Design of a gearbox test-rig and a novel approach to machine monitoring

Abstract

Consulta en la biblioteca ETSI INdustriales (8852)

[EN] This thesis presents a novel methodology for condition monitoring of machinery: the Evolving Clustering Method (ECM). This is a multivariate statistical method able to predict machine performance from a mathematical model created during a training period. After that, deviations between the actual and the expected valued can be detected, in order to diagnose a problem in the machine. The main possible benefits of this method are the possibility to analyse any kind of performance data, and the capacity to detect failures even in machines working under variable load conditions. This is still not a widely used method for condition monitoring, and this thesis shows an application example of the method to real machinery performance data. In order to prove the validity of the method for fault detection, it was applied to performance data from two different machines, one of which had a failure affecting one of the measured variables. This data comes from sensors measuring different parameters including temperature, pressure, electric power... where the samples were taken every 10 minutes during more than one year. To achieve satisfactory results, it was necessary an exhaustive pre-treatment of the available data, but finally it was possible to isolate and identify the failure by the application of ECM and the related fault detection tools. Furthermore, a preliminary design of a planetary gearbox for condition monitoring is proposed, which will be useful in the future to carry out test in the laboratory. Different damaged parts will be mounted on it, in order to improve ECM and other condition monitoring techniques, and find patterns of behaviour that can be helpful in the early detection of defects in this specific type of gearboxes.