Abstract
Glass fiber forming is a complicated process in which many factors could affect the quality of fibers. The forming machine has many fiber-forming tubes that are close to each other and arranged in several layers. The closeness results in inadequate lighting and unwanted video signals. An anti-causal zero-phase filter was employed to remove noise with insignificant pixel location shift or distortion. In addition to the noise, the unwanted video signals constantly moving from one place to another also presented a challenge in image analysis. These signals were identified by a trained neural network that classified patterns. The unwanted signal identification through instant pattern classification made online inspection possible. During the fiber drawing process, the diameters of glass forming tubes and the profiles of glass melting cones were closely monitored and measured online in order to control the final fiber diameter. The accurate diameter measurements were accomplished by the noise removal along with a subpixel-resolution based edge detection technique. The results thus obtained for noise removal and unwanted video signals identification were quite good. The fiber diameter measurements were performed online, and the entire inspection process was automated with the aid of a programmable logic controller.
