This paper discusses the framework of a knowledge-based expert system environment to design aerospace structures under structural and aerodynamic constraints using ASTROS (Automated Structural Optimization program). ASTROS is a synthesis tool built around the NASTRAN finite element program. The knowledge base capabilities are discussed for synthesizing in statics, normal mode, steady and unsteady aerodynamic disciplines. A description of the two ASTROS advisor modules, the Editor/Bulk Data generator and Post-processor, is included. Experiences and issues involved in hierarchical representation of knowledge as menu options at different levels of abstraction are presented. Illustrative examples of the advisor in designing airframe structures are also included.

Numerical simulation of partial differential equations (PDEs) plays a crucial role in predicting the behavior-of physical systems and in modern engineering design. However, to produce reliable results with a PDE simulator, a human expert must typically expend considerable time and effort in setting up the simulation. Most of this effort is spent in generating the grid, the discretization of the spatial domain that the PDE simulator requires as input. To properly design a grid, the gridder must not only consider the characteristics of the spatial domain, but also the physics of the situation and the peculiarities of the numerical simulator. This article describes an intelligent gridder that is capable of analyzing the topology of the spatial domain and of predicting approximate physical behaviors based on the geometry of the spatial domain to automatically generate grids for computational fluid dynamics simulators. Typically, gridding programs are given a partitioning of the spatial domain to assist the gridder. Our gridder is capable of performing this partitioning. This enables the gridder to automatically grid spatial domains with a wide range of configurations.

KANT is a knowledge-based system designed to diagnose Olivetti personal computers connected as remote terminals to a host computer through a SNA link. KANT is a collection of a few knowledge-based units: some of them operate in the field, and some operate back at the home office. They are configured in two blackboard systems which exchange data via the SNA link. The first blackboard runs on a cheap personal computer, only employs shallow knowledge, and performs the diagnoses that can be achieved in the field. The second blackboard runs on corporate mainframes, employs deep knowledge, and supports the more sophisticated analysis that is required from the project team for fixing complex problems.

We are working on using machine learning to make the numerical optimization of complex engineering designs faster and more reliable. We envision a system that learns from previous design sessions knowledge that enables it to assist the engineer in setting up and carrying out a new design optimization. We have performed initial experiments for two aspects of setting up an optimization: selecting a prototype to serve as a starting point for the optimization and selecting a reformulation of the search space. Both choices can dramatically affect the speed and the reliability of design optimization.

This paper extends previously developed generalized set propagation operations to work over relationships among an arbitrary number of variables, thereby expanding the domain of engineering design problems the theory can address. It then narrows its scope to a class of functions and sets useful to designers solving engineering problems: monotonic algebraic functions and closed intervals of real numbers, proving formulas for computing the operations under these conditions. The work is aimed at the automated optimal design of electro-mechanical systems from catalogs of parts; an electronic example illustrates.

In this paper a unique design methodology known as 1stPRINCE (FIRST PRINciple Computational Evalualor) is developed to perform innovative design of mechanical structures from first principle knowledge. The method is based on the assumption that the creation of innovative designs of physical significance, concerning geometric and material properties, requires reasoning from first principles. The innovative designs discovered by 1stPRINCE differ from routine designs in that new primitives are created. Monotonicity analysis and computer algebra are utilized to direct design variables in a globally optimal direction relative to the goals specified. In contrast to strict constraint propagation approaches, formal qualitative optimization techniques efficiently search the solution space in an optimizing direction, eliminate infeasible and suboptimal designs, and reason with both equality and inequality constraints. Modification of the design configuration space and the creation of new primitives, in order to meet the constraints or improve the design, are achieved by manipulating mathematical quantities such as the integral. The result is a design system which requires a knowledge base only of fundamental equations of deformation with physical constraints on variables, constitutive relations, and fundamental engineering assumptions; no pre-compiled knowledge of mechanical behavior is needed. Application of this theory to the design of a beam under torsion leads to designs of a hollow tube and a composite rod exhibiting globally optimal behavior. Further, these optimally-behaved designs are described symbolically as a function of the material properties and system parameters. This method is implemented in a LISP environment as a module in a larger intelligent CAD system that integrates qualitative, functional and numerical computation for engineering applications.

In preliminary design, the details of a structure are insufficient to warrant the use of numeric tools traditionally used in structural analysis. However, an accurate prediction of the behavior of a structure and its components in the preliminary design phase can have a significant effect on the final design process in reducing the number of alternative solutions, avoiding the costly design revisions, and improving the quality of design. Presently, there are few tools available for preliminary analysis of structures. This study represents an initial effort towards the development of a tool that can be used in the conceptual design stage to qualitatively evaluate the behavior of a structure.

This paper describes a prototype system, QStruc, for qualitative structural analysis, which combines first principles in structural engineering and experiential knowledge of structural behavior. The purposes of QStruc are: (1) to generate qualitative models from the schematics of a structure; and (2) to infer the qualitative response of the structure in terms of deflected shape, moments, and reactions. The qualitative analysis strategy employs: (1) a greedy depth-first approach that tries to expand the derived response as much as possible from known parameter values; (2) a causal ordering mechanism, which enables the system to identify the solution path for the qualitative analysis; (3) qualitative calculus, which enables the qualitative evaluation of the physical quantities of the causal model that describes the behavior of the structure; and (4) Quantity Lattice (Simmons, 1986) which enables the system to reason about partial ordering among physical quantities and to reduce some of the ambiguous conclusions caused by the impreciseness of the information. Examples are provided to illustrate the effectiveness and limitations of the prototype system.

Although the case-based reasoning (CBR) process is domain dependent, certain aspects of it can readily be captured into a generic framework which in turn can be applied to various engineering domains. One such exercise that has been carried out is described here. In this paper, we present the notion that CBR can be formalized and applied in a specialized framework in an integrated knowledge-based environment. We first analyze the CBR process to abstract the steps involved in the development of a CBR system. We then propose a framework in which most of these steps are formalized so that they can be applied in a domain-independent manner. The salient features of this framework, called CASETOOL (CASE-based reasoning TOOL-kit), are then described. The highlight of this approach is the use of a concept called design criticism in the CBR process. The versatility of the tool is demonstrated through an application from the bridge engineering domain.

The “sketch” drawn by a human designer represents a shape class of wider variability than can be captured by traditional CAD models; these typically work with parametrizations based on a nearly finished shape. Traditional Qualitative Reasoning is also unable to model this degree of ambiguity in shape. Cognitively, shapes are often represented in terms of an axial model. In defining 2D contours, such an axial representation is called the Medial Axis Transform or MAT. By perturbing the parameters of the MAT—length, link angle, and the node radius—one can define a shape class. Unlike the contour-to-MAT transform, which is well-known to be unstable, the MAT-to-contour process is an integrative process and is very stable. The variation in these parameters can be controlled by defining a suitable discretization over the parameter space. This leads to a broad class of similar shapes from which an optimized shape can be obtained for a given set of criteria. The optimizing criteria may involve the boundary description for each shape; the axial model is only used for generating the shape class. This Qualitative MAT model has been tested in several design optimization contexts, using Genetic Algorithms, and we show results for Automobile contours, IC engine parts, building profiles, etc.

PANDA, the Pumper Apparatus Novice Design Assistant, is a case-based design system developed to assist firefighters who wish to design their pumper engines. In contrast to other such systems, PANDA addresses the unique needs of a novice, non-specialist who performs design in a highly specialized domain, where the design is decomposable into elements which each fulfill their own identifiable function. In PANDA we study how to create an interactive case-based design assistant that can understand the needs and desires of a non-specialist designer and can translate them into formal specifications; that can provide assistance by using case-based design methodologies; that can deal with non-functional design criteria such as aesthetics and traditional practices; and that can guide the novice designer by discussing alternatives, tradeoffs, and adaptations. Our prototypical system verifies our methodological approach to supporting novice design activities.

According to Simon (1983), machine learning is a process that improves the performance of an intelligent system. The performance task of a design system is to arrive at a structural description of a device given its functional description. Machine learning can occur, to improve the performance of the design system, in a number of ways. The learning can make the process of arriving at a known structure to function mapping faster, either by improving the control strategy or by learning new world knowledge needed for the structure to function mapping. It can also improve the performance of the design system by allowing it to come up with new structure to function mappings. The latter kinds of learning leads to innovative designs. The work presented in this paper, EnviroAdapt, belongs to the latter category. EnviroAdapt has the performance task of designing devices along with their operating environments. The EnviroAdapt learns about the novel environments and designs devices for these environments. This learning process is enabled by adapting the previous designs of devices along with their operating environments. This adaptation process makes use of abstracting general mechanisms from the previous designs, then instantiating them in the context of new design problem. Both of these processes are driven by the characteristics of the new design problem that requires a device to operate within a novel operating environment.

This paper develops a philosophy for the use of Artificial Intelligence (AI) techniques as aids in engineering project management.

First, we propose that traditional domain-independent, ‘means–and’ planners, may be valuable aids for planning detailed subtasks on projects, but that domain-specific planning tools are needed for work package or executive level project planning. Next, we propose that hybrid computer systems, using knowledge processing techniques in conjunction with procedural techniques such as decision analysis and network-based scheduling, can provide valuable new kinds of decision support for project objective-setting and project control, respectively. Finally we suggest that knowledge-based interactive graphics, developed for providing graphical explanations and user control in advanced knowledge processing environments, can provide powerful new kinds of decision support for project management.

The first claim is supported by a review and analysis of previous work in the area of automated AI planning techniques. Our experience with PLATFORM I, II and III, a series of prototype AI-leveraged project management systems built using the IntelliCorp Knowledge Engineering Environment (KEE™), provides the justification for the latter two claims.

The paper describes a novel means of performing calculus operations on poorly understood engineering functions using networks of radial Gaussian neurons. The network architecture and training algorithm used for this purpose is described briefly. Once trained, a network can be converted into a form that provides the differential or integral of the learned function, by a simple substitution of the type of activation function used at the hidden neurons. A range of substitute activation functions, for conversion to first- and second-order partial differential and integral forms of the network, are derived. Following this, the technique is tested on a selection of calculus operations for two example civil engineering problems: assessing motion in tall structures; and modeling moisture penetration in porous materials. The converted networks produced in these experiments provide accurate models of the actual differentials and integrals of the function the original networks had been taught. A method of improving the accuracy of results by training the original network beyond the region in which the converted networks operate, is described. The paper concludes by identifying some areas for further development of the technique.

Since the inception of research on machine learning (ML), these techniques have been associated with the task of automated knowledge generation or knowledge reorganization. This association still prevails, as seen in this issue. When the use of ML programs began to attract researchers in engineering design, different existing tools were used to test their utility and gradually, variations of these tools and methods have sprung up. In many cases, the use of these tools was based on availability and not necessarily applicability. When we began working on ML in design, we attempted to follow a different path (Reich, 1991a; Reich & Fenves, 1992) that led to the design of Bridger (Reich & Fenves, 1995), a system for learning bridge synthesis knowledge. Subsequent experiences and further reflection led us to conclude that the process of using ML in design requires careful and systematic treatment for identifying appropriate ML programs for executing the learning tasks we wish to perform (Reich, 1991b, 1993a). Another observation was that the task of creating or reorganizing knowledge for real design tasks is outside the scope of present ML programs. Establishing the practical importance of ML techniques had to start by addressing engineering problems that could benefit from present ML programs.

Most computer-based design tools assume designers work with a well-defined problem. The traditional treatment of design as two discrete phases; problem formulation and solution synthesis, is challenged by recent research. Though the view on discrete phases may be applicable to simple and/or well-defined design tasks, current research (Jonas, 1993; Logan & Smithers, 1993; Gero, 1994; Smithers et al., 1994) has shown that design is an ill-structured problem and the discrete phases view is not a good description of the process during which design alternatives are generated. A potential role of machine learning techniques is to provide a computational model of the changing representation of the design problem in response to the search for design solutions.