Statistics in Industrial Quality Control

Why Statistics in the Factory?

No bolt comes out at exactly 10.000 mm every time. Some measure 10.02, others 9.98. This variation is natural — but it is the enemy of quality. Industrial statistics separates acceptable variation from real problems that demand action.

Mean and Standard Deviation

Mean locates the center: μ = (x₁ + x₂ + ... + xₙ) / n

Standard deviation measures spread around that center: σ = √(Σ(xᵢ - μ)² / n)

Example: five bolts at 10.02, 9.98, 10.01, 9.99, 10.00 mm. Mean = 10.00 mm, σ ≈ 0.014 mm — tight production.

Normal Distribution and the 3σ Rule

Thousands of measurements cluster around the mean in a bell curve called the normal distribution:

68.27% within μ ± 1σ
95.45% within μ ± 2σ
99.73% within μ ± 3σ

This is the 68-95-99.7 rule. If tolerance is 10.00 ± 0.05 mm and σ = 0.014, then 3σ = ± 0.042 mm — safely inside specification.

Control Charts

Invented by Walter Shewhart at Bell Labs in the 1920s. Plot measurements over time with statistical boundaries:

UCL = μ + 3σ    (Upper Control Limit)
CL  = μ         (Center Line)
LCL = μ - 3σ    (Lower Control Limit)

Any point beyond UCL or LCL means the process is out of control — stop and investigate.

X̄-R Chart

Used for continuous data (lengths, diameters, weights). Small samples of 4-5 parts are drawn at regular intervals:

X̄ chart: tracks sample mean — detects center shift
R chart: tracks range within each sample — detects increased spread

If X̄ rises above UCL, the cutting tool may be worn. If R widens, the fixture may be vibrating.

P-Chart for Proportions

Used for pass/fail data — tracks the fraction defective per batch:

UCL = p̄ + 3 × √(p̄(1-p̄)/n)

A PCB line normally at 2% defect rate jumps to 5% — the P-chart flags it immediately.

Process Capability: Cp and Cpk

Can the process meet specifications? Capability indices answer this:

Cp = (USL - LSL) / (6σ)

Cp = 1 means variation fills tolerance exactly. Cp = 2 means it uses only half — excellent.

Cpk adds centering: Cpk = min((USL - μ)/3σ, (μ - LSL)/3σ)

Automotive requires Cpk ≥ 1.33. Aerospace demands Cpk ≥ 1.67.

Six Sigma: 3.4 Defects per Million

Six Sigma was developed by Motorola in the 1980s. The goal: specification limits at 6σ from the mean.

3σ  →  66,807 defects/million  →  93.32%
4σ  →   6,210 defects/million  →  99.38%
6σ  →     3.4 defects/million  →  99.99966%

The methodology follows DMAIC: Define, Measure, Analyze, Improve, Control. A factory producing one million solder joints daily sees 66,807 defects at 3σ but only 3.4 at 6σ.

Summary

Industrial statistics transforms quality control from random inspection into a scientific system. Mean and standard deviation describe the process, the normal distribution predicts defect probability, control charts monitor stability in real time, Cp and Cpk measure capability, and Six Sigma sets the ultimate target.