Here's how the author challenged accepted norms that are degrading the performance of a system.
When a diamond is irradiated with x-rays it luminesces (or fluoresces, I cannot tell the difference). This property is used to separate diamonds from gravel in the mining industry. Once the rock is mined, it is crushed to a gravel of small stones and diamonds, and then the gravel mix is fed into a sorting machine. The first stage is a vibrating bed to spread the gravel evenly on to a short conveyor belt about four feet wide. At the end of the conveyor, the gravel is launched into the air through an x-ray beam and above this zone are six photomultiplier (PM) tubes ranged evenly across the width of the machine to detect the light emissions from the diamonds. When a flash is detected, an air blast is fired in the same channel associated with the specific PM tube. The blast takes out the diamond and some of the gravel around it.
Since diamonds are the target of theft, the temptation was removed in development by using "tracers", small cubes that floures… ah, lumines… ah, emit light when irradiated by x-rays. While investigating a problem a colleague and I climbed through a hatch to have a closer look. When we came out we discovered that the x-ray generator had been powered up the whole time. Even though even back then life was cheap in South Africa, we were rushed to a clinic for blood tests. An interlock was also fitted to the hatch immediately, in a classic case of shutting the stable door after the horse has bolted. However, this is merely an aside and a warning to all not to make assumptions on the safety of any machine.
Each channel control on the unit consisted of an 8751 microcomputer that monitored the output of a PM tube and fired the blaster when the PM output crossed a programmable threshold. Bill Schweber covered the technique used for the analogue part of the circuit in his blog "Peak Detector: A Classic Analog Circuit Still in Wide Use". There were six of these controllers and an 8032 central controller board that supervised the six controllers and provided a user interface. The motor drive for the belt and the interlocks were done through a PLC with no interface at all to the electronics.
*Figure 1: Double Eurocard 8032 based controller board. I only had a A3 plotter (11.7" x 16.5" and so the board was developed as two single Eurocards and then manipulated to a double Eurocard. The flat cable in the middle linked the one half to the other. In production the board was modified to remove the external linkage. (Source: Aubrey Kagan)*
*Figure 2: The system front panel. Note the six channels connected to the PM tube outputs via BNC connectors. The controller board of Figure 1 is seen mounted on an extender card on the right. (Source: Aubrey Kagan)*
The problem that led to my irradiation was that an air blast would occur simultaneously on several air "guns" for a single tracer. For all the years that this style of machine had been produced, no one had ever thought of doing the test in this detail and the problem existed in all production machines to that date. In order to improve performance, we mounted a tracer on a rotating disc. The disc could be moved along an axle across the width of the sorter. The locus of the tracer approximated the trajectory of the tracer launched from the end of the belt, and it gave us a definitive location and repeatability for our measurements. I measured the reactions of the PM tubes gradually across the breadth of the machine and created a profile and it became obvious there was significant overlap between PM tubes. The solution was to provide collimators for each PM tube and the results improved dramatically.
On a diamond mine (actually the word "mine" my be misleading, since in the case of alluvial diamonds the extraction method is simply to use a bulldozer to shift the sand around) several of these machines are placed in series, gradually reducing the amount of gravel surrounding the diamond. Human operators are placed at the start of the process to visually detect large diamonds although I suspect that a mine does not want too many large ones—there are only so many people who will fork over millions of dollars for a useless chunk of crystal.
Have you ever had to challenge accepted norms that were degrading the performance of your system?