Using genetic algorithms to estimate confidence intervals for missing spatial data

Gas turbine blades, which come in many shapes and sizes, must meet strict engineering specifications. The current manual blade measurement system is slow and labor intensive. As part of the development of an optical measurement system, an approach for characterizing missing data was required. A novel technique for conditional spatial simulation using genetic algorithms (GAs) was developed. The problem is encoded using the "random key genetic algorithm" (RKGA) approach. The RKGA allows the use of a sampling distribution for missing measurements that can accommodate values uncharacteristic of the area surrounding the missing data, while still allowing realizations of the missing data with reasonable directional semivariance characteristics to be developed. A unique optimization approach was used, consisting of a crossover-only GA, followed by a hill-climbing phase. Each phase addresses different parts of the problem (the low and high special frequencies, respectively). This spatial simulation technique can be used to characterize regions of missing data in regularly sampled measurements. The proposed technique is much faster than simulated annealing, the current state of the art in spatial simulation. An application of this technique to determining confidence intervals for missing data in optical measurements of gas turbines is described