Arduino & ClearPath Servo: Moving 34 lbs FAST! WW131 

@Morgan Pablo yeah let me guess someone tried it for 20 minutes and it worked perfectly??

@Morgan Pablo ooh you changed it to 10 minutes lmao

Thanks mchiodox69. Tuning the "pesky gains" can be a double-edged sword to a user with limited experience. Used appropriately, they provide superior performance. There is a detailed response about the auto-tuner listed under the post for snikkeldak that you may find interesting.

Thanks Colin MacKenzie. We look forward to working with you. Some details about the auto-tuner can be found under the post for snikkeldak.

John, how did you do in Calculus and Diff EQ's in school? If it wasn't for you I'd avoid a high level controls class (I have a BSME and the math was killer)

Ted Saylor hahaha so true, that was the most math I think I've ever used in my life. Brutal honestly, but when you finally get to the end and have the "aha moment" it's all worth it.

Well it does get quite messy. One of the things that I realized much too late is that mathematics is a special language to describe abstract concepts. Like any other language, it takes a long time to learn to think in that language; and then to move onto the different dialects. There isn't enough time to acquire a language as part of a 4-year course that deals with a thousand other things, if one isn't gifted/cursed. Believe it or not; I found a use for second order, partial differential equations in the real world. (I was dong static footing design because the senior Civil Engineer's brain had exploded.) Well; I reduced the solution space down to exactly one; which was solved for particular cases by numerical methods. 30 years later; not sure that I could replicate the work from scratch without half a year to "warm up".

Bernd Felsche yeah, even though I've been through all the math of PID and could probably still get it done with a little brushing up, I'm thankful for companies like clearpath that make it easy to get the job done. That's what sets a company apart.

NYC CNC Yeah that looks like a nice motor but I need something for a tenth of the price.

Max Maker look into mechaduino. its a board that you put on the back of a stepper that turns it into a servo. They are not very expensive. Steppers have high low speed torque but it is very jerky. Microstepping helps, but there are a lot of factors. you might look into getting a stepper with a gearbox. Look for motors with a high rated voltage, like 24v. They will take more power(with a higher supply voltage) than a low voltage motor at the same current. I have also found the TI DRV8825 based driver modules to be the smoothest, in 32x mode. The reality with motion control is you pay for torque and power, unfortunately. Even a big NEMA 34 stepper and appropriate driver will cost as much as a small clearpath.

Max Maker the other question to ask yourself is do you NEED computerized motion control? You might just be able to use a DC gear motor and a cheap speed controller. If it needs to move between multiple positions you could use micro switches and an Arduino to sense when the mechanism is in position and turn the motor on and off.

You are a true inspiration to a lot of guys wanting to get into machining and cnc work. I probably would have never converted mine over to cnc if it wasn't for all the videos and help that you get out to us with your videos. THANKS SO MUCH.

Thank you for your suggestion. Maybe John will do another video showing such a test. What you would see in such a test is a blip in the tracking accuracy whenever the weight was instantaneously changed (i.e. when a block was picked up and when a block was released). The magnitude of the blip would depend on how extreme the test was (i.e. how fast and how much difference in weight), but in any case, the motor would quickly compensate for this error and be able to adapt for the change in weight. (Comment continues) In real-world applications, many axes do not have well-balanced, non-variable loads. Most CNC applications do not have a consistent load weight and/or cutting forces throughout operation. This is why ClearPath has Variable Load and Inertia Matching Technology. This technology allows for high inertial mismatch and also allows for variable load weights and forces (within reason) with virtually no change in performance. In order for this technology to work optimally, we recommend that the motors are tuned for worst case scenarios (heaviest load weights and forces, and highest inertia mismatches).

As the reply from @Teknic Inc was not made public here it is: Thank you for your suggestion. Maybe John will do another video showing such a test. What you would see in such a test is a blip in the tracking accuracy whenever the weight was instantaneously changed (i.e. when a block was picked up and when a block was released). The magnitude of the blip would depend on how extreme the test was (i.e. how fast and how much difference in weight), but in any case, the motor would quickly compensate for this error and be able to adapt for the change in weight.

I think you'd probably get the same result. It would simply come to a stop more slowly as it accounted for the load. These are very smart drives. They do the calculations for ramp up and ramp down before they even move, and are able to determine when they have to start slowing down in order to prevent overshoot.

NYC CNC thanks!

Autotuning is always a quite involved process which sets all parameters of a controller, in this case the motor controller. For such applications like servos you usually use a so called PID controller which is very effective when tuned properly. Every controller tries to eliminate (respectively minimize) the error of your system. The error is difference between your set point (value to reach) and the actual value of the controlled variable (plant output) and may or may not occur due to disturbances in the system (plant). To achieve zero error, you put a controller in front of your plant which manipulate the signal based on the set point and error and your system. Your controller knows the set point, can measure the actual value and therefore calculate the error (= negative feedback loop). Its task now is to control and adapt the actuating signal (plant input) in such a manner so that your plant exactly behaves like you want it to (plant output = set point). The beauvoir of your plant in most cases is described by a so called transfer function, basically mathematical description of reality, but that isn't the point of interest now, so let it be a black box. Furthermore it is important that the measured value (plant output) is the same type as the set point (plant input). Now the fun part begins... how does it work? Well, in case of a PID controller, the controller "consists" of three parts (P, I and D). All parts eventually are added to the plant input u. P (proportional) takes the value of the error (e) and multiplies it with a specific constant value cp. P is a first rough approach to get in the area of zero error. It's quite accurate if the system doesn't change frequently. e is getting smaller every cycle and finally is zero, or is it? I (integral) sums up all errors in a certain period of time (lets call the sum i) and multiplies it with a specific constant value ci. The P part often can't handle very little errors sufficient enough because of several reasons. The I part though sums up all those little errors over time and is able to counteract those. D (derivative) calculates how fast the error changes (lets call the speed of change d) and multiplies it with a specific constant value cd. The D part is crucial in frequently changing systems. By knowing how fast and in which direction the error changes the controller can adjust the plant input to avoid upcoming and growing errors! It can compensate big errors before the even occurred. Pretty cool, right? As a result: u = e*cp + i*ci + d*cd u now is fed into the system and the errors hopefully is almost zero. But how near to zero it is, is depending on the constant values (also called gains) in your controller and your system. That's the point where Autotuning kicks in. Every system is different. And there aren't universal values for your controller, you have to find the perfect ones for every single application. But since the change of one controller part effects the behavior of the other controller parts as well, you not only have to adapt those constant values to your system but also to themselves! Every change of any variable will effect the performance of the whole controller. You can imagine: trying to figure out the perfect values of a controller for a application is a very involved process and nearly impossible to do by hand. Well, there are certain approaches that help finding those values by hand, but to find the PERFECT ones for your application, not only good ones by hand.. it's almost impossible. The most efficient way is numerical iteration. But depending on how good your algorithm works, the performance of the whole system is good or bad. That's what Autotuning does. It optimizes the variables of the controller just to fit your application perfectly. Hope that helps!

So yes. It even matters for low speed applications. For instance during high accelerations with heavy loads there is a pretty good chance that your system is hunting (real value is hunting set value, your system reacts slower than it should) if the controller (in this particular case especially the D part) isn't properly tuned.

The controller has to take the signal from the arduino as well as the reading from the position encoder and send some amount of current to the servo motor for some amount of time in order to make the "stage" move to the position specified by the control signal (all the while slightly adjusting the current or time based on continuosly monitoring the encoder) The "tuning" sets a bunch of parameters that let the controller know how much current and for how long. This will change depending on such factors as the load, the friction, the power of the motor, the pitch of the lead screw (if one is used), etc. As others have suggested look up PID Control Theory and find an article that walks through P, then PI, then PID or just google for "Control Theory" for deeper explanations (-8

Max Maker, that is a great question. There is a detailed answer listed under the post for snikkeldak.

NYC CNC indeed, a laser cutter needs extremely high performance as I primarily cut paper with fast speeds and acceleration, minor errors are easily reflected in your final product, I need my servos to be with in an error of 5 counts or less. the auto tuning gave me something like 25 error counts on the initial overshoot and the move settled in around 100 ms, manually tuning them I was able to keep my initial overshoot with in 5 counts and have the rest of the motion oscillating within 1 or 2 counts with accelerations of 4600 mm/s^2 which to me is just fantastic

Abraham Saenz, I am glad to hear that the ClearPath is working well for you. There is a more detailed response about the auto-tuner (and your situation) under the post for snikkeldak that you may find informative.

Teknic Inc Thank you, your product is truly remarkable, this product really blows yaskawa, Panasonic and others out of the water for CNC applications, they may have a lot of advanced features but most are useless for CNC, on top of that your product is much sliker, smaller, lighter, easier to use and on top of all that cheaper. The more I learn about servos the more I realize how great the Clearpath's are.

Are you still wondering moron?

Klaufmann I noticed that little error too. 1 micron = 0.000 001 meter = 0.001 millimeter. 0.001 mm / 25.4 mm/in = 0.000 0393 7 in. x 25 = 0.000 984 25 in. Like you said, 25 microns is very close to 1 thousandth of an inch.

steppers only loose steps when you specify something to weak for its purpose. You can also get encoded steppers so you r argument is biased and false.

wow, that's great to hear. I have not working knowledge of servos, but I do know I like the built-in feature of those parts.

Thanks dirkblanston. The ClearPath actually uses PIV control and not PID. Details can be found under the post for snikkeldak.

No one cares

ummm, there are alot of very important parts between fusion 360 gcode and Clearpaths you might mount to your mill. Oh, and "beta testing" might get expensive.

Hi, my name is Tom and I'm an applications engineer at Teknic. snikkeldak has provided some good insight into the auto-tuning approach and there have been several comments on both ClearPath’s auto-tuner and its servo algorithm (a.k.a.: servo compensator), so I'd like to offer some input. The ClearPath auto-tuner uses square wave commands (instantaneous changes in velocity and position) to set the servo gains for the torque loop, the velocity loop, and the position loop. One reason for using square wave commands is to “excite” the mechanics at the machine’s natural resonant frequencies. Once these frequencies have been identified and processed through an FFT analysis, the tuning algorithm will raise the gains as high as possible to provide a robust tuning profile but leave enough design margin to keep the mechanics from resonating. Square waves are also used to drive the servo loops into saturation to make sure that the servo remains completely stable under the worst-case scenarios. (If you’re an experienced servo user, you may find it surprising that we don’t use a “small signal” stimulus like most servos. There are a number of heuristics built into ClearPath’s servo algorithm to effectively deal with these large-signal, square wave inputs which drive the servo into non-linear control regions.) The ClearPath auto-tuner compensates for a wide variety of loads, frictions and mechanical designs. Although it’s true that manual tuning will sometimes result in better performance (more on that in a minute), the opposite is also often true. The auto-tuner often produces a tuning file that is so good that even an experienced Teknic engineer cannot improve upon it. (This is actually a humorous source of frustration for some of our applications engineers!) Also, even when the manual tuning produces better performance (e.g., tighter position tracking), that better performance may be at the expense of robustness (i.e., good performance as the machine wears, or repeatable performance over hundreds of machines built over time as in the case of OEM machine builders). The ClearPath auto-tuner tests for robustness, and does so over the entire given stroke of the axis. The experienced manual tuner can, of course, also do this, but often does not. And finally, there are also a few scenarios, as Abraham has mentioned, that require a specialized tuning approach. An application requirement for high dynamic motion control bandwidth (such as a high speed laser), combined with a mechanical design that can accommodate higher servo gains, and tuned by someone with motion control experience, may be one of those scenarios. That being said, after almost three years of experience with the auto-tuner, we have found that it works well, and is often optimal, for nearly all of the applications out there. Note also, that for users who would like to see if the auto-tuner’s performance can be improved upon, ClearPath has a “fine tuning” control that allows the user to easily and safely adjust the tuning after the auto-tuner has run. This allows you to tweak the tuning for the specifics of the application without having to manually tune. Typically, fine tuning provides improved dynamic and static disturbance rejection/stiffness for a moderate increase in audible noise (or vice versa). When your axis has varying loads that it must deal with when running, it is important to auto-tune your ClearPath motor with the heaviest load/highest inertia for that axis. This will provide the best overall results. Regarding the servo compensator, ClearPath uses a PIV compensator as opposed to a PID version (the "V" term denotes an embedded velocity loop as opposed to the "D" or derivative term within PID). It also uses multi-derivative feedforward gains to reduce tracking errors (preemptively, before they even occur) based on the motion command. Designing a PIV compensator is more difficult than designing a PID, which is why the PID is much more common, but it is higher performance, and the tuning is not iterative like PID tuning. This makes tuning much easier and more deterministic whether the tuning is done manually or by an algorithm.

The 2-cent lesson: Potential Energy (PE) is that which is stored in an object; be it from its position and being able to fall, or in e.g. a spring that's compressed. For the test blocks: PE = m × g × h Where m is the mass, g is gravitational acceleration and h is the height. Mass is how much there is of an object; weight is the Force that that mass exerts under gravity (gravitational acceleration). i.e. F = m × g Raising or lower an object over height h changes its potential energy corresponding to the difference in height. [ I work in SI ("metric") units so don't want to confuse you with the wrong erm... Imperial units. ] The block and the moving part of the slide is being lifted or lowered about half a metre. The mass of the slide is about 20kg (34 lbs + bits). Gravitational acceleration is 9.8 m/s² - near enough to 10 for a first estimate. To change potential energy, "Work" is done. Work in these "kinematics" can be represented by the distance moved against a given force: W = F × s Where s is the distance. The work that is done changes the potential energy. The work to change the potential energy of the test blocks with slide is therefore W = 20 × 10 × 0.5 = 100 joules This is the same as the useful Work done Power is the rate at which work is done. i.e. the work divided by the time that it took: P = W / t When the test block and slide are raised/lowered in e.g. 1.5 seconds total; then the average power is simply: P = 100 / 1.5 = 67 W If we know the peak speed (800 ipm?) of about 0.4 metres/second, we can do some algebra as velocity is distance over time v = s / t and perhaps deduce that P = m × g × v = F × v i.e. that power is equivalent to the force, multiplied by the velocity. So peak power would be P = 20 × 10 × 0.4 = 80 W Note that these calculations don't include losses that produce e.g. heat and noise. They represent the minimum amount of work done or power needed to do it; at 100% efficiency; no losses. Armed with these few formulae, you can estimate the minimum electrical power requirements to move an axis against a resisting force, be it due to gravity of machining loads. NB: If this is wrong, what'd you expect for two cents? ;-)