Evolutionary fuzzy logic system for intelligent fibre optic components assembly

This paper presents a new evolutionary fuzzy logic system for use in the assembly of optical bre components. The system optimizes the light output from a bre by applying a gradientbased algorithm enhanced with momentum information. The parameters of the algorithm are adjusted on-line by a fuzzy controller according to the progress of the alignment process. The control knowledge base is automatically generated via a new evolutionary algorithm. The algorithm divides the population into three subgroups, each concerned with a different level of knowledge-base optimization, and employs a new adaptive selection routine that aims to keep the selection pressure constant throughout the learning phase. The resulting fuzzy logic controller demonstrated a robust performance with alignment times and accuracies comparing favourably against those obtained using a manually designed controller. Moreover, the evolved knowledge base was expressed in a transparent format that facilitated the understanding of the control policy.