Subjects · Engineering & Technology

Engineering & Technology: Parameter sweep

Know which way things move and how fast when you change something, which is design intuition, and is what separates an engineer from someone who can evaluate a formula.

What you'll be able to do: Know which way things move and how fast when you change something, which is design intuition, and is what separates an engineer from someone who can evaluate a formula.

Why this is design, not analysis

Analysis answers: given this, what happens? Design asks: what should I change, and by how much?

The second requires knowing the shape of the relationship. Deflection goes with the cube of depth, so doubling the depth divides deflection by eight, which means depth is the lever and width barely matters. That single fact is worth more than the deflection formula, because it tells you where to spend material.

Students hold these relationships as algebra and not as behaviour. Ask them what happens if the span doubles and they reach for a calculator. An engineer says "sixteen times the deflection" immediately, because span is to the fourth power and they know it as a fact about beams rather than as an exponent in an equation.

The loop

  1. Fix everything, choose one parameter.
  2. Predict: direction first, then the power. Linear? Squared? Cubed? Inverse?
  3. See it.
  4. Note whether you were wrong about direction, rate, or the extremes.
  5. Push to the boundaries deliberately. What happens at zero, at very large, at the point where the model stops applying? The edges are where the scope conditions live and where the interesting design decisions are.
I'm learning [relationship]. Walk me through what happens as [parameter] increases, one step at a time. Before each step, ask me to predict the direction and the power, and wait. Then tell me what happens and whether my REASONING was right, not just my answer.

The relationships worth knowing cold

These are the ones practising engineers carry as intuition:

Structural. Deflection ∝ L⁴ and ∝ 1/d³. Buckling load ∝ 1/L². Stress in bending ∝ 1/d². This is why beams are deep and why long columns are a different problem from short ones.

Fluids. Pressure drop ∝ v² in turbulent flow, ∝ v in laminar. Pump power ∝ Q³ in a fixed system. Pipe capacity ∝ d^2.5-ish. Doubling a pipe diameter is a much bigger change than people expect.

Electrical. Power loss ∝ I², which is the whole argument for high-voltage transmission. Impedance and frequency, RC time constants.

Thermal. Radiation ∝ T⁴, this dominates at high temperature and is negligible at low, which is why the design changes character. Conduction ∝ 1/L.

Chemical. Rate roughly doubles per 10 K, which is why cooling failure is catastrophic rather than gradual (article 04). Surface-to-volume ∝ 1/L, which is the entire problem of scale-up.

Rotating machinery. Fan laws, flow ∝ N, pressure ∝ N², power ∝ N³. A 10% speed increase is a 33% power increase.

Software and systems. Latency under load is not linear; it's flat then vertical. Anyone who has only seen the flat part will size the system wrong.

The scale-up question

The most valuable version of this method in engineering, and one students rarely meet:

What happens to [system] if I make it ten times bigger? Which quantities scale with length, which with area, which with volume, and what breaks first?

Square-cube behaviour is why a scaled-up model of a working design frequently doesn't work: strength goes with area and weight with volume, so the thing gets relatively weaker as it grows. Heat generation with volume and dissipation with area is why big reactors have cooling problems that small ones don't.

This is a genuinely engineering-specific insight and it comes out of a parameter sweep on length.

Pitfalls

  1. Watching without predicting. Then it's an animation.
  2. Varying more than one thing. Nothing is attributable.
  3. Predicting direction only. The power is where the design value is.
  4. Staying in the comfortable range. Boundaries are where the model changes.
  5. Trusting the sweep past the model's validity. Laminar correlations applied to turbulent flow produce smooth, confident, wrong curves.
  6. The tell: you can describe the relationship and reach for a calculator to answer "what if I double it".

Try this today

Take a relationship you use. Ask to be walked through one parameter increasing, predicting the direction and the power before each step.

Then ask the scale-up question: ten times bigger, what breaks first?