DESARROLLO DE LOS ELEMENTOS DEL SISTEMA DE DIAGNÓSTICO DE UNA TURBINA DE GAS

Abstract

In recent years the use of gas turbines has grown tremendously, because they are an important part in national economic development. They are considered like complex engineering systems due to continual technological changes. Thus, gas turbine can be affected; this can be seen in the increase of accidents in industry. Mechanical faults occupy a high percentage of the causes of these accidents. To ensure a necessary reliability of gas turbines, many techniques and diagnostics tools have been implemented. They ensure control of the condition of gas turbines. A wide variety of condition monitoring techniques of machines has been improved in the recent years such as: vibration monitoring, oil debris, visual inspection, monitoring noise, and monitoring environmental pollution. These help better determine the machinery operation condition in industry. These also are used for gas turbine condition monitoring. Automated systems of gas turbine monitoring are based on parameters measurement and recording. For this operation such systems need not shutdown and disassembly of the engine. Therefore these systems work in real time and provide a diagnostic analysis online. Into this monitoring system there is a specific technique for gas turbines, called Gas Path Analysis (GPA). The gas path analysis has been chosen like a representative approach for gas turbine diagnosis. Gas turbine condition monitoring, based on parameters measurement of engine gas path, includes stages of problem detection, fault identification and prognosis. It also requires one preliminary stage of computing deviations between measurements and a baseline. In the present work we focus mainly on development of gas turbine diagnostic system elements, for two stages of monitoring, the preliminary stage of computing deviations and the second stage of faults identification. For both cases an important tool of Artificial Neuronal Networks (ANN) is widely used in recent years. The ANN`s have proved to be very complex tool and very useful for gas turbine diagnosis, considering the growth that has been suffer since they appeared. Although Matlab have several functions to perform different neuronal network, in this thesis we have developed our own algorithms. Then one of them was realized in Fortran After. As a result, we have executable program modules ready to include in a real monitoring system.