Machine Kinematics
solved, then drawn
Theme

The thread: Exact and approximate, part 3 of 3

Watt's straight line is not straight and Peaucellier's is. The difference is measurable, it took ninety years to close, and the error curve of an approximation is a more interesting object than the approximation.

273 essays in 3 parts: part 1 · part 2 · part 3

3 handovers, and 3 cuts where the shared ground stands empty. Four in turn: each tool's lift against the cam angle, from its own cam, with a thick bar under it while its front is inside the shared ground. The dashed lines are the fewest cuts that clear every pair of tools at once, 3, at 114.5°, 204.5°, 294.5°, and the shaded bands are the windows each may move in, 65° to 115°, 155° to 205°, 245° to 295°: the times when no tool is inside the shared ground. The pairs need 1 cut at worst and 6 between them; the station changes hands 3 times after the turn's seam. Links with a width

A cam machine is cut at its handovers

In one symmetric linkage, every pair of parts crosses its shared ground on either side of the same few moments, so the whole machine needs no more cuts of its drive than its worst pair. Four tools driven by four cams into one work station do not share their moments. Each pair needs one cut, the pairs need six between them, and the machine needs three: one for each time the station changes hands. Each cut can move only within the time the station stands empty, which is the cam diagram's own timing margin, and when a schedule is squeezed until that margin closes, the tools collide.

 A seventh leg does no work on the free twist only when it leans as the six do. At home, level at the dead yaw, the seventh leg's signed work on the twist the six legs cannot resist, against how far round the axis its platform attachment sits from its base attachment. It is nought at exactly two leans, 82.5° and 97.5°, the two leans of the six legs themselves (ticks), positive between them and negative everywhere else, most negative at 278°. The best leg of the earlier placement map leans 285° and does work −0.506, near that extreme of −0.507. Several legs, one platform

A seventh leg must lean its own way

Seven legs on a Gough platform fail only where the twist the six leave free does no work on the seventh. Near the paired platform's dead yaw that set is decided by one angle: how far round the axis a leg's platform attachment sits from its base attachment. The six legs all lean 82.5° or 97.5°. A seventh at either lean joins them and is singular at every level position; a seventh at any other lean is singular at none. Tilted, the failing leans spread out from the six's, to 74.5°–107° at four degrees of tilt and 44.5°–137.5° at eight, and never reach the far side of the axis, where the best leg leans.

 The smallest springs that balance the lamp, and the one that is not needed. The tilting-head desk lamp with one parallelogram, its springs of no free length placed to make the total of their peak pulls over the working range least, inside an outline of brackets 0.15 m high, attachment points 0.08 m past the wrist and along the head's 0.15 m. Peak pulls 22.7, 0.0, 23.2, 17.5 N (lower arm, upper arm, elbow bracket to head, wrist), 63.4 N in all and 23.2 at most. The head spring meets the head at 0.0615 m, the threshold, and the upper arm's spring has no stiffness at all: it is left out. Machines you have met

The lamp needs three springs

A desk lamp with a tilting head, balanced exactly with springs of no free length, was built with four springs whose peak pulls add to 162 N. Searched over every attachment point inside the same outline, the least total is 63.4 N, and it is reached with three: the head spring hung at the threshold where the upper arm's spring stops being needed, 77% of the way to the head's centre of mass. Beyond the threshold a second spring is the first one split in two, and splitting a spring never makes it smaller. A design with no parallelogram and six springs is never close: 213 N at its best.

 Every boundary touches the circle on the rod. The coupler's plane in the rod's units, crank pin at (0, 0) and rocker pin at (1, 0). Shaded: the coupler points whose three departing double points are all real at a ratio of rocker to crank of 1, the region where R = λ⁴A + λ²B + C is negative; 8.1% of a window from −1.4 to 2.4 along the rod and ±1.5 across it. Dotted: the circles one rod from each pin, the boundary at the family's two ends. Dashed: the circle with the rod as its diameter. Every boundary in the family touches that circle and never crosses it; at this ratio it touches at (0.50, ±0.50), the points where −B/2A = λ². Inside it all three are real at every ratio. How many answers

Three discs decide the real double points

A four-bar growing into the elliptic trammel loses three double points, and how many of them are real depends on the coupler point and on the ratio λ of rocker to crank. The boundary of the region where all three are real is one polynomial, R = λ⁴A + λ²B + C, quadratic in λ² and short when written in the squared distances from the two pins. Its sign is the count at every one of 23,100 cells tried. As a quadratic in λ² its discriminant is the circle with the rod as diameter times a figure of eight, so that circle is touched by every boundary and crossed by none, and a coupler point's whole history along the family is settled by three discs.

 Along one line the nearly free curve stays nearly free. The arm's pivot is fixed at the centre of its reach (dashed); the quintic x⁵ + 0.4x²y − y = 0 is drawn with its own point — where the catalogue put the pivot — moved 0.2 of a link along the line x = 0, which is the same as moving the pivot the other way. Heavy: the arc priced, the part of the quintic within 1.4 of its own point. Thin: the nearest curve built from at most four cosines of the arm's angles, found by trying every set of four: 6.0% of the reach away, the line through the pivot bent by the square of the radius. The equation itself costs 22 cosines here. The curve as an equation

The quintic's cheapness runs along a line

A quintic whose equation costs a compiled machine eighteen cosines is drawn to a few per cent of the arm's reach by four of them, and its exact count jumps to 22 or 30 the moment the arm's pivot moves. Priced by the nearest curve instead, the two directions part completely. Along the line x = 0 the nearest four-term curve stays within 7.3% of the reach out to 0.4 of a link and the nearest six-term within 3.4%. Across it the four-term price goes from 3.2% to 18% between a twentieth and a tenth of a link, and from 0.15 on four terms buy nothing over three. The move along adds four cosines of amplitude 0.2 and 0.4 times the move; the move across adds twelve, sixty to ninety times larger in total.

 Past the play, a caged joint shares its load and a grooved one leans on one ball. Two six-ball Rzeppa joints bent 20°, at an input angle of 15°, each twisted 0.03 of a window clearance past its free lag, with every contact a spring of the same stiffness and the cage and balls settled where the stored energy is least. Bars: each ball's share of the total contact load. Left, grooves offset 0.1, where the cage sets the play: 0.17, 0.20, 0.19, 0.16, 0.14, 0.15, 5.91 balls' worth. Right, grooves offset 0.7, where the grooves do: 0.00, 1.00, 0.00, 0.00, 0.00, 0.00, 1.00 balls' worth. Dotted: an even sixth. What a joint is

A tight groove leans on one ball

A six-ball Rzeppa joint's free play is set by the tighter of its cage and its offset grooves, and the two limits look alike from outside. They are not alike once the play is taken up. With every contact made a spring of the same stiffness and the joint twisted past its free lag, a joint whose cage binds spreads the first load across four to six balls' worth and, a third of a window clearance on, gives every ball between 0.17 and 0.19 of it. A joint whose grooves bind puts the first load on one ball alone over most of a ball pitch, and two clearances on it still spreads it over fewer than six. At the same torque its hardest-pressed ball carries 1.5 to 3.0 times the caged joint's. Tightening the tracks to cut the play buys the play with uneven loading.

 Where two coefficients vanish, a curve; where three do, two points. The second leg's three joints each turned about the shared line by ψ₁, ψ₂ and ψ₃. Every point of the curve makes the shake's second- and third-order coefficients nought, σ₂ = σ₃ = 0, found by continuation from the earlier two-joint tuning (black) both ways, over an arc of 7.25 rad, until it reaches configurations where the two legs come close to sharing a second twist (crosses), where every coefficient runs off to infinity. Left: ψ₂ against ψ₁; right: ψ₃ against ψ₁. The curve is drawn in one colour where σ₄ is negative and another where it is positive; it changes sign at the two dots, (106.31°, 140.35°, 24.66°) and (136.39°, 57.24°, 18.34°), where σ₄ is nought too. Marked: 50% of the way along, at (107.0°, 141.4°, 25.2°), where σ₄ is positive. Out of the plane

Three tuned joints stop at the fifth order

Two legs of three revolute joints that share a line of force let their platform shake, and every joint of the second leg turned about that line clears one more order: the square of the push, then the cube, then the fourth power. A third tuned joint was expected to buy the fifth. Read off the two legs' own Taylor series instead of a push test, the set where the second- and third-order coefficients vanish is a curve in the three tuning angles, and the fourth-order coefficient crosses nought on it twice. At both points the gap falls as the fifth power of the push, confirmed independently with slopes of 4.97 and 4.89. The fifth-order coefficient is −0.0042 at one and 0.109 at the other, and with three angles spent there is nothing left to clear it: three tuned joints stop at the fifth order. And a higher order is not always more travel: in a millionth of play the second of them moves less than the fourth-order pair.

 A spring that pulls the same at every pose holds the band for less. The five prescribed poses of the rocking slide's body, with the arm point the earlier essays found lightest, (-0.60, -1.20) in the body's frame, and the spring that keeps the block on + − + − + for every working load from nought to the design load. The shaded sliver is every attachment point from which any spring that pulls can choose those walls at all. A constant pull of 44.6 from (3.10, 2.05), no stiffness, pulls at most 44.6 times the unit load. The problem backwards

A constant pull holds the band for less

A spring with free length was the obvious way to make the rocking slide's spring cheaper, and it does nothing: offered one, the lightest design takes none, at every arm point tried. The spring's law never mattered to where it may hang, because the walls depend only on which side of each crank pin the spring's line passes, and that cuts the attachment points to a polygon whatever the spring is. What the law decides is the price. Three poses bind at the lightest attachment point, and a spring whose tension follows its length must pay for the longest of them. A pull that is the same at every pose pays for none of them, and holds the band from nought to the design load for 44.6 units against 52.0. Designing for the band directly halves the hundred units the earlier essay read off by scaling.

 One bar made 0.0001 too long, one bar at a time. Every bar of the machine compiled from a lemniscate lengthened by 0.0001 in turn, the machine re-solved, and the polynomial read at the tracing point. It is no longer zero anywhere. The worst bar takes it to 9.1e-3 — an amplification of 91 — and the median bar to 2.8e-4. The bars that matter are the reflectors, which are the cheapest part of the machine; the translators, which are most of it, barely move the answer at all. Size and fragility live in different parts. As built

Exactness a micron destroys

Lengthen one bar of a compiled machine by a ten-thousandth and its tracing point leaves the curve. On the smallest machine the error comes out smaller than it went in; on a fifty-bar one it comes out ninety times larger — and the bars that matter are the reflectors, which are the cheapest part of the machine.

 The same part, dimensioned two ways. Three holes at 0.00, 1.40, 3.50, each located to ±0.010 by the process. Dimensioned as a chain, each hole is placed from the last, so the two gaps are ±0.010 each and the errors accumulate over the whole span to ±0.0141. Dimensioned from a baseline, every hole is placed from H0, so the span is ±0.010 and the gap between H1 and H2 — which touches the datum at neither end — is ±0.0141. The two schemes are exactly √2 apart in opposite places. Nothing about the part decides this and nothing about the process does; the drawing does, and a tolerance analysis that starts from the lengths has already thrown the information away. Numbers that were measured

The same part, dimensioned twice

Three holes, one process, two drawings. Dimensioned as a chain the errors accumulate and the span is ±0.0141; dimensioned from a baseline the span is ±0.0100 and the gap between the second and third holes is ±0.0141. Exactly √2 apart, in opposite places, and nothing about the part or the process decides which — the drawing does.

 A rod on three blocks, on three slides through one point, and it moves. The double slider of slides at 70° and a rod of 2, with a third block on the rod at a point of its rolling circle (radius 1.064, drawn at this instant) and a third slide along the line that point draws, through the same crossing. Five bodies and six joints of one freedom each: the count gives 3·4 − 2·6 = 0. The rod turns anyway, all the way round, every pin on its slide, because each point of the rolling circle runs on a line through the crossing and the third slide is that line. The paths points trace

A third slide moves only through the crossing

Every point of a double slider's rolling circle runs on a line through the slides' crossing, so a rod can ride a third block on a third slide there, and a fourth and a fifth. The count gives nought, then minus one, then minus two; the rod turns freely every time, because each slide past the second is implied by the first two. Build the third slide slightly wrong and the count wins, in a way that depends on the error's kind: moved parallel, the rod turns freely inside the block's clearance and does not assemble outside it; turned by ε about the crossing, it moves only in two windows a turn, of 2·arcsin(c/(2r sin ε)) each, where the third pin passes through the crossing.

 Each family carries one self-stress, and the price is how hard the errors load it. Dixon's framework pushed 0.5 along B₂'s line, on the family it starts in. Pinned and pushed it has eight free coordinates against nine bars, so it carries exactly one self-stress: bar forces that balance at every joint. Each bar is drawn with a width in proportion to its share of that stress, tension solid and compression dashed: B₁W₁ -0.126, B₁W₂ 0.210, B₁W₃ -0.140, B₂W₁ 0.270, B₂W₂ -0.395, B₂W₃ 0.350, B₃W₁ -0.333, B₃W₂ 0.587, B₃W₃ -0.337. A framework as built with joint errors d pays, per unit ρ, |ω·e| / Σ|ωᵢ|, where e is the bar-length error its joints make. Framework 4 pays 0.0293 on the near family and 0.0267 on the far one; framework 7 pays 0.0131 and 0.0023. What can move

The price is one self-stress

A Dixon framework built with its joints a little out of place has to be carried past its crossing by play, and the play it needs was measured by bisection, framework by framework. It has a closed form. Pinned and pushed, the framework carries exactly one self-stress on each family, and the least play is the errors' bar misfits weighted by that stress, over the stress's size. The formula matches every measured price to 0.6%, says why two frameworks of ten prefer the far family, and finds that one in twelve does.

 A 45° rise: the slide's cam and the arm's, each as small as it can be. The quick programme with its rise of 20 shortened to 45°, cut for two flat-faced followers at the smallest base that keeps each cam convex, drawn to one scale. Left: a sliding face, base 185.56. Right: a face on an arm pivoted 90 from the cam's centre, lifting against the cam's turning and stood off −44.9, base 23.26, 8.0 times smaller. At that base the arm's cam stays convex for bases up to 1.0% larger, and no further. Prescribed motion

A steep rise spends the arm's room

A flat face on an arm that lifts against the cam's turning needed a fifth of a sliding face's base on a 90° rise. Steepen the rise to 45° and the slide's base grows 5.6 times, to 185.6, while the arm's least base grows only 3.2 times, to 23.3. But the arm's least base sits where its cam is about to go concave on the return as well as the rise: the base can be raised by only a hundredth before the return fails. Ask for a tenth's room above the base and the arm's base grows 5.1 times, nearly the slide's rate, and at 35° no stand-off gives that room at all.

 The line in contact swings, and the least of it is what a tooth carries. A helical pair at a transverse contact ratio of 0.4 and an overlap ratio of 3.5. Left: the field of action, 0.4 of a base pitch across and the face up, with the contact lines at the moment the total length in contact is least; each line crosses the face while advancing 3.5 base pitches, so only a short piece of each lies in the field. Right: the total length in contact over one base pitch, as a share of its mean, 0.857 at its least and 1.143 at its most; the closed form b/(εβ cos βb)·[εα εβ − min(nα nβ, (1 − nα)(1 − nβ))] gives the least exactly. Dashed: the spur optimum at εα 1.2, which carries its whole face for four fifths of a pitch and twice it for one fifth, 0.833 and 1.667 of its mean. Teeth

The face carries the least line

A helical pair that gives up transverse contact to cut its root's sliding was priced by the face it needs to carry the spur optimum's mean length of contact line: 2.65 times as wide, at a third of the contact on a 30° helix. But the length in contact swings through every base pitch and the peak load stands on its least value. That least has a closed form in the fractional parts of the two contact ratios, exact against the swept field to 10⁻¹³. The spur optimum's own least is its face, a sixth below its mean, so least against least the factor is 2.21 when the overlap ratio is a whole number. Real faces land between whole numbers and need 2.21 to 2.87 times the spur's, 2.39 on a spur face of ten modules, and above 2.65 only on faces under ten. And a steeper helix is not always more line: at a fixed face, raising the overlap ratio from 1 to 1.5 takes a third of the least line away.

 Re-chosen attachment points buy back a third of the tilt at four times the springs. The tilting-head lamp with one parallelogram and the smallest springs of the earlier essay. Every spring 2 cm free and every attachment point re-chosen for the least worst moment, all springs still pulling: 0.0611 N·m, a third less, with peak pulls 29.6, 0.0, 41.6, 195.1 N (lower arm, upper arm, elbow bracket to head, wrist), 266.3 N in all. The upper arm's spring is dotted where it has no stiffness. Machines you have met

A free length is absorbed where the springs attach

Give every spring of the smallest balanced desk lamp 2 cm of free length and refit the stiffnesses, and the lamp is left with 0.093 N·m of unbalance, growing in exact proportion to the free length. The spare fourth spring that ideal springs had no use for recovers under three per cent of it, and moving the head spring beyond the threshold only makes it worse. What absorbs a free length is a spring whose length barely changes. Re-chosen attachment points buy back a third of the unbalance that way, and pay for it with four times the pull.

 One tyre pressed harder: its patch grows and averages the road it used to ride. 0.6 m of a class C road built from 110 sines from 1 cm to 90 m, heights in millimetres and drawn far taller than they are, under a 0.3 m tyre. The tyre is a circular tread on a bed of vertical springs; at each position its centre height is the one at which the springs carry the load. The grey line is a rigid wheel's lowest point, riding the peaks. The coloured line is the level the loaded tread's patch rests at, 9 mm deflection, a patch 146 mm long on a flat floor: it passes over the short texture as an average of it instead of following its tops. Wheels, and where they may not go

A load reads the road a radius cannot

Press a tyre harder and its contact patch grows, and a patch averages the road where a rigid wheel rides its peaks. Averaging is linear, so the overcount it leaves falls as the patch's first power and scales with the road's roughness in exact proportion, not as the two-thirds power three wheel sizes read. One 0.3 m tyre at deflections of 3, 9 and 27 mm fits that law and extrapolates the distance to 3 × 10⁻⁷ on a smooth road and 6 × 10⁻⁶ on an average one, against 2 × 10⁻⁵ for three rigid wheels. On the roughest roads its lightest patch lifts off in places, and it does worse than they do.

 Half a turn on, the six legs lean alike again, and the seventh still the other way. The paired Gough platform from above, level, turned half a turn from its dead yaw: the base circle of radius 2.2, the platform circle of radius 1.1 (dashed), and the six legs, each now leaning 262.5° or 277.5° round the axis, two angles with the same sine, so the six are singular here at every position. The seventh leg (heavy) leans 105°, with the opposite sine, and does work 0.373 to 0.560 in size on the six legs' free twist across the level workspace: never nought. Several legs, one platform

A seventh leg has no dead yaw of its own

Seven legs can only be singular where six already are, since a 7 × 6 matrix that loses rank takes every 6 × 6 minor with it. Level, over a whole turn of yaw and the whole workspace, the paired Gough platform's six legs are singular at two yaws only, the dead yaw and the yaw half a turn on, and at both a seventh leg fails exactly when it leans as the six do. The best seventh leg leans the other way at both, so it has no dead yaw of its own. Its singular set exists, and it hugs the six legs' dead yaws, but it comes in only at tilts of 15° and more.

 Tighten the windows and the cage's limit drops below the grooves'. The six-ball Rzeppa joint bent 20° with its grooves offset 0.7 and a track play of one: across a ball pitch, the free lag the grooves alone allow (solid) and the cage alone allows with windows of clearance 0.5 (dashed). The joint's play is the smaller of the two. The cage is the limit wherever its curve is the lower, here at every input angle; it becomes the limit everywhere once the windows' clearance is below 0.559 of the track play. The largest play over the pitch is 0.163. What a joint is

A tight cage spreads the load until the tracks join in

A Rzeppa joint whose offset grooves set its play loads one ball first, and at equal torque its hardest-pressed ball carries up to three times a joint whose cage sets the play. Tighten the cage's windows below 0.559 of the track play and the cage becomes the limit at every input angle: the first load spreads over 3.6 balls' worth or more everywhere, the play falls, and at low torque the worst ball carries no more than a caged joint's. The cage does not stay in charge. Once the twist reaches the gap between the two limits the tracks engage, unevenly, and the worst ball climbs back towards the untightened joint's. A tighter cage widens the gap and pushes that torque up, so the windows' clearance is a choice about the torque at which even loading should end.

 The smallest springs that balance the lamp, and the one that is not needed. The tilting-head desk lamp with one parallelogram, its springs of no free length placed to make the total of their peak pulls over the working range least, inside an outline of brackets 0.15 m high, attachment points 0.08 m past the wrist and along the head's 0.15 m. Peak pulls 22.7, 0.0, 23.2, 17.5 N (lower arm, upper arm, elbow bracket to head, wrist), 63.4 N in all and 23.2 at most. The head spring meets the head at 0.0615 m, the threshold, and the upper arm's spring has no stiffness at all: it is left out. Drawn wrongly

Six things a balancing spring is not

More springs share the load, a spring that must push wants a compression spring, a stiffer spring balances better, a refitted stiffness repairs a free length, least squares finds the balance, and the best spring is one of no free length. Six things said about springs that balance a mechanism, each the natural reading of something true, each answered by a number the essays on the desk lamp and the rocking slide measured. They fail the same way. Each treats the spring as the thing to be sized, and a balance is decided first by where the spring is hung and how its error is judged.

 The cheap pivots lie in a valley tilted off the line x = 0. The price of the quintic's arc drawn by the line through the pivot bent by the square of the radius, four cosines, as a share of the arm's reach, at pivots a two-hundredth of a link apart across and a twentieth along, within 0.16 and 0.45 of the catalogue's point. Darkest: within 5%; then within 10% and 20%. The cheapest pivot at each height (dots) lies on a line at dx/dy = -0.172, 9.8° off the line x = 0 (dashed), with prices from 1.4% to 2.7%; on x = 0 itself the price reaches 7.3%. The curve as an equation

The quintic's cheap pivots lie off the line

Four cosines draw the quintic within a few per cent of the arm's reach from its catalogue pivot, and the earlier measurement found that cheapness surviving along the line x = 0 and dying across it. Mapped over a grid of pivots, it does neither exactly. The cheap pivots form a narrow valley tilted ten degrees off that line, about a twentieth of a link wide. In the valley four cosines hold the arc to 1.4–2.5% at every height out to 0.4 of a link; on the line itself, at the same heights, they miss by 6–7%. At two valley pivots every set of four cosines and every set of six was tried, and none beat the bent line. The amplitudes of the cosines a move adds predict a tilt a seventh as large, so they do not explain it.

 A growing four-bar walks away from its own double points. A four-bar with a rod of 3, its crank and rocker both of length t, pivoted so that both pivot circles pass through the slides' crossing, with the coupler point at (0.3, 0.2) in the rod's units. Drawn in units of t: the two pivot circles (dotted), their second meeting (ringed), the real coupler curve the machine draws (solid), and its real double points, filled where two real configurations meet and hollow where none reaches. At size t = 7 the curve's real double points are isolated (2.02), isolated (3.32), crossing (0.93); the number in brackets is how far the line of rod orientations through each passes from the unit circle's centre, a crossing below one and isolated above. As t grows the curve shrinks onto the crossing and the second meeting and the double points stay where they are. How many answers

A growing four-bar reaches none of its departing double points

A four-bar growing into the elliptic trammel loses three double points, and wherever they are real the question is whether the machine passes through them. In the limit it passes through none: a configuration that reaches a point a crank's length away needs its rod along an isotropic direction, which no real rod has. How fast each point leaves the machine's reach is a ratio of four distances, √(1 + λ²)·|Z|·|Z − Zₛ| / (2b·|Z + 1|·|Z + iλ|), and it vanishes only at the pivot circles' second meeting. Every crossing is lost at a finite size: through a cusp, when the law predicts it, or by meeting another crossing and leaving as a complex pair, when it does not.

 The pin opposite the third runs on the perpendicular slide, a quarter turn behind. The double slider of slides at 70° and a rod of 2, with a third block at a point of its rolling circle (radius 1.064) and a fourth at the opposite point, each on the slide its pin draws through the crossing. The two extra pins are the ends of a diameter of the rolling circle, so their slides are at right angles and their positions along them are 2r cos(φ − φ₃) and −2r sin(φ − φ₃): here 1.864 and 1.027. Four blocks and the count gives minus one; the rod turns, and the fourth pin passes the crossing a quarter turn after the third. The paths points trace

A fourth block can only take freedom away

A rod on a double slider can carry a third block on a third slide through the crossing and still turn, and a slide turned slightly wrong leaves it free only in two windows a turn. The hope was that a fourth block, at the point of the rolling circle opposite the third, would travel in opposition and cancel the error. It cannot, for two reasons. The opposite pin is the other end of a diameter: it runs on the perpendicular slide, a quarter of a turn out of phase, so its windows fall exactly between the third's. And whatever its place, a block only adds a condition the rod must also meet, so the rod's freedom is an intersection and can only shrink. With the double slider's own blocks exact, the third and the opposite fourth turned 0.0066 rad each lock the rod, where the third alone still moves at any error.

 A ratio of 3 costs both stages forty degrees. The largest quick-return ratio of a shaper's drive against the worst transmission angle it must keep, from 30° to 60°: an offset slider-crank alone, 1.448, 1.360, 1.286, 1.224, 1.172, 1.128, 1.092; a drag link alone ahead of a centred slide, 2.867, 2.632, 2.443, 2.284, 2.130, 1.968, 1.809; and the two in series, both stages keeping the angle, 3.975, 3.427, 3.006, 2.696, 2.432, 2.184, 1.957, against the product of the separate ceilings, 4.153, 3.580, 3.143, 2.796, 2.496, 2.221, 1.976. The train reaches 3 at 40.0°. Marked: 40°, where the train reaches 3.006. Linkages

A ratio of three costs forty degrees

A shaper's two-stage drive, a drag link ahead of a slider-crank, could not be given a return three times as quick as its cut while its drag link kept 40° and its slide 45°. Drawn as a frontier, the largest ratio against the worst transmission angle both stages keep, a ratio of 3 costs 40.0° in both. The two stages in series reach 95.7% to 99.0% of the product of their separate ceilings at every angle from 30° to 60°, so two stages relax one stage's bound very nearly by a product. Shared unequally, every degree the drag link gives up buys the slide between one and a half and three: the drag link is the stage to let lean. And the trains on the frontier cut badly, the ram's fastest speed in the middle of its cut 3.2 times its slowest at a ratio of 3. On the way, the drag link's own ceiling at 40° turns out to be 2.443, not the 2.365 a single search had found.

 The same 45° rise on an arm, with the quick return and with a return chosen for the arm. A cycloidal rise of 20 over 45°, cut for a flat face on an arm pivoted 90 from the cam's centre and lifting against its turning, each at the least base that keeps the cam convex. Left: the quick programme's return, 170° cycloidal, base 23.26, with 1.0% of room before the return goes wrong. Right: a 170° simple harmonic return, base 18.08, with 4.9% of room before the rise limits it. Prescribed motion

A gentle return cannot widen a steep rise

An arm lifting against a cam on a 45° rise had its least base pinned by the return: raise it a hundredth and the return went concave. Give it a return chosen for the arm, slower or gentler, and the least base falls from 23.3 to 18.1, a fifth less. But the window above it only grows from 1% to 4.9%, because once the return is out of the way the rise closes the window itself, and the arm's own reach closes the next one. With a tenth's room the best return saves 8% of base, and no return of nine gives a quarter's room. Constant acceleration, the law with the lowest peak deceleration, is the worst return of the three.

 Where on the field the root band sits, and when it is loaded most. A helical pair at a transverse contact ratio of 0.4 and an overlap ratio of 1.5. Left: the field of action, the path across and the face up, with the bands a tenth of the path wide at each end shaded, where the sliding is worst, and the contact lines at the moment the share of the load in those bands is largest, 1.36 times its even value. Right: over one base pitch, the share of the load in the root bands over its even value (solid), and the total length in contact over its greatest (dashed), least 0.667 of its mean. Teeth

The least line carries the root

A helical pair's length of contact line swings through every base pitch, and the peak load stands on its least value. The roots, where the sliding is worst, sit at the two ends of the path of contact. The share of the load carried there swings too, and in the limit of a thin root band it reaches its peak while the line in contact is least: half as much again as an even share at overlap ratios just above one, a quarter above two, a sixth above three. A whole-number overlap ratio removes both swings at once. So the overlap that makes the least line largest is also the one that spreads the root's load evenly, and a face sized on the least line alone understates the root's worst moment by the root's own swing.

 Allowed 100 newtons, the lamp spends them on its attachment points. The tilting-head lamp with one parallelogram, every spring 2 cm free, its attachment points chosen inside the earlier outline for the least worst moment on any joint with the springs' peak pulls capped at 100 N in total. Worst moment 0.0620 N·m; peak pulls 22.5, 28.7, 13.9, 34.9 N (lower arm, upper arm, elbow bracket to head, wrist), 100.0 N in all. Each spring's zigzag is drawn heavier the harder it pulls, and dotted where it pulls less than half a newton. Machines you have met

The first pull goes to the spring the lamp did without

Give every spring of the balanced desk lamp 2 cm of free length and let it have more pull, and the worst moment it is left with falls from 0.086 N·m at 62 N to 0.052 at 227 N, the most any attachment points inside the outline can do. Half of that improvement costs the first 18 newtons; the second half costs the next 147. Spent on stiffness alone, with the attachment points held, extra pull buys nothing at all. And the first place it goes is the upper arm's spring, the spare that ideal springs had no use for, hung lower on its bracket while the head spring moves out past the threshold. The wrist spring, which carried almost all of the earlier re-attached design's cost, takes the largest share only after 140 N.

 A clearance that has a sign, over a tolerance box. The same machine at each of the sixteen corners of a ±0.02 band on its four lengths, with the smallest gap over a whole drive computed at each. The nominal machine clears by 0.1041; the worst corner clears by 0.0763 and the best by 0.1321, a band 0.0558 wide from a tolerance of 0.02 on each length. Every other quantity computed from the lengths becomes an interval when the lengths do. This one has a sign, and an interval that reaches zero is not a wider answer to the same question — it is a different answer, because on that side of it the parts do not go together. 24 random draws inside the box beat no corner, which is the check that the extremes are where they are assumed to be. As built

A clearance inside a tolerance box

Every quantity derived from the lengths becomes an interval when the lengths become ranges. This one has a sign, and an interval that reaches zero is not a wider answer to the same question — it is a different answer, because on that side of it the parts do not go together.

 A slide turned about a point is free in a band round that point. The double slider of slides at 70° and a rod of 2, with a third block and a fourth at the opposite point of the rolling circle (radius 1.064), each on the line its pin draws through the crossing. Each extra slide is set to pass through its pin at one common pose of the rod, 45° from either pin's crossing, so a turn error of 0.01 rad pivots it about that pin's place then: -1.505 along the third's line and -1.505 along the fourth's. With a clearance of 0.01 the block is free while its pin is within c/sin ε = 1.000 of the pivot (the shaded bands). At the angle shown the third pin is at -1.505 and the fourth at -1.505: the rod is free. The paths points trace

Fit every extra slide at one pose

A double slider's extra slide, turned slightly wrong, leaves its block free in a band of the slide centred on the point it was turned about, and the rod free in windows where the pin passes through that band. Turned about the crossing, a third and an opposite fourth block have their windows a quarter of a turn apart and lock the rod from 0.0066 rad. Set each slide through its own pin at one common pose of the rod and every error becomes a turn about the place its pin has then, so every window contains that pose and the rod can never lock, with two blocks or six. Each slide pivoted where it alone is freest does better while the errors are small and locks from between 0.02 and 0.05 rad, where the common pose 45° from either crossing comes within a tenth of the best pivots there are.

 Five degrees below forty buy back an even cut at a ratio of three. At each worst transmission angle kept by both stages of a drag link ahead of an offset slide: the evenness of the cut, the ram's fastest speed over its slowest across the middle four fifths of the working stroke, for the train with the largest time ratio at that angle (dashed), which reaches 3.98, 3.75, 3.52, 3.43, 3.33, 3.24, 3.16, 3.08, 3.01 and cuts at 6.11, 5.21, 4.70, 4.35, 4.14, 3.90, 3.65, 3.46, 3.25; and for the most even train that still reaches a ratio of 3 (solid): 1.22, 1.26, 1.32, 1.42, 1.64, 1.90, 2.53, 2.73, 2.97, at 30, 32, 34, 35, 36, 37, 38, 39, 40°. The shaded band is 1.12 to 1.47, the best ratio-2 trains of the earlier essays. Linkages

Five degrees buy an even cut at a ratio of three

A shaper's two-stage drive reaches a quick return of 3 only by keeping no more than 40° of transmission angle in both stages, and the one train that does cuts unevenly, its ram 3.25 times faster at one point of the cut than at another. Searched for the most even cut that still returns three times as fast, the trains at 40° do little better, 2.97. But the price falls steeply below it: 1.90 at 37°, 1.42 at 35°, 1.22 at 30°. By 35° a ratio-3 shaper cuts as evenly as the best ratio-2 trains did. The even trains leave the long, nearly folded drag links that buy the most ratio for compact ones, and not one of them takes a longer rod than its angle requires.

 A number that runs away. The Denavit–Hartenberg offset of a pair of nominally parallel axes, against how far from parallel they actually are, for a fixed out-of-plane tilt of 0.05°. The marks are extracted from the geometry by finding the common normal; the line is the closed form −A cos α cos β sin β / (sin²α cos²β + sin²β), and the two agree to 2.4e-16 relative over three decades. At 30° of twist the offset is 0.0030 of a link length; at 0.01° it is 1102. The dashed line is the worst case over the tilt, which sits at β = α and is exactly A/2α. Nothing about the machine has changed by as much as a degree. Numbers that were measured

A number that runs away

A hundredth of a degree of unintended twist on a nominally parallel pair of joint axes puts the Denavit–Hartenberg offset at −1,102 link lengths. The extraction from the geometry and the closed form agree to 5 × 10⁻¹⁶ over three decades, and the worst case over the tilt is exactly A/2α.

 The bores, and the one line that has to pass through all of them. A hinge of 6 knuckles, its bores drawn at the distance each was made from the nominal axis in units of the bore tolerance. The leaf is a rigid body, so its pins are on one straight line — two parameters of position and two of direction — and it assembles when some line passes within the clearance of every bore. The line drawn is the one whose largest miss is smallest, and that miss is 0.875 of the tolerance. 2 of the 6 bores are at that distance and hold the fit; the rest are slack and could have been bored anywhere inside it without changing the answer. As built

A piano hinge is not forty door hinges

A three-knuckle hinge works because the misfit its bore errors create is smaller than the play already in its pins. A piano hinge has forty knuckles and thirty-nine of them are redundant, so the obvious reading is that it needs thirteen times the play. It needs two and a half times, and it can never need more than the bore tolerance itself — because a rigid leaf has one axis and a line through the middle of the errors misses every bore by at most the largest of them.

 Which numbers have a size. Twelve quantities from eight fields of mechanism kinematics, each scaled by taking every length in its mechanism up and down together and fitting the power its value follows. A shape lands on zero, a length on one, a curvature on minus one, an area on two, and every row lands on an integer to 3.3e-11. The sorting is not a units argument — it is measured from each field's own library, by asking that library the same question. What it says is that most of what kinematics computes is a shape: a mobility, a Grashof class, a transmission angle, a contact ratio and a velocity ratio are all unchanged by making the machine bigger, so all of them are recoverable from a measurement that cannot recover a single length. The row worth stopping at is the tolerance band: held to a fixed ±0.01 it lands on minus one, because ±0.01 is a length and a bigger machine held to the same absolute tolerance is a better machine. Written as a percentage the same band lands on zero. One quantity, two drawing conventions, two different kinds of number. Numbers that were measured

Which numbers have a size

Take every length in a mechanism up and down together and fit the power each computed quantity follows. A transmission angle lands on zero, a coupler point's speed on one, a path curvature on minus one, an enclosed area on two — and a tolerance band held to a fixed ±0.01 lands on minus one, which nobody would guess.

 What each instrument recovers. The same twenty poses of the same four-bar, read three ways. A protractor on the output link recovers 3 of the four lengths and leaves the fourth exactly invisible, because its readings are dimensionless in the lengths and scaling the machine does not move them. A coordinate machine on the tracing point recovers all six parameters — the four lengths and the two that say where the tracer sits — at a condition number of 162.3. Using both recovers the same six at 26.1, 6.2 times better, which is the case for putting two instruments on one machine: not more parameters, better-conditioned ones. Numbers that were measured

What this field cannot measure

Every field of this collection has a boundary and this one has three: a direction the readings cannot span, an alternative no derivative detects, and a model nobody thought of. The first is computable exactly, the second needs a search, and the third is not detectable from data by any method at all.

273 essays in 3 parts: part 1 · part 2 · part 3

All themes