New perspectives for the intelligent rolling stock classification in railways: an artificial neural networks-based approach

Abstract

In railway operations, several factors must be analyzed, such as operation cost, maintenance stops, failures, and others. One of these important topics is the analysis of the Hot Box and Hot Wheel due to the failure of these components. It can compromise the entire operation, resulting in serious accidents, such as train derailments. Thus, the use of a method that is able to classify a failure is essential for accident prevention. The innovative use of Multilayer Perceptron combined with Set-Membership for the Hot Box and Hot Wheel binary classification problem enhances failure prediction and contributes to accident prevention. Unlike the reported models in the literature, Set-Membership Multilayer Perceptron excels in learning from the non-linear and intricate patterns of this dataset. In addition, as aforementioned, its ability to update representations and patterns with new data avoids frequent retraining, ensuring a more efficient and adaptable solution. Besides that, the proposed work presents a better performance in terms of Accuracy and other metrics compared to other literature works. To validate the performance, we compare twelve algorithms applied in eight datasets, seven of which are benchmarks, and one is composed of Hot Box and Hot Wheel problems.