C Conferentia Proceedings
CILAMCE2017-1021 COMPUTATIONAL INTELLIGENCE TECHNIQUES FOR OPTIMIZATION AND DATA MODELING

Failure Classification in Electric Switch Machines Using Symbolic Concepts and Computational Intelligence

Nielson Soares1; Leonardo Goliatt da Fonseca1; Eduardo Pestana de Aguiar1

1 Federal University of Juiz de Fora

doi:10.20906/CPS/CILAMCE2017-1021

Resumo

Switch machines are electromechanical equipment that is of great importance in a rail network, as they are used to operate railroad switches, allowing the wagons to be guided from one road to another, such as a rail junction. With the growth of the Brazilian railway sector, these machines have been more and more used, which tends to increase the probability of failures, which in turn, can represent a great cost for the companies. An early diagnosis of failures that may occur in a switch machine can mean a reduction of these costs and an increase in the productivity of the company responsible for the railway network. Usually, the amount of data that is available is big, which can make the analysis an expensive process when performed by a specialist. An alternative is the use of techniques of extraction and selection of characteristics, in order to obtain a subset of data that represents the original data. However, this practice may lead to a loss of information. Another alternative is to perform a symbolic data analysis (SDA) that allows to adequately represent the variability and uncertainty present in the raw data. This article aims to reconcile the SDA technique with supervised and unsupervised learning methods. The supervised methods used were Random Forests (RF), K-Nearest Neighbors (KNN) and Support Vector Machine (SVM). The unsupervised K-Means method was employed with the intention of identifying and separating the different faults that may occur in switch machines. The data set was provided by a Brazilian railway company and covers four possible switch machine states like (i) normal operating state, (ii) failure due to lack of lubrication, (iii) lack of regulation and (iv) Malfunction of a component. The results presented show a high accuracy regarding the classification and identification of these faults. However, the same could not be observed regarding the grouping of the defects, indicating a correlation between the faults that make proper separation of the faults difficult.

Palavras-chave: Computational Intelligence; Symbolic Data Analysis; Failure Classification; Switch Machine

Como citar

Nielson Soares; Leonardo Goliatt da Fonseca; Eduardo Pestana de Aguiar. “Failure Classification in Electric Switch Machines Using Symbolic Concepts and Computational Intelligence”. XXXVIII Ibero-Latin American Congress on Computational Methods in Engineering. CILAMCE2017. 2017. DOI: 10.20906/CPS/CILAMCE2017-1021