Artificial Intelligence in Design

The majority of the computer-aided design (CAD) systems developed to date are not true design systems. They are in most cases mere draughting or analysis packages lacking the intelligence and creative faculty of the human designer. Due to the recent availability of massive computing power at relatively low cost, opportunities have arisen for building CAD systems with more genuine design abilities [1–12]. These systems apply techniques drawn from the branch of computer science known as artificial intelligence (AI). The most promising techniques are those of expert systems or intelligent knowledge-based systems. Several of these techniques will be discussed in different parts of the book. They, together with others, will be assembled and overviewed in this chapter. The chapter contains two main sections. The first deals with techniques underlying intelligent knowledge-based systems in general. The second is devoted to techniques applicable to intelligent knowledge-based systems for design.

Geometric information processing is a core topic in the computer-aided design of mechanical products, common to a variety of tasks such as draughting, geometric modelling, finite-element analysis and production planning. The development of adequate software tools that support a wide range of geometry-related applications is therefore of major interest in view of the economy of the design process and the quality of the envisaged products. This concerns the user interface as well as the manipulation of internal product descriptions.

Geometric computation has been widely used for over 20 years in computer-aided design (CAD), but until fairly recently, the emphasis has been on the end user deciding what geometric constructions to make. The main mode of use has been to treat a CAD system as being to geometry as a calculator is to arithmetic — the CAD system can perform various geometric manipulations and draw the results, but it has no built in knowledge of geometric theorems and concepts, just as a calculator has no knowledge of number theory.

The process of computer-aided engineering design is characterized by interactive decision making between a designer and a computer so that design objectives are satisfied under various design constraints. The design process involves the recognition, formulation and satisfaction of constraints. They are continually being added, deleted and modified throughout the design process. The constraints are often numerous, complex and contradictory [1]. A designer must manage these constraints so that design objectives are well satisfied. However, it is difficult to manage them in the case of designing a large and complex engineering system.

Since no mechanical part can be manufactured with exact dimensions and perfect shape, tolerances, together with dimensions, are used to specify acceptable variations of part geometry. To ensure design functionality, assemblability and manufacturability, tolerances must be specified and checked. Currently, dimensions and tolerances are usually placed on design drawings of CAD systems in accordance with ANSI [1], ISO [2] or other national standards. Tolerance analysis is used to check the stack-up of related tolerances. The process of tolerance accumulation is modelled, and the accumulated tolerance is verified. Tolerance synthesis (or tolerance design), on the other hand, is the process of allocating tolerance values derived from design requirements in terms of functionality or assemblability among related design tolerances. It is a process of distributing a few known tolerance values.

Designing products for easier assembly is recognized as an important part of “simultaneous engineering”: the process of making products meet functional, manufacturing, quality and cost targets at the early stages of design. Considering the assembly aspects of existing and new products may lead, for example, to the use of simpler robots with fewer tools and grippers, and less costly fixtures, or indeed, to abandoning robots in favour of alternative process equipment.

This chapter presents the concepts and philosophy underlying an intelligent product design and manufacturing (IPDM) system developed at the Centre for Flexible Manufacturing Research and Development at McMaster University. The overall objective of the IPDM research program is to: (1) investigate the basic issues involved in integrating the design and manufacturing task planning activities; and (2) demonstrate the feasibility and potential of applying artificial intelligence, expert systems and feature-based modelling concepts to these activities. The project addresses the fundamental problem(s) of establishing an effective link between computer-aided design (CAD) and manufacture (CAM).

Engineering design problem solving, in general, is ill structured, i.e. high diversity, complexity and uncertainty are involved. Different strategies and techniques are employed in various design domains, and even within the same domain diversely by different designers. Design heuristics are difficult to organize in a form suitable to be coded into computer systems. The question of what underlies empirical and experiential expertise and skills in specific domains leads to a need for research on developing a general model of the design process. There seems to be little possibility however, in general, of achieving this goal without first experimenting with specific domain problems. A complete model of a manufacturing system must contain information describing both plant and product, reflecting the design process in systematic terms (both organizational and technological).

Process planning is recognized as a critical bridge between designing and manufacturing stages [1]. In this process, manufacturing information required for realizing the product is determined, based on the complete machine part definition. Therefore, computerized process planning is said to be a key technology for constructing computer-integrated manufacturing (CIM) systems. Recently many types of automated process planning systems have been constructed. Expert system techniques are introduced for such systems in order to utilize expert production engineers’ knowledge for generating process plans. However, such “expert” process-planning systems still have some problems in achieving higher levels of automation.

Developments in artificial intelligence (AI) and knowledge engineering have led to task-oriented products and the so-called expert systems. Different methods of building expert systems have been proposed [1]. The task-oriented approach offers expert systems in very different areas of human activities of which those in engineering are of special interest at the present time.

Process planning is one of the most important procedures in production because of its interfacing function between the design and the manufacturing procedures. The manufacturing procedure consists mainly of three core activities, i.e., machining, assembly and inspection. Accordingly, there are various kinds of process planning depending on each activity; however, research and development are so far concentrated on process planning for machining. This trend may be attributed to the fact that machining is deeply concerned with part manufacturing, which is fundamental to producing the goods, appliances, equipment, devices and so on required by society. In this chapter, thus, process planning for machining is described.

With the eventual goals of reducing labour costs and increasing quality in small-batch manufacturing, several research institutions and industrial organizations have been developing research methods that:

Forging is the process whereby the shape of a workpiece (usually metallic) is changed by pressing or hammering the workpiece between two or more dies, with or without the application of heat. Complex shapes are usually forged from the original stock material in a number of stages. Quite often, the dies for several stages are grouped together in the same forging press or hammer, the workpiece being transferred from die cavity to die cavity (manually or by an automatic mechanism) in between successive blows of the machine.

The design of jigs and fixtures is a highly complex and intuitive process. There are many mathematical and scientific formulae which can be used to calculate cutting forces, deflection of structural members, tolerance analysis of locating datum, etc. However, many of the good design features of jigs and fixtures such as correct proportioning, ease of loading, safety considerations, ingenuity in securing workpieces, etc. come from the experience and skill of the designer. As a result of the demand for both engineering analysis and craftmanship, the automation of fixture design has not been considered possible in the past. Recent developments in artificial intelligence techniques and, in particular, knowledge representation and inference procedures have opened up great opportunities in automating this field. Computer-aided jigs and fixtures research started in mid-1970s and towards the early 1980s, several prominent research institutions began work in this area.

Jigs and fixtures are devices used in manufacturing processes such as machining, welding, bonding and assembly. They are employed to locate and hold the workpiece firmly in position and to ensure that it is in a state of stable equilibrium (geometric control) and that dimensional accuracy is maintained during the manufacturing operation (dimensional control). Jigs and fixtures are generally referred to as “fixturing systems” or simply “fixtures”.

At present there are few well-considered systematic theories and determinate methods for designing hydraulics. Engineers in various fields mainly rely on their own practical experience and use the manual method of design. The quality of designed projects largely depends on the ability of designers. However, experts are limited in number, so it is necessary to research the design theory and practice of hydraulics, sum up the experience of experts and use the computer to simulate experts’ design procedures.

The design of control systems for industrial automation requires knowledge of both mechanical configuration and control logic. A design tool, named rule-based controls development system (RBCDS), was developed to apply knowledge-based system technology to the engineering discipline of control system development. RBCDS aids developers of automated equipment, such as machine tools, material-handling equipment, and industrial manufacturing processes, to design and test software-based control systems.

Manufacturing industry has witnessed significant developments in recent years, as measured by the increase in the number of automated systems in use. The key to the success of these systems is effective exploitation of available resources such as machines, tools, fixtures and material handling systems. Proper use of these resources has increased productivity. At the same time, the design problems related to the modern manufacturing systems have become more complex. Designers and users of automated manufacturing systems have attempted to develop new tools to cope with these complexities. Knowledge-based systems represent a class of modern tools that have been applied to improve the design and management functions in automated manufacturing systems.

The systematic design of intelligent manufacturing systems (IMSs) is a complex task that requires an understanding of the nature and structure of manufacturing systems, as well as the models used to represent such systems. This chapter discusses some critical issues in making precise the representation and utilization of knowledge for intelligent manufacturing design. The representational issues include the analysis of the structure of manufacturing plants and the representation of these systems as automata. The utilization issues relate to two conceptually separable clusters: decision rules and performance metrics.