Graphing Axis Errors

I understand you got caught by displaying this in a way you did not like. I could take a look at this, because it seem not to be a math issue, but only reorganizing things a little bit. If you desire, send me a copy of your file and I can examine it. Please use email bhrescobar@gmail.com .

But, before anything else, I understood that your original tolerance was 0.001". Per your description, your robot is expected to perform within this range. So, why are youtrying to push so hard (by measuring and controlling things down to 100 times less than the original spec)? I see no practical reason for doing that. Please, clarify us, maybe we could help better.

GA

I am interested in following this blog as I am into CNC etc..

My question is are you using Steppers, Servos or Something else? (the 3 "S"s), as "repeatability" is difficult on steppers if you are not driving them correctly. They need a good ramp up, a good ramp down and not to go too fast in between, especially not to go into the problems of some frequencies.....as if the wrong frequencies are used and damping is incorrect, they can just sit there and make funny noises, instead of moving!!!

Also, is the Robot loaded up sometimes and sometimes not?

May we ask what sort of Robot it is?

They are servo controlled DC motors. These are wafer handling robots so there is really no load on them. Just the mass of the arm it self.

What type of motors????

You need to supply much more detail, really!

DC motors. What else do you need they are have two brushes and equipped with 2500 encoders. Or 2500 counts per revolution with on index.

I could help you but I do not know how your test is organized. Could you make a sketch? It will help a lot to see which is the best way to obtain your systematic and probabilistic errors. Please show how the measuring devices are positioned with respect to the start and targets. To make a correct analysis it is necessary to know how the measurements are related to the position (as geometrical bonds) they measure.

Here is robot with the test end effector at station c. The Z sensor is the top one, the one in back or in front of the end effector is the R and the T is one one on the side. The testing sequence again is to move to a station near the red bottles that is lower and then one to the right of the sensor and higher then back to station c. When I set up I move to this position manualy and then set the end effector is 10cm from each sensor and then set zero reading of each is about 10. Then the PLC then moves the robot in the sequence of a, c, b, c, a .... for 100 cycles. Does this help?

What errors (magnitude) are you seeing?

Your new sensors are accurate to 0.0001" if I remember correctly, and you are measuring at the end of an arm, that can be placed to within 1/2500 of 360°.

Per axis, how many axis do you have as the picture was remarkable in just how little it informed me in this area!

We also need to know how long the arm(s) are as this affects the size of the (possible) error dramatically.......

Can you supply all of the missing infos and either a drawing (that can be easily seen in spite of the lack of resolution on CR4) or better pictures.

Will try to draw you some pictures later. As the description under the picture said this is a 3 axis robot. T is the rotation of the center black tube. R is the arm extending and retracting and the Z is the height of the arm from the big lack base plate. The software takes the encoder counts and for the R and Z axis sets them to 0.01" movements so a move of 1000 would be one inch for each of the R or Z axis. The spec for these in repeatability is 0.001". For the T axis it is divided in degrees. And the spec is 0.01 deg. New robots will do better than this with no problem. Its when they get old that I want to be able to graph the error and then after repair graph again for the improvement. You asked about the arms length. They can be 1, 2 or 3 link arms and the second link can have different lengths. This is compensated for by the software again. So as the arm moves out the T axis also moves to make it move in a linear straight line. There are acceleration and deceleration steps built in to the software and PID's. Each movement is expressed in T, R, Z by the controller as absolute positions. In the picture it is as follows: T = 797, R = 5919, Z = 2445. These positions are calculated by the controller from a Zero point. This is robot dependent as no all robots have there Zero points in the same location. This will not effect the position in the sensor as it is a taught location for each robot. And for that robot every time it is told to go to Station c it will go to those three taught corrodinates. What I want to graph is ok the robot is at this station c what is the drift if any for each axis. Then we move the robot to some where else and back to station c and again if you ask the robot for its position it will give you the same response, but is that really where it is. Will draw this out and attach it next.

Hope this helps with the other comments.

WOW! Drawings are good enough, many thanks.

But I somehow now feel that you need to decide on how much error you are going to allow yourself at full extension and simply measure to see if you can stay with in it and also how much repeatability you have.....

I did not realise just how many levels of movement you actually have.......

I cannot see how any of us, without hands on access, can help you further.....sadly.

But of course I could be completely wrong (it happens!), someone else may have some great ideas for you. In fact I sincerely hope so.....

Best of luck.

Glad that I was able to get you enough info to see what I am doing. What I need is some help from some math guru to plot out the data that I collect and help me put it in a good graph maybe a spider plot to how from the ref point where each cycle was. Then with a gird showing the spec and did it fall in or out of spec. I am a old duffer that grew up when calculators showed up and thus I learned how to use them to get thru math class. Now don't have the basics to get a handle on this simple graph. Hope to find a helping hand. Thanks.... robert

Ok Robert,

Now I can help you without any difficulty!

1- The best is to use EXCEL to process the data since in it you have also a graphic display of the results and you can chose the one from a wide choice to have it say as much as you want.

2- As I see you have 3 sensors and collect for each positioning the 3 coordinates with respect to the position yo named "zero". Those are coordinates with respect to the sensors so that for your further analysis you should make the differences between what the sensors indicated in your first "zero" setting and following measurements. Those differences are what you are interested in since they give the deviations and via a simple processing the dispersion on each axis. It is iportant for you to make a difference between drift and -dispersion: the drift will be the mean value change versus time (or even better versus number of operations the robot did between the tests you make since wear is proportional to number of cycles in first approximation) and the dispersion will be the differences of the 100 measurements with respect to the mean value computed for each test.

In excel you have already the function "mean valuue" or "average" and the other which is the "standard deviation". This second funtion will give you a measure for the dispersion which is correct under the limit that the distribution has to be a "normal" one called also gaussian distribution (the bell type one). This means that for being sure that your distribution is such or near to a bell type you should verify it.

This can as well be done with excel where you have an other function giving the sum probability depending on the distance the value is from the average.

If using the private message channel you send you mail address I can send you such a xls file so that you have only to adapt it to your data.

3- Now to make a good analysis you should have a graph for the X-Y plane as a shooting target since it will show as well the distances as the angular errors.

4- For the "Z" axis the same can be done.

It is possible with more complex software packages to make a vectorial 3D graph and combine all data in a space picture but as far as I know Excel does not have such a function. I shall have a look if a free soft could be available for this which is of course most interesting to analyse evolution.

From the data knowing the distances you can deduct the angular positioning errors and thus see if the controllers are still of or not any more. One factor which can disturb repeatability and increase errors is friction in the bearings. Its presence influences the angles via the load variations for the drivers and depending on the integral part of the control loop. There are models and procedures to compensate it but it is not easy.

That would great. you can email to robert@intlsearcorp.com. Thanks so much.

Nick, You have been very helpful and value your support. Would never have gotten all the data graphed with our your and the other help. Thanks a lot CR4'ers. Robert