I checked with my co-worker. He worked at General Motors in the 1970s. He was using the average range method in 1975 and recalled that GM had empirically derived the criteria using simulations. GM later developed the ANOVA approach as well. This matches my memories of having been exposed to the method by GM in the 1980s.
There is absolutely no statistical basis for the criteria. In fact, the entire notion of using ratios of standard deviations is mathematical nonsense because standard deviations are not additive. Dr. Donald Wheeler has many excellent articles on this topic in Quality Digest and on his own
SPC Press website.
Having said that, the methodology does work despite its mathematical shortcomings. It was developed during a time when most people did not have access to computers, so it had to be simple and tradeoffs were made to make it simple. Today, we have better alternatives (see Wheeler's
An Honest Gauge R&R Study), but AIAG has firmly entrenched the original method into most industries.