Rodin 9422707 Design of composite materials to have desired properties is a task of great engineering significance. Presently this is usually accomplished through trial and error laboratory synthesis and ad hoc analysis with the result that non-optimal design is commonplace. The dramatic increase in computational power available for mathematical modeling and simulation raises the possibility that computationally-based design can play a significant and complementary role to laboratory development of compmsite materials, as has been the case in other engineering design problems. The computational problem is to determine the overall properties ( not necessarily linear elastic) of the composite material from the constituent properties. The goal of the proposed research is to create a toolbox of computational components with which engineers can design and analyze composite materials directly form the microstructural level. The approach taken in this proposal is the adapt the emerging methods for high performance parallel computation, being developed for computational modeling in other domains, to micromechanical analysis of composite materials. The application of large-scale computation to this calls of problems has been impeded by the fact that neither available computation power nor available computational methods were adequate for realistic micromechanical models. The methods to be adapted in this project are: (i) parallel h-P ada tive fineite element and boundary element methods for computational fluide dynamics and acoustics, and (ii) parallel fast multipole methods emerging form many-particle dynamics. The most significant properties of these methods are: Superalgebraic convergence rates, permitting effective simulations with models requiring a minimum computational effort to achieve acceptable accuracy's. Amenability to ca common infrastructure for efficient parallel implementation. Development of an initial version of this infrastructure under pervious research is one cr itical enabling factor for this proposal. Natural applicability to micromechanical modeling of composite materials since they perform similar roles in their current domains. The methods are complementary in their domains of application to composite materials. The development of the methods and of the toolbox will be empirical and example driven. The algorithms and methods will be evaluated by application to known composite materials supplied by industrial collaborators. the toolbox will consist of libraries of object classes: (i) infrastructure for distributed dynamic arrays, (ii) h-P adaptive finite element modules, (iii) h-p adaptive boundary element modules, (iv) parallel linear algebra modules, (v) fast multipole modules, and (vi) and interface which facilitates composition of applications form the libraries of modules. The toolbox is to be initially built around the PHLEX computational kernel which provides object-oriented data management of hp-meshes and adaptive methods. This is truly interdisciplinary research. It is no to be expected that the adaptation of the parallel adaptive methods will be trivial. The interactions in composite materials are quite complex and substantial redesign and extensions of the methods to be effective for composite materials is anticipated. The Research team includes expertise in micromechanical modeling of composites (Rodin), H-p adaptive methods (Oden), high performance parallel computing (van de Geijn and Browne), data structures (Browne), and application development environment (Browne). It is also very high leverage results since research accomplishments form other fields, with decades of development effort behind them, are being exploited in this project.
