Ian C. Taig British Aerospace plc Military Aircraft Division
SUMMARY
Finite element analysis computer systems are available to many
users who have little knowledge of the underlying theory. Expert systems can be developed to assist such users to model their problems and interpret their results properly. The FEASA project evolved a prototype system which shed light on the potential of such systems and the methods available to develop them. Indications are given of future developments and their integration with conventional pre- and post- processor systems.
THE DOMAIN: THE ENGINEERING APPLICATIONS
Finite element analysis methods are among the most widely
available and universally used computer systems. A single software package can solve almost any structural analysis problem from the stress distribution in a single bolt to the strength and stability of an aircraft wing or the survivability of buildings and structures during an earthquake. To cope with such a wide range of problems, the systems are necessarily complex and possess a wide variety of options for representing structures, modelling the loading and other external conditions and formulating and solving the relevant equations. Pre- and post- processor systems, often linked to graphics - based computer - aided design, make it relatively simple to prepare the input data and interpret the results. However, despite the relative ease of handling the "mechanics" of the system, a major problem exists in assuring that the analysis properly represents the physical problem: the inexpert user can perform analyses which, at best, are vastly expensive and, at worst, disastrously misleading. Two kinds of expert system can help to avoid serious misuse.
D. Sriram et al. (eds.), Applications of Artificial Intelligence in Engineering Problems
The user of a finite element system such as MSC-NASTRAN (MacNeal Schwendler Corporation) must be able to find the particular program facilities, data entry procedures and solu- tion methods which are appropriate. The documentation of the system may run to 10 or 12 volumes and up to 10,000 pages, of which something like one quarter may be directly relevant at any one time. An intelligent front end system starts from a user's broad description of a finite element analysis and guides him/her to the appropriate facilities and data entry methods. It may also assist in the selection of solution methods and the preparation of formal instructions to implement them.
The first known finite element expert system, SACON (Bennett
et al, 78) tackled solution method selection and more recent systems such as NASCON (Corlett et al, 84) address the much wider problem.
Specification and modelling aids
The foregoing systems help the user who already knows, in some detail, how to represent the physical problem in appropriate finite element terms. A second type of expert system helps the user to identify salient features of the real-world physical problem and guides him/her in the important modelling decisions such as choice of suitable representation of, say, a shell structure (membrane, facet plate or curved shell; mathematical refinement; type of symmetry), representation of joints and sub-scale physical features, mesh density and local refinement or application of physical constraints.
This paper describes the development and status of the first
known expert system of this type, FEASA (Finite Element Analysis Specification Aid). It records experiences gained during development of the system and points to a wider range of systems spanning both categories, which can be developed to interface directly with (and ultimately to be subsumed into) conventional graphics pre- and post- processors.
FEASA - A PROTOTYPE EXPERT SYSTEM
As the compiler of NAFEMS' Guidelines to Finite Element
Practice (Taig 84) and a member of B.Ae's management team setting up a Knowledge-based Systems function, the writer conceived the FEASA project as a natural candidate for exploring some aspects of expert systems technology and satisfying a widespread need. The project has 3 main aims:-
Develop a prototype expert system and evaluate it, both as
a user aid in its own right and as a pointer to a family of similar systems.