Engineering Lessons from the Past

Written By Ed Tate

Technical reviews fail when experts don’t understand what is possible and rely on assumptions, consensus conclusions, and incorrect heuristics. A great example from the past is ENIAC.

In 1945, ENIAC was the first programmable, electronic, general purpose digital computer [1]. When it was first proposed by John Mauchley and J. Presper Eckert as a faster way to compute artillery firing tables, the peer review of the design concluded that it was impossible to be reliable enough to do anything useful. They were wrong.

By 1956, ENIAC contained 18,000 vacuum tubes, 7,200 crystal diodes, 1,500 relays, 70,000 resistors, 10,000 capacitors, and approximately 5,000,000 hand-soldered joints. Each of these elements was a point of failure that would require hours of ‘debug’ and troubleshooting. Of all these parts, vacuum tubes were the worst. Their combined reliability was the reason the original proposal was rejected.

Using typical values for a switching vacuum tube reliability [3][4], the failure rate of a single tube would be about 250 failures per one million hours. This sounds pretty good … except all 18,000 vacuum tubes needed to work continuously together. At that failure rate, ENIAC would have less than a 50% chance of running 10 minutes [6]! Assuming it took more than 10 minutes to find and replace a tube, ENIAC would never operate. It would be turned on and a red queen’s race [5] would commence where parts fail faster than they were replaced. Using these assumptions, ENIAC could never work and should not be built.

But it was built and ran up to 116 hours without a failure [1]. Using the consensus estimate of reliability, the odds of this happening would be about 1 in a Googol squared [7]. Something was seriously wrong with the peer review estimates….

The reviewers didn’t consider that reliability of vacuum tubes could be radically changed by operating away from their rated conditions. They used incorrect heuristics based on existing hardware.

The ENIAC designers beat this heuristic by testing every available tube at different conditions to find they could tradeoff unnecessary electrical performance for the reliability they needed [8]. This hack moved the reliability from minutes to days [1] and proved the reviewers wrong.

But ENIAC would not have happened if there was not an unsolvable problem. In the early 1940’s every new piece of artillery required a firing table. [9] These tables took longer to generate than it took to get the guns into service. ENIAC was commissioned against the peer review recommendations because it might solve this problem. If not for this, ENIAC would probably never have happened.

With other technologies, similar stories exist. In technical reviews of wireless power transfer (WPT) and space based solar power (SBSP), incorrect heuristics and assumptions lead to the wrong conclusions. Creativity and hardware will dispel them.

Ed Tate
Previous

Economics Lessons From Microwaves
Next

Art & Engineering