Probabilistic Design and Reliability | Module 8 of 10

Robust Design

A robust design does not merely perform well at the nominal point. It keeps acceptable performance when loads, dimensions, materials, and manufacturing conditions vary.

01

Readiness check

Tick only what you can do closed-notes before using this module for design decisions.

  • Explain sensitivity as output change caused by input change.
  • Compare two designs using mean and standard deviation.
  • Recognize a tolerance as allowed variation.
  • State a trade-off among cost, performance, and manufacturability.
  • Use a small parameter sweep.
0 or 1 weak itemsContinue with this module.
2 weak itemsReview optimization formulation in Optimization, Module 1.
3 or more weak itemsReview tolerances in Engineering Graphics and CAD, Module 13.
02

The core idea

Robust design chooses variables and tolerances so performance is acceptable and insensitive to variation. It balances mean performance, variation, cost, reliability, and manufacturability.

robust objective example: minimize mean(loss) + k var(response)signal-to-noise idea: larger useful signal, smaller noise responsetolerance trade-off: tighter variation costs money

Nominal optimization can produce a design that looks excellent at one point and fragile everywhere else. Robust design asks a different question: how does performance move when the real inputs vary? A design with slightly worse nominal performance can be better if its output scatter is much smaller or if it stays farther from a failure boundary. Robustness is not the same as reliability. A robust design can be consistently unacceptable if its mean is on the wrong side of the limit. A reliable design can still be sensitive but have enough margin. Good design often needs both: move the mean away from failure and reduce the spread. Taguchi methods introduced useful signal-to-noise language, but they are not the only robust-design method. Modern robust design also uses sensitivity analysis, tolerance allocation, probabilistic simulation, and optimization.

The skill works when: the failure event, units, assumptions, distributions, and checks are visible.
The skill breaks down when: a probability is reported without the mechanical model and evidence boundary.
The concept. Robust design chooses variables and tolerances so performance is acceptable and insensitive to variation. It balances mean performance, variation, cost, reliability, and manufacturability.
03

The skills, taught in order

Compare mean and scatter

Judge designs by both expected performance and variation.

Find controllable variables

Separate design choices from noise variables such as loads, material scatter, and environment.

Reduce sensitivity first where possible

A geometry change that reduces sensitivity can be cheaper than tightening every tolerance.

Allocate tolerances deliberately

Tighten only the dimensions that strongly affect performance or failure probability.

Keep reliability visible

Robustness metrics must be checked against the actual failure boundary.

Engineering connection: use these skills to turn variability into a design decision, not just a plot.

04

Worked example 1: spring stiffness variation

Two spring designs have target stiffness 100 N/mm. Design A has mean 100 and standard deviation 8 N/mm. Design B has mean 97 and standard deviation 3 N/mm. The acceptable range is 90 to 110 N/mm. Which design is more robust, and what must still be checked?

  1. ProblemTwo spring designs have target stiffness 100 N/mm. Design A has mean 100 and standard deviation 8 N/mm. Design B has mean 97 and standard deviation 3 N/mm. The acceptable range is 90 to 110 N/mm. Which design is more robust, and what must still be checked?
  2. AssumptionsStiffness distributions are approximately normal and centered as stated.
  3. Modelrobustness screen: smaller output standard deviationreliability check: P(90 <= K <= 110)
  4. SolveDesign B is more robust by output scatter because its standard deviation is 3 instead of 8. Its mean is lower, but still inside the acceptable band.
  5. CheckA robustness label is not enough. Compute probability outside 90 to 110 if the decision matters.
  6. ConclusionDesign B is less sensitive to variation, but final selection requires an acceptance probability and cost check.
Result. Design B is less sensitive to variation, but final selection requires an acceptance probability and cost check.
05

Worked example 2: tolerance versus geometry change

A shaft stress model has tau proportional to 1/d^3. A team can either tighten diameter tolerance by 50 percent or increase nominal diameter by 2 percent. Which action directly improves mean margin, and which mainly reduces scatter?

  1. ProblemA shaft stress model has tau proportional to 1/d^3. A team can either tighten diameter tolerance by 50 percent or increase nominal diameter by 2 percent. Which action directly improves mean margin, and which mainly reduces scatter?
  2. AssumptionsTorque and material strength are unchanged.
  3. Modelmean stress falls when nominal d increasesdiameter scatter contribution falls when tolerance tightens
  4. SolveIncreasing nominal diameter reduces mean stress and moves the design away from failure. Tightening tolerance mainly reduces stress scatter from diameter variation. Both can help, but they act differently.
  5. CheckIf failure probability is driven by low diameter tail, either action may help. Cost and packaging decide the better move.
  6. ConclusionRobust redesign separates mean-shifting decisions from scatter-reducing decisions.
Result. Robust redesign separates mean-shifting decisions from scatter-reducing decisions.
06

Python activity

Engineering question: Compare two candidate shaft diameters under manufacturing variation.

Assumptions and units: Use the units shown in the code comments. Keep the random seed fixed while learning so the result is reproducible.

from math import pi
from random import Random
from statistics import mean, stdev

rng = Random(23)
n = 100_000

for nominal_d, tol_sd in [(28.0, 0.35), (29.0, 0.35), (28.0, 0.18)]:
    margins = []
    failures = 0
    for _ in range(n):
        torque = rng.gauss(400_000.0, 35_000.0)
        strength = rng.gauss(220.0, 18.0)
        d = rng.gauss(nominal_d, tol_sd)
        tau = 16.0 * torque / (pi * d**3)
        g = 0.58 * strength - tau
        margins.append(g)
        if g <= 0.0:
            failures += 1
    print("d", nominal_d, "sd", tol_sd, "mean g", round(mean(margins), 2), "sd g", round(stdev(margins), 2), "Pf", round(failures / n, 5))
Numerical checks: Increasing nominal diameter should improve mean margin. Tightening tolerance should reduce margin scatter.
Limitations: This simplified study does not include weight, packaging, cost, fatigue, or stress concentration.
07

Misconceptions and diagnostics

MistakeDiagnostic questionCorrection
Reliability and robustness are the same.Ask: is the design insensitive to variation, or is it unlikely to cross a failure boundary?Correct the assumption, then rerun the calculation or rewrite the report.
Robust design always means tighter tolerances.Ask: can geometry or architecture reduce sensitivity more cheaply?Correct the assumption, then rerun the calculation or rewrite the report.
Taguchi methods are the only robust-design method.Ask: what sensitivity, tolerance, or probabilistic method is appropriate here?Correct the assumption, then rerun the calculation or rewrite the report.
The best nominal design is the best real design.Ask: how does it perform under input variation?Correct the assumption, then rerun the calculation or rewrite the report.
08

Practice ladder

Recognize

State why a design with perfect nominal performance can be fragile.

Solution guidance

Write the governing equation, state units and assumptions, then compare the answer with the relevant limit state or design decision.

Calculate

If output standard deviation drops from 8 to 4, by what factor did variance change?

Solution guidance

Write the governing equation, state units and assumptions, then compare the answer with the relevant limit state or design decision.

Diagnose

A team tightens all tolerances equally. Explain why this may waste cost.

Solution guidance

Write the governing equation, state units and assumptions, then compare the answer with the relevant limit state or design decision.

Design

Propose one geometry change and one tolerance change that could improve shaft robustness.

Solution guidance

Write the governing equation, state units and assumptions, then compare the answer with the relevant limit state or design decision.

09

Working with AI, and proving it yourself

Useful AI support

Ask for alternative explanations of terminology, candidate code tests, or a checklist of assumptions to verify.
Ask it to challenge a derivation, then prove the result with units, limiting cases, and an independent calculation.
Do not let AI invent material distributions, fabricate failure data, claim standards compliance, or make final safety decisions.

Portfolio evidence

Create one short reliability note for this module. It must include the engineering question, assumptions, input definitions, equations, calculation, checks, interpretation, and limitations.

Verification required: dimensional analysis, a benchmark case, convergence or sample-size check, and a plain-language residual-risk statement.
10

Retrieval and spaced review

Closed notes. Answer out loud, then reveal.

What is robust design?

Designing for acceptable performance with low sensitivity to variation.

How is robustness different from reliability?

Robustness concerns variation sensitivity; reliability concerns probability of satisfactory performance.

What are controllable variables?

Design variables the engineer can choose.

What are noise variables?

Sources of variation such as load, environment, material scatter, and manufacturing variation.

Why keep cost visible?

Reducing variation often costs money or manufacturability margin.

TodayFinish the review and first two practice levels.
+1 dayRebuild one worked example from a blank page.
+3 daysChange one assumption and predict the effect.
+7 daysConnect this module to the capstone bracket.
+30 daysReuse the method in a new mechanical component.
11

References and source discipline

This course synthesizes original MechCompass explanations from established reliability engineering sources. Confirm current editions and standards before using them for formal design approval.

Topic in this moduleReference direction
Robust and reliability-based mechanical designLe, Reliability-Based Mechanical Design, Volume 1
Design under uncertaintyHaldar and Mahadevan, Probability, Reliability, and Statistical Methods in Engineering Design
Optimization trade-offsNocedal and Wright, Numerical Optimization, for optimization foundations