Condition monitoring in the optical fibre drawing process through the use of neural networks
It is well understood that optical fibre quality can be affected by the conditions present during fibre drawing. fibre tension , drawing speed, furnace temperature, the atmosphere within the ddrawing zone, viscosity of the glass, pre-form neck down shape and vibration in the drawn fibre are all factors which may affect the drawing process, and consequently the properties of the optical fibre. Most of these factors can be affected by the behaviour of the drawing furnace. Therefor monitoring furnace condition and predicting furnace decay are very important to the optical fibre manufacturing process. However, furnace conditions are very complicated and cannot be described by traditional mathematical models. In order to implement condition monitoring more effectively, the effects of process parameters and vibrations on the porperties and geometry of germanium-doped silica-core single-mode fibre, the subject of the current study were investigated by experiments and literature survey. These results can be used to determine the optimum processing conditions and select feature parameters for a condition monitoring system. Sources of vibration were analysed in order to provide information on which subsequent work to minimise vibrations on the drawn fibre could be based. The feature parameters which relate to furnace decay were extracted from the selected drawing parameters and are presented in this thesis. A new non-contact tension measurement system was devised. Different kinds of neural networks and their application in the furnace condition monitoring systems were investigated and the results are reported. A neural network software with fast training speed and a data exchange interface was developed to meet the needs of this furnace monitoring system. This thesis oresents a novel on-line condition monitoring system for drawing furnaces in the optical fibre drawing process. This system utilises the feature parameters extracted from drawing parameters and a neural network as the learning and decision making component. It can monitor the performance of the drawing process and give a pre-warning when furnace decay occurs or drawing parameters exceed the allowed working range. Hence, fibre properties can be enhanced, the production yield can be improved and machine utilities can be increased. This system has been used on a production optical fibre drawing tower at Pirelli Cables Limited and a high success rate for recognising furnace condition was achieved.