LLNL, Livermore, California, USA
To optimize laser performance and minimize operating costs for high energy laser shots it is necessary to locally shadow, or block, flaws from laser light exposure in the beamline optics. Blockers are important for temporarily shadowing a flaw on an optic until the optic can be removed and repaired. To meet this need, a combination of image analysis and machine learning techniques have been developed to accurately define the list of locations where blockers should be applied. The image analysis methods extract and measure evidence of flaw candidates and their correlated downstream hot spots and this information is passed to machine learning algorithms which rank the probability that candidates are flaws that require blocking. Preliminary results indicate this method will increase the percentage of true positives from less than 20% to about 90%, while significantly reducing recall – the total number of candidates brought forward for review.
