<title>Statistical analysis methods for strength distributions in optical fibers</title>

Abstract

In this paper we present the results of computational studies on a set of experimental data concerning the strength of commercially available silica optical fibres. We compare various statistical methods for characterising the distribution of fibre strength and assess their validity by comparing random sampling from a set of experimental data with results generated using a Monte Carlo computer routine. The variation of calculated Weibull parameters with sample size is compared for experimental data and for computer generated data. We show that for two-point bending tests, and for greater than 50 samples, the maximum likelihood method is the most appropriate for calculating Weibull parameters. We also show that the experimental data are in very good agreement with our Monte Carlo simulation. We present data obtained from both bending and tensile experiments and argue that while the Weibull distribution can be shown to describe the statistical variation of strength in bending very well, it is not so convincing for tensile results. In addition it is clear from our analysis that the distribution of strength is strongly affected by variations in fibre diameter and comparison of our results from bending and tension show that variations in the diameter may be sufficient to account for the observed deviations in fibre strength.