Question 3 of 7: How Design Activity Has Changed, and What Comes Next

Question 3: How Design Activity Has Changed, and What Comes Next (15 marks)

Question text not reproduced: the examination questions are © Engineers and Geoscientists BC. Open the official past paper (linked at the top of this page) to read the question, then follow the worked solution below.

Part A — Five ways design activity has changed (10 marks)

1. Design separated from making. In antiquity and through the medieval guilds, the designer and the maker were the same person, and the design existed only in the artefact and in the craftsman's hands. Knowledge passed by apprenticeship, and improvement was incremental because there was no medium in which to reason about an alternative before building it. The decisive change was the drawing: scaled orthographic projection in the Renaissance, formalised by Monge's descriptive geometry in 1795, allowed a design to be examined, criticised and transmitted before any material was cut. Design became an activity with its own output — the drawing — and, unavoidably, a division of authority between the person who decides and the person who makes.

2. Standardisation replaced fitting. Through the industrial revolution, parts were fitted individually to their mating parts; a replacement had to be made to suit. Interchangeable manufacture, screw-thread standards such as Whitworth's, and eventually national and international standards bodies changed what a designer produces from an instruction to fit into a specification with a tolerance. This is the birth of tolerancing as a design discipline and, with it, of the idea that a part is correct if it falls within a stated band rather than if it happens to work in one assembly. It also made supply chains possible: a specification can be sent to a supplier who has never seen the assembly.

3. Prediction replaced trial and error. The rise of engineering science — strength of materials, thermodynamics, fluid mechanics — and then of numerical methods from the 1950s and 1960s allowed a designer to know how a design would behave before building it. The economic effect is large and quantifiable.

Given. A design study with 5 factors at 2 levels each. A physical build-and-test run costs CAD 8,500 and 12 days; a validated simulation run costs CAD 350 and can be run 8 at a time.

Find. The cost of the full factorial study by each route, the saving available from a fractional factorial design, and the elapsed-time effect.

1. Cost the physical and simulated studies. A full factorial requires $2^{5}=32$ runs, so $$\begin{aligned}C_{\text{physical}}&=32\times 8{,}500=272{,}000 \\ C_{\text{simulated}}&=32\times 350=11{,}200\end{aligned}$$ in CAD, a factor of $\boxed{24.3}$ reduction, and the elapsed time falls from $32\times 12 = 384$ days run sequentially to $32\times 0.5/8=2.0$ days.
2. Note that statistical design of experiments compounds the saving. A resolution-III fractional factorial $2^{5-2}$ needs only $2^{3}=8$ runs, taking the physical study to CAD 68,000 at the price of confounding main effects with two-factor interactions. The modern practice — screen with a fractional design, simulate the promising region densely, then confirm physically — is only available because prediction became trustworthy.

4. Design became digital data rather than a document. Two-dimensional CAD in the 1970s merely automated the drawing board, but 3D parametric solid modelling, and then model-based definition and product lifecycle management, changed the nature of the artefact: the master is now a controlled data set from which drawings, tool paths, inspection programmes, service manuals and simulations are all derived. A dimension changed once propagates everywhere. The designer's work is now inseparable from configuration management, and the failure modes are new ones — an uncontrolled model, a broken feature tree, a supplier working from a stale release.

5. Design became a distributed, concurrent, lifecycle activity. The final shift is organisational rather than technical. Design is no longer a phase performed by designers and handed downstream; it is performed concurrently by a cross-functional team in which manufacturing, service, purchasing, regulatory and sustainability specialists participate from the start. Design for X — manufacture, assembly, environment, disassembly, serviceability — formalises obligations that used to be somebody else's problem, and product responsibility now extends past the sale to the end of life. Design activity also became global and asynchronous, which introduces the communication problems Question 6 addresses.

Part B — Two technologies that will facilitate the design process (5 marks)

Technology 1: generative design coupled to additive manufacture. Generative design inverts the designer's role. Instead of proposing a geometry and analysing it, the designer states the loads, the constraints, the keep-out volumes, the manufacturing process and the objective, and an optimiser proposes geometries. Coupling this to additive manufacture matters because the process constraint that has always disciplined design — that a shape must be mouldable, castable or machinable — is relaxed, so the optimiser's answers become buildable. The reason a search algorithm rather than an engineer must do this is simply the size of the space.

Given. A parametric model with 12 independent parameters discretised to 8 levels each. One finite-element evaluation takes 40 s of processor time, on a 64-core workstation.

Find. The elapsed time for an exhaustive search of the design space.

1. Size the exhaustive search. The number of combinations is $$N = 8^{12}=6.872\times 10^{10}$$ so the elapsed time is $$t = \frac{N \times 40}{64}=4.295\times 10^{10}\ \text{s}=\boxed{1{,}361\ \text{years}}$$ Exhaustive search is not merely expensive, it is impossible, which is why gradient-based topology optimisation and heuristic search — which sample a vanishing fraction of the space — are the only viable route, and why the designer's remaining job is to pose the problem well and to judge the candidates.

The challenges are real and professional. A generated geometry has no design intent a reviewer can follow, so verification cannot lean on engineering judgement about the shape; it must lean on independent analysis and test. The optimiser will exploit any error in the load case with enthusiasm, so a wrong boundary condition produces a confidently wrong part. And an engineer still seals the design: responsibility does not transfer to the algorithm, which means the burden of proof rises rather than falls.

Technology 2: the connected product and its digital twin. Instrumented products returning duty-cycle data from the field close the loop that has been open since design separated from making. The specification for the next generation stops being an estimate of how customers will use the product and becomes a measurement of how they actually did: real load spectra replace assumed ones, so safety factors can be reduced where the assumption was conservative and increased where it was optimistic; failures are diagnosed against the exact history of the unit that failed; and features can be validated by observing whether anyone uses them. A digital twin — a model of the individual unit kept current with its telemetry — extends this to predictive maintenance and to remote calibration.

Its challenges are equally real: personal-data and consent obligations under Canadian privacy legislation, security of an update path that must be maintained for the whole hardware life rather than the software's, a dependency on cloud services that demands a graceful-degradation requirement in the specification, and the design discipline of deciding what to measure before the product ships, because a sensor omitted at design freeze cannot be added later. Both technologies point the same way: the designer's leverage shifts from producing geometry to framing problems, specifying constraints honestly, and judging evidence.