Ultra Low Dropout Regulator for flicker free LEDs

Cheap 12V LED lighting strips are very sensitive to small voltage changes. This is because they just have multiples of three white LEDs in series plus a small resistor to set the current. The voltage across the LEDs is more or less constant at 3 × 3.2V, so the resistor sees the difference between that and the supply voltage, around 2.4V. This means that a 10% change in supply voltage produces around 40% change in current and hence brightness.

ATX PSUs have poor regulation on the 12V rail but this doesn't affect the rest of a 3D printer because everything is regulated downstream. The stepper drivers are constant current and the heaters are temperature controlled.

Here is a dip in the 12V rail caused by the 10A bed switching on. As you can see the swing is 12.2 to 11.6V, around 5%. This results in about a 20% reduction in light which is very noticeable.

One way to fix the problem would be to make an accurate 12V from the unused 5V rail using a small boost converter. I didn't want to cause any EMC issues, so I decided to use a low drop out (LDO) regulator to regulate to just below 11.6V. I thought it would just be a single chip solution, but after a lot of searching, the lowest drop LDOs I could find were 0.5V and I didn't want to waste that much.

I decide to roll my own using a low RDSon MOSFET as the series element. The LEDs only take about 0.5A, so the drop out voltage with a MOSFET with an RDSon of a few milliohms can potentially be a few millivolts, hundreds of times better than any LDO chip I could find.

In practice I used a BTS134 MOSFET I had lying around with an RDSon of 60mΩ. To put that in perspective that is less than the voltage drop in the wires I used.

This is the circuit I came up with: -

I regulate the negative side of the LEDs so that I can use an N channel MOSFET. Virtually any logic drive MOSFET should work. I wired this after my RPi switching circuit described here: hydraraptor.blogspot.co.uk/2014/06/lights-camera-action.

I use a TL431 adjustable precision reference as the control element. These are great little chips that can do anything from replace a zener diode to being an audio amplifier. The green LED acts as a level shifter to ensure the cathode voltage of D1 is higher than the reference voltage, a requirement of the chip. A white LED would give more margin.

VR1 should be initially set for the maximum output voltage and then gradually reduced until the flickering stops when the bed is switching. Handily the green LED flickers when the LEDs are flickering aiding adjustment without being blinded by the light.

C1 prevents the circuit oscillating at about 50kHz by providing negative feed back. I found its value by trial and error as there wasn't any phase shift versus frequency data in the data sheet. Basically R3 and R4 form an RC network with the gate capacitance of the MOSFET. This will produce a 90°phase lag at high frequencies. The TL431 must also create a 90°phase lag around 50kHz. To prevent oscillation C1 has to reduce the loop gain to less than one when the total phase shift is 180°. The circuit was stable with 1nF but had a bit of ringing. With 10nF it was a bit slow to respond.

The scope view below shows the performance of the regulator. The yellow trace is the 12V supply to the LEDs. The cyan trace is the negative supply to the LEDs (shown at a bigger scale). The purple trace is the difference, i.e. the voltage the LEDs see. It is virtually constant and the drop across the regulator goes as low as 40mV.

The maximum voltage across the MOSFET is only about 600mV (the size of the dips in the 12V rail). That makes the dissipation about 300mW, which is easily dissipated without a heatsink. Don't set VR1 too low though as the dissipation will go up rapidly.

Here is a veroboard layout done in VeeCad: -

If there is enough interest I could make a PCB to combine this with the RPi interface and also add a 5V in connector to avoid the need for the micro USB plug.

Mendel90 GitHub catch up

I finally found time to update GitHub with some Mendel90 changes that I have had in the works for a long time. The problem with releasing them sooner was that they were all not quite finished and / or would make unintended knock on changes to the kits I was producing. In particular the changes I did to make a Huxley90 in a hurry for the TCT show and the E3D mods kindly contributed by Philippe LUC that conflicted greatly with it, so needed a lot of work to merge.

OpenScad

I also updated to the latest version of OpenScad. The upside was that hull and some of the 2D operations are much faster. I was also able to replace all the calls to minkowski with offset as I was only using it for 2D offsetting. The net result is it is now four or five times quicker to generate the preview and the STL files. The downside is that the 2D sub-system now uses fixed point coordinates but the rest of OpenScad doesn't. This makes it difficult to get 2D and 3D geometry to match up. For example, an extruded circle now has slightly different vertices to a cylinder of the same size. This created a few degenerate triangles requiring that I changed the way I constructed some objects in order to get nice clean STL files.


The solution in the case above was to make the cylinder slightly bigger than the circle used to make the pointer.

On the up side it seems OpenScad has got better at handling unioning exactly coincident faces since I first wrote Mendel90, so I could remove some of my small offset bodges to avoid z-fighting.

Another benefit is that the X end brackets now slice correctly in Slic3r, as the bug that caused internal faces to point the wrong way has now been fixed. Skeinforge doesn't care about face orientation, it just counts edges to work out what is inside and what is outside. Other slicers got confused and filled in the nut cavity.

Along the way I discovered that, although OpenScad now has trig functions that are accurate for multiples of 90 degrees, etc., it doesn't use them in rotate, or vertex creation for circles and cylinders. It converts to radians and uses the library trig functions. Degrees can never be represented accurately as radians in floating point because Pi is irrational, not to mention transcendental. To get round this I now override the built in rotate with a user space version that uses the accurate sin and cos degree functions.

module rotate(a)
{
 cx = cos(a[0]);
 cy = cos(a[1]);
 cz = cos(a[2]);
 sx = sin(a[0]);
 sy = sin(a[1]);
 sz = sin(a[2]);
 multmatrix([
  [ cy * cz, cz * sx * sy - cx * sz, cx * cz * sy + sx * sz, 0],
  [ cy * sz, cx * cz + sx * sy * sz,-cz * sx + cx * sy * sz, 0],
  [-sy,      cy * sx,                cx * cy,                0],
  [ 0,       0,                      0,                      1]
 ]) children();
} 

Not surprisingly every STL and DXF file generated is now slightly different numerically but hopefully not dimensionally. I made a stable branch to record the state before these global changes, just in case. GitHub has some excellent image and STL comparison views but unfortunately it gives up if more than a handful of files have changed and there are hundreds in the Mendel90 tree.

Wade's Block

After a few people started to report broken or cracked Wade's blocks I strengthened it a bit around the bearing block. I also made the bearing sockets a bit bigger so there is less stress created pressing them in. Kits from around March 2015 have shipped with this version.

X Carriage

When Philippe LUC created the E3D branch he fixed a few bugs. One of these was that the X carriage top was only 2mm thick, when the design intent was 3mm. This was due to the fan duct using the same variable name. Whoops! I have updated it now in the main branch. I also made the nut traps for the fan bracket screws deeper to allow for longer screws and to allow them to be withdrawn further without loosing the nuts. This makes it easier to remove and replace the duct. Simply removing the washers is an alternative.

E3D Hotend

I temporarily parked Philipp's mods in an E3D branch until I could merge them. I have now updated the master branch to support E3D V5 and V6 hot ends with this one line change to the config file. The generated files for V6 that are different from standard build are in new folders dibond_E3D and sturdy_E3D and I have deleted the temporary E3D branch.

There is no room for the right hand wing nut because it clashes with the hot end's fan. Fortunately the carriage has always had nut traps to allow the screws to be inserted from below. A plain nut above can then be used to secure the extruder.

Primarily the things that change are the Wade's block, the fan duct and the fan bracket. The Wade's block has no extension to avoid losing more Z build height than necessary and a plain screw hole on the right end instead of the hex socket.

The fan duct has to slope downwards to avoid the E3D heatsink. That creates a sloping bridge that is also skewed horizontally. I haven't found a slicer that handles this properly yet, having tried Skeinforge, Slic3r, Cura, Kisslicer and even paid for Simplify3D! I have blogged about their failings in another post here: hydraraptor.blogspot.co.uk/2016/01/a-bridge-too-far. Any other slicers I should try?

Another bug Philippe noticed is that there was almost no clearance between the fan and the belt. Fortunately the belt is twisted so it actually does clear the fan. I have added more clearance as
Philippe did. It makes the fan bracket and fan duct 2mm longer. If you print either from the new files be sure to print both or the duct will be misaligned.

I also improved the internal shape of the duct a bit. From this: -

To this: -

It probably doesn't make a lot of difference but a comparative test of various fans and ducts will be the subject of a later post.

Even with the shortened Wade's block the E3D V6 hot end is 4mm lower and the V5 is a bit longer still. If you retro fit it to an old machine you will lose 4mm Z travel. If you are building a new machine then there are alternative files which add 4mm to the height of the frame and lengthen the Z smooth rods and threaded rods on the bom. That also has a knock on effect on the shape of the spool holders and the dust filter. If you use the larger sheets be sure to get the correct size rods and use the correct spool holder parts to match the frame.

New Lighting Options

I redesigned the lighting system I described here to work with some commonly available LED light strips. These consist of an aluminium PCB strip that slides into an aluminium extrusion with plastic end caps, which I discard. Instead of printing a bar to hang the lights and camera from I now add printed end caps to the light strip and uses those to hinge it from the frame edge clips. I then hang the camera from the strip with its own hinge.

The strips come in 500mm lengths but they can be cut at discrete points between every third LED. They are described as "50CM 5050 SMD 36 LED Warm White Aluminium Rigid Strip Bar Light Lamp" and I bought them from bgood2010 on eBay.

I got some from another seller and although the eBay picture looked the same the extrusions where actually not as deep. The STL files on GitHub are for 8.6mm deep extrusions and are generated by light_strip = RIGID5050_290. Setting it to Rigid5050_290 generates the clips for 7mm deep extrusion. Other sizes can easily be accommodated as long as they are rectangular. The definitions are here.

Rather than waste the off-cut I mount it above with a second pair of end caps that clip onto the main light strip. These are set back just far enough to avoid the build volume in the unlikely event you print something tall at the back edge of the bed. This is calculated by the model with lots of trig and Pythagoras maths. Set show_rays = true to see this view showing that the camera and lights are pointing at the centre of the bed and the build volume is clear.

Another light strip that can be selected is this one: FSRP3W, discovered by Alzibiff.

Again the end caps are removed and replaced with printed ones that clip into the screw channels in the extrusion. There is no room for the plug so I just solder the wires on.

It looks neater and gives a more diffuse light but is not as bright as the double strip of 5050 LEDs and is more expensive. I bought it from www.ledlightingandlights.com.

The only problem with these light strips compared to my original Sanken ones is that they are unregulated, so they flicker when the bed switches on and off. I described how I fixed that here. I also need to update the mounting for the Raspberry Pi to accommodate the plethora of new Pis that have appeared since my original design.

Huxley90

The Huxley version is scaled down in the same way as the Sells Mendel was scaled to make the Huxley. It has a build volume of 150mm cubed and uses NEMA14 motors, 6mm smooth rods and M3 fasteners for the frame. There is a good photo of it alongside the full sized machine on Ivor O'Shea's blog post.

The NEMA14 motors have about half the torque of the NEMA17s when driven with the same current. The Y carriage and bed have about half the area hence half the mass, so that is about right. Also a NEMA14 has half the mass of a NEMA17, so the X carriage also has about half the mass.

I believe the flex in the middle of the rods is proportional to the length cubed times the weight divided by the bar radius to the power of four. The length of the X rods is almost exactly 75% of the Dibond version and the diameter is obviously 75% as well. The relative flex then boils down to 0.5 / 0.75 = 0.67. So going down to 6mm rods is justifiable as well. Everything scales very nicely physics wise.

As the design is fully parametric shrinking it should have been easy, but because vitamins don't scale perfectly lots of snags arose where things clashed. A typical example was the x_motor_bracket. The NEMA14 motors are smaller but the raised boss around the shaft is the same size. This makes the bracket a different shape and it then needs a support to print it.

Half a truncated teardrop with a crutch!

The heated bed was made with veroboard and coincidentally has the same resistance as a full sized Prusa PCB, so the machine takes the same amount of power but heats up about twice as fast. There is no room on the frame for an ATX PSU, so I used an external XBOX 200W PSU. I couldn't find a spec for the 5V standby rail but it seems to supply enough current to power a Raspberry Pi.

Direct Drive Extruder

The extruder is where the scaling fell down a bit. The original Huxley used a Bowden drive to make the carriage small. I didn't fancy that but I didn't want to have the carriage as big as a geared extruder would need, so I went for direct drive with a NEMA14 and 1.75mm filament. 

The filament needs about one third of the force to feed and a Wade's has roughly 1:3 gearing, so a direct drive NEMA17 is about equivalent. The NEMA14 has half of the torque, so it is a bit under powered. I used the smallest drive pulley I had which was a mini hyena from Laszlo Krekacs' Indigogo campaign. Unfortunately I don't think the small diameter version is available now. I could probably make one from a hobbed bolt if I needed to or hob one from scratch.

It feeds PLA fine at 200C but isn't able to pull it off a spool. I will try a spool holder with a central bearing rather than the rim bearings to see if that is low enough friction. If that doesn't work I might try a powered filament supplier like the one on the first Up printer preserved here. Or maybe even try Bowden drive.

The design is parametric so there is a NEMA17 version suitable for Mendel90. I just need to adapt it for a commonly available drive pulley. It should just be a matter of adding a description here.

It can also use the E3D hot end but that doesn't fit between the bars on a Huxley90.

So that is Github up to date and hopefully correct although I haven't tested a lot of these changes.

I noticed that Blogger is now a lot worse than it used to be. Headings and pictures are now a nightmare.

End Game

Since my last post there have been many expressions of sadness in the comments here and here. I'd be lucky to get so much mourning at my funeral I think! Yes it may be sad for potential customers that missed out, but not for us. Fortunately we are at a stage in life where time is more important than money to us. It does feel a bit weird though as the house is now silent after having three or four machines printing solidly for about five years.

The harsh reality is we always planned to stop selling kits before demand dried up completely because otherwise we would be left with unsold stock. In order to to get volume discounts and cover long lead times we needed to carry around £12,000 - 15,000 of stock.

As we didn't hire premises or staff, and had enough capital to cover cash flow, the only business risk we had was being left with stock. It was very hard to predict future demand because it fluctuates wildly with currency changes and, with a new 3D printer coming out nearly every day, there is a risk sales could suddenly dry up if a better cheaper kit emerged. For example, a £40 Prusa I3 from China with 2 rolls of filament and LCD, bargain!

In the end the issue with Dibond made the tricky decision of when to stop for us. This happened a few months ago and we synchronised most of our stock to all run out together.

Yes we could have used other sheet materials, solid aluminium and steel were suggested. However, these are more expensive to buy and machine, much heavier and therefore more costly to ship. Although they are a bit stiffer there is no real advantage to the frame being stiffer because the forces due to accelerating the axes is along the plane of the sheets, where almost any sheet is very stiff indeed. As long as the machine sits on a solid work surface it doesn't flex at all when printing, unlike some of the triangular prism machines.

The good thing about an Open Source product is that the design and instructions will always be available, so people can still self source, or another company could pick up the baton. We will always be able to supply printed parts and, fastener kits, wire and sleeving, hobbed bolts and the extruder PCB. See the RepRap forum for prices. Everything else can be self sourced.