#1678 1924 · Western Electric / Bell Telephone Laboratories (Walter A. Shewhart) · Manufacturing / quality management
Shewhart's one-page memo showed that correcting every deviation makes quality worse
问题
Managers adjusted the process after every off-target reading, and the adjustments made output vary more, not less
背景
By the early 1920s Western Electric's Hawthorne Works was manufacturing telephone equipment at a scale where inconsistency was expensive: parts had to be interchangeable across a national network, and the standard response to a measurement that came in off-target was to correct the machine. The discipline looked like diligence — watch the output closely, and adjust whenever it drifts.
Walter A. Shewhart, a physicist working in the inspection engineering department, saw that this instinct was the problem rather than the cure. Every process fluctuates for reasons that have no identifiable source, and adjusting in response to those fluctuations does not cancel them; it adds a second source of variation on top of the first. What was missing was not more vigilance but a rule for telling the two kinds of variation apart.
换别人会怎么做
Watch output more closely and correct the machine every time a measurement comes in off target, on the reasonable-sounding logic that faster correction means tighter quality — which is exactly what the factories were already doing, and exactly what was generating the excess variation.
他们看到了什么
Reacting to every deviation is not diligence, it is tampering — correcting for random variation injects more of it. Shewhart drew limits saying when a wobble is noise to leave alone and when it is a signal to act on.
那一手
On May 16, 1924, Shewhart sent his superior a memorandum roughly one page long, about a third of which was a simple diagram — recognized today as the first control chart. The text around it set out the essential principle: variation in a process comes either from chance causes, which are inherent to the system and produce a stable, predictable band of fluctuation, or from assignable causes, which are genuine disturbances with a findable source. Plotting measurements over time against statistically derived limits makes the distinction visible: a point inside the limits is noise and must be left alone, while a point outside them is a signal worth investigating. The counterintuitive consequence is that acting on ordinary fluctuation, later called tampering, actively degrades quality by injecting the operator's corrections as new variance. The method spread from Hawthorne through Bell Labs, and because the War Department had seen it work at Western Electric it recommended and then required these methods in wartime production. W. Edwards Deming, who worked with Shewhart, carried the approach to Japan after the war, where it became a foundation of Japanese manufacturing quality and, in turn, of statistical process control, Six Sigma and the modern quality profession. The Journal of Quality Technology marked the chart's centenary in 2024.
为什么管用
The chart works because it converts a judgement call that humans reliably get wrong into a visible, pre-agreed threshold. People are strongly disposed to find a cause for every fluctuation and to act on it, and each such action adds its own error to the process; the control limits remove the discretion by defining in advance what the system's ordinary behaviour looks like. That also makes the rule usable by an operator on the floor rather than only by a statistician, which is why a one-page method survived a century.
值了多少
The 1924 chart became the basis of statistical process control, mandated in US wartime production and later carried to Japan by Deming.
什么时候会失灵
The limits describe the process that produced the historical data, so a chart built on an unstable or badly measured process encodes that instability as if it were normal, and will then wave through the very problems it should flag. It also assumes the cost of investigating is real; where a rare excursion is catastrophic — a safety-critical failure, a contamination event — waiting for statistical confirmation is the wrong trade, and the chart should not be the only alarm.
后来呢
Statistical process control became standard practice across manufacturing worldwide and the direct ancestor of Six Sigma and modern quality management; the Shewhart chart is still taught and used unchanged in fields far from the factory, including hospital quality improvement, where the Institute for Healthcare Improvement teaches it as a core tool.
资料来源
- [1]The 100th anniversary of the control chartJournal of Quality Technology, 2024tandfonline.com
- [2]Walter A. ShewhartAmerican Society for Quality, 2020asq.org