High performance implementation of probabilistic damage tolerance analysis

Periodic inspection and repair are necessary to maintain safety of aging aircraft structures. Because of the costs associated with inspections and repairs, airworthiness certification requirements must be carefully considered to mitigate hazard conditions such as metal fatigue while avoiding excessively frequent maintenance actions. Knowing when and where detectable fatigue cracks are likely to occur provides the necessary data to set optimal inspection schedules and support certification requirements. Over the past 50 years, fatigue crack growth models and computer programs capable of simulating crack growth under random cyclical loading were developed, making it possible to apply Damage Tolerance analysis, a method of predicting the number of cycles a component with a small initial crack in a critical location can endure before fracturing, to aircraft structures. When combined with probabilistic methods accounting for variations in material properties, dimensional tolerances, and applied load levels, Probabilistic Damage Tolerance Analysis (PDTA) computes the probability of failure (POF), a useful risk metric for planning inspection schedules and fleet maintenance actions. The Federal Aviation Administration (FAA) sponsored development of SMART/DT at The University of Texas at San Antonio. SMART/DT is a comprehensive PDTA software which interfaces with NASGRO, AFGROW, and FASTRAN to run multiple crack growth analyses incorporating variations in material properties and component geometry to calculate the POF over time, and models the effects of multiple inspection and repair actions. In order to process the millions of computationally expensive fatigue crack growth evaluations required to calculate the low probability levels necessary for PDTA, SMART/DT makes use of an adaptive kriging surrogate model which was further developed in this work. In this research, high performance computing methods are applied to develop a parallelized version of SMART/DT to take advantage multiple core processors available in general purpose computing systems today. Complementary efforts are also undertaken to improve the serial execution performance, and optimize performance of the parallel adaptive kriging algorithm. With the combination of serial performance optimization and parallel execution, speedup of nearly four orders of magnitude is achieved in the adaptive kriging algorithm. Several PDTA analyses are presented demonstrating the performance and accuracy of the parallel PDTA software.