Machine Learning and q-Weibull Applied to Reliability Analysis in Hydropower Sector

Brushes are critical components in power generation equipment. The interruptions in brush operation because they have failed in their service can cause financial losses avoidable by proper maintenance planning. Therefore, this article aims to offer a methodology for estimating the reliability parameter of brushes used in hydroelectric generators through machine learning concepts and through a statistical distribution compatible with complex phenomena. The method uses six selection patterns by graphical plotting of lifetimes and brush lengths to recognize problems of recording wear information. Then, the information is separated into three data sets according to the failure mode. The brushes reliability prediction uses an artificial neural network with assisted learning to predict a cumulative distribution function based on the operating time extracted from the equipment’s hour meter. The method compares the statistical models $q$ -Weibull, Weibull, $q$ -exponential, and exponential with the prediction function. Three measures of goodness of fit were calculated, the logarithm of likelihood, coefficient of determination, and mean squared error. Most of the values found point to an advantage in the use of artificial neural networks over the use of the $q$ -Weibull distribution. The method compares density and failure rate functions. The application of artificial neural networks in reliability analysis can have a significant impact on reducing maintenance costs, as it leads to results closer to reality. This article presents artificial neural networks for the first time compared to a distribution based on non-extensive statistical mechanics in the context of hydroelectric brushes.