Intelligent CAD Systems I

Assumptions for CAD are discussed, pointing to a distinction between human knowledge and machine representations of knowledge. Implications for future useful CAD systems are considered. A strategy for “mechanistic” symbol processors is presented, employing “mechanisms” of formal logic to manipulate written and drawn expressions of designers9 knowledge.

An intelligent CAD system is a tool box of automated problem solving aids that allow designers to conceive, evolve, and document their designs. Our research involves CAD systems for mechanical applications where three-dimensional geometric shapes play a significant functional role. In this paper, we consider the design process itself as a problem solving activity. Computer systems suitable for assisting different modes of problem solving must be based on different assumptions about users’ intentions and their dialog with the system. We identify three modes of problem solving based on their implications on a system’s underlying paradigm. Because of the important role of geometric information in design, a prerequisite for intelligent CAD systems is more explicit representation and manipulation of geometric knowledge. We discuss a paradigm for an intelligent CAD environment specifically suitable for geometric information and its implication on knowledge representation.

The battle for the best knowledge representation language seems terminated, not because anyone had found one but because fairly obviously there is none. We propose to turn our attention to general architectures of knowledge representation. If theories to be represented become specified, we can use such general architectures in the development of individual ones. Theories are generally used by some agent for specific purposes. Agents need to communicate with each other in order to make use of their theories. We claim that a theory of agenthood should describe communication and conversation among agents and the way they interact with the environment. A functional architecture for intelligent systems interconnection (ISI) is proposed. We identify functional and hierarchical layers of representations and theories. We attempt to show the road to bring together represented conscious and unconscious as well as not-represented inherent knowledge. We do it in order to combine intelligence with effectiveness.

Expert systems are more and more commonly used in CAD. Their performances are better if the problems are limited. However in CAD several designers cooperate for the design of a sophisticated product. In the expert system approach a super-expert system which is able to deal with all aspects of design seems unrealistic. Thus a multi-expert system, which is a federation of smaller expert systems seems more appropriate. In this paper we study this multi-expert system approach mainly from the view point of strategies for cooperation, which are related to the design process model. We also present a knowledge management system which assures information management for the whole system and supports the exchange of information between different tools including expert and classical systems.

Intelligent computer-aided design (CAD) emulates the human activity of design so that production planning decision making and inventive design can be performed by computers. Based on the history of human experience in engineering design, a formalized approach to design methodology should include procedures from (1) conceptual design, (2) layout design, and (3) numerical optimization design. The highest design level in such a system should be responsible for generating skeleton structures of entities within the design process which are eventually to be specified uniquely, and to be optimized. Planning plays a key role in such a system. Planning has been utilized as a tool for process organization within the knowledge domains of chemical engineering electrical engineering and manufacturing as well as for general problem formulation and solution. State estimation, subtask scheduling and constraint propagation are factors of prime importance in this type of problem. A methodology for planning and the problems associated with it within large scale design are discussed. The generality of our hypothesis for the design optimization process is examined within the context of a prototypical mechanical design model. An example which demonstrates the applicability of this approach to mechanical power transmission design and which is representative of a large scale design problem, is provided.

This article describes a CAD system that allows for declarative specification of drawings, in contrast to the construction procedure that has to be carried out in conventional CAD. As a smaller part this fits into a picture of design as a series of refinements from a “functional” description at the beginning to a description sufficiently detailed for manufacturing. Declarative, but non-complete, specifications are also convenient for representing and recognizing standard form-features of geometric products. In future the level of this kind of descriptions has to be increased and formalized in the direction of functionality. For performing efficient geometric reasoning about spatial relationships between elementary geometric entities, e.g. line segments and arcs, we propose a system architecture combining a production mechanism and a verification mechanism, both based on rules. Central in this architecture are new terms defined for an intermediate representation of how entities are related.

This paper deals with the characteristics of two different methodological approaches to intelligent CAD systems. The underlying paradigms are introduced and basic concepts, such as method, intelligence, and CAD systems are defined and elaborated on. One methodology is based on a decision structure oriented toward product-users. In conclusion, requirements are formulated with respect to the implementation of the two different methodologies as a language.

A workbench for acquiring design and analysis knowledge to build knowledge-based systems is presented. Eliciting and modeling such knowledge from a human designer is a major problem when building knowledge-based systems for CAD problems. Aquinas, an extended version of the Expertise Transfer System (ETS), combines ideas from psychology and knowledge-based systems research. Aquinas interviews design and analysis experts and helps them model, analyze, test, and refine their knowledge. Expertise from multiple designers or other knowledge sources can be represented and used separately or combined. User consultations are directed by propagating information through hierarchies. Aquinas delivers knowledge by creating knowledge bases for several different expert system shells. Help is given to the expert by a dialog manager that embodies knowledge acquisition heuristics. Aquinas contains many techniques and tools for expertise transfer; the techniques combine to make it a powerful testbed for rapidly prototyping portions of many kinds of complex knowledge-based systems.

We define ICAD (Intelligent Computer Aided Design) as the instantiation of the interactive relation between designer and a variable, category-based, extendible machine, based on a parallel configuration of Intelligent Processors. The paper provides an outline of a computational design theory, followed by a set of observations and recommendations as specifications for an ICAD system. Issues of methodology and configuration conclude the conceptual analysis. In place of ‘expert systems for design,’ implemented as partial solutions to the design problem within limited domains, we propose ‘expert systems of design,’ meta-expert systems containing knowledge of the design process. Expert systems of design can be built using current technology, and can be used to organize and guide the development of conventional, domain-specific, expert systems modules. The design team is the metaphor around which we developed the conceptual model of a multifunctional intelligent design (MIND) machine.

The aim of this research is to study how high-level mathematical abstractions could be used for reasoning about hardware design. These abstractions will be used as the foundation for a CAD tool for VLSI. Of course, different algebraic abstractions are appropriate on different levels of the design hierarchy. Uncovering formalising and integrating these will be difficult but is necessary for the development of future CAD tools. We have developed a general algebra, applicable at all levels of the hierarchy, for composing small modules into large designs. Having a common composition algebra throughout allows transformations between levels (both refining transformations downwards and verification and abstraction transformations upwards) to be algebraic morphisms. In this paper, we present a sketch of the composition algebra, and closely follow a design at the behavioural level. It is hoped to give an indication both of the use of composition and of the kind of verification that can be done at the highest levels.

The goal of our research is to open new perspectives in CAD by identifying different techniques and theories of Artificial Intelligence (AI) that can be used to create the next generation CAD tools. Therefore the design task has to be better understood and models of it must be built. Our work is based on studies we made in a specific design domain, but our model of the designer will be at a high level of abstraction and will cover a wide class of problems. Good designers think about problems from a global viewpoint. Starting from assumptions they build different worlds containing different possible solutions. They evaluate and compare that limited set of possible solutions by using common sense and domain knowledge. Truth maintenance systems and time reasoners have been used to construct such a model of the designer. This paper relates our way towards a solution to this problem.

We present a model for future CAD systems which aid the designer by combining conventional CAD techniques with expert problem solving and design planning tools. Dialog with these tools is directed at the application level and all tools provide explanation. We describe how two models of the design process, the goal tree model of Mostow and problem solving by debugging almost right plans of Sussman and Steele, Jr., can be integrated to support both high and low level planning of engineering design. We suggest a set of paradigms for integrating the model.

The paper describes design phases and types of design. Design phases define the function and functionality related to graphics presentation. In a design process four characteristic graphic presentations are produced; a scheme, a rough drawing a drawing (project drawing), and a manufacturing drawing. Four design modes define the designing activities; new design, innovative design, variation design and adaptation design. Connecting the design phases with the design modes makes systematic design possible.

A generic, multiparadigm user interface for intelligent CAD systems for mechanical design is presented. The needs for an interface are examined from the engineer’s standpoint as well as the standpoint of integration with CAD systems. A general architecture for such an interface is discussed. As an example, a frame-based implementation for a mechanical component design is presented.

This paper examines the criteria for ‘intelligent’ CAD systems for VLSI design in the light of practical requirements. It identifies the importance of the total support environment in creating a system which is sufficiently integrated and flexible to provide the design engineer with an intelligent assistant. The role of knowledge-based tools in CAD for VLSI is also examined, considering their use as individual ‘expert’ tools and as rule-driven intelligent adjuncts or controllers of conventional CAD applications. The SIDESMAN Design System is then used to illustrate some of the techniques which can be employed to produce an intelligent CAD system.

In the past few years, the use of computer aided design for electronics has become widespread. However, automated testing support from design is only now beginning to take place. In this paper, we describe a system Synergist, which derives automatic testing procedures from the computer aided design description of a circuit. Synergist uses the Artificial Intelligence technique of structural isolation in order to automatically derive testing procedures for printed circuit boards.

Differences between human intelligence and knowledge engineering are observed. Based on a brain model developed in physiology, a structure for learning and creative expert systems is proposed. The model is related to general design theory. Generative grammars are pointed out as a tool for applications.

We present in a unifying framework the basic notions of I DDL (Integrated Data Description Language) to code design knowledge in the IIICAD system. IIICAD is an intelligent, integrated, and interactive computer-aided design environment we are currently developing at the Centre for Mathematics and Computer Science.