How Do You Assess Whether an Attribute Gage Can Detect a Defect? (Signal Detection MSA)

In manufacturing, attribute gages (go/no‑go gauges, visual inspectors, or automated pass/fail sensors) are often treated as simple binary tools. But a critical question remains: how sensitive is your inspection to the defects that matter? If an inspector or gauge fails to catch a nonconforming part, the cost appears downstream. Signal detection MSA answers this by measuring the probability of detection for attribute inspection systems.

What It Is

Signal detection MSA (also called susceptibility assessment or defect detection capability analysis) is a method from the AIAG Measurement Systems Analysis (MSA) manual for attribute inspection systems. Unlike variable MSA (which studies bias and repeatability on continuous data), signal detection focuses on whether the inspection method reliably identifies a known defect when it is present.

The core output is a detection probability — the percentage of times a known defective part is correctly rejected. This is not about "how big" the defect is, but about the system's ability to sense a signal (the defect) against noise (normal process variation, lighting, inspector fatigue, etc.).

How It Works / Steps

The AIAG approach for signal detection typically follows this procedure:

1. Select samples — Obtain parts that are known to be good (conforming) and parts with a specific, real defect type. The defect should be representative of what the process actually produces.
2. Blind evaluation — Present the parts to the inspector or gauge in a random order, without telling the operator which are defective.
3. Record decisions — Each part is judged as "accept" or "reject."
4. Compute detection probability — For each defect type, calculate:

\[
\text{Detection Probability} = \frac{\text{Number of defective parts correctly rejected}}{\text{Total number of defective parts evaluated}} \times 100\%
\]

1. Interpret — AIAG recommends a detection probability of at least 90% for a capable attribute system, with 95% or higher preferred for critical characteristics. Values below 80% indicate the inspection method is likely missing real defects.

A Worked Illustrative Example

Example data (illustrative only):

A visual inspector checks 50 parts known to contain a surface scratch defect (the "signal"). The parts are presented in random order among good parts. The inspector correctly rejects 43 of the 50 defective parts.

\[
\text{Detection Probability} = \frac{43}{50} \times 100\% = 86\%
\]

Interpretation: At 86%, the system is below the AIAG recommended 90% threshold. The inspection method is missing 14% of defective parts — a risk that should trigger improvement (better lighting, clearer work instructions, or a different gauge).

Common Pitfalls

• Testing only extreme defects — If you use defects that are obvious to anyone, the detection probability will look artificially high. Use marginal or realistic defects.
• Informing the operator — If the inspector knows a "test" is underway, behavior changes. Keep the evaluation blind.
• Ignoring good‑part errors — While signal detection focuses on catching defects, also record false rejections (good parts scrapped). A system that rejects everything will have high detection but terrible efficiency.
• Small sample sizes — With fewer than 20 defective parts, the confidence interval is wide. Use at least 30–50 per defect type.

Closing

Signal detection MSA turns a simple pass/fail check into a measurable, improvable process. If your attribute inspection is missing defects, you will see it in the detection probability — and you can act before the customer does. To run the calculations and track detection rates quickly, try the free signal detection MSA tool at https://www.6sq.com/tools/sigdet/.