prediction of a compressor failure, caused by the integration of manually entered and potentially erroneous data. In further steps we point out the challenges of processing event-based diagnostic data to create useful features in order to train a classifier for the prediction task. Finally, we give an outlook on the tasks we yet have to accomplish and summarize the work we have done.

Journal or Publication Title: Proceedings of the LWA 2015 Workshops: KDML, FGWM, IR, and FGDB
Divisions: 20 Department of Computer Science
20 Department of Computer Science > Knowl­edge En­gi­neer­ing
20 Department of Computer Science > Telecooperation
Event Title: KDML: Workshop on Knowledge Discovery, Data Mining and Machine Learning
Event Location: Trier
Date Deposited: 21 Oct 2015 13:38
Export:
Suche nach Titel in: TUfind oder in Google
Send an inquiry Send an inquiry

Options (only for editors)

View Item View Item