In one line: Make the product’s performance insensitive to the variation it will meet in manufacturing and use, instead of trying to eliminate the variation itself.
Core idea
Real products face noise: variation in parts and materials, manufacturing processes, environment, usage and wear. A robust design delivers its target performance despite that noise. The idea, associated especially with Genichi Taguchi, is that the cheapest way to quality is often not tighter tolerances but better design parameter choices, settings at which the output barely changes when the noise does [1]. That makes robust design a central strategy for Reliability.
How it works
- Identify the performance metric to optimize, and its target.
- Classify the factors:
- control factors: design parameters the team can set
- noise factors: variation the team can’t control, or can only control at high cost
- Design the experiment. Use an orthogonal array to test many factors at several levels in a small number of runs (see Design of Experiments). Expose each run to the noise.
- Run the experiment and record results.
- Analyze:
- compute the main effect of each factor on the mean;
- compute its effect on variability, often as a signal-to-noise ratio;
- choose settings that hit the target with the least variation.
- Select and confirm the settings with a confirmation run.
In practice
The program’s lab makes this tangible with paper airplanes. Four factors (weight position, stabilizer flaps, nose length and wing angle), each at three levels, would need 81 combinations to test exhaustively. A nine-run orthogonal array estimates each factor’s effect on flight distance and consistency in a fraction of the effort. Readings on reliability and statistics make the broader point: failures are usually caused by a lack of robustness or by mistakes, and variation, not average performance, is what customers experience as poor quality.
Questions to ask yourself
- Which noise factors does my product face in the field, and have we tested against them?
- Are we trying to buy quality with tighter tolerances when a better design point exists?
- Which design parameter most affects our product’s variability?
Connections
- Method: Design of Experiments. Purpose: Reliability.
- Foundations: Statistics in Engineering. Complements: Failure Mode and Effects Analysis, Six Sigma.
- Capabilities exercised: Realizing the Vision, Inquiry, Decision Making.
Sources
- Ulrich and Eppinger - Product Design and Development [1]
- Davis - Science, Engineering, and Statistics [2]
- Clausing - Reliability [3]