Hi @madgrizzle and thanks for taking the time to do such a nice write up.
The main design target, as it stood, was to add as few parts and expense to the machine as possible so that it is more accessible to the community. However, I personally appreciate the mechatronic approaches to the problem. If one could execute your plan for say 25-50 (even $75 imho) dollars and hold the corner areas as tight as the rest of the tolerances on the machine, I think its a win for sure., and well worth any efforts.
I think the best one could do with the cable approach is compare chain let out to the tension using a load cell and potentiometer, and combine that with a geared DC motor to keep the tension as a function of let-out (closeness to corner) chain – doing that for both corners, and on a separate controller. We would then need to add a program function in GC that would re-calculate the chain sag variable so that it isn’t thrown off as tension is applied during the cut.
I personally believe that changing the physical characteristics/geometry of the machine is the best path forward, rather than trying to compensate for it’s current limitations, but I always want to assist in any exploration of other avenues!
What do you guys think about doing all of the slack compensation in software. If you had a software that could be taught manually by the user software location vs physical location. I am not saying this would be an easy algorithm to write but theoretically you should be able to have a compensation coefficient that is specific to the system that would incorporate organically such variables as drag, chain slack, weight of sled,…etc. (These variables should remain constant for each system…after a fashion at least)
You could have a set training set of shapes or positions that the router would have to hit. then you measure reality versus software and the delta should be a starting point for the software to understand the system.
Has this already been addressed? If so i apologize…
I’m struggling 2 years that the software tells me where the centre of my sheet is in Y-axis.
It never got it right. On X is was always spot on.
I wish i could have centred the bit in a same sized hole in THE REAL CENTRE and was able to tell the software that this is my Y0 and nothing else.
I like the concept, but there is an inherent difficulty in this that would have to be overcome (and it’s not isolated to what you are suggesting, I’ve experienced it in what I’ve tried to do).
What I’ve learned about all my work on the Maslow is that you can’t improve the accuracy of something unless you can measure to something better than what you are trying to achieve. For instance, if your goal is accuracy to < 0.5 mm, then you have to be able to measure better than that to maybe at least 2x times better based upon a gut feeling, but maybe much more is needed… I’m sure if I did some research and read some books on the topic, this has been figured out already.
So, with that said, if you want to put markers on the board that you move the router to and compare actual chain lengths vs. what you calculate, you have to place the markers extremely accurately. Across a 4x8 sheet of plywood, placing the center of a marker down to no greater than 0.25 mm accuracy will be incredibly challenging in my opinion. If you look at my work here: Optical Calibration Demo and Three Hours Working on a Bug you will ultimately see I used a 4x8 printed pattern on a banner. That pattern isn’t perfect in itself because the printer the company used isn’t perfect. So then what do you do?
This isn’t necessarily true. There is a whole field of statistics which covers this topic. In summary, inaccuracies in your measurement system can partially be compensated by making multiple measurements. You measure the position of one point 12 times, each time ensuring that the measurement is not subject to the same bias. Statistically, the average of the 12 measurement points is more accurate than each single measurement.
My comments about statistics do not address this issue.
I think this is called a Measurement System Analysis. The purpose is to identify how accurate your system can measure something.
I agree through statistical analysis of multiple samples you can improve your measurement results. I do that in the optical calibration by processing 10 different camera images and taking the mean value of the results that are within two standard deviations of the initial mean. I do that to make sure I don’t use a garbage measurement because the edge detection went wonky, but I also think it improves accuracy as you suggested.
However, I think the qualifier “each time ensuring that the measurement is not subject to the same bias” is one of many challenges to overcome. You’ve got the accuracy of the tape measure itself, the accuracy of how you place both ends of the tape measure, the accuracy of your eyes and how well you can read a tape measure and then finally, the bias of reading something and not just using the last value you read because you can’t be really sure its any different.
I’ll rephrase my original statement though
“What I’ve learned about all my work on the Maslow is that you can’t improve the accuracy of something unless you have measurements that are better than what you are trying to achieve.”
How about using laser pointers and a optical imaging system. you could even project a laser grid onto the work area and use a “laser bit” instead of a end mill in your router . Use a red laser for the grid and a green one to give you an x/y line that marks the center of the router and then write a learning algorithm to analyze the delta in the corner…NOT that I have even the faintest idea how to do this. You could have this run over night taking hundreds (or thousands) of automatic measurements with extreme detail. The teaching data set would give you a statistically meaningful data set that could overcome data errors.
Using a high MP camera and laser would give you much greater accuracy than you are trying to achieve…if you can calibrate it correctly
I feel that a one size fits all approach will not work. Since you probably even have to take into account the temperature of the chain to calculate slag (A colder chain will be stiffer than an warm chain…etc)
I am just trying to help with ideas/concepts…but I could be totally wrong…or this could be a monumental task…
This happened to me today cutting a square in the far left corner. On the counterclockwise downcut. same trajectory as well (puling the bit in towards the center. In my case it may have been a lose machine screw that had come loose from the bottom of my metal sled. causing to much friction on my sled - or the projecting screw could have been trying to ride in the grove at the top of the cutout… hard to say. Im still testing on my first sheet. Its just full of trials and test… I may try this again with a plain o square and see if the miscut happens again.
So I just recently picked up my maslow and had this idea to complete the loop of the chains to solve corner tension as well as slack issues at the top when I came across this thread. My understanding from the posts is that having the tension coming from both sides does not improve the cutting. Maybe this idea has already been posted, but do you think this could solve this by creating more tension on the side the sled is closest to? Or am I misunderstanding the issue?
Welcome to the Forum!
Oh wow! That looks great!
The issue I see is that you would want all colours attached to the sled the same height over the sheet to keep the sled parallel. Same height should be at the 2 motors and the first 2 rollers at the bottom counting from the sled. With some chain guides to avoid them touching at the crossings, could work. Some more hardware I guess. Thanks for sharing.
This is the part that makes a passive system like this difficult.
My opinion is also that a closed 4 chain system like the first one you proposed also requires too much tension to be safe or achievable on a wooden frame. It seems to me like this would only work if you drastically decreased the weight of the entire system including the chains.
An idea that I have bantered around for 3+ years, but never achieved, is a 3 chain system with a passive counterweight. In theory with three chains, nothing is ever over constrained. I have always thought of this design as a “vectored gravity approach” in a way moving the angle of gravity on the sled to keep both chains in tension. My hunch is that an angle in between the two chain angles is ideal, but I am not smart enough to know how to calculate this.
Anyways, here are two drawings of the idea. I am happy to discuss it if anyone wants. The vector of the passive counterweight is controlled by a 4th motor on a slide.
I like it. Alot of thought went into that, but Im not sure I could handle a build like that though I agree there are probably those on these forums that could. As far as the tension for my idea, I would definately select cording with enough stretch to be safe and requiring as little tension as possible. My initial thought is that the chain as well as attached cords, represented in dark blue and yellow will need minimal stretch properties so that the 2 sided pulley movement distance up and down is predictable. Because this part is doubled over, when 8-9ft of chain is let out or retracted to cross plywood it should move up or down around 4-4.5ft. The tension will be applied by stretch of the orange and light blue cords of the apparatus. These will require a double pulley system at top and bottom so that when not stretched the cord measures about 8-10ft and can reach across the work surface(orange). and when pulley moves up 4-4.5ft, they will stretch to 16-20ft(light blue). In my head it makes sense, but I will get some better measurements and keep updated. For safety factors I can initially hide some additional slack on the bungee side until I can get some initial tension readings.
The problem with this approach is positioning the slide across the
bottom (remember sawdust and chips will be falling on it) and how much does this
complicate the math of figuring out where the sled actually is based on the
motor positions (since you now don’t just have gravity pulling on the sled, you
have this additional force)
our movement speed is limited by the available force of gravity in the bottom
corners, but our accuracy is limited by the accuracy of our mathmatical model of
the machine. I’m not convinced that we are properly modeling chain sag/stretch,
adding an additional force complicates the model more.
the problem is that weights provide constant tension, if you have constant
tension, then lines to the bottom corners won’t work because the angle to the
far corner is going to be flatter than the angle to the near corner.
This will produce more pull towards the far side than to the near side (break
the tension down into a vertical and horizontal component and you will see what
I mean), this will pull the sled in the wrong direction.
Correct. My idea with weights is for weight to be lifted only when sled approaches the bottom corner. The opposite weight would be resting on the ground at this point and be out of play providing tension only to side that requires it. But you are right at the point when the weight is lifted, it would be constant tension
how do you lift the weight only when there is less line fed out?
Also, the point where you start lifting now introduces an abrupt change in the
tension, how do you measure exactly where that happens so that you can add it to
the model and know when you need to run the motors more to compensate for the