Tag Archives: galvo

Precision VS accuracy : why I sold my Form1

Disclaimer: I had one of the first Form1 printer sent to backers so my point of view might be biased and I’m sure Formlabs have perfected their calibration process and maybe even changed some mechanical parts of the printer to answer some of the point raised in this article. Take all of this with a grain of salt, read some of the forum posts and make your own judgment!

It’s been a while since I posted on this blog but wanted to update my (few) readers on my current 3D printing status. I’ve finally sold my 3D touch a few weeks back, I had sold my Form1 previously in October to people that needed one to develop alternative resins. I’m waiting for a Zim dual head printer that should be there is a few month, so I’m not abandonning the 3D printing world. It’s just a pause 🙂

So why posting on an old subject?

After a few months of use, I reached the conclusion that the Form1 with all its promises was not going to work for my needs (more details under). But at the time, the number of printers in the wild was small and I couldn’t say if my experience was a fluke or more an intrinsic issue of the design. Then it happened: more and more posts in the forum talked about bad galvo calibration, or printer that would lose calibration after some time. The final stroke was a post made by Kevin Holmes (PDF) were he had to manually adjust the analog PID controllers of each axis to correct his axis scaling and another one (PDF) where he had to replace a galvo on his printer… There is also a growing concern to know if the printer can be used for engineering purpose.

When I pledged on Kickstarter, my idea was to use the Form1 to design brain teasers. In other words, mechanical assembly where all the pieces have to slide and fit perfectly with reduced tolerances. But what the Form1 brings to the table is slightly different: the galvanometer system is analog and moves continuously drawing very  smooth curves. In other word while the system is precise, it does not mean the result will be accurate.

Precision is not Accuracy

First, the difference might seems like a semantic war (see how the “Does zero padding help in improving frequency resolution?” discussion on the DSP group of Linkedin turned into a flame war). The topic is also relevant to my area of work when it’s important to know the range of possible steps/pixel a cursor of a touch screen can move to (precision or quantization) and what is the maximum deviation from the real position (accuracy). To make it more clear the schematics under should explain what I mean:

Precision VS Accuracy

So Precision is akin to the number of steps the laser point can move, and in a Form1 you could consider it as nearly infinite as the analog PID are effectively smoothing the output and the mirrors have a finite frequency response. Now, even if the laser is moving smoothly from one point to the next, it does not mean the laser is aiming to to right location on the bottom of the tank. These accuracy / linearity errors can have many causes.

Accuracy errors in the Form1 cause and mitigation

The challenge of creating a cheap stereo-lithography printer is the temptation to use cheaper elements. From my point of view, the current design can be improved in these areas:

  1. The pealing process is really creating tension force that leads to deformation or requires heavy supports that cannot be cleanly removed from the part. Adhesion to the build platform and damage to the PDMS are also sources of failure. Some of the professional printers are using a dive-in structure were the part is sinking in the resin and the new layer is cured from the top but you need to maintain a large tank of expensive resin.
  2. To make is more compact, the printer is using a mirror to bend the laser path. The beam is now touching the bottom of the tank with a variable angle. As the tank is composed of multiple material, the refractions will deform the laser spot and introduce non-linearity on the border of the print area that could be challenging to correct.
  3. The printer does not include any optics to correct of the barrel effect (F-theta lens), as the laser paths will be varying in length, the focus spot size might change depending on the location.
  4. Last but probably the most important element are the galvanometer amplifiers. Choosing a analog PID controller was really terrible as the user cannot really adjust or calibrate the printer without opening it and fiddling with variable resistors on the controller board (which will void the guaranty). Digital PID controllers (or cheap FPGA with DAC/ADC converters) wouldn’t have been much more expensive and would have given a very wide range of calibration options.

Note that apparently Formlabs has included an option to scale the X and Y axis in Preform 1.2.1 to compensate for some of the calibrations errors, so they are making effort answering the user-base but these correction are likely to be global and won’t correct local non-linearities that I’ve observed on the border of the platform.

Form1 from an engineer point of view

So in its current contraption, I would classify the Form1 in the top left quadrant (Precise but not accurate). Depending on your requirements, this might be perfectly acceptable. So for example if you model figurines or art shapes where surface finish is more important than dimensional accuracy then the Form1 might fit your needs. For mechanical assemblies with small tolerances having a 1.5% discrepancy on the scale of one of the axis (error that I had on mine) makes any circle an ellipse and render the pieces not usable.

For mechanical parts, I would be tempted (but I have no experience) to use either:

  • DLP stereo lithography printers that at least guaranty that each voxel (pixel projected) is at the right location. The calibration of these printer seems to be more straight-forward as you only have to make sure that the projection area is rectangular.
  • An other option might be the ink-jet printers were the XYZ moving stage is more accurate in positioning the print head

Unfortunately there is no free lunch in this word, so I will go back to FDM low cost printer that at least fit my budget with an acceptable accuracy and finish for the price of the consumable. If you can recommend any other solution please send me a message or comment!

[<<Prev. -Preform  not compatible with JewelCAD]    [3D Printer articles]   […- Next>>]

Advertisements

Episode 2 : Form1 calibration strikes back

[March 11th edit : these calibrations issues were enough to convince me to sell my Form1 printer, for details see this post]

In a previous post I was showing my Form1 was not really well calibrated and circles were turned into ellipses. After filling a bug report to Formlabs, their answer was on the line of “thanks for your concern we are working on it…”

I’m confident they will find a solution, but I started to wonder how a user could calibrate his printer himself. This is important as the warranty on the printer is only valid for 3 months and as any mechanical machine, I’m expecting the printer will require some calibration from time to time. Last but not least, this will be only possible if Formalbs is actually giving us access to more parameters than the currently closed/dumb down version of PreForm we have (Hint! Hint!).

Laser+Galvo based printer calibration challenge

A normal user has limited resources and not access to complex measure instrument, so the procedure should match these requirements:

  • Simple steps: if possible it should be as simple as print test objects, measure them, compute the corrective factors (Excel Spreadsheet/integrated to PreForm), enter the new parameters and re-print the test objects to validate the calibration.
  • Not too long and costly: printed test object should be small, we don’t want to waste time and resin
  • No expensive measure instruments: I’ll be only using a digital caliper available for less than $50 on Ebay/Amazon

My main assumption for this exercise is that the calibration required is only affecting linear parameters. I will not go into galvanometer fine tuning. We will see at the end of the article that unfortunately non-linearity are present and will limit your end result quality.

The main challenge with laser base printer is that it’s difficult to measure precisely the position of a moving dot at the bottom of the vat (absolute measurements). So one solution is to do relative measurements of printed object features. We will use a digital caliper to get a good number of data points to build a robust average estimation of the calibration.

In the graph under, you can see that the laser spot is covering each each slices of the object using two kind of paths. The perimeter/skin-paths are tracing the outline of the object, they will be responsible for the smooth finish. The In-fill paths are straight and parallel paths to cure the inside of each slice. Using simple shapes like circles and square will let us measure independently the performances of both X and Y axis galvanometers.

calibrationExplained1

The final dimension of the Slice is affected by both the axis amplification (deflection angle range) and the size/shape of the laser spot. When tracing the measures VS the theory graphs, the amplification correction is measured by the slope of the curve, while the shape of the laser spot is creating a constant offset.

calibrationExplained2

As PreForm software is not open source, I cannot be sure how to use these correction factors. For the amplification, it will probably mean the angles transmitted to the printer will need to be multiplied by the correction alpha found (different on each axis) . The offset is a consequence of the spot size mismatch. PreForm must use a boundary offset to draw the perimeter/skin paths. A positive offset means that the diameter of the spot need to be reduced by half of the value. Also if the laser spot is not exactly round it might be trickier to program…

Calibration object and Procedure proposal

The shape I used for my calibration is a simple series of square and circle extrusions of various small size and one large for the base. Each instance of the object provides 20 points of measures:
– Circles: [1, 2, 3, 4, 23] mm diameters, X and Y axis
– Square: [3, 4, 5, 6, 25] mm side length, X and Y axis

Cylinder_dimensions

To know where each object was printed on the platform, I’ve added a number engraved. That could be useful later if we are tracking non linearity issue depending on the location of the objects.

[STL & Inventor files for all squareCylinders Test] [5x .form file] [9x .form file]

I’ve used the 5 duplications file as it’s already providing 100 measure points. The base of the object is 3mm thick as I was hoping I could print it without supports to save time and resin. It turn out the Form1 is curing multiple times the slices in the first 2mm for the base. It creates an adhesion issue for the next layers, and I ended up with unusable objects with holes so I’m now generating support structures.

x5Cylinder_platformPosition

Here are my measures, with the 5x file, transparent 100micron material profile and default supports configuration. My first prints were done with the 0.8.1 software but reading that the 0.8.2 is supposed to have an improved laser tracking I decided to reprint everything. (Un?)fortunately the measures are consistent and the 0.8.2 didn’t improve the calibration…

Measure_x5Cylinder_0-8-2

The full spreadsheets with regression formulas are here [0.8.1] and [0.8.2].

x5Cylinder_calibration_results

From the regressions results we can see that the calibration required is consistent in both PreForm version. My printer has a 1.4% error in X axis and 0.35% error in Y. The laser spot diameter is also off by 0.2 mm.

For some reason the PreForm 0.8.2 is less consistent than the 0.8.1, the calibration is only reducing the overall error by a factor of 4 while it was reduced by 5 in the previous version…

Ain’t you doing anything about outliers?

When doing a root cause analysis on anything, you have to solve the first order issues that are dominating your calibration errors before looking at the rest. My Form1 is displaying non linearity on the X axis on geometries close to the back of the platform (2 = back-right and 5 = back-left).

cylinderX5_nonLinearity

When you look at the measures table, both objects are very different from the rest. I have no explanation yet for that phenomenon, but I hope that once I will be able to correct the offset and scale in PreForm I can build a new test to tackle this issue.

Until then, if you have a Form1 I would be glad to ear if your printer has the same calibration issue as mine. More reports will help Formlabs building a great printer!

[<<Prev. – Extract STL from PreForm Files]    [3D Printer articles]   [Preform & JewelCAD – Next>>]

The Form 1 might need calibration after all…

July 12th edit : I have now a Form1 Calibration procedure.

These past few days some very nice Form 1 tests have been publish by TJ and Gregg. These tests have been concentrating on the shape and small features. I’ve not yet seen a test on accuracy / linearity.

The Form1 is supposed to have a 300 microns (0.3mm) diameter laser spot which limits the size of individual features but the galvanometers (galvos) can be oriented very precisely. I think I recall reading that someone in Formlabs was talking about a few micron of positioning accuracy (position of the center of the spot on the bottom of the vat). The galvos are a dynamic system using PID controllers and tuning them to get a perfect positioning takes time (see my post on the ILD test cases).

Recently I had to print a mechanical assembly for someone and even with the 300 microns tolerance, it was difficult to slide the cylinder into the tube. So I decided to investigate a bit more: I was also wondering if the peeling was not introducing some deformation in the piece.

Simple Linearity Test

The test I designed is simple: It’s a series of 20mm cylinders that are horizontal/vertical and with a 45 degrees angle. The piece is placed perpendicular to the platform to measure independently the X and Y galvo. As any mechanical system, the mirrors are not jumping from one location to the next instantly. So even if the calibration is perfect in static condition (after some resting time), the dynamic properties (damping and oscillations) might be off. These configurations are also axis dependent as the mirrors might have a different shape/mass and each axis has his own amplifier. In the ILDA test, this dynamic behavior is checked with the circle inside the square.
Male_torture_test

And the results are not as expected… As you can see the cylinders are not really round! So either the preform sampling space is too large (not enough points on the path / movement too fast) or my ‘peel’ axis amplifier board needs some adjustment.

Linearity_measure_XY

My caliper measurements are:

Horizontal cylinder:
Peel axis = 19.52 mm
Orthogonal axis = 19.99 mm

Vertical cylinder:
Peel axis = 19.5 mm
Orthogonal axis = 19.87 mm
Vertical axis = 19.96 and 19.92 mm

45 degrees cylinder:
Peel axis = 19.51 and 19.71 mm
Orthogonal axis = 19.91 and 19.75 mm

So conclusion while the Z (thickness) and Y (orthogonal to peel) axis seems to be close, my printer seems to have an issue on the third dimension. I hope Formlabs will have a calibration procedure to correct this issue.

Location Influence on the Print Quality

The other main issue I’ve encountered while printing technical piece was reliability of the print quality. In some occasions the peel process is making loud “clacks” noise when the surface to remove is large. As the tilt is starting from one side, the strength created on the prints is different depending where the geometry is placed  on the base.

The picture under is showing two pieces that I had to reprint twice to get an acceptable result. The peel strength issue triggers supports break and non printed/delaninated walls.

Location_print_error

The only difference between the two prints is the location on the platform. I didn’t had to regenerate the supports nor changed the material setting (grey resin 50 microns). On my first print (left side), only the C part came out well. So for the second try I removed the C part and moved back A and B as far away from the peel side as possible. Note that I had to print the pieces horizontal as it’s not possible to generate internal supports yet and any other orientation would have resulted in non supported features…

Prefom_locations

I’m not sure what to think about this issue, maybe reducing the peel speed when there is a large area to remove could help? If this issue is proven, that will mean the silicon layer will get damaged faster on one side triggering a replacement more often due to fogging…

Anyway I’ll keep on exploring the Form1, if you have any question please ask I’ll make more tests. Cheers!

[<<Prev. – Preforn 0.8.1 Evaluation]    [3D Printer articles]   [Extract STL data – Next>>]

Differential Amplifier for the galvos

I’ve made some research on how to convert a single ended analog signal to a differential output to drive the galvos for the scanner. The requirements are:

  • Galvos input voltage range is +/- 15V (and there is a +-15V power supply connector available on the driver board to pick)
  • Signal bandwidth is limited to audio range, so no need to use very high speed products
  • DIP package would be great for breadboard prototyping
  • The sound card level has probably a limited voltage output so some amplification would be good to get a nice mirrors angle deviation
  • Cheap and in stock in local store

I’ve found quite a few full differential amplifiers but the best match so far is the DRV124 from TI. Not only it seems simple to use and compatible with my available power source, but the DIP package is convenient. It’s not available locally, but you can request free sample on TI website, so that’s perfect to try 🙂

Companies like TI are publishing very nice white papers on the theory of fully differential amplifiers (Here and there). I’ve added a few more amplifiers in the sampling cart so I might be able to use some of this information.

It’s alive! And first blood…

A few days ago I received the galvo kit and tonight the laser arrived with a USB sound card and my new soldering iron. So I’ve decided to give a try and check if I could make the thing move.

With a signal generator I tried to draw lines as I could only drive one channel at a time. The lasers worked between 3 to 5V as advertised and the luminosity is somehow decreasing until 2.5V. It will probably be relatively easy to do a blanking mechanism, but controlling the power might be tricky. Also the first lesson was theses beasts are very sensitive, as I fried one without any real obvious mistakes… First conclusion: I will order some cheap heads to tune everything before switching to the small spot laser.

Here are some boring single axis tests, changing the frequency makes the spot appears as a continuous line or a moving dot.Drive_frequency_tests

Also I’ve tried to see how fine the beam could be focused at a 50cm distance. I’m quite satisfied to see that if everything goes right, the line could be around 1/2 mm wide. This picture was taken with a flash, so most of the speckle is gone. The apparent width of the beam with naked eye is much larger. That probably mean I will have to find a filter to put in from of the webcam.

beam_width Anyway I think the next step for this weekend is to build a Sound-card to galvo driver adapter to be able to draw some figures!

[<<Prev. – First steps]    [3D Scanner articles]   [TBD – Next>>]

First steps of the scanner building

After looking at what it would take to build a 3D scanner,  the most complex HW element is probably the galvanometer scanner. Nicely there are kits sold on Ebay that includes power supply, galvo mount, 2 * (PID driver, galvo) and cables for less than $120. It seems a safe bet to have a working system faster.

galvo

The kit is a based on a close loop PID controllers/galvo that apparently allow a 20k points performances at an opening of 20degrees. I’m not sure about the real perfomances of these cheap parts, but as I won’t be trying to get a very fast and complex animation they should do the job.

The manual is not very detailed on the tuning of the system, but once again with some search, the online community is very helpful. There are multiple tutorials (A or B) on the principles and what all each element of the ILDA test vector is testing.

For the laser, I have no experience on what would work better so, I found that Quarton offers relatively cheap modules with a specified divergence (Most of the no-name have no datasheet available). I’ve selected 2 modules to start:

  • A small dot module (VLM-650-11 LPA) with a 2.5mm dot at 5m and a 0.25mRad of divergence which should give me a ~1mm dot at 50cm
  • A adjustable focus module (VLM-650-02 LPA) to see if I can get a better dot size for the laser

The module here is very convenient to compute the diameter size.

In a future post I will explore the driving signal generation for the galvos and the laser command.

[<<Prev. – ILD file format]    [3D Scanner articles]   [TBD – Next>>]

Scanner 3D Architecture

Let go over the scanner architecture in more details and try to lay a framework for the task.

Note that I have never built anything like this and all these notes are preliminary. I’m sure many challenges will show up and I may change completely my plans mid-way. But that’s the nature of the beast with technical projects. If anyone would like to share the adventure with me and try to build the scanner and solve the challenges please comment!

The principle is to use two galvanometers (galvos) to control a laser spot position and project lines on the object being scanned. The lines will follow the shapes of the object and their deformations are recorded by a camera. Once acquired an algorithm process the images to extract the lines and reconstruct the object geometry.

The diagram under shows the relation between all the pieces.scanner_architecture

I’ve not yet talked about this Fesnel lens in the middle. The role of the lens is to project parallel laser rays on the object. I actually don’t know yet if the lens is required to ease the reconstruction task, maybe the software could compensate for the divergence? Fresnel lens can be cheap even for a large diameter, but they may introduce too much artifacts on the line projections to be usable. We will see at the build…

Also the object could be placed on a motorized rotary table (not in the schematics) to scan multiple angles automatically.

Now let’s see what pattern we could use with the scanner. One great challenge is how we can keep the scanning time low. So we can try to use the video capability of the camera to acquire as many frames/line position per second. The idea I have is described under. The duration for one line position is timed to last ~2 frames (1000 * 2 / 24 = 83ms). During the time the same path can be repeated multiple time depending on the speed of the galvos. Also multiple horizontal and vertical lines could be projected at the same time to reduce the total scan time. The last element is a ‘code-bar’ that can be decoded by the software to mark the line position. Think of it as a counter to uniquely tag the frame. To avoid confusion of the software, the laser should be blanked during the transitions.scan_frame_structure

The reason this frame code is interesting is that it allows a complete resynchronization of the laser movement and the camera video. So the scan sequence after calibration could go like:

  1. Build the glavo movement and blanking sequence
  2. Launch recoding of the camera
  3. Play the laser movement file
  4. Stop camera recoring
  5. Go though the video and using the frame codes to resynchonize and modelize the object (could be done in parallel of 3 if processing power requirement is not too large)
  6. Rotate the table and restart at 2 for the new angle

During a frame scan, the horizontal lines and vertical line will sweep across the field area to cover the whole object:

sweep_scan

One parameter difficult to judge before hand is how many line the galvos will be able to display correctly. As it’s a mechanical device with inertia, the maximum speed while keeping strait lines is limited. So it might be needed to make a tradeoff between accuracy and total scan speed.

Scan_Speed_change

From this architecture we can already see challenges an maybe start mitigation actions:

  1. Overall scanner principal is wrong/not working [low risk]: there is quite a lot of literature on the subject that describes exactly this methodology.
  2. Galvos cannot draw the lines properly [low risk]: one again, the online laser show community is a great source of information. The patterns are not too complexes and should be achievable.
  3. Reconstruction is difficult : see point 1, as we are not using exactly the same setup details may vary like ray divergence, camera lens correction…
  4. Cost too high : at this point I have no idea what will be the final BOM cost of the scanner, but if we can keep away from complex optics we should be good. The goal shoud be well under $1k.
  5. Poor accuracy : I’ve not talked about the performances of the scanner, but as it’s oriented toward the scan of small objects, the reconstruction has to be quite accurate to be useful (mm range?). Things like the laser speckle (size of the dot) will limit the scale of the details the system can capture. I’m not sure if the laser will require focalization or the use of special filter in front of the camera to reduce the visible speckle.
  6. Schedule slip [no risks!]: yeaah for once there is no project leader and gantt diagram so this project will go at its own pace!

That’s all for now, if you have any remark, comments or question ask away!

[<<Prev. – Introduction]    [3D Scanner articles]   [ILD File reader – Next>>]