What Is Normal Distribution — and Why Does It Appear in Manufacturing?
Normal distribution (also called the Gaussian distribution or bell curve) is the natural pattern of variation that appears when you try to make something to a target value.
For example: manufacturing pen shafts to a target diameter of 10 mm produces results like 10.01 mm, 9.98 mm, and so on. Values far from the target — 8 mm or 12 mm — are extremely unlikely. The closer to the target, the higher the probability; the further away, the faster the probability falls. That shape is the normal distribution.
Fig. 01 — Normal distribution: symmetric around the mean μ
The bell curve. Most values cluster near the mean; extreme values become exponentially rarer.
Why manufacturing creates normal distributions
The act of aiming at a target value — and having many small, independent sources of variation push results slightly above or below — is precisely the mechanism that generates normal distribution. Conversely, data that deviates from normality is often a signal of something abnormal: tool wear, a material lot change, or a process shift.
σ and Probability — What the Numbers Mean
The power of normal distribution lies in this: once you know the mean (μ) and standard deviation (σ), you can calculate exactly what proportion of data falls within any given range. This is the foundation on which control charts and process capability indices (Cp, Cpk) are built.
| Range | Data within this range | Outside (defect rate estimate) |
|---|---|---|
| μ ± 1σ | ~68.3% | ~31.7% |
| μ ± 2σ | ~95.4% | ~4.6% |
| μ ± 3σ | ~99.7% | ~0.3% (3 per 1,000) |
Connection to control charts
Control chart upper and lower control limits (UCL / LCL) are typically set at ±3σ. A point outside those limits represents something that should only happen 0.3% of the time by chance — statistically, a signal that something has changed in the process.
Standard Normal Distribution — One Ruler for All
There are infinitely many possible normal distributions — each defined by its own mean and standard deviation. But a single transformation converts any of them into a common reference: the standard normal distribution (mean = 0, σ = 1).
The standard normal distribution has a lookup table — the z-table — that gives the probability for any z-score. Whatever the original mean and standard deviation, once standardized you can read off the probability directly from that table.
Why normal distribution is called the "universal tool"
Know μ and σ → standardize → look up the z-table → get the probability. This chain is why normal distribution underlies process management, inspection, and experimental design. But the entire chain only works if the data is actually normally distributed — which is exactly what needs checking.
The Histogram Trap — Why Visual Inspection Is Dangerous
The most common approach to checking normality is drawing a histogram and eyeballing it. This is unreliable — and in some situations, actively misleading.
Skewness ≠ 0
Kurtosis ≠ 3
A common floor mistake
"I drew a histogram and it looks like a hill — normal distribution confirmed." This logic is flawed. A hill-shaped histogram can hide significant departures from normality in its skewness or kurtosis. Calculating control charts or process capability indices on that assumption leads to systematically wrong conclusions.
Why QQ Plots Work
A QQ plot (quantile-quantile plot, also called a normal probability plot) is significantly more reliable than histogram inspection for assessing normality.
The principle: plot the theoretical quantiles of a normal distribution on the horizontal axis against the actual quantiles of your data on the vertical axis. If the data follows a normal distribution, the points fall along a straight line.
Combining the QQ plot with a normality test:
After visual inspection with the QQ plot, add a normality test such as the Shapiro-Wilk test for a more objective judgment. A p-value below 0.05 indicates that the data is likely not normally distributed. Use both together — the plot for visual understanding, the test for the number.
Summary
Knowing μ and σ gives you the probability for any range. This single property underlies control charts, Cp/Cpk, and most of statistical process control.
99.7% of data falls within ±3σ. A point outside this range is a "0.3% event" — treated as an alarm signal in process management.
Histograms are unreliable. Use a QQ plot to check visually, and a normality test (e.g. Shapiro-Wilk) for an objective p-value. Use both together.
Normal distribution is powerful precisely because it unlocks so many analytical tools — which is exactly why checking the assumption matters most. Using powerful methods on a false premise leads to wrong conclusions. QQ plots are available in any free statistics package. Make normality verification a standard step before applying any normal-distribution-based analysis.