Identificação de falhas em trilhos de ferrovia baseado em corrente parasita e inteligência artificial

Abstract

Rail transport is widely recognized as a vital sector for a country’s economy, offering advantages over road transport, especially for long distances, as costs per volume are lower in railways and accident rates are lower. Therefore, it becomes essential to invest in improving the reliability of railway assets, such as locomotives, wagons, power supplies, signaling, and permanent way. The permanent way refers to the set of supporting infrastructures such as sleepers, rails, ballast, and other fastenings. Consequently, improving the reliability of these assets, combined with a logistics strategy, is crucial to increasing the speed and productivity of transport. In this context, this work presents an embedded solution that contributes to increasing the safety and efficiency of permanent way maintenance, impacting the increase in reliability, productivity, and improving the performance of maintenance planning, consequently reducing emergency stops and accidents with personal and economic impacts. The proposed system locates and identifies surface defects in rails through eddy currents, the Wavelet transform, and a convolutional network. The system was embedded in a railway vehicle for field tests that demonstrated the feasibility of the proposal. Field tests were also performed to enable the design and evaluation of the fault identification method, presenting an accuracy of around 98%. Comparison with other classification techniques demonstrated that the convolutional network is the most suitable for the task, presenting higher accuracy and precision than the other techniques considered.