Skip to content
Quality Tools | Part 5 of 8

The Histogram: Center, Spread and Shape

How to build a histogram from measured values, how to read its shape and why splitting it by group often shows more than the overall picture.

A check sheet with marks says good or bad. A measured value says how much. Where a characteristic can be measured at reasonable cost, a small number of measurements shows more than a large number of marks. The tool that makes measured values readable is the histogram: bars that show how many values fall into each class.

Build it

  1. Collect at least 50 values, better 100, and record the factors you may want to sort by.
  2. Choose 6 to 10 classes of equal width with round limits.
  3. Count the values per class and draw the bars without gaps.
  4. Draw the specification limits as vertical lines.
  5. Draw one histogram per group on the same scale and compare.

On powder coating line 2 the specification for coating thickness is 60 to 120 µm. During one week of the baseline, 120 parts were measured, 40 from each hanging row.

Class in µm All 120 parts
50 to below 60 8
60 to below 70 25
70 to below 80 36
80 to below 90 34
90 to below 100 14
100 to below 110 3

The mean is 77.1 µm and the standard deviation 12.1 µm. The smallest value is 50, the largest 107 µm. Eight of 120 parts, or 6.7 percent, are below the lower limit.

That last number is worth a second look. On the check sheet, thin coating ran at 3.0 percent of all parts. The inspector finds a thin part by eye when the base material shines through. The gauge finds more. Measured data shows the real size of a problem that counting underestimates.

Read the shape

The overall picture is one wide peak with a tail to the left. Common shapes and what they suggest:

  • One peak, symmetric. One stable process. Compare center and spread with the limits.
  • Two peaks. Two processes in one data set: two machines, two shifts, two positions. Split the data.
  • A long tail on one side. A natural limit on the other side, or occasional disturbances.
  • Cut off at a limit. Someone sorts or reworks before the measurement. You do not see the real process.
  • A comb with gaps. Rounding, or a measuring device with too coarse a resolution.

Split it by group

The stratification in part two had already pointed to the bottom hanging row. The same split on the measured values:

Hanging row Below 60 µm Mean in µm
Top 0 of 40 83.3
Middle 1 of 40 80.1
Bottom 7 of 40 67.8

Drawn on the same scale, the three histograms tell the story at a glance. Top and middle row sit comfortably inside the limits. The bottom row has the same shape and almost the same width, but it sits about 15 µm further left, with its lower end hanging over the limit.

This distinction decides what to do. A distribution that is too wide points to disturbances and instability, which are hard to find. A distribution that is simply shifted points to a setting, which is easy to change. The bottom row is not unstable. It gets less powder.

What the histogram cannot show

A histogram has no time axis. A process that drifts slowly during the week and a process that is stable can produce the same picture. Always look at the same data in time order as well, which is the subject of the next part.

After the change

When the cause was found and the setting of the lower guns changed, the measurement was repeated with another 120 parts. The mean of all parts rose to 83.6 µm, the standard deviation fell to 9.3 µm, and no part was below 60 µm. The mean of the bottom row was 84.1 µm, level with the other two rows. Laid side by side, the three histograms were now hard to tell apart, which is exactly what the team wanted to see.

Newsletter

One Lean idea.
Every two weeks.