Swarm intelligence algorithm combined with type-2 fuzzy logic system for the classification of trends in hot boxes and hot wheels

Abstract

Hot box and hot wheel problems on trains are significant threats in any rail operation because they increase fatigue and wear processes, resulting in failures, costly train stoppages and even derailments. For this reason, there are many hot box and hot wheel detectors to monitor those components distributed along the railway, and when they detect overheated components, a warning is given to alert to the impending failure. However, some situations such as solar incidence, misalignment, or sensor defect can lead to improper warnings, that is, warnings that actually do not exhibit the problem. Therefore, this work develops a new model for the classification of proper and improper warnings in hot box and hot wheel problems, by the analysis of the temperatures measured in the wheels and bearings of the train when the warning was given. To this end, the discussion focuses on the use of a metaheuristic optimization algorithm into the training phase of an upper and lower singleton type-2 fuzzy logic system, in order to increase the convergence speed and the accuracy of this procedure. Thus, the performance of the new model will be based on the real data set composed of train wheel and bearing temperatures, provided by a Brazilian railway transportation company, covering several situations of proper and improper warnings. Additionally, this work presents performance analyses based on well-known data sets provided by Knowledge Extraction based on Evolutionary Learning Repository, aiming to evaluate the proposal. For comparison purposes, the results of the proposed classifier will be compared to well-know deep-learning classifiers as well as the original fuzzy model it is based on. The performance analysis of the new model is discussed in terms of classification ratio and convergence speed. The reported results show that the proposal achieved higher classification ratio and proved to be an appropriate option for the presented problem.