Home / Courses / Probabilistic Design and Reliability / Why Deterministic Design Is Not Enough

Probabilistic Design and Reliability | Module 1 of 10

Why Deterministic Design Is Not Enough

A nominal safety factor says what happens at one assumed point. Reliability asks how often the real population of parts, loads, strengths, and operating conditions crosses a failure boundary.

01

Readiness check

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

  • Explain a factor of safety as resistance divided by load effect.
  • Read a mean and standard deviation from a small data set.
  • Identify a mechanical failure mode for a shaft, beam, or fastener.
  • Separate a design requirement from a calculation assumption.
  • Use units consistently in stress and strength calculations.
0 or 1 weak itemsContinue with this module.
2 weak itemsRevisit safety factors in Machine Elements, Module 1.
3 or more weak itemsReview statistics in Mathematics, Module 15 and stress in Mechanics of Materials, Module 1.
02

The core idea

A deterministic design check uses one load, one strength, and one geometry. A probabilistic design check treats those quantities as populations, defines failure, and estimates the probability that a real part falls into the failure region.

n = R_nominal / S_nominalg(X) = resistance(X) - load_effect(X)P_f = P[g(X) <= 0], reliability = 1 - P_f

Safety factors are still useful because they give a quick margin against a selected design case. They do not, by themselves, say how likely failure is. Two shafts can share the same nominal safety factor while one has tightly controlled material and load variation and the other has broad scatter. The first can have a much lower failure probability even though the deterministic number matches. Reliability adds the missing population view: define the failure mode, describe uncertainty in the inputs, propagate those inputs through the performance function, and interpret the resulting probability together with consequence. Robustness is related but different. A robust design changes little when inputs vary. A reliable design stays on the safe side of a defined failure boundary with acceptable probability. Risk combines probability with consequence, so a rare failure with severe consequence can still require action.

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. A deterministic design check uses one load, one strength, and one geometry. A probabilistic design check treats those quantities as populations, defines failure, and estimates the probability that a real part falls into the failure region.
03

The skills, taught in order

Nominal value versus population

A nominal value is a chosen representative number. A population contains the actual spread of strengths, dimensions, loads, temperatures, and use conditions.

Factor of safety versus reliability

The factor of safety compares selected values. Reliability integrates the overlap between load effects and resistance across their distributions.

Failure mode first

Reliability has no meaning until the failure event is defined: yield, fracture, excessive deflection, leakage, seizure, loss of preload, or missed function.

Robustness and risk

Robustness measures sensitivity to variation. Risk adds consequence to probability. Do not collapse these into one word.

Evidence boundary

A calculated failure probability is credible only inside the data, assumptions, and model checks that support it.

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

04

Worked example 1: same nominal safety factor, different reliability

Two steel rods carry a nominal axial load of 40 kN and both have nominal yield resistance of 80 kN, so each has n = 2. Design A has resistance standard deviation 4 kN and load standard deviation 3 kN. Design B has resistance standard deviation 12 kN and load standard deviation 9 kN. Estimate failure probability assuming independent normal resistance and load.

  1. ProblemTwo steel rods carry a nominal axial load of 40 kN and both have nominal yield resistance of 80 kN, so each has n = 2. Design A has resistance standard deviation 4 kN and load standard deviation 3 kN. Design B has resistance standard deviation 12 kN and load standard deviation 9 kN. Estimate failure probability assuming independent normal resistance and load.
  2. AssumptionsResistance R and load L are independent normal variables. Failure is R - L <= 0. This is a first screen, not a certification calculation.
  3. ModelM = R - Lmu_M = mu_R - mu_Lsigma_M = sqrt(sigma_R^2 + sigma_L^2)P_f = Phi(-mu_M / sigma_M)
  4. SolveBoth designs have mu_M = 80 - 40 = 40 kN. Design A has sigma_M = sqrt(4^2 + 3^2) = 5 kN, so beta = 40/5 = 8 and P_f is about 6.2e-16. Design B has sigma_M = sqrt(12^2 + 9^2) = 15 kN, so beta = 40/15 = 2.67 and P_f is about 0.0038.
  5. CheckThe deterministic safety factor is identical, but the margin distribution is much wider for Design B. Wider scatter makes the lower tail cross zero more often.
  6. ConclusionThe same nominal factor of safety can hide orders of magnitude difference in failure probability.
Result. The same nominal factor of safety can hide orders of magnitude difference in failure probability.
05

Worked example 2: turning a safety check into a performance function

A bracket is acceptable if maximum bending stress is below yield strength. Nominal yield is 250 MPa and nominal stress is 125 MPa. Define the deterministic factor, the performance function, and the failure event.

  1. ProblemA bracket is acceptable if maximum bending stress is below yield strength. Nominal yield is 250 MPa and nominal stress is 125 MPa. Define the deterministic factor, the performance function, and the failure event.
  2. AssumptionsYield failure is the relevant first mode. Local stress concentration and fatigue are outside this first model.
  3. Modeln = S_y / sigmag(X) = S_y(X) - sigma(X)failure: g(X) <= 0
  4. SolveThe nominal factor is n = 250/125 = 2. The probabilistic model does not use n as the random output. It uses g = S_y - sigma, with failure when the random stress exceeds the random yield strength.
  5. CheckIf both S_y and sigma were fixed at nominal values, g = 125 MPa and failure would be impossible in the model. That is exactly why uncertainty must be represented.
  6. ConclusionThe deterministic check becomes a limit-state model once inputs are allowed to vary.
Result. The deterministic check becomes a limit-state model once inputs are allowed to vary.
06

Python activity

Engineering question: Compare two designs with equal nominal safety factor but different scatter.

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 erf, sqrt

def normal_cdf(x):
    return 0.5 * (1.0 + erf(x / sqrt(2.0)))

cases = {
    "A": {"mu_r": 80.0, "sd_r": 4.0, "mu_l": 40.0, "sd_l": 3.0},
    "B": {"mu_r": 80.0, "sd_r": 12.0, "mu_l": 40.0, "sd_l": 9.0},
}

for name, c in cases.items():
    mu_m = c["mu_r"] - c["mu_l"]
    sd_m = (c["sd_r"] ** 2 + c["sd_l"] ** 2) ** 0.5
    beta = mu_m / sd_m
    pf = normal_cdf(-beta)
    print(name, "beta =", round(beta, 3), "Pf =", pf)
Numerical checks: The nominal safety factor is two in both cases. The reliability result changes only because the scatter changes.
Limitations: The normal and independence assumptions must be justified with data or engineering evidence before use in a safety decision.
07

Misconceptions and diagnostics

MistakeDiagnostic questionCorrection
A safety factor automatically tells me the reliability.Ask: what scatter in load, strength, geometry, and environment did that single number include?Correct the assumption, then rerun the calculation or rewrite the report.
Reliability is one minus the safety factor.Ask: is the factor a probability? It is not. Reliability is one minus failure probability.Correct the assumption, then rerun the calculation or rewrite the report.
A low calculated failure probability proves the design is safe.Ask: does the model include the governing failure mode and credible input distributions?Correct the assumption, then rerun the calculation or rewrite the report.
Uncertainty can be removed simply by collecting more samples.Ask: is the uncertainty aleatory variation, epistemic model uncertainty, or both?Correct the assumption, then rerun the calculation or rewrite the report.
08

Practice ladder

Recognize

State the difference between a nominal load and a load distribution.

Solution guidance

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

Calculate

For R = 120 kN and L = 60 kN, compute the nominal factor of safety.

Solution guidance

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

Diagnose

A report says n = 1.8, therefore reliability = 0.44. Identify the error.

Solution guidance

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

Design

Choose one mechanical component and define one failure event that could be analyzed probabilistically.

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 the performance function used in this course?

g(X) = resistance(X) - load_effect(X), with failure when g(X) <= 0.

What does a factor of safety compare?

Selected deterministic values of resistance and load effect.

How are reliability and risk different?

Reliability is probability of satisfactory performance; risk combines probability and consequence.

Can two designs with the same safety factor have different failure probabilities?

Yes, if their input scatter or distribution shapes differ.

What should be defined before any reliability calculation?

The failure mode and the consequence context.

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
Reliability framing and performance functionsHaldar and Mahadevan, Probability, Reliability, and Statistical Methods in Engineering Design
Engineering statistics and distribution checksNIST/SEMATECH Engineering Statistics Handbook
Risk and design reliability contextDer Kiureghian, Structural and System Reliability