Is a positive average safety margin enough?

No. Variation can still place some demand above capacity.

Is failure probability the common area under two density curves?

Generally no. It is the probability of the defined ordering event.

Can correlated load and strength be analyzed?

Yes, but their dependence must be included in the joint model.

AI Glossary / Statistics and probability / Intermediate

Statistical interference

Definition

Statistical interference, in reliability and tolerance analysis, concerns the possibility that random demand exceeds random capacity, or that two varying quantities cross a critical ordering. A common example is stress-strength interference: failure occurs when applied stress is greater than available strength. The analysis uses their joint probability model to calculate that event's probability. This specialized engineering term is distinct from statistical inference, which concerns learning from data, and from interference between participants in causal studies.

What it means

Comparing average stress with average strength is not enough when both quantities vary. A favorable average margin can coexist with a meaningful chance that a particular demand is unusually high or a particular capacity unusually low.

Plots of the two distributions can help explain the concern, but the probability of stress exceeding strength is not generally the geometric area shared by their density curves. It depends on the distribution of their difference and on their dependence. Two systems with identical marginal stress and strength distributions can have different failure probabilities when the pairing between them differs.

How it works

Define the failure criterion and the operating context. Let S denote strength or capacity and L denote load or stress, measured on compatible scales. For the simple ordering criterion, define margin M = S - L. Failure occurs when M < 0, with equality handled according to the physical definition when it has nonzero probability.

Specify or estimate the joint distribution of S and L. If they are independent normal variables, M is normal with mean mu_S - mu_L and variance sigma_S squared + sigma_L squared. The failure probability is Phi((mu_L - mu_S)/sqrt(sigma_S squared + sigma_L squared)), where Phi is the standard normal cumulative distribution function.

For jointly normal dependent variables, subtract twice their covariance in the margin variance. Other distributions or more complex failure surfaces may require integration or simulation. Assess uncertainty in distributional parameters and the validity of tail assumptions. A static stress-strength calculation does not automatically describe repeated loading, fatigue, degradation, repair, or every failure mechanism of a complete system.

Three practical examples

1. A component designer compares a distribution of material capacity with a distribution of operating load under a specified failure rule.

2. A manufacturing team evaluates whether variation in a shaft and hole dimension creates an unacceptable fit, using the relevant difference and dependence assumptions.

3. A capacity planner studies the chance that uncertain demand exceeds available resources, treating the calculation as an analogy that still needs a suitable operational model.

Think of it this way

Think of a bridge clearance and the height of vehicles passing underneath. Average clearance exceeding average vehicle height does not guarantee every passage is possible. The relevant question concerns the chance that a particular height exceeds the available clearance under the actual pairing and conditions.

Worked example

Consider an original simplified model with independent normal strength S having mean 100 and standard deviation 10, and load L having mean 80 and standard deviation 5. Both are expressed in the same arbitrary engineering units. Failure is defined as L > S.

The margin S - L has mean 100 - 80 = 20 and variance 10 squared + 5 squared = 125. Its standard deviation is sqrt(125), approximately 11.1803. The standardized zero margin is (0 - 20)/11.1803, approximately -1.78885.

The failure probability is therefore Phi(-1.78885), approximately 0.03682, or 3.682%. Reliability for this single specified comparison is approximately 96.318%. The positive mean margin does not make the failure probability zero.

If the same marginal variables are jointly normal with correlation 0.5, their covariance is 0.5 times 10 times 5 = 25. The margin variance becomes 125 - 2 times 25 = 75, changing the failure probability to Phi(-20/sqrt(75)), approximately 1.046%. This comparison demonstrates that dependence matters; it is not a claim that positive dependence always improves every reliability problem. The normal assumptions and parameters are stipulated solely for teaching.

Illustrative case study

In a fictional equipment company, engineers present a component with average capacity comfortably above average load. A reviewer asks about variation, common environmental influences, and whether extreme loads occur repeatedly.

The team builds a stress-strength model for one clearly defined condition, tests its assumptions against relevant data, and separately evaluates degradation and repeated exposure. Management receives a conditional probability estimate with uncertainty and scope limitations. The analysis supports design decisions without turning one simplified interference calculation into a comprehensive safety claim.

Common misconceptions and mistakes

Do not confuse statistical interference with statistical inference. Do not estimate failure probability by visually measuring the overlap between two density curves; that overlap coefficient answers a different question. Define the event and calculate its probability from the joint model.

Independence is an assumption, not a consequence of plotting separate distributions. Normal distributions may be poor models for positive quantities or extreme tails, even when they approximate central variation reasonably. A fitted failure probability can also be very uncertain when few tail observations are available. Finally, a probability of failure must be paired with the relevant exposure and consequences before it can inform a broader risk decision.

Business use

Interference analysis helps organizations evaluate whether reducing demand variability, increasing capacity, or changing dependence could improve performance. It supports sensitivity studies and prioritization of better data collection. State the units, exposure, failure criterion, distribution assumptions, and parameter uncertainty. For consequential engineering decisions, combine the statistical calculation with validated physical models and qualified engineering review.

Check your understanding

Independent normal capacity and demand have standard deviations 6 and 8. What is the standard deviation of their difference?

Answer: It is sqrt(6 squared + 8 squared) = 10. With dependence, the covariance term must also be included.

Related terms

Reliability
Covariance
Monte Carlo simulation
Stress-strength analysis
safety margin
limit-state function
tail probability
tolerance analysis

Learn next

Learn distributions of differences, then compare independent and dependent stress-strength models. Study parameter uncertainty and repeated-exposure models before applying the method to engineering decisions.

Sources

Timothy Adams, NASA Kennedy Space Center, How to do Quantitative Reliability, static stress-strength analysis: https://extapps.ksc.nasa.gov/Reliability/Documents/Presentation_How_to_do_Quantitative_Reliability.pdf

The primary NASA teaching material was consulted. The normal-margin derivation, numerical examples, and fictional case are original. This entry uses the engineering interference meaning and does not assert product safety or a validated failure rate.

Previous: Statistical hypothesis test Next: Statistical learning theory