Electronics products, especially computers, have been offering truly astounding rates of improvement. Prices are declining fast, making products more and more affordable; they are easy, if not easy enough, to use and each year they are smaller and use less power.

There are only a few hundred thousand electronics engineers world-wide - not many more than 30 years ago, but the volume of new products these engineers design has increased, on any reasonable measure, by two orders of magnitude. This gain in design productivity is enabled by a continuous revolution in the design process, supported by generation after generation of Electronic Design Automation software tools.

In the early 70s, electronic design focused on documenting the layout of silicon or printed circuit boards, then testing a prototype of the design. By 1980, the process had moved to design of the logic needed for the product. Automated tools generated the physical layout from logic diagrams and net lists that represented the design logic. The resulting products were simulated before they were manufactured. By 1990, design had moved to a higher level of abstraction. Product requirements to be implemented in silicon could be described in a high level language, then simulated and checked. Both the design logic and physical layout could be synthesised and verified as meeting the design requirements. Today, innovation continues with still higher levels of abstract models. This frees designers to work on the things they are good at, and eliminates the need for designers to keep abreast of the latest manufacturing technologies. The design of products at an abstract level lets designers cope without needing to visualise and understand the actual, very complex, layout of circuits on a chip.

When we think about applying the same ideas to mechanical design, we immediately hit barriers. Mechanical design is inherently more complex. Electronics is all about binary decisions - 0 or 1? Mechanical functions are continuous. We can damp a movement either with a spring or with a piston - the same effect but quite different implementations. Maybe mechanical design automation is simply too hard?

I don't believe so. I think we have to try harder. As Peter Marks of Design Insight has pointed out, a lot of mechanical design is a repeat of past designs. Just as production measures itself by inventory turns, design should measure itself by knowledge turns, reusing past design. For years, super-users of mechanical design authoring tools have written macros in languages like GRIP and BaCIS to generate geometry for some family of parts from a series of parameters. But these are procedural languages to drive the CAD package, not modelling languages to describe requirements in abstract terms and record manufacturing rules to control the geometry produced. Now, there is renewed interest in mechanical design automation, sometimes referred to as Knowledge Based Engineering. Late last year, Dassault Systèmes purchased the British privately held company, KTI. Dassault had been developing and promoting "Knowledgeware", a modernised procedural approach. KTI's ICAD product has been around for about 15 years. It uses a rule based approach.

This is the visible tip of an iceberg of a new generation of companies offering various mechanical design automation tools. Invention Machine, also partly owned by Dassault, helps engineers re-state the design requirements, so that they can consider alternative design solutions. Granta Design helps engineers consider designs using different, innovative materials. Engineous helps optimise trade-offs between different simulations for stress and vibration. Flomerics similarly helps the trade off between thermal cooling and electro-magnetic radiation in electronics enclosures. Driveworks, IcedCAD, and Rulestream capture the processes used to create complex existing designs, and then help re-apply them to to new requirements. A Driveworks example is a custom tanker body built on a standard truck chassis. An initial version of this engineer-to-order design, including calculations, costings, material take offs and construction drawings, can be generated after an interactive dialogue between designer and customer.

Historically, it has been difficult and expensive to create rules of design. Now, approaches like wizards have simplified the user interface. By narrowing the problem space, either to a specific discipline like mould design, or to a specific industry like mechanical handling, it is possible to use the context to make rules easier to generate. Within these constraints it is possible to use tools that optimise specific design problems - such as Moldflow to obtain quality plastic moulding solutions.

But today, all these tools are unconnected. Each one only helps part of the process. They don't fit into the kind of revolution in design process that electronics has seen. If we are to make serious use of mechanical complexity, such as nano-technology, improvements are necessary.

Many mechanical design departments are already working to improve their design flows. Best practices include:

" Separating research from development. When product development schedules are disrupted by unpredictable research outcomes, profits are often hit by delayed product launches. By focusing research on building blocks that correspond to particular components of the product architecture, the latest generation of the product can simply use the latest proven version of the component.

" Creating a clear product architecture with a platform product, and modules to provide a range of functions and options.

" Designing for continuous improvement. For example, to take advantage of the fact that the next generation of drives will be a few per cent smaller and use less energy.

" Developing a product portfolio plan to ensure that development projects in design come to fruition at appropriate intervals. Conducting reviews to make sure that the projects will deliver appropriate functionality.

The challenge is to incorporate these best practices into a design flow which is moving towards mechanical design automation. We are not going to get there in one step. Alan Grayer, the A of the ACIS kernel solid modeller, observed that mechanical design automation is rather like computer chess: standard openings (Invention Machine/ Granta Design); brute force and pattern recognition in the middle; algorithmic end games (ICAD, Driveworks, IcedCAD, Rulestream)

It is this middle section where new thinking is required. We need new model description languages and notations. They would support optimisation processes where different design trade-offs are analysed and simulated to find the best match to the design requirements. If we can start in narrow, better defined areas to establish the principles, we can expand our techniques later to a wider range of problems. We can then expect to see a variety of improved products - from lawn mowers to printing presses.

Mike Evans

A version of this article was first published in the July 2003 issue of EAR

back to top