In one line: Reliability is failure-mode avoidance: designing so the product keeps delivering its ideal function, for every customer, in every condition, over its whole life.
Core idea
Don Clausing defined a failure mode as “any customer-perceived deviation from the ideal function of the product” [1]. On that view, reliability isn’t a separate statistic computed at the end. It is quality viewed over time, and it is mostly decided in design. Failure modes are either soft (performance degrades) or hard (the function stops), and they have two root causes:
- Lack of robustness: the design is too sensitive to noise, such as variation in parts, environment, use and wear.
- Mistakes: design errors, wrong assumptions, or misunderstood requirements.
How it works
- Attack robustness with parameter design: choose design settings that are insensitive to noise (see Robust Design), and test under realistic stresses rather than nominal conditions.
- Attack mistakes with disciplined processes:
- failure-mode analysis (see Failure Mode and Effects Analysis)
- design reviews
- checklists and mistake-proofing
- learning from prior failures
- Measure in the language of the customer. Field returns, warranty data, mean time between failures, and failure rate over life all matter. The classic bathtub curve shows three phases:
- early “infant mortality” failures, often from manufacturing defects
- a flat useful-life period with a low, roughly constant failure rate
- rising wear-out failures at end of life
- Test to learn, not to pass. Accelerated and stress testing exposes failure modes early. Tests designed only to demonstrate compliance tend to miss them.
In practice
Engineers often meet reliability as a specification number assigned late in a program. Clausing’s framing moves it upstream: every design choice either adds or removes sensitivity to noise. A practical habit is to ask, at each design review, which noise factors the design has been exposed to so far, and which failure modes the team has actively looked for, not just which tests it has passed.
Questions to ask yourself
- Which noise factors (temperature, supply voltage, part tolerance, user behavior) has my design actually been tested against?
- Are our tests designed to find failures, or to show we have none?
- Which past field failure would our current process catch, and which would it miss?
Connections
- Main strategies: Robust Design, Failure Mode and Effects Analysis.
- Tools: Statistics in Engineering, Design of Experiments. Why it’s hard: Complexity.
- Capabilities exercised: Responsibility and Urgency to Deliver, Inquiry, Ethical Actions and Integrity.