Inspection of defects in optical fibers based on back-propagation neural networks

Abstract

As the application of optical fiber is boasting, higher quality optical fibers are required in some applied fields. Some pinhole defects such as air bubbles, caused during the manufacture of optical fiber, may have severe (or fatal) impacts on precise instruments because of affecting optical fiber's toughness and signal propagating along the fiber. Therefore, it is very important to detecting the defects in optical fibers using an effective, non-touched, on-line and fast method. In this paper, a novel defect-detection method based on back-propagation (BP) neural network is proposed. Methods on preprocessing and extracting feature from the high-dimension source data obtained by the optical device are investigated. A setting threshold operation and a cosine transformation method are proposed being used to filter the background noise from and reduce dimension of the source data respectively. The experimental results exhibit the processing methods are very effective on doing them. Resilient BP method is considered as network training algorithm because of its fast convergence. The experimental results given for training sets and testing sets denote that the method with BP neural network is competent for optical fiber detection.