Nombre: VINICIUS MARTINELLI SPERANDIO
Tipo: MSc dissertation
Fecha de publicación: 20/07/2015
Supervisor:
Nombre |
Papel |
|---|---|
| ANSELMO FRIZERA NETO | Co-advisor * |
| MARIA JOSE PONTES | Advisor * |
Junta de examinadores:
| Nombre |
Papel |
|---|---|
| ANSELMO FRIZERA NETO | Internal Examiner * |
| EVANDRO OTTONI TEATINI SALLES | Internal Examiner * |
| MARIA JOSE PONTES | Advisor * |
| PAULO SERGIO DE BRITO ANDRÉ | External Examiner * |
Sumario: Equipment installed in areas considered classified due to the existence of likelihood of gas mixtures in an area, need to be protected in some way. In such areas there is a permanent risk of initiating an explosion caused by a short circuit or even cause high temperature, should meet international standardization. One of the most used sensor in the industry process is pressure, which can also be used to infer level, flow and temperature. In this context, the use of optical fiber with low power laser used becomes quite attractive to read pressure values. Optical fiber associated with the use of
low-power laser, can reduce the risk of explosion in hazardous areas without protection required in electrical equipment by international standardizations. To perform pressure measurements with optical fiber in this Master dissertation, a low-power laser (range of mW) was coupled through a polymer optical fiber (POF) replacing the pointer of a bourdon gauge. The light is propagated through the fiber and reflected in rough area that generates speckles patterns containing the information of turning angle. The speckle patterns is a granular image that is formed when a coherent light beam through a medium (in this case, the multimode fiber) with hundreds of propagation modes and presents variations in refractive index. The output image of multimode fiber coupled to the gauge is recorded to measure the angular displacement of the observed
area, which should be replaced by the pointer. The study and interpretation of speckle patterns generated experimentally were made by software, correlating the images with the values of the reference gauge. A morphological filter routine previously applied eliminated the noise of the image due to vibrations produced by the very means that the prototype was installed. With this, the image undergone less variation means to then be processed in two ways: first was applied to Wiener-Khintchine technique, acquiring the peaks of cross-correlation between images of video. After applying this technique was also developed a neural network with mean square error values such that enabled its practical application. The results obtained through the Wiener-Khintchine technique with the implementation of fast fourier transform were enough to show the dependence of the pressure variation, but the readings taken were still infeasible for the implementation in field. The readings had not produced sufficient resolution, as were 18 degrees with a variation of 3:2%. After the development of the neural network, using feed-forward backpropagation algorithm multilayer, it was possible to achieve resolution for field application, with fewer errors than 3 degrees.
