Técnica de diagnóstico de falhas em motores a combustão interna utilizando aprendizagem de máquina

Abstract

The modern industrial scenario denotes the increase of the industrial competitiveness, due to the complexity of machines and equipment - products of high demand -, the costs of industrial installations, allied to the concern with aspects of industrial safety and environment. This trend induces large global industries to increasingly invest in devices, technologies and tools designed to previse and predict failures due to non-conformities and breakdowns in machines, equipment and industrial facilities. In this context, the field of action dealing with predictive maintenance, forecast analysis and fault diagnosis has gained prominence place, as well as several investments in research and development, mainly with policies aimed at the design of the industry 4.0. With the industry 4.0 approach, it is possible to analyze dynamic mechanical components and real-time responses without the need for stationary equipment, which is directly related to the reduction of costs and production time. The aim of this thesis is to present a new methodology for the detection and monitoring of tribological faults in internal combustion engines through the use of unsupervised learning methods and Big Data using signal processing techniques allied with to algorithms of Artificial Neural Networks (RNA) and analysis of clusters, creating an intelligent system capable of identifying fault patterns from the conditions of these and the variation of mechanical load in Otto cycle internal combustion engines.