SMM @AMLD 2022 in EPFL
Cyprien Hoelzl presents on data-driven railway assessment
AMLD Presenttaion
On Board Monitoring for Railway Infrastructure Condition Assessment
with Cyprien Hoelzl
14:15 - 14:35, March 29th
Track: AI & Mobility
Railway operators pursue a shift of the current condition assessment and monitoring paradigm from a reactive to a predictive mode. During the external page AI & Mobility track of the AMLD 2022 event at EPFL, we present the latest developments in the field of acceleration-based assessment, developed in the context of OMISM, a joint research project between ETHZ and SBB.
Data harnessed from on board monitoring obtained in a continual fashion from in-service trains is used to develop methods for more accurately diagnosing condition and faults of railway infrastructure, as well as providing complementary information regarding the state of the rolling stock. We demonstrate the classification of rail faults and rail components using machine learning algorithms applied on axle box accelerations. Outlier analysis is used to identify potentially faulty rail components, which are in a second step validated by experts. We discuss the challenges in establishing a cross-business feedback loop process for data validation in order to integrate expert know-how in machine learning pipelines.
The detection of faults on rail components, enabled by expert labelling, will be used to improve the rail component based flaw detection. The combination of single measurement diagnosis with temporal evolution information, will in future support the automated track inspection by increasing availability while reducing life-cycle costs.
read more external page here