Classic Mini · Arden 8-port · progress report, October 2026

45 mm of manifold

Between the carburettors and the cylinder head of this engine there is room for an intake manifold about 45 mm long. In that distance the air has to squeeze from a 45 mm carb throat into a 31 mm port and, on one cylinder, step 33 mm sideways.

This is a record of trying to design a better manifold for that gap, with open-source CAD and flow simulation on a laptop. Everything you can drag on this page is driven by the real geometry or by one of more than 100 simulations. The part below is where it starts: the existing manifold design, two pieces, one per carburettor.

The engine is a 1293 cc A-series with an Arden 8-port head, which gives every cylinder its own intake port. Two Weber 45 DCOE carburettors feed it, one throat per cylinder. It made 142.7 hp at 8150 rpm on the dyno, so at full song each port swallows air at around 100 metres per second.

The box it has to fit in

On an 8-port Mini the intake faces the front of the car. Three things are fixed. The bonnet runs along the top of the carbs, so they cannot move up. There is no room ahead of the trumpets, so the manifold cannot get longer. And the throttle linkage sits in a 30 mm gap between the two carbs, so they cannot move together.

The head adds its own twist: the four ports are 65, 75 and 79 mm apart, while the carb throats are evenly spaced. Seen from above, every runner has a different sideways distance to cover. The off-the-shelf manifolds for this head are 60 or 90 mm long. Try those lengths below, then come back to 45.

Two rules of thumb for ducts say that a bend's radius should be at least 1.5 times the bore, and that a runner should arrive square at its port. Switch the figure to the S-bend and they collide. At 45 mm, cylinder 4 needs a bend radius smaller than half its own bore, and the inner wall folds over itself. At 90 mm the same runner is merely tight.

The existing manifold takes the other option: each runner is a straight, slanted, tapering tube that meets both faces at an angle. The question for this project was whether anything does better.

What the air does at a corner

To find out, each runner was simulated on its own, the way a flow bench would test it. A straight pipe stands in for the carb, another for the port, and air is pulled through at an average of 100 m/s in the port. The figure shows a slice through the result, seen from above. Colour is speed. The streaks are tracers carried along by the computed flow.

On cylinders 2 and 3 the air follows the walls and fills the port. On cylinder 4 it cannot make the corner where the slanted runner meets the head. It peels away from the inside of the turn, and a slow strip runs down one side of the port. The tracers there barely move. That strip is wasted port: the air that does get through has to squeeze past it faster, and the pressure pays for it.

So the job was clear. Leave the two middle runners alone and fix that corner on cylinders 4 and 1.

Two angles

The first family of shapes kept the runner round and varied two numbers: the angle at which it meets the head, and the angle at which it leaves the carb. Zero for both is the S-bend from before. Large values for both give back the slanted tube. Every cell in the grid is its own simulation of cylinder 4.

Set the carb angle to zero and watch the middle of the runner. Leaving the carb dead straight forces a violent turn halfway down, and those shapes are worse than the slanted tube they were meant to beat. The best corner of the grid is a small angle at the head with a moderate one at the carb, about a quarter better than the existing runner.

One idea did not survive this round. Keeping the bore wide for longer, so the air turns while it is still slow, made every shape worse. A further fourteen variations around the best cell failed to beat it. Round runners were exhausted.

Shaping the walls

Look at the flow pictures again and the trouble always starts in the same kind of place: a corner where a wall turns away from the air. Each side wall has exactly one. So the second approach shapes the two walls separately and lets the cross-section become whatever it must to keep its area. The runner comes out narrower sideways and taller instead.

A tall, narrow duct makes a bend feel gentler to the air than a round one of the same area, because the distance across the turn is smaller. Step through the shapes below. The first is the existing runner. The next two were a guess that rounding the corner at the carb end would help. The last three are wall-shaped.

Softening the carb-end corner did nothing useful: the wall then has to turn harder further down. The wall-shaped runners are a different matter. The best one loses about 30% less than the existing runner, and a version held to 45 mm in height, no taller than the carb throat, keeps most of that.

A flattering mesh

A simulation chops the air into cells. To try dozens of shapes quickly, the search ran on a coarse mesh of about 0.3 million cells per runner, a few minutes each. The winners were then run again with five times as many cells, an hour or more each. Here is the same runner on both.

The improvement was real but smaller than the quick runs had promised. On the coarse mesh this shape removed about three quarters of the loss that could be avoided on cylinder 4. On the fine mesh it was closer to 60%. For cylinder 1 the gap was wider still: 75% promised, 35% delivered. The coarse search had been quietly favouring shapes with a sharp corner at the carb. From then on every decision was made on the fine mesh.

Those are the fine-mesh numbers. The gap between the best and the worst cylinder shrinks from about 830 Pa to about 300 Pa. Cylinder 4 stays the worst, and 33 mm sideways in 45 mm is probably why. The model also has a floor to its eyesight: a test runner with no sideways step at all scored 1000 Pa, so differences of less than about 90 Pa mean nothing.

A fine-mesh case takes about an hour on an eight-core laptop, and a batch takes a night. Power management cost more time than the physics did. One batch was lost when the laptop dozed off on battery. The first overnight run survived a flat battery only because the machine hibernated and carried on when it was plugged back in.

What the photo showed

Every simulation so far assumed that the port carries straight on into the head. A manufacturer's photo of the same head, taken square onto the intake face, says otherwise. Through each open port you can see the guide of that cylinder's inlet valve. The guides are evenly spaced. The ports are not.

A camera makes things that are further away crowd toward the middle of the picture, so a guide deep inside a port does not appear straight behind the opening. The figure does that sum. Rings mark where each guide sits in the photo. Dots show where it would appear if the valve were straight behind its port. Three dots are already home. Move the fourth.

The dot lands in its ring at about −11 mm. Cylinder 4's port opening sits 11 mm inboard of its own valve, which is exactly why the port spacing is uneven. Now look at the lower half of the figure. The manifold carries the air 33 mm one way, and the head then takes it 11 mm back. The simulations only ever saw the first half of that zig-zag.

The same photo gives the height of the ports above the bottom of the head, and that tells a second story, this time from the side.

The port dives at around 35 degrees, starting right behind the face, while the manifold arrives level or slightly rising. No simulation so far contained that turn. It may also turn the tall runner from a good idea into a bad one: its floor climbs steeply into the port just where the port's own floor falls away.

The recommendation to freeze the shapes had been written that morning. It was withdrawn the same day.

Against a port that dives

So the simulation got a new port: one that falls 50 mm over 68 mm and, for cylinder 4, leans 11 mm back toward its valve. It has no valve and no guide. Its only job is to give the manifold the right kind of turn to feed. Every candidate was run again on the fine mesh. The pressure is now counted down to a point below the bend, so these figures include the port's own turn and cannot be set against the ones further up the page.

Two things are different from before. The bend inside the head now costs every cylinder roughly 750 to 1100 Pa, more than the manifold in front of it. And the air comes off the short inside of that bend and never quite settles, so the totals for cylinder 4 wobble by about 100 Pa from one stretch of the calculation to the next.

Against that port the tall runner, the winner of everything so far, is barely better than the existing one. It still cures the slow strip at the head face, but the port then loses as much as it did before. The runner held to 45 mm is a different matter: a fifth less than the existing runner. Now step to the one held to 50 mm. The gain is gone again. The switch sits between a runner that never gets taller than the carb throat and one that does.

Three smaller results came out of the same batch. The 11 mm lean of cylinder 4's port costs nothing that can be measured. Putting the tall runner's extra height on top, so that its roof comes down into the port while the floor stays low, made it worse instead of better. And arriving squarer at the head, now that the port leans back, did not pay either.

Checking the soft number

A gain that lives in the bend of a port whose shape was read off a photo deserves a second look. The soft number in that estimate is the distance from the head face to the valve. So the existing runner and the 45 mm runner were both run again with the valve at 60 and at 80 mm, and the six cases that matter most were run for twice as long to average out the wobble.

The gain holds at all three depths, at about a fifth each time. The longer runs moved no result by more than 40 Pa. One more case settled what does the work: the tall runner's own walls, held to 45 mm, scored within 10 Pa of the 45 mm runner. It is the height, not the wall angles.

Why the height matters is not established. The best guess is that a taller runner's floor has to climb further into the port, just where the port's own floor falls away. And a gain that appears in full at 45 mm and not at all at 50 is a switch between two states of the flow, not a smooth trend, in a port that has no valve in it. What speaks for it is that it survives three port depths, and that the same runner was also good with the straight port: 1300 Pa against 1740.

What is frozen

Cylinder 1 got the same treatment. With the diving port its reshaped runner, uncapped, had come out level with the existing one. Held to 45 mm it is slightly ahead, by less than this model can resolve. What it does do is remove the strip of backward-flowing air at the head face and bring cylinder 1 level with the two middle ones.

Those are the numbers with the diving port. The air passages are now fixed: wall-shaped runners for cylinders 4 and 1, never taller than the carb throat, and the two middle runners as they were. The spread between the best and the worst cylinder shrinks from about 450 Pa to about 100 Pa. The title of this page was meant to describe the length of the manifold. It turned out to describe its height as well.

Room for the tools

A fixed air passage is not yet a part that can be bolted on. Two tools have to get past these runners: the spark plug socket, and a spanner on the nuts that hold the manifold to the head. Both turned out to want space the passages were using.

The plugs sit between the ports and lean about 40 degrees. Follow the axis of the plug next to cylinder 3 and a socket passes 6 mm from the air inside that runner, one of the two that were supposed to stay as they were. And the reshaped cylinder 4 runner now sweeps right over one of its own studs: the corners of the nut would have sat inside the air passage.

So both passages were pushed in locally, with soft edges, leaving the round port and throat alone. The plug relief is up to 6.8 mm deep and takes almost a tenth of the section at its worst point. It costs about 30 Pa. The stud relief costs nothing the simulation can see. After a hundred runs spent chasing tens of pascals, these two came nearly free: the narrowest point of the whole passage is the port, and neither relief is anywhere near that small.

One correction came from the owner and not from the simulation. The first stud relief made room for the nut and none for the spanner that has to turn it. Checked properly, with an open-ended spanner swung through the 30 degrees it needs, six of the eight nuts were fine and two were not. Deepening the relief was the obvious fix. Lowering the nut's seat by a millimetre and thinning the wall above it to 1.5 mm worked better: the relief shrank from 4.5 to 2.8 mm and an ordinary spanner fits.

Two numbers in this are assumed. The socket is taken as 20 mm across, which is what the plug pockets in the head measure in the photo, and the plug axis was read off the old part and is good to about 2 mm. Projected into the head photo it runs through the middle of the machined pocket, which is reassuring but is not a measurement.

From air to metal

Until here the part was a hole. The metal around it has now been drawn: titanium, to be printed, 2 mm thick wherever the air touches it, measured square to the wall so that a leaning runner is no thinner than an upright one. Where a runner meets a flange it widens into a blend. Then everything that needs room is cut away again: three plug sockets, eight nuts on the head studs with the swing of their spanner, and the nuts behind the carb flange.

The outside did not get to choose much. Around every stud the flange is thinned to a flat lug, so that the nut sits 4 mm above the head, 3 mm at two of them, and the blends stop where a spanner has to pass. The wall is 2.0 mm over nearly all of the surface. Over the end of one stud it is – mm, as agreed earlier, in a spot a few millimetres across.

Drawing the metal showed something the air passages could not. No nut can be dropped onto its stud once the manifold is against the head: a runner or its blend stands over all eight. Each nut has to slide in sideways at seat level while the manifold is still held about 10 mm off the head, and be wound on as the manifold goes home. The existing design works the same way at two of its studs, and the owner is content with that. But it was found by measuring the finished solid, not by planning it.

Most of the work in this step was a fight with the geometry kernel. A cut that fails does not always say so: twice an operation handed the part back untouched and reported success. Every cut is now checked twice, by the volume it removed and by testing that no point of the result still lies inside the tool. Every clearance quoted here is measured on the finished part, not inferred from the passages.

The part will be printed in metal powder, layer by layer, and a printer cannot build a ceiling in mid-air. Any surface that faces down and lies flatter than about 45 degrees needs a support under it, and a support leaves a scar where it is broken off. Inside an air passage that is the last thing wanted. So every way of standing the part in the printer was tried, 4000 directions per body. Flat on its head flange the passages need no support at all, but the whole underside of the carb flange does, 50 to 65 cm² per body. Tipped by about 47 degrees the passages still need none, and only about 10 cm² of the outside does. So the passages can be printed exactly as the flow study left them. The file for the printer carries 1.2 mm of extra metal on the two sealing faces, to be machined flat afterwards.

Three kilograms on two short pipes

There is no bracket under the carburettors. Each one, about – kg with its linkage, hangs on two runners 45 mm long. A first sum on the bare passages had said they were lightly loaded, about 1 MPa of bending for every g of acceleration. A finite-element model of the finished part agrees in general and disagrees in one place. Over the runners the tension stays below – MPa per g. In the pocket for the nut under the cylinder 4 runner it reaches –: the wall is thinnest there. Switch the figure above to “how hard it works”: a ring marks the spot. Turn the part over to see it.

That number took two attempts. The first model said about 12 and was wrong. The figure quoted here is from the corrected one, with every inside corner of the pockets rounded and 1.2 mm of radius in the loaded pocket, the most its nut leaves room for.

What that is worth depends on how hard the engine shakes its carbs, which nobody has measured here. With typical published strengths for printed titanium, used as it comes out of the printer, that one spot reaches its fatigue limit at about – g of steady up-and-down shaking and the other body at about – g. Bending it for good would take around – g. So static strength is not the question. The margin against fatigue at one rounded pocket is, and it waits for the print service's own figures.

The more interesting result is about rhythm. A carb on two short runners is a weight on a springy stalk, and like a swing it has rhythms of its own: shake it in one of them and small pushes add up to a large motion. That is resonance. A four-cylinder engine shakes hardest twice per revolution, so its rhythm climbs with the revs: from 5000 to 8300 rpm it is – to – Hz. If a carb were bolted solidly to this manifold, its two lowest resonances would sit right in that range.

Bolted solid, both carbs would pass through a resonance in the revs this engine uses. So would the existing design: in the same model, in aluminium, it lands within about twenty hertz of the new part, because both are set by the same 45 mm of runner. That shows equal stiffness and nothing more. The existing design has only been printed in plastic to check clearances and has never run on the engine.

Neither part will be bolted solid. The carbs sit on O-ring soft mounts, which are sold to stop the fuel frothing in the float bowls. Here they do a second job: a carb floating on rubber no longer drags the manifold's own resonance, – Hz, down into the rev range. How well that works depends on the mounts' stiffness and on how tight the nuts are, and neither is in the model. So the finding is that rigid bolting would be wrong for this manifold, not that the soft mounts have been shown to be right.

The structural model is linear: no bolt preload, no contact, no rubber. The carb's centre of gravity is assumed 107 mm from the head face. The strengths are typical published values, not the print service's.

Still open

The shape study cannot go further on a port that is itself an estimate, and the structure cannot go further without the carb mounts. What comes next:

• real measurements from the head, and a putty mould of one port, to replace the photo estimates;
• a flow bench comparison of a printed 45 mm runner against a printed copy of the existing one, which is the test that can confirm or refute the flow result;
• a plastic test print, with the same cut-outs as the metal part, tried on the engine with the real plug socket, spanners and nuts, because the socket size and the plug axis are assumed;
• a tap test with the carbs on their soft mounts, to see where they really ring;
• the print service's own fatigue figures for titanium, in place of textbook ones;
• a check with pulsing flow at several engine speeds, because everything here is steady flow.

Geometry: CadQuery. Flow: OpenFOAM, steady, k-omega SST, one case per runner. Structure: Gmsh and CalculiX, linear. The modelling, simulation set-up and this page were written with Claude Code; the car, the measurements and the overnight runs are the owner's. The flow figures replay slices of the computed velocity fields. They are not live simulations.