Scientific Data and Knowledge Management in Aerospace Engineering

Kurzfassung

A key technology in aerospace research and development is high-resolution parallel simulation on supercomputers using sophisticated numerical algorithms and optimized codes. Examples for current large-scale simulations are: • The complete simulation of all flow phenomena throughout the entire flight envelope including the multidisciplinary simulation of all involved disciplines of space and aerospace vehicles. • The multidisciplinary optimization of the overall aircraft design as well as the design of major parts, such as the turbine engines. The goals are to analyze the aerodynamic and aero elastic behaviour of the aircraft and its parts and the numerical prediction of aircraft performance and handling qualities prior to the first flight. Similar goals are also apply to other industrial sectors, for example the ship building or the automobile industry. For these kinds of complex simulations, two distinct technologies are need. First, highly sophisticated and optimized numerical simulation codes for each involved discipline (for example, codes for computational fluid dynamics, structural analysis, or flight mechanics). Secondly, an efficient simulation infrastructure and well-designed supporting tools. This talk focuses on the infrastructure and the supporting tools, especially for managing both the data resulting in large-scale simulation and the necessary knowledge for conducting such complex simulation tasks. Examples of recent developments in the fields of data and knowledge management to support e-Science in aerospace research are presented. Grid Computing is an important technology for utilizing large distributed computing and storage resources and for collaboratively working between different research institutions and industry. The current developments of the German national Grid initiative D-Grid are presented, especially result from the D-Grid project AeroGrid. The AeroGrid project aims at providing an efficient Grid based working environment for the aerospace research community. The AeroGrid environment allows virtual organizations to cooperate in research and development projects, to always use up-to-date program versions, data, and compute resources across all locations, and to document and trace the detailed history of a computational process that leads to a certain result (“Provenance”). AeroGrid is used to use distributed compute resources by industrial and academic users, to collaborate during the design of new engine components, and to further develop CFD codes (such as the DLR code TRACE) cooperatively. The main user interface for data management in AeroGrid is DataFinder. The DataFinder is an easy-to-use data management client for Grids that primarily targets the management of scientific technical data. Data can be attached with individual meta data that is based on a free-definable data model to achieve data structuring and ordering. Moreover, the system is able to handle large amounts of data and can be easily integrated in existing working environments. The system is based upon a client-server-architecture and uses open, stable standards. One of its main features is the extensibility using Python scripts. The server side consists of the meta data server that stores the meta data as well as the system configuration. The server is accessed with the standardized protocol WebDAV by the client side. Data can be stored on the same server as the meta data or separated from it. The concept of separated meta data and data storage allows the flexible usage of heterogeneous storage resources (for example, FTP, GridFTP, Amazon S3, Storage Resource Broker, operating system data interfaces, or WebDAV). The DataFinder can be customized to the specific application. It can also be integrated into existing working environments, which is an important prerequisite for a successful deployment. The DataFinder can be used as simple client software for just storing data in Grids, but even more important is the ability to structure existing data and combine it with meaningful meta data. The customizing steps usually include an extensive requirements analysis for the different engineering task, which leads to a logical view to all involved data files. This allows representing connections between distributed stored data, which allows engineers to work more effectively. For more complex simulation tasks involving large dynamic workflows, a more general tool could be used. For example, the Reconfigurable Computing Environment (RCE). RCE is service oriented software framework to manage collaborative engineering processes. It hides the complexity of heterogeneous and distributed IT systems behind a common user interface and hereby enforces secure and uniform access of data and services. RCE is application independent and can easily be adapted to a variety of application domains. It is based on OSGi (Open Services Gateway initiative), the industry standard for modular dynamic Java applications, therefore RCE is platform independent and can thus be used on any architecture from laptops up to mainframes. The seamless integration of Grid-technologies into RCE allows the transparent access to resources. RCE can be easily extended by application specific services which are added and managed as plug-ins, the central mechanism known from the Eclipse universe. Non-Java code, like C or FORTRAN, can be integrated via code wrapping technologies, which are designed to integrate existing code very easy. It provides services for data management, data access, or workflow management. Since many scientists in aerospace engineering are using Microsoft Excel for many tasks (for example, “small calculation” or visualisations) and because a lot of input data for calculations is stored in Excel spreadsheets, RCE has an interface to Excel. This allows engineers to either access data in Excel sheets or include Excel a component into the simulation workflow. An important aspect for the industrial usage of numerical methods are standardised procedures, which assures the reliability and comparability of results from different simulation runs. These procedures are based of experience of knowledge of previous simulations. Therefore a tool for managing and querying expert knowledge is very helpful to support complex and error-prone simulation tasks. For example, knowledge about strengths an weaknesses of certain simulation codes for different applications or guidelines for choosing meaningful parameters of CFD codes are a huge advantage. This is true especially for new and inexperienced engineering stuff or for experienced engineers which are not familiar with a certain CFD code. As an example, the DLR expert systems XPS4CFD is presented, which is developed based on the rules engine JBoss Drools and RCE. In the end, a big picture combining the described software tools is presented. It shows how to link expert knowledge with Grid environments to support large-scale complex aerospace simulations. This includes the use of data and meta data within the semantic context of aerospace engineers and the use of expert knowledge to generate simulation workflows being executed in a Grid.