A few thoughts about further exploration and tuning of the loop. The cable elasticity is likely the major issue, but the cog belt and table suspension are also in the loop (and variable part mass as others pointed out). One idea is to mount an MEMS 3 axis G sensor (or 6 DOF) to the table and see exactly what is going on there with the tuning. The dynamics may include vertical table motion, as it's unlikely the cable force will be at the center of mass. Then try adding a part to be ground. That is ultimately what matters for grinding to a tolerance & finish. Another perspective on RAS is it controls the frequency spectrum of applied force. Reducing jerk and further derivatives reduces the higher frequency inputs. If you don't put them in, they don't need to be controlled out. Mechanical structures usually have a huge number of resonance modes, and stable control of them is the classic control loop challenge.

The primmer on acceleration, jerk, and application to motion systems was really nice. Thank you.

I expect that RAS setting is going to need to increase once jigs / work is added to the chuck. These will increase the moving mass and thus the "slip" on the spindle + stretch in the cable drive.

We're using ClearPath motors (SC-HP series) on our project (restoring a 12.8m satellite dish for radio astronomy). The API we use (sFoundation) has an API function that lets you query the encoder resolution so you don't have to wire it in to your code. We're taking our project very slowly--the dish weighs in at a svelte 5.6 tonne. We had a crash a few weeks ago--not due to the motors, but due to a failure in our mechanical system design. Kinda scary, but the only thing really damaged was our egos, and the 50+ year-old crash buffer at the bottom of the travel. We're using fairly modest motors (a roughly 3/4HP on azimuth, and 1.3HP on elevation), but with extra gearboxes to boost torque. It's working well, although it will never win any slewing-speed contests :)

I don't know if it was intentional but "Something about fitting a fourth order polynomial spline, that sounds great" sounded so saucy and had me rolling

James, you absolutely work and think on another level but provide a wonderfully clear and detailed description even for the complex math and physics behind it. Keep doing what you do

James - Great video. Your research took me back to when I was 17 and learning basic calculus. "This is hard, where am I actually going to use this?" Then one day we learned the relationship between displacement, velocity and acceleration as derivatives and integrals of each other. Seventeen year old nerd mind blown! Fast forward 41 years and most of an engineering career and at age 61 you point out that there are at least two more useful derivatives to be had when it comes to springy systems. Old nerd mind blown once again! Before you started showing ClearPath's solution to this problem, my mind was running towards a solution in the control software. Since your stops won't be moving during a run, my thought was to run one slow pass at the start of the run to learn the positions of the stops, then use those known positions as part of your control loop and predictively change velocity and acceleration as the movement approaches the stop. Move 95% of the distance at full rates, then reduce acceleration for the last 5%, then same going the other way, something like that. Just check the end stops on each loop to make sure that we didn't miss a count somewhere and as a safety, but since those are servos, the only error might be mechanical slippage. One last thought - the table and vise are fairly massive, but do you think different tuning might be needed with a heavy object added to the system? I'm thinking a large angle plate or sine bar setup plus some heavier part to grind might add enough mass to the far end of the sprung system to mess with the control parameters.

I am having flashbacks to Mech Eng and Mechanics of Machines and Engineering Calculus from nearly 40 years ago. Lucky I am old and have forgotten most of the horror 🤣

You could always design in a 3rd shaft for the X axis pulley drivetrain with the new shaft down below. Make the servo go from its small pulley down to a big one, that big one is on a shaft with a small pulley on the other end, and have that new small one go to the existing X axis large pulley that goes into the machine. This would double your gear ratio which gives you more torque and will let you easier control the motion inertia with a smoother accel/decel curve. You aren't using all the motor RPM right now, so you can afford to cut the max RPM in half to double the ratio. When I did this project I had to use a higher ratio than I thought due to these similar inertia problems of the weight of the table and the springiness of the wire. By doubling the ratio the motor can take more time ramping speed down which does help with the inertia. Eventually you might want to do what I did and replace the wire underneath with a leadscrew and lead nut. The lifetime of the drive wire became too low with a servo due to the higher speeds we tend to run over what you do by hand. The leadscrew fixed this lifetime issue. I first started with a 3/8 screw, because its weight limit was higher than the table. However, the inertia was so high, it overloaded the lead nut too much and caused the nut to fail and strip out after a year of use. I replaced the screw and nut with a 3/4" one and so far (4 years now) I haven't had to replace a single thing again. The lead screw is mounted on each end to the end of the surface grinder table, so it doesn't turn. I have two bronze lead nuts mounted in a timing pulley, with the ends of the nuts sticking out of the pulley. The motor has a pulley on its output shaft that connect to the pulley holding the lead nuts with a timing belt. I have two mounting plates mounted to the table with ball hearings mounted in them, and the ends of the lead nuts are held by the ball bearings. This keeps the lead nuts and pulley centered properly on the lead screw and keeps the pulley from being able to travel left or right, since we want it to stay in place and drive the lead screw through it. I used a 3/4" .250" lead screw, which with my motor (3000RPM max) gives me a 12.5" movement per second on my surface grinder X axis. This was enough for me, but if I was going to do it again I think I would use a 3/4" .333" lead instead to have a but more room to play with things.

Always enjoy content that leaves me knowing more about a subject than when I started watching. This video was most enjoyable !

You can see that torque in the belt when it reverses. Very impressive. Also, they need to remove the comma after "proprietary" on page 199.

Wow! Just today I auto-tuned some Nema 34 clearpath servos for my CNC milling machine and noticed the RAS 16ms default. Now I know what that's doing! Thanks for the education!

I’m quite spoiled using Kinetix motion servos as a System Integrator. Do a Motion Axis Move to a position in x amount of time, set it to S-curve and let it rip. A the hiperface absolute encoder makes indexing bulletproof. Throwing excessive acceleration at a wet noodle axis doesn’t always work, but it’ll be interesting to see how the dynamics of the system changes when a workpiece is added and the wheel is engaged because the resonance frequency shifts. Also the forward and back directions will see different forces because of the grinding wheel engagement. Will be great to see the final results

If you're writing in C++, you might also mark the external state variables as volatile. Otherwise, the optimizer might make assumptions, which could lead to odd (and potentially dangerous) bugs.

I would attach a stop for each of your stops. Basically make it so that you don't have to rely on seeing the marks you added, as you would be able to just move your switch stops to a stop that doesn't ever move.

Motion control is an incredibly cool subject to look into. For a project I have going at the moment I'm using AC servo drives which have absolutely terrifying trapezoid performance due to the winding voltage, instead of doing jerk limiting within the motor PID I'm doing it as part of the trajectory generation, this allows for optimizing within the context of the planned path. So you can have a highly damped control system where you can have the motor track a target of arbitrary error and acceleration and all of the motions between those errors are smooth and jerk free. You can also modify those constraints on the fly, so in the case of your surface grinder you would switch to a dynamics profile optimal for the turnaround point when the limit switch is triggered and then after the velocity is reversed switch to a purely trapezoid profile so that the target velocity is hit much faster, would allow grinding a part at the full width of the table working area without creating hot spots at the edges.

Easter morning and I've learned something!!! Snap, crackle, pop! Fabulous.

Always a good day when you learn something.

In college, in a cam design course, we looked at jerk, and its derivative ( we called it jounce) in order to smooth the entry ramp on the cam curve. The goal was to smooth the entry into the cam, reducing the float of the cam follower.

Couple more notes regarding the ClearPath: * If not, then run auto-tune again with the new motor, and see, if it finishes. There have been instances when the auto-tune has "gotten stuck" at certain percentages, in such cases I have just applied some force to the mechanics to allow it to overcome it * Try fine tuning. You can find it under Setup -> Tuning -> Fine Tune. There you have a slider which you may try moving to either side to see how it affects your use case. * Use higher input resolution. You have set it to 800 in the MSP software, but the motor's native resolution is 6400 which would result in (a little) better movement since the RAS algorithm has a higher input frequency. CleraCore can handle such high output frequency very well thanks to the clever internal "clock"

I have been thinking of doing the same thing with my Landis 618. Thank you so much James - you probably saved me weeks of faffing about when I finally get around to it. I will only likely do the x axis. Awesome channel.

James thank you.....I learn so much from you.

Success always feels good. Thanks for the video keep on keeping on.

So much insight into the black art of motion control tuning. Teknic's videos makes it look unbelievably easy, your explanation makes their results understandable. Now I know specifically what I'm trying to affect with RAS/jerk/jerk derivative. Thanks James.

Hooo - Boy! James, you are really showing me how uneducated I am. I am a retired airline pilot. I was really good at carrying folks safely in really bad weather. That skill now looks like kindergarten level. Thanks for explaining.

perfect job for a viscous coupling :)

Disney talks about the visual impacts of this if you ever see the behind-the-scenes tours or videos. They refer to it as Audio-Animatronic Compliance in regards to addressing jerky limb movement when trying to make their attractions move faster.

I learnt a lot from this video. Many thanks.

Excellent work, sir! 👍👍

I don’t own a surface grinder but if I ever decide (and have the Benjamin’s) I would consider this model because you have already worked out all the bugs to automate its function. Another fantastic video.

@patchvonbraun now that’s a series of videos I would love to watch

Thank you very much for this video!

This went way over my head.

Cool project.

As someone who knows almost nothing about physics, you did a really great job of explaining everything. I think im actually on the same page as you on all this haha.

Another great video James. Something to be aware of at that fast of a speed on the X-axis is the speed of the grinding wheel vs longitudinal movement. If it's too fast you will not leave enough time for the wheel to do its work. Think of climb cutting on a mill at a rapid feed rate. Every "automatic" grinder I've ever used had a much slower x- axis feed rate. Also, as you may already know, things can go wrong in a hurry when surface grinding, so an external control outside of the e-stop would be helpful to vary the speed.

Very interesting

Nice! I wish I had this info a couple of years ago when I built a couple of machines for a local manufacturer. It had a springy x axis (6’ of 0.25” wide timing belt). It had very little mass/speed/force but I had a lot of problems with resonance. A combination of a much stiffer belt and auto tuning finally took care of it, but I would have much rather just swapped to a RAS-able motor if I knew about it.

Im honestly supprised you didn't add a jerk paramiter to your code so you could tune it on your own. On one of my 3D printers I compiled my own marlin firmware and apent some time messing with the jerk setrings and it was easier then i thought it would have been

Thanks for the educational video today. It occurs to me that the surface grinder presents a special case where the motion is repetitive for each job, so after the initial (or a few) traverses are made, you could take advantage of that by allowing the motor to learn when it needs to slow down prior to the reversals to prevent the torque excursions by using an algorithm in your code. Here is what I'm thinking: Sum the number of motor steps for each motion for the first, say 5 cycles to take advantage of statistics, and calculate the average for each direction. Then, following the 80/20 rule, start ramping the velocity down at the 80% of the full average number of steps (for that direction) to some chosen velocity that the motor can accurately accommodate with minimal excursions. I think that would allow both a faster velocity and acceleration while maintaining good torque profiles and reducing the stress on the motor. Just a thought.

Estops should always control things physically by cutting power (or a controlled stop through a safety rated system) with software being a secondary thing.

Very nice! I understand that some professional cnc machines use the linear scale to determine the position of the axis, instead of closed loop servos. That might also help in this situation?

Snap Crackle Pop! Gotta love nerds naming terms 😄

hilariously, I would have coded the RAS myself in the controller. not too hard to implement. especially if the accel can be directly coltrolled by code and not just constants

Those motors are amazing. Could not help wondering about the peak currents at accel/decel points. Must be quite a bit, for a short while.

How hard would it be to get a DRO scale feeding back into the servo control? Move the feedback out to the end of the line and you can easily tune a PID for the springiness.

It will be interesting to see how much the scope curves change as you place different mass on the chuck to grind

Definitely worth the extra money. I guess another option would be mechanically damping the torque spikes with a hydraulic torque limiter. One out of the cooling fan on an engine could be retrofitted to work. The damping effect is mechanically variable and controlled by a bimetallic thermostat which could be replace with a lockable thumbscrew adjustment. For extra credit, you could use a centrifugal actuator to mechanically emulate those rounded corners on your torque graph making it a purely electromechanical system with no controller.

Interesting viseo as always but I did have look at the wikipedia page to understand what, Snap, Crackle and pop were all about. 😊

Make 2 stop sensors on each end, one for slowing down table and second one for dead stop. With this setup, table/motors/belts etc. at the moment, is going under a very heavy load because there is a high speed and a rapid change of direction. Just make it that table slows down gradually before changing directions.

Always impressed with the details. Jusr curious; Did you test it with a heavy weight on the bed?

Hey James, not sure if you considered this...Maybe an initialization routine that first determines the locations of the left and right stop, then you could ramp up and 'brake' between the two locations. My intuition is telling me that you would have to ramp up before making contact with the part and brake after leaving the part to maximize surface quality uniformity.

Great Video! Have you tried a belt tensioner? Looks like some of the inertia springiness might come from the belt tensioning. Maybe even the wrap angle on the driven wheel? Just food for thought - really enjoying this series and your way of going about things 😊

This is a great tutorial, great understanding, great tenacity & great outcome. What would be the icing on the cake is a test piece, that has an area (strip) of grind at same depth etc but with different motors and RAS settings to see if the quality of traverse actually shows or is measurable in the real world or if the mass of the table hides it. Thanks James, this is perhaps the best (most revealing) project you have done.

Very nice work! I'd be interested how it's going to change with a increase load on the table. Have you run experiments loading the table?

It would be very revealing to compare the motor’s encoder position to the table’s position as determined by a linear encoder scale on the table. However, to do so would rapidly degenerate into a research project. As it stands, James has enlightened most of us with the novel RAS solution. From a design standpoint, it is a great solution because it avoids the expense of a linear encoder and it’s mounting. If this was a ground-up design (vs James’ retrofit) the better approach would be to replace the cable drive with a ball screw. But even then, this is a surface grinder’s X axis: It doesn’t need precise positioning, only smooth velocity.

That's an impressive improvement James. My Rockwell Delta bed sits on ball bearings and there's nothing preventing the bed from lifting as it changes direction. I wonder if your grinder is subject to any mechanical lift during the directional change. If there is, it would affect you surface finish. You might want to check that out. Gilles

Your software could start each run by identifying where the limit switches are located, then just run back and forth within them, to get highly precise travel without worry of overrunning the stops. If you start hitting the limit switches, you know your cable is stretching, has slipped, or has loosened. Better performance and extra safety - win-win.

Would be interesting to see if there will be any difference in thr torque oscillations when you put some weight on the bed...

Nice job! The cost of the new motor is very little... compared to the performance you achieved. I would set the acceleration to about half of what you ended with, the mechanical of the machine will appreciate that... and last way longer. What I usually say... it is just like driving your car with the peddle all the way to the floor. You may get away with it for a day or two. But the car will be destroyed in short order. Yet if you drive is like a grandma... it will last forever. Machines are much the same. Treat them nice, and they will be nice back. Food for thought.

I was only half way through watching and my first thought was you need to provide a delay between direction changes to allow the system to settle. It will be interesting to see what you did...

Fascinating to see how the RAS fixed up the oscillation. I wonder what the difference in the actual servos are - I suspect it is just a software difference - The hardware is identical.

Thanks for a very interesting video. Appreciated. What happens when you put the force of the surface grinding wheel on a metal surface on top of the table? Best, Annie

Nice! Have you thought about using dro-style linear scales for feedback and having a real closed loop position system?

I know it's a personal preference, but I would have made the limit switches virtual. Leave one physical switch at the far end of travel to home, then set start and length parameters in the SW. You could also build a cool interface like the lead screw controller with speed, start and stop displayed. That way you can dynamically control it during the job without have to stop and physically adjust the end stops. Then as the controller knows where the two points are, it can accelerate and decelerate in SW to the exact positions. You don't need the more expensive servo, and you can stop on a dime without any overshoot. You could even leave the two physical end stops at the two extreme positions as estops too late now. You bought the expensive servo. But you have to live with the overshoot... Nice project, though. :-) I'm still jealous ... :-)

Looks good, I am curious though what happens when you introduce a 10kg piece to surface grind and the grinder running as it will change the dynamics of the table. I'm guessing there might be a need for a new session of tuning when that happens.

Great mod, is there a failsafe if the limit switch fails or cams break and the bed progresses infinitely?

Is there any reason to not do the jerk limiting in software? It would require more programming yes, but it also seems like you would have better control over it.

How much will this tuning be affected by the mass of your vise and part? It is the table significantly more massive than anything you'd clamp onto it?

I though that snap/crackle/pop were early April’s fools … but apparently it’s an old April’s fool.

Great video! You may have covered this but do you have a plan to stop the motor if (when) a limit switch fails?

Just a thought. You could have accomplished almost the same thing in two different ways. 1. Count steps between the limit switches and then apply a two step decel. 2. Added two more switches that act as a warning and start your decel on the first switch. Your decel and direction change should take place after the wheel has cleared the part. That way you will not see the direction change or decels/accels in the surface finish of the part. Spent too many years doing this by hand.

I was going to reply to a nasty comment about your code, some nonsense about not needing a state machine. I think his post got deleted. I am glad it did. It wasn't particularly productive.

My grandfather used to “jerk” the accelerator, on and off when he was driving. He said we had Kangaroo petrol 😂

Things will change when you put heavy part on table - inertia and load will rise. You have calibrated empty table

Quoting Wikipedia, "The terms snap, crackle, and pop‍-‌for the fourth, fifth, and sixth derivatives of position‍-‌were inspired by the advertising mascots Snap, Crackle, and Pop." You live, you learn.

Very nice video James! If you are programming the motor movement in C++, it seems you could have programmed in the control system for the motor to control the jerk/snap/... instead of purchasing the enhanced version. Just a thought from another controls guy.

What if you put a bellcrank at each end of travel to lift a weight up against gravity? That might soften the stop and help with the acceleration.

Great video. I came into it frankly thinking that it would likely be boring. I mean, how entertaining could 26 minutes of smoothing X-axis motion be? But it was! And very educational too! (Now I know about "jounce", as just one takeaway.) I'm wondering why you decided on the mechanical lever switches rather than the proximity switches. From your earlier video comparing the two, I really thought the proximity switches would be the winner. Heck, if you switched to using the proximity switches, you could add the mechanical switches as WAY-overkill to trigger eStop if the table went beyond limits.

Awesome video and great project. Can the Clear Core controler be programed in a language other than C++? I haven't used C++ and it's appears daunting to learn.

Big Brushed DC servo would be ideal for the X axis, nice big heavy rotor, hard switching, loads of grunt particularly near shaft stall.

Will resistance, introduced by griding wheel and object friction, affect the smoothness of the x axis?

Could the progressive acceleration (anti-jerk) required not have been implemented in controller software? Along with anticipation of travel limits, based on tracking of the most recent limit switch triggers?

Fiddling with the G stop should help, the trouble is finding the G stop.

nice work James, is there a line of code to make it stop at the end ( Left or Right) and not on the job your grinding

Do you think the added mass of the work piece will require retuning?

James. In many PAST videos you made. I'm now retired from my career in Michigan. I was very involved in automation with huge transfer system, cycling. On and on! What / why are you planning a surface grinder for, production??

I'm surprised that you don't have "stopped" as a third state in your finite state machine. I'm also wondering about "accelerate left", "decelerate left", "accel right", "decel right" but haven't thought long enough about it yet!

Did you do any experiements with the enhanced motor, with the STEPS_PER_REVOLUTION defined as 6400, which I think you mentioned in the spec sheet ?

James, I'm curious if you think the tuning issue was more about the 'squishiness ' of the cable drive on the table or an inertial mismatch?

Looks like all those settings depend on the physics of the machine, including the weight of the table. Does that mean you'd need to adjust things when you put a heavy item on the chuck? Or can you set things low enough to allow some spare capacity before the springiness causes issues? It might be interesting to test the same torque graphs with different amounts of material on the table.

The RAS enabled motor has clearly worked. Would you be able to program different acceleration and deceleration values to smooth out the start of motion if RAS did not work?

I have a headache, good vid, need a lie down...

Clearpath, if you're reading this, not having a 90 day return/swap exemption when a support engineer says "you need this other thing" is a little lame. I'm happy to pay a premium for a nice product with nice support, but I'm also the kind of person who will order motors months before I can start on a project (and discover I bought the wrong ones).

Possibly a newbie question - how much does the table/object speed matter when surface-grinding? The wheel's surface speed is far higher so does it matter if the part is moving slow or fast ?

I was hoping to see you add some weight to the table and then observe the changes.

I've never thought of that limiting jerk limits change of force, because F=ma and dF/dt = m da/dt = mass * jerk.

James, I think this wire spool system is not great. What about if the table slid such that slack developed in the wire? If the table got bumped in the direction it was already headed, there would be slack and no control of the table until the tension in the string caught up. It was as if someone slipped or was horsing around in the shop and pushed or landed on the table in the direction it was already going. The current machine also lacks protection against this edge case. Conversely, the grinding wheel could put tension in the string system if it added a push in the opposite direction of table movement. These issues made me think perhaps a rack and pinion, lead screw, or in the very best case, a ball screw would be better for motion in the Y and Z (did I get my labels correct?). By design, they each resist backlash. I can not imagine it would happen often, but if the grinding disc had just the proper engagement, it could push the table, so it was outrunning the wire, creating slack. It reminds me of when I push my kids on a swing; at the apex, there is no tension in the chains, and you can see them go slack until the kid and swing seat catch up and grab at the chain again. It is perhaps not a great analogy. You are very concerned with safety, so I thought I'd mention this possibility. I wonder what they use on real automated machines?

James, Your work and intuitive deductions are outstanding, there may be several sets of factors that you have entirely missed and may effect the performance of the electric motor in a measurable way. #1 the fit of the molded rubber drive belt into the drive cog and any slop that may result or increase from wear, ie: unlike a metal driving gear system there IS movement and stretch in rubber; and does the belt wear, and or increase in size due to speed and friction and thus increase friction/drag on components? #2 the compression factors and rebound time of the rubber belt itself loading and unloading the forces as they switch back and forth ie: compress/rebound These factors absolutely will effect the minute timing parameters and increase in variability as speed increases.....and therefore influence how your motor will display/chart its data. Should a distinct measurable variability in the motor performance be evident in using the rubber belt cog drive , it may necessitate using a solid gear tooth drive system or a crankshaft type drive to achieve your optimum electric motor performance. Just saying !

How about a dial indicator on the surface of the table to quantify any bounce better than eyeball?