Modelo para detecção de anomalias na quantidade de sódio encontrado em amostras de óleo lubrificante de locomotivas para suportar a tomada de decisão para manutenção/inspeção

Abstract

The preventive maintenance of locomotives presents itself as a crucial task for the proper functioning, mainly in relation to unforeseen and premature problems that generate equipment unavailability and can affect the entire operational planning and schedule of the fleet. There are several indications of anomalies in the equipment that, if properly monitored, can promote a performance within the expected for the equipment, including increasing its useful life. This work aimed to develop a decisionmaking prototype in relation to machines in order to serve as a complementary tool to support the specialist. The model was developed using the sodium level present in a historical series of equipment measurements. It is basically based on the distortion between the value predicted by a machine learning strategy or statistics in relation to the value obtained in the test. This distortion served as a reference for the determination of rules, together with the specialist, for decision making at different levels of alert. Using linear interpolation, regression and auto-regression models allowed a simple evaluation of the most adequate pattern for the prediction process, including indicating methods that present better efficiency