CHA's - CAD Product Directions (1988)

This paper discusses the direction Computer Aided Design (CAD) products are expected to take toward the end of this decade, (1988-1990), with a view towards helping users plan. The discussion covers the strategic implications of the direction of data communication, management and structure standards. It also covers the tactical effects of emerging technologies on CAD products' accessibility, responsiveness, flexibility, friendliness, and productivity.

The major shift in CAD is that its usage is becoming strategic. CAD, accompanied by its computer aided cousins (CAE, CAPP, CAM, CAQ, etc.), is leading the information revolution in engineering and manufacturing organizations. Already it is evident that this revolution will greatly improve the responsiveness, flexibility, precision, and control of these organizations' operations and products. Consequently, how well this information revolution is planned and executed will have a major effect on the competitiveness and even the survival of users' organizations.

No longer is CAD primarily the automation of an isolated function (drafting, for example) which is justified on a purely financial basis. Now it is a requirement that all automation which bears on any product design, manufacturing, and/or service function must be considered in the context of (its integration path to) overall engineering and manufacturing automation. This requirement would be stultifying were it not for the adoption of standards as a major product direction.

Despite the major shift towards strategy, the cost of CAD automation is so high that consideration of tactics can have major financial consequences. We believe that awareness of the product directions of personal computers, workstations, networks, computational devices, applications, and artificial intelligence can help users avoid fruitless investments.

This paper is done from the perspective of a CAD market vendor with the purpose of helping those who are planning on being extensive CAD users. Nonetheless, these are opinions, not facts and are subject to all the errors attendant with predicting the future. All future times apply to the end of this decade, 1988 through 1990.

In the late eighties, progressive CAD engineering and manufacturing organizations will be marked by:

	a)	a much larger population of users most of whom will be using personal computers (rather than more powerful engineering workstations)
	b)	heterogeneous computing systems used by many disciplines (e.g., design, documentation, testing, planning, fabrication, assembly, ... ) and connected by a local area network.

	a)	allow one discipline's data to be used by other disciplines automatically and instantaneously
	b)	manage and control the access and changing of data
	c)	eliminate doing any CAD data entry, edit or usage without computer assistance
	d)	optimize the design process for "downstream" requirements of reliability, availability, quality and price/performance.

Consequently, the direction of CAD usage will be governed by two distinct goals, the first being strategic, the second being tactical. These goals are to:

	a)	fully integrate CAD technology (and its cousins) to achieve "computer aided product management" from design concept through product manufacture and service
	b)	fully utilize and exploit CAD technology to be highly responsive to rapidly changing market demands for competitive products.

The remainder of this paper deals with the way in which product directions will help and hinder users' efforts to achieve these goals. The next section, Strategic Directions, discusses the effects of data communication, management, and structure on CAD integration. The following section, Tactical Directions, discusses the requirements for widespread data access, fast response times, powerful customizing, easy usage, and productive applications necessary for full and widespread utilization of CAD technology to occur.

The key to full integration is data: data communication protocols, data management techniques, and data structure standards

To achieve full integration of CAD technology means that CAD data serves users' organizations by providing information which:

	a)	meets business needs (how much will a product cost, when will it ship, what is its status, who is responsible … )
	b)	meets operational needs (what are its dimensions, what does it look like, what is needed to make it, how is it made, how is it tested, what changes have been made, … )
	c)	meets developmental needs (what are the limits of what it can do, how does it perform, what are the effects of tradeoffs, … ).

To meet these diverse informational needs in the context of large user populations across a network of heterogeneous computer systems means that there must be good protocols, techniques, and standards for data communication, management and structure. To date, progress is good, but even a minimal set of protocols, techniques and standards is not widely used (perhaps because CAD technology is developing so rapidly).

Data communication protocols (under the ISO-OSI model) are good, but computationally expensive.

The projected status of data communication standards is very promising. The ISO-OSI (International Standards Organization - Open Systems Interconnection) model provides a framework for the communication of information and is specifically structured for "mixing and matching" different protocols at different levels of functionality. However, be forewarned that the computational cost of ISO-OSI layering materially contributes to the reduction of (process to process) network communications to under 10%" of network transmission capacity (e.g., 10 Megabit/second Ethernet might only effectively transport 100-250 KiloBytes/second). As accommodations to this computational burden, we expect the short circuiting of rigorous compliance between successive layers within a single process to become widespread (but with little negative consequences on users).

A transitional data communications course could be adoption of TCP/IP (Transmission Control Protocol/ Internet Protocol) and Ethernet to get immediate, multiple vendor supported standards in place. These protocols and standards will, in al likelihood, migrate graciously to full ISO-OSI compliance. The consequence of good data communication protocols (even as they exist today) is that heterogeneous systems on standardized networks can be created with the full assurance that data can be moved anywhere in the network (although usage of the moved data will depend on good data management techniques and good data structure standards).

Data management techniques offer real excitement about filling users' information needs

The prospects for data management are very exciting as the significant advances of the past decade (relational files, object oriented programming, and relationship specifying) are exploited to serve the users' interface and information requirements. Relational files support freely formed queries and provide extensibility. Object oriented programming greatly amplifies user intent by endowing language (e.g., "insert wall") with rich context and usage considerations with automatic reconciliation (e.g., walls are inserted orthogonally, have standard thickness, do not encapsulate a room, usually need power, … ) thereby eliminating much "obvious" tedium. Relationship specification (either as an entity model or in the form of Lisp or Prolog statements) can automatically flag and/or resolve data procedures, such as revision control or configuration management. Finally, usage of natural language as data query language promises a de facto standard of the most universal and acceptable kind (to casual users).

While these data management advances are emerging, users should capture data in relational files (due to their extensibility, communicability, and purity). Separate files (or Lisp or Prolog statements) can be created by users to define the data relationships and become the basis of CAD data validation (for configuration management or revision control). Object oriented programming (and hierarchical relational files) can be added later as data base technology matures.

The data structure of the IGES standard offers hope for a neutral CAD data file, but non-geometric needs are largely ignored

The likely prospect for a neutral CAD data file standard in the next five years is IGES (Initial Graphics Exchange Specification). Computer Graphics Metafile (GCM) is a promising standard primarily for the storing and reproduction of pictures (especially for output spooling), but is not intended to act as a neutral CAD data file (which is why it is called a metafile). This metafile and its companion Computer Graphics Interface (CGI) facilitate the mixing and matching of output (and input) devices attached to networks. GKS is an expensive graphics interface standard in that it slows the most critical portion of graphic terminal processing, but will become increasingly important in the development of applications software.

Unfortunately, IGES is still evolving and does not adequately address the growing non-geometric data needs of CAD. Vendors and users will have to design their own non-geometric data structures (in conjunction with the data management techniques they want), rather than wait for standards.

The keys to full utilization are: ready access, responsiveness, flexibility, ease of use, and productive applications. To achieve full utilization of CAD technology products requires that:

	a)	access must be throughout users' organization
	b)	response times must be fast
	c)	customizing must be powerful
	d)	usage must be easy
	e)	applications must be directly productive.

Evolving computer technology is accelerating products in precisely these directions.

Personal computers and fast networks assure widespread access to CAD data

Widespread access is assured by the trend in personal computer products towards high performance, high resolution, bit mapped displays coupled by the extraordinary communications capacity growth made possible by fiber optics communications technology. By the end of the decade, personal computers will have tables with 8 bit planes (permitting 256 colors in any single frame), 1-10 million bytes of resident memory, and optionally, some local, removable disk storage. Fiber optic network broadband backbones will have at least 100-1000 million bits per second capacity, at affordable prices. Coaxial cable cluster (token ring) networks will have a tenth or less of the fiber optic capacity; however, these coaxial cables will not be the communications' bottleneck.

Computation will be much faster, but geometric precision cannot increase so fast and access on shared disks will be slow

General purpose computation will be an order of magnitude faster than today and algorithmic progress will also be notable (e.g., the use of adaptive step size with successive refinement and the organization of solutions to take advantage of parallelism). Since most CAD computations are of higher order than 1 (e.g., orders 2 to 6 might be: 2D, 3D, kinematics, holography, ray tracing), a 10-20 fold computational speed increase might translate into only a factor of 2 reduction in interpolation step size for the same responsiveness as today.

However, new concurrent or parallel computer architectures will offer 50 fold increase in computation power per dollar (over DEC's VAX 780, for example) which might appear as a 5-7 fold increase in precision of second and third order CAD problems (e.g., mesh size of finite element analysis). Higher order problems (e.g., holography, ray tracing) are not crucial to CAD goals, and even so will be solved by special purpose hardware.

However, a real bottleneck arises during the simultaneous random access of data on shared disks. Today, particularly in UNIX, such accesses are computationally bound, and are likely to remain so as users impose additional computational burdens associated with distributed file systems. The use of intelligent caching, extensive indexing, and concurrent accessing (on multiple spindles) will offer user transparent avenues to alleviate the bottleneck, but the users' appetite for virtual file access features will advance faster than product solutions.

Powerful graphic languages are underway, customizable data management tools are also needed

Customizing the processing of CAD data is the major task facing users. Each organization has a unique set of evolving requirements which are a result of historical accident, personal preference, and competitive c onsiderations. Vendors are slowly coming up with the powerful data management languages necessary to accommodate customizing by users. The product direction is widespread and progressing steadily. The integration of dependency relationships and the administrative process for their management and control is barely even recognized as the major utilization requirement that it is. It is likely that, until users call for this capability, vendors won't offer it. As mentioned earlier, the implications go to the heart of the data structure and so CAD users should insulate themselves from the upheaval this will cause when vendors finally address it. One tactic for users to use is to resist the temptation of imbedding administrative requirements in applications, but rather employ data validation applications which can be discarded when vendors finally offer this capability.

Products' user interfaces are advancing in the direction of intelligent icons, menus and windows

CAD products' user interfaces are employing many of the innovations which originated at Xerox PARC during the late seventies and which now can be seen in Apple Computer's MacIntosh and Lisa product lines. This is in response to the recognition that intelligent, object oriented, non-stop operation (disallowing the use of off-line references or learning) is a vital requirement of widespread CAD data usage by casual users (i.e., all of the business users and the bulk of the operations users). The development of these intelligent user interfaces is one of the few immediate applications of Artificial Intelligence (AI) in CAD that will meet with widespread success. There is no doubt that this is the prevailing direction, that it facilitates widespread usage, and it will be done successfully by most vendors.

Direct user productivity will occur immediately, whereas synergistic benefits (and expert systems) will occur slowly)

No matter how suitable the systems environment, the users' CAD data usage and applications must yield direct benefits for CAD products to become popular because it is human nature to use products which directly benefit the user. Intelligent data access methods for the casual user and powerful engineering workstations with local disk storage for the intensive user will excite these two constituencies about CAD product usage. The four direct areas of CAD usage will experience the following gains:

	a)	input: increased use of incremental, parametric, approximate and inherited input as contrasted to tedious and explicit input
	b)	processing: dynamic, alternative, iterative, and convergent processing rather than static and cumbersome processing
	c)	editing: automatically checked, coached, and controlled editing, rather than error prone and happenstance editing
	d)	output: filtered, simplified, enriched, and high quality rather than the coarse, complete, busy data of today.

During the next five years, CAD applications will make significant progress in the simulation and analysis of "downstream" product activities such as product activities such as product fabrication, assembly, testing, and, especially, kinematics. However, since full integration is going to take users many years, some of the synergistic "downstream" productivity enhancements will not be dramatic and may not result in immediate gratification and increased usage.

The major discrepancy between expectations and likelihood is the near term contribution of expert systems to productivity. To date there have been perhaps a thousand expert system attempts with, we speculate, fewer than 100 actually put into practice by users. Expert systems' applications must be carefully selected, and then generally take two years to put into use plus two more years to productize. Short term, dramatic productivity gains will not be coming from expert systems, although the long term potential is very significant. At Computervision, we expect that successful expert system applications will have the following attributes:

	a)	algorithmic solutions are non-existent or are very costly
	b)	good solutions are readily recognized
	c)	the consequences of bad decisions are not catastrophic
	d)	the utility of automating decision making is high and users won't object to automation
	e)	human experts are rare or too costly
	f)	available human experts can systematize their expertise
	g)	the domain of the data and rules is contained
	h)	the data must be easily representable, editable and subject to a stable rule base.

We believe that AI will, sometime in the next two years, experience a reduction in investment growth. However, we also believe that AI will alter the industrialized world over the next half century.

The viewpoints expressed here are borrowed liberally from our colleagues at Computervision and contributors to the literature to whom we owe great thanks. A few (of the more controversial) views are our own, and we take personal responsibility for them. These views are not necessarily the views of Computervision.

PRESENTED AT: The 4th Int'l MICAD Conference on March 1, 1985 in Paris, France