Detection and Clustering of Neutral Section Faults Using Machine Learning Techniques for SMART Railways

Abstract

Fault detection and diagnosis plays an important role particularly in railways were abnormal events are detected and a detailed root causes analysis is performed to prevent similar occurrence. The current method used to detect immediate and long-term faults is through foot inspections and inspection trolleys fitted with cameras proving to be inefficient and time consuming when analyzing the data. This paper examines the smart fault detection system on the overhead wires by applying machine learning techniques for accurate assessment of the neutral section before and after failure thereby grouping the events into fault bins. Modern computational intelligence has enabled the fault diagnostic and fault detection to be accurate from the data generated and sensors. The interaction between the pantograph and contact wire will be monitored using accelerometers and non-contact infrared thermometer sensors were should there be a deviation from the normal signal spectrum it will be detected. The measured data from onsite will be conveyed to ThingSpeak for cloud computation thereby providing notifications in real-time which allows the end user to visualize, analyze and act on data online. A prototype has been built and tested which shows that the system works reasonably with data collected from sensors.