Imageamento térmico do estator de geradores de alta potência empregando sensores óticos distribuídos

Abstract

This work presents a new method of thermal imaging for stator of high power generators, using distributed optical sensors based on Raman scattering, which can identify areas with relatively high temperatures in any part of the structure. Regions with above 100 ºC may indicate wear on the insulation structure, which when not identified in the initial stage increase the risk of failures that can cause serious damage to the generator, such as short circuit in the stator windings. The method is based on a software that combines the temperature information and spatial position of the sensor with a three-dimensional stator model, for generating thermal images according to the temperature distribution in the structure. The main advantages over conventional stator instrumentation, which uses electrical transducers, can be highlight the immunity to electromagnetic interference, the use of only one sensor element, and the possibility of identifying hot spots in any part of the structure. The method performance was evaluated by experimental testing with a hydroelectric generator 200 MW. In the test, it was possible to evaluate different thermal dynamics of the stator, such as heating and cooling due to the generation or absence of active power, respectively. The generated images showed great performance to monitor the stator temperature distribution according to the power of the generator. However, a limiting factor of the proposed method is the spatial resolution of the distributed sensor, which makes the system unfeasible to detect hot regions in dimensions smaller than one meter. As a solution, signal/image reconstruction algorithms were developed based on methods to solve the inverse problem. The algorithms use a distributed sensor acquisition model and Total Variation deconvolution. The thermal images are generated through sparse representation of a hotspot dictionary, constructed through simulation with a multi-physical model of the structure. Due to the difficulties of reproducing insulation faults in the real stator structure, experimental tests were performed using a stator prototype. The results demonstrate the ability of the algorithms to reconstruct images of hot regions up to 15 cm, representing a resolution gain of six times when compared to the spatial resolution of the sensor. The application of the proposed algorithms may be a promising solution to improve the accuracy and reliability of the imaging system. Continuous monitoring of the stator temperature can be a fundamental tool of predictive maintenance, reducing the risk failures that result in serious damage to the generator, thus contributing to the operational availability and the machine performance. In addition, other benefits such as maintenance costs reduction and losses with fines for lack of energy supply may be considered. The efficient use of generators also contributes to the national electrical system, which nowadays requires the most of their energy capacity.