Rotoforge 2023-11-30 Cold Working and Other Problems

 7 months of implementations later...

I thought I grasped the physics. I was wrong. well not completely, but at least 50% wrong, which might constitute not even wrong.  Or perhaps its just the process of searching an engineering implementation space for a real system which reliably harnesses relatively complex underlying physics...

Regardless of the reasons, it has been a long time since my last post. This is largely due to the impending end of my PhD program and the beginning of a new chapter in my life, and due to the significant number of engineering implementations I and others in the project have attempted in order to address three problems I have encountered in the course of the Rotoforge project to the present day.  Listed below:

1. Cold work (specifically fatigue strain failure) in the wire feedstock that has not yet entered the rotating die/motor assembly (see rotating die extrusion). You might think of this like cold creep. The opposite of heat creep in FDM, it is the generation of internal stress by cold working in a wire segment that is not above the annealing temperature of the material such that the wire fails in brittle fracture along a slip plane at some critical fracture stress.  Literally sudden brittle failure induced by cold working.
2. Simultaneous temperature and strain control of the wire inside the rotating die, or barring that, reliable measurement of temperature at the deposition region or at least in the die. Without this kind of information it is nearly impossible to know how one needs to adjust die speed, extrusion rate, or linear feed rate to accommodate stable extrusion. Fortunately, a good friend and collaborator, Andrew Shield's (ShieldsExperiments on the discord channel) cooked up a lovely non contact magnetic temperature(and possible strain) sensor which I will link to the repository for, and discuss in as much detail as I am able in the following paragraphs. It operates on the temperature dependence of magnetic inductance using the H13 tool steel die as a core of inductor, and potentially as a Barkhausen noise measurement device(strain/extrusion pressure measurement)
   
3. Die wear (mostly by mechano-chemical means) is severe in most materials other than H13 tool steel. even H13 has its limits as feed stock hardness and printing temperature increases.
   

The structure of this post will be as follows for those looking to skip the long read:
I have also (poorly) shown and talked about some of the ideas in this post and the following post in a YouTube video back in august of 2023.

1. a discussion with images and cartoons of the failure modes we have encountered.
2. a illustrated discussion with links to device designs we have attempted to implement to overcome these problems.
3. a brief introduction to some alternative approaches that may be simpler in practice and some lead in to the post following this one.
   

Cold Work is best served hot-

As the engineers and materials scientist among you likely already know, materials subjected to plastic deformation at temperatures below their annealing temperature, tend to retain their shape and in the case of most metals undergo work hardening. In the rotoforge print head, conditions that result in cold work are intrinsic to the design, if one opts to use a constricted die design, instead of an open, keyed die as per additive friction stir deposition.  Below is a series of figures which illustrate the basic process which is the cause of problem number 1.

Metal wire feedstock travels down rotating hollow shaft, contacts rotating die and begins to heat due to friction.

Flash forms as frictional heating reaches peak. Plastic deformation begins in the flash until strain rate synchronizes with the rotational frequency of the die within the limits of the metal on metal friction.

I have added strain markers further up the wire, away from where frictional heating can raise the temperature of the metal above its annealing temperature. Continuing rotation between die and flash and continued shearing of the flash causes some rotational coupling (think plasticized metal fluid coupling) with the rest of the relatively colder wire.

Rotational coupling between colder wire section and plasticized wire causes steady rotation in the cold section at a rate limited by the coupling efficiency of the "plasticized metal fluid coupling". Probably described by some complex terms and viscoplastic friction.

As rotation continues, the colder wire exceeds its elastic limit and undergoes continuous plastic deformation at a reduced rate, and without heat available to drive dynamic recrystallization. This results in grain shrinkage, and a steady growth in shear strength (and shear stress just as in a torsion spring!)

Eventually the shear stress exceeds the ultimate strength of the cold worked wire, and a crack propagates in brittle fracture along a shear plane in the metal. Exactly where seems to be somewhat inconsistent but rest assured it will eventually occur.

To overcome this problem the process essentially requires that the wire be unconstrained to rotate with the fluid coupling at the die. Perhaps at a reduced total RPM based on the efficiency of that coupling, but still at a speed of a similar order of magnitude. This necessitates some method of applying a thrust to a rotating assembly.

There are many ways to accomplish this task. However, as the scale of the rotating assembly shrinks toward single millimeters, and the RPM increases to greater than 10K RPM, the number of available off the shelf choices approaches zero. So i took the liberty of designing a few of my own solutions to the problem.

So what did we try and how does it work?

Solution 1 is a combined linear-rotary flex shaft bowden cable like actuator.  Essentially, it operates by supporting a long wire of a relatively high modulus material that will resist galling, such as Kanthal or H13 tool steel, in a brass cup or similar connection. This brass cup is welded, glued or brazed to the H13 or kanthal wire and the cup itself is seated in a combined radial-thrust ball bearing assembly.

 

Essentially the way this is meant to work, is by pushing the wire which is constrained in the thrust axis by the brass seat, but is free to rotate on the thrust bearing included in the figures above. The thrust force applied by the linear actuator pushes the push rod, made of copper, kanthal, or H13 steel, down a PTFE guide tube, where the push rod then contacts the rear end of  segment of wire (chopped from a spool) inside a PTFE tube. This enables a slick connection between push rod wire and the wire being pushed, and prevents welding between them due to the unconstrained rotation of the push rod wire.  The problem with this system is ensuring that the pusher rod and bowden cable system is rigid enough not to flex under the applied load, and thus reduce the total thrust applied to the wire. I was unable to find a satisfactory solution to this problem.

Solution 2 is essentially a remix of a wades type pinch wheel extruder, using an solid carbon fiber filled PEEK(for its high wear and temperature resistance, and low friction properties) pad in place of one of the typical pinch wheels, and with the addition of omni wheels in place of the typical hobbed pinch wheel that grabs the filament. This modification, in combination with a filament chopper, such as the one from the smuff project, enables applying a thrust to a relatively repidly rotating wire or filament, by pinching the wire between the omni wheel and PEEK pad, and rotating the omniwheel with a stepper motor or gear box drive just as in a standard pinch wheel extruder. The big problem with this implementation is that the omnis are under significant load and this increases friction between the rollers and the axle of the rollers, which generates heat and increases rolling resistance against the wires rotation which can break the wire at high rotation rates.  So i got to designing and modeling how the pats would need to go together to test one of these in practice...

 

 

Ok now that we have an idea of how it can go together, with all the relevant parts, lets actually try building one. naturally the one we actually fabricate ends up quite different from the one we designed, due to limits of materials and fabrication capability, and the realities of tolerance stackups and real world material strengths and stiffness.  This meant that the omni wheel had to be about twice the diameter of the design, and ultimately trying to get two printed omnis at the size needed with the tolerances required (because no one makes omnis this small, its nearly impossible to find omni's smaller than ~35 mm OD) and the cost of having them made by a contract house would be huge, I ended up just designing them based on an existing design on thingiverse which was quite convenient. making use of 1" key rings as axles for the omni rollers and m2 stainless steel washers as rollers.

 

So lets get this thing mounted and tested

Unfortunately, the washers as rollers concept works well for low rotary speeds of the wire, but not well for high speeds. Further development and refinements to the designs of the omniwheels will be required to make them reliable.

So what about temperature and strain control?

   Without reliable material feeding, temperature and strain control can only offer so much. But, none the less, once the problems with material feed have been resolved, it will be necessary to setup some form of closed loop control of the temperature and strain rate of the material being deposited through the rotating die. In the same way it proves necessary to use PID to control the temperature of the hot end in FDM. Since material viscosity in solid phase processing depends strongly on temperature, force and strain rate, and the micro structure of the product depends strongly of on the processing history, it is critical to develop reliable probes for these three parameters that can work in situ, in real time, and with the highest possible accuracy to the actual values encountered in the processed material.

Unfortunately, there are some problems with existing solutions to measuring temperature, force and strain rates in very small rapidly rotating assemblies. Contact thermocouples are generally highly dependent on surface contact state, and have wear life limitations associated with high rotation rates of rotating assemblies like our rotating die. More importantly because the volume of material we are working in Rotoforge is so small, most thermocouples have masses comparable to our worked material and so significantly skew the results of the measurement(think measuring the velocity of a tennis ball by hitting it with another tennis ball). So direct contact methods are out for now...

Non contact methods are really the only practical means we have of extracting temperature data. Among the most common non contact methods, pyrometers, and bolometers (thermal cameras) are first to mind. However, the temperature range capabilities of cheap devices rarely reach above ~450 degrees C, which is too low for the printing temperatures of most metals, and may be a bit close to the margin for aluminum alloys. Further, pyrometers and bolometers require accurate calibration, and few such accurate calibrations exist for a material undergoing solid phase deformation at a high rate in a small volume. There is still little understanding of how surface emissivity may evolve with defect density in metals, IE how severe plastic deformation may skew the temperature measurement; to say nothing of the challenge of measuring the temperature of small spots accurately due to the typical low resolution of most modern thermal sensors (especially for high temperature measurements). Thermal microscope cameras do exist, but are typically relatively bulky and expensive and optics for high magnification of IR radiation are similarly priced.

This situation sounds hopeless until one considers all the properties of a metal which change with temperature substantially enough to be measured, and repeatably enough to be useful. From the first principles, a reason metals being severely plastically deformed are able to bond with their underlying substrates(and with the die they are forced through) and are able to rapidly change shape is due to the liberation of electrons participating in the metallic bonds which hold the solid metal together and make it rigid. When energy is introduced to the metal in the form of heat from friction, or as mechanical shear deformation, these bonding electrons are freed in some numbers in proportion, and this facilitates the breaking of bonds and thus the change in viscosity into the semi-solid regime. 

Fortunately, the liberation of electrons also alters the charge state of the metal atoms in the material, and changes the effective electrical resistivity of the metal (and other metals in contact with the deforming metal whose temperature rises by conduction heating, like our die). While making electrical contacts robust to rapid rotation and high temperatures is difficult, this change of electrical resistivity also makes it more difficult for electrons to form eddy currents in response to a changing magnetic field, as per faradays law of induction. Essentially, if we consider the deforming metal, or a die in contact with it as the core of an inductor, it becomes possible to observe the temperature dependence of the inductance (really of the saturation current in the core) by applying a magnetic field to the "core" with a coil, and then measuring that magnetic field as it "rings down" in the core. Put simply, the current induced by the electromotive force applied by the initial magnetic field of the driving coil, subsequently induces a magnetic field of its own, whose intensity depends on the saturation current that was initially generated in the core by the initial magnetic field, and which itself depends on the electrical resistivity of the "core" at the temperature of interest and the mechanical plastic deformation state.   The following cartoon illustrates somewhat how this drive and response interplay occurs in cross section of the rotating shaft/die and copper inductor coil assembly.

The copper coil generates a magentic field in response to a supplied current from the driver board.

The magnetic field from the coil induces a saturation current in the die and wire and shaft, according to their relative permeabilities. H13 tool steel die is the most permeable by a factor of 50 or more. Thus it contributes primarily to the response.

Drive coil then turns off, and listens for the "response" magnetic field of the saturation current in the die as it rings down from the initial drive coil magnetic field pulse.

The response magnetic field generates a measurable return current in the drive coil, and this return current magnitude can be converted to a temperature value of the die based on either calibration data (provided by shields) or a known temperature dependence of the permeability of the material being measured.

Serendipitously, H13 tool steel (our favored and most wear and gall resistant available die material that is easily workable)  has a curie temperature around 880 degrees C.  (Wikipedia on Curie temperature for the unintiated)This implies that its magnetic response behavior, IE its saturation current has a relatively linear dependence on temperature proportional to its temperature coefficient of electrical resistance which is large enough to be easily measurable with affordable electronics as Andrew has demonstrated with his circuit and coil design and subsequent testing. Additionally, this technique is potentially sensitive to barkhausen noise, if a frequency component of the response magnetic field is measured. This barkhausen noise can be used to measure strain (thus applied force) and deformation, that is, strain rate in the deforming material. 

Above the curie temperature, measurements can still be made but the response is nonlinear and this complicates further measurements. Additionally, the magnitude of the response declines toward zero as the order of the magnetic domains is disrupted at higher temperatures still. So the practical upper limit of temperature measurement depends greatly on the stability of the magnetic response character (the permeability) of the material you are using as the inductor core. 

This particular development (largely with the assistance of Andrew Shields(Shieldsexperiments), Rob Herc, Paul(Parkview), Sam (Sparkgap) and others from our discord channel. is potentially very useful in a wide variety of rotating assembly measurement situations and can be found, schematics, test data and all as well as a brief user manual,on the electronics section of the Rotoforge discord, and github pages as well as at Shield's personal github page.  There may be a wide variety of applications for this technique, in many fields beyond our little project as it bears a resemblance to NMR/MRI and many other techniques. I am extremely grateful for their assistance and consultation in building it according to the needs of the project and solving this problem for me and future developers.

I have included the circuit diagram and initial calibration data (from 30 to about 450 degrees C) for a H13 tool steel die (both courtesy of Andrew shields) in their current entirety for completeness and thorough documentation purposes in the event of future catastrophe here as well. I have also included all the data sets I have collected on the previous friction die extrusion experiments... and the data that this graph for the thermomagnometer was pulled from courtesy of Andrew Shields.

 

So what else can we do?

In the interim a commercial company released a rather nice CNC mill with a tool changer and auto leveling which appears to be an excellent opportunity to implement serial friction surfacing as a method of metal additive manufacturing on the desktop at somewhat higher cost than I would like but very nearly in reach. One could easily imagine a machine like the makera grabbing individual rods of constantly reloaded sections of feedstock from the tool holding area, and depositing those rods onto a build plate one at a time as in friction surfacing, to form 3D objects. Indeed, I know of one person who has done this with some success already in a startup context. Alas, 5K USD is a bit too rich for my grad student blood. 

So what is next?

So we shall move on...

So, seeing that these approaches were becoming increasingly mechanically complicated and prone to failure as a result, i opted to test a simpler hypothesis based on a system we all know well by now.
FDM.

In the next post I will explain why I have opted to pursue a metal FDM via conduction heater approach after struggling with the friction welding technique for so long.