Don’t underestimate the stealth variable
This was well before microprocessors or even digital processing was available, and it was to be a completely analog device. The system had to be usable in the field, self-contained, including its power, with sunlight readouts indicating computation results. (We used D’arsonval meters marked GO/NO GO.) The parameters to be measured were the pressure in a fuel tank and its temperature. From these results, we were to make a computation of gas volume.
Using the universal gas equation, PV=nRT, in the form of V=nRT/P, where P is the pressure, V is the volume, n is the number of moles, R is the universal gas constant, and T is the absolute temperature, the system was to calculate the volume of gas in the fuel-tank system to determine whether any liquid fuel had leaked (been lost) from the four interconnected, sealed fuel tanks. These tanks, distributed along the length of the missile and joined together to function as one volume, were to be partially filled with a given quantity of fuel.
After the missile had been deployed for months or longer, it was intended that this measurement system was to be plugged into the missile and would indicate if the quantity of fuel onboard was correct (by measuring the volume of the residual gas in the system). The volume of gas in a properly fueled missile was a known quantity. If it was indicated to be different from what was specified, then fuel must have leaked to allow that expansion.
It had been planned that this measurement could be made on the flight line, in the field, under whatever temperature conditions the assessing persons happened to find the missile. It took me several meetings to convince our customer that if there was any thermal gradient in the missile, the system could not work accurately. The reason: We made only a single temperature measurement, and there were three other locations, different fuel tanks, in which residual gas could be present. In due course that application limitation was conceded and a requirement was added that the missiles were to be in a stabilized temperature environment for a few days before this measurement was to be made.
As you might guess, there were state-of-the-art requirements for accuracy. Errors of either pressure or temperature might have a disproportionate effect on the computation of gas volume. We proposed that the temperature measurement be made using a platinum temperature sensor, which is almost a secondary standard for measuring temperature and very stable. The pressure measurement was to be made using an LVDT (linear voltage differential transformer) absolute pressure transducer—also capable of very high accuracy. I came up with a cute way of creating directly the function T/P by inversely exciting the pressure transducer with a potential directly proportional to absolute temperature. This method was an error-free computation to allow us to meet the requirements of the contract.
During the development and assessment of the design, we, as typical prudent engineers, were always looking for trouble. Was everything always as expected? Was the noise appropriate? Was it stable? Was it steady? Was it repeatable? Could we explain everything that we observed? Was it within specification over the temperature range? What if the temperature was changing? Was it still sufficiently accurate?
During one of these careful observations, we observed a slow, very slow, drift. The construction, by the way, was “welded cordwood,” where the components with axial leads were mounted between Mylar wafers with welded, nickel interconnects. After two days of meticulous work, this drift was tracked down to a tantalum capacitor that had been installed backwards—quite a subtle error and very difficult to tag. The nagging worry was constant that the source was a subtle design error.
The LVDT pressure transducer measured absolute pressure with “infinite” resolution, meaning that in the air and noise permitting, it was sensitive to any variation in atmospheric pressure. One day, as we carefully monitored the pressure transducer output, we noticed that it would occasionally take a very small jump and slowly recover to the previous stable value. The movement could be either positive or negative, but always with a slow recovery to the original value. The monitoring was done on an analog meter (probably a Fluke differential dc voltmeter) rather than a digital meter, where the effect might have been masked by a dithering digit. Anyway, it was quite a small effect but repeated again and again at random intervals. Sometimes the effect would repeat in less than a minute; sometimes many minutes would go by without a waiver. It was like shot noise, but not an offset and with a recovery time constant of four or five seconds.
The laboratory area in which we were working was quite large, involving many other projects and engineers and technicians. We were concerned and puzzled—was it some interaction with our surroundings? It wasn’t a big effect, but I had found, over the years, that anything noticed and unexplained may well occur later and at a much more inconvenient time. So we checked and rechecked the setup, the instruments, the power—what could it be?
Then somebody, facing in a fortuitous direction, noticed that as one of the doors to the lab was opened, the reading jumped. There was no fault after all! The transducer was responding perfectly to the very slight fluctuation in air pressure as the door to the lab was opened and closed. Problem solved. We exhaled and moved on.
R. Heathcote Russell is an adjunct staff engineer at Dartmouth Thayer School of Engineering and vice president of engineering at Garmire Russell Associates.