Monte Carlo simulation of the R/1 automated damage test

In this paper, a Monte Carlo computer analysis of the R/l automated damage test procedure currently in use at LLNL is presented. This study was undertaken to quantify the intrinsic sampling errors of the R/l ADT method for various types of optical materials, particularly KDP and fused silica, and to provide a recommended minimum number of test sites. A gaussian/normal distribution of 10 J/cm ampersand sup2; average fluence (μ) was used as a damage distribution model. The standard deviation (σ) of the distribution was varied to control its shape. Distributions were simulated which correspond to the damage distributions of KDP (μ/σ = 5-10) and fused silica (μ/σ - 15). A measure of the variability in test results was obtained by random sampling of these distributions and construction of the cumulative failure probability S curves. The random samplings were performed in runs of 100 tests with the number of samples (i.e. sites) per test ranging from 2 to 500. For distributions with μ/σ = 5-10, the study found an intrinsic error of 3 to 5% in the maximum deviation from the distribution average when using 100 site tests. The computations also showed substantial variation in the form of the CFD for any given test. The simulation results were compared to actual data from eight 100 site R/l automated tests on a sample from rapidly grown KDP. It was found that while each 100 site damage probability curve could be fit to a gaussian distribution reasonably well, the 800 site cumulative damage probability curve was better modeled by a lognormal distribution. The differences observed in the individual CFD curves could be accounted for by sampling errors calculated from gaussian models