-
Leonardo Pessoa Freitas e Silva
Programa de Pós-Graduação em Instrumentação, Controle e Automação de Processos de Mineração, Universidade Federal de Ouro Preto (UFOP) e Instituto Tecnológico Vale (ITV), Ouro Preto, MG
-
Glauco Ferreira Gazel Yared
Departamento de Eng. Elétrica da Universidade Federal de Ouro Preto (UFOP), Rua 36, 115 - Loanda, 35931-008, João Monlevade, MG
-
Agnaldo José da Rocha Reis
Depto de Eng. de Controle e Automação da Universidade Federal de Ouro Preto (UFOP) - Campus Morro do Cruzeiro s/n, 35400-000, Ouro Preto, MG
Keywords:
Machine learning, steel sleepers, defect detection, pattern analysis, railway safety
Abstract
The rail system is essential for commercial activities and transport of people in several countries, playing an important role in improving economic indicators. In order to ensure the reliability and safety of rail transport, it is becoming increasingly important to monitor the conditions of the railway and to execute planned maintenance. The defect in the sleepers can cause an overload on adjacent sleepers, accelerating the structure fatigue of such components, contributing to the occurrence of new defects and finally affecting the track gauge. In this context, one proposes a new method for detecting defects in steel sleepers from the permanent way geometric signals, based on signal processing and machine learning. Three classifiers with different learning characteristics were trained: Artificial Neural Networks (ANN), Suport Vertor Machines (SVM) and AdaBoost. In addition, a multiple classifier system was implemented to improve system accuracy.