The Bimetallic Compensation Balance

Take your balance-and-spring oscillator and time it cold and warm — first with the whole assembly heated, then with only the spring warmed while the wheel stays cool.

Expect warming everything to slow it, and warming the spring alone to produce most of that slowing.

Record the split.

This is the measurement that tells you what to compensate. Most of a watch's temperature error lives in the SPRING's stiffness, not in the wheel's size — which is why the cure has to be applied to the wheel in the opposite direction rather than by simply making the wheel out of something stable.

Bond a strip of brass to a strip of steel along their length — riveted, soldered or clamped at intervals — and warm the pair.

Expect it to bend, curving towards the steel side, because the brass grows more and the bond forces them to stay together.

Measure the deflection against temperature and check it is repeatable on cooling.

You have made the fundamental sensor of an entire century of control engineering. Two materials with different expansions, joined, convert temperature into MOTION — and motion is something a mechanism can use.

Form the bimetal into a circular rim on a two-armed cross, then cut the rim through near each arm so you have two free-ended curved arms rather than a closed circle.

Warm it and watch the free ends.

Expect them to move inwards.

Now try it uncut and warm it again.

Expect almost nothing: a closed ring simply grows slightly, because it cannot curl without somewhere to go.

The cut is not a detail, it is the mechanism. A compensator needs a free end, exactly as the gridiron needs its rods actually in the load path.

Fit small weights on the free arms and time the watch cold and hot. Move the weights towards the free ends and repeat; then towards the fixed ends and repeat.

Expect weights near the free ends to give strong compensation and weights near the arms to give little, because the free end travels furthest.

Adjust until hot and cold rates match.

Note what you have just done: you did not remove an error, you introduced a second error of equal size and opposite sign. The watch is now wrong in two ways that cancel — which is a completely respectable engineering answer and worth being honest about.

With the balance set to agree at, say, 5 °C and 30 °C, now time it at 17 °C, halfway between.

Expect it not to agree — expect a small error in the middle of the range.

The reason is that the spring's stiffness falls almost linearly with temperature while the bimetal's curl does not, so two curves that cross at two points must diverge between them.

Matching a curve at two points does not match it everywhere. This residual, called middle temperature error, was the last significant defect of the mechanical chronometer, and it was never fully solved by mechanism — only by inventing a spring alloy that barely changed at all.

It grew out of the chronometer trade, and its parentage is shared. John Harrison used a bimetallic strip as a temperature compensator in his sea clocks; Pierre Le Roy, John Arnold and Thomas Earnshaw developed the cut bimetallic rim into the form the trade actually used. Earnshaw and Arnold argued publicly about priority for years. As with the balance spring, the geometry is unambiguous and the credit is shared.

It is what made chronometers a PRODUCT rather than a masterpiece. Harrison's H4 was a unique object requiring its maker. A compensation balance, a detent escapement and jewelled bearings could be made to a pattern by a trade of specialists — and by the middle of the nineteenth century chronometers were made in hundreds and sold to any shipowner. The invention that matters commercially is usually the one that makes the first invention repeatable.

The same strip went on to run the twentieth century's thermostats. A bimetallic strip that bends with temperature and closes a contact is the thermostat in the incubator blueprint, in an iron, in a kettle, in a car's choke and in an oven. One idea — join two materials with different expansions — appears here as a compensator and there as a sensor, and it is worth being able to see that they are the same object used two ways.

Then Guillaume made it unnecessary, twice. Invar gave a balance that barely expands, and Elinvar gave a balance SPRING whose elasticity barely changes with temperature — removing the dominant error at its source. A plain monometallic balance with an Elinvar spring outperforms an elaborate cut bimetallic one. Second time in this batch that a beautiful compensating mechanism was retired by a new alloy; when you meet an intricate compensator, ask what material would delete it.

Honest limits. It cancels only at the two temperatures it was adjusted for, leaving middle temperature error. It is a cut, weighted, delicate structure that shock can distort and that adds parts to the most sensitive component in the watch. Its adjustment is skilled hand work, done by timing at temperature over days. And it does nothing at all for the other errors — position, magnetism, shock, and the ageing of the spring itself.