Abstract
Abstract. An investigation into the use of machine learning for fault detection and classification of Neutral Section faults on a railway system. Two methods are addressed, K-means for condition clustering and Support Vector Machines for condition classification. The fault monitoring system proposed uses data col-lection via wireless sensors as opposed to typical labor intensive approaches such as manual inspections and observation via camera-fitted trolleys which are proven to be inefficient, expensive and time consuming practices. This paper in-vestigates the feasibility of deployment of the smart fault detection system on the overhead wires. The system applies machine learning on the remotely collected data for categorization of events on the neutral section in order to determine anomalous behavior. The system comprises accelerometers and infrared-based temperature sensing to record vibration and temperature information relating to the interaction between the pantograph and contact wire. The system is intended to monitor and detect any events outside the normal limits thereby allowing for triggering of alarms in real-time. The measured data from onsite is conveyed to the ThingSpeak for cloud computation thereby providing notifications in real-time allowing the operator to visualize, analyze and act accordingly. In this paper, the built prototype is tested and preliminary analyses of the applied techniques on the collected data is carried out. Experimental results show that these methods are feasible on the railway and can be deployed to detect anomaly and classify events accordingly.