Learning From Failure

by Don Friedman on February 22, 2017


I first heard the joke in 1982: How dumb are engineers? We can’t learn from success, rather we only learn by destroying something. Thirty-five years later, it occurs to me that the reason I never thought the joke was clever is that I had played Clue and Mastermind when I was a kid and both games are most easily won by getting your intermediate guesses completely wrong.

Success and failure are not mirror images, are not symmetrical. When a design succeeds, we don’t know for certain why. We may think we know, and the process of design is in part a process of removing obvious locations of doubt, but we cannot be certain.  When a design fails, we can usually find out exactly what went wrong. All it takes is patience in examining the wreckage, careful analysis, and honesty. There is far more evidence to be found in a failure than in a success, which is why engineers learn more from failure. The history of building codes since 1900, for example, includes the reaction to various catastrophic failures. The most famous example may be the requirement, after the Cocoanut Grove fire, for outward-swinging egress doors adjacent to revolving doors.

Since waiting around for failures to occur as a way to advance our education is both time-consuming and morbid, engineers have created forensic engineering journals and conferences to collect the available stories. One of the best collections is the CROSS and SCOSS website* run by the Institution of Civil Engineers, the Institution of Structural Engineers, and the [British] Health and Safety Executive. There are hundreds of descriptions and analyses of failures, ranging from unforeseeable loading conditions to outright stupidity. The site accepts anonymous reports, which is an effective way to broaden the reach of analysis. Reading the reports is less fun than Clue but more informative.


* CROSS stands for Confidential Reporting on Structural Safety; SCOSS stands for the Standing Committee on Structural Safety.

Previous post:

Next post: