Efficient computation of dynamic response of large flexible spacecraft

A new approach for efficient, accurate and robust computation of dynamic response of extremely large, detailed structural models was evaluated for application to large flexible space structures. Automated Multi-Level Substructuring (AMLS) was developed to meet similar challenges in the automotive industry and has become an automotive industry standard. The current aerospace industry standard, MSC NASTRAN using a LANCZOS eigensolver algorithm, requires more time and resources for identical response computations with large automotive models. This paper compares AMLS computation of response for large models of ultra-lightweight aerospace structures to response computed using the LANCZOS algorithm implemented in NASTRAN. Model sizes range from several thousand to hundreds of thousands of degrees-of-freedom. Further comparisons include computation using AMLS on a single-processor workstation and a single processor of a supercomputer. Unlike LANCZOS, AMLS is amenable to parallelization, so performance is also evaluated on a multi-processor supercomputer. Timing studies were conducted for the evaluations. Results demonstrate that AMLS requires less computational time and effort than LANCZOS for these models. Parallelization using up to eight processors significantly reduced the time to compute the response. Specific studies included thin-film reflector dynamic response with variation of curvature and inflatable torus support structures with self-stiffening designs. An additional aspect of this effort was development of model generators to create large models.