Abstract:
This paper introduces a framework for Multidisciplinary Problem Solving Environments (MPSEs), which are designed to combine various specialized Problem Solving Environments (PSEs) to solve complex problems with numerous interrelated components, such as the design of a gas turbine engine. The core of this framework is a net-centric, agent-based architecture that assumes a computational grid as its hardware foundation. This architecture allows for the simulation of multi-physics applications (MPAs) by treating the physical system as a network of collaborating numerical objects, each with its own solver agent. The paper describes a prototype MPSE called GasTurbnLab for simulating gas turbine engines, which utilizes existing legacy software like ALE-3D, KIVA-3V, and PELLPACK. The framework's design enables the use of these computational tools by encapsulating them within stationary agents or servers, allowing for flexible and robust integration with the agent-based system. A key method for solving these complex problems is domain decomposition, where a large problem is split into smaller sub-problems, each handled by a dedicated solver agent, with a mediator agent managing the interfaces between them to ensure a converged solution. This approach is demonstrated through prototype simulations of a gas turbine engine, showing how different legacy codes can be successfully coupled to model parts like the compressor and combustor. The framework's layered architecture, including the graphical user interface, middleware, and computational layers, facilitates the integration of various tools and technologies, such as the IRIS Explorer system and the Grasshopper agent platform, to enable rapid prototyping and distributed high-performance computing.
