HELP???PID tuning preferences?????Comp Saturday!

[josh_siegel]

This is my first year using PID and I have found that most people use two different methods of manual tuning and I am just wondering what are benefits/drawbacks or if 1 is better than the other or whatever there is to know. These are the 2 methods I found, with my understanding of them.
Method 1: Set Kp, Ki, and Kd, to 0. Then slowly increase Kp until oscillation. Once it reaches oscillation set Kp to half that value. Then increase Ki slowly to get rid of any steady state error. Then increase Kd to speed it up.
Method 2: Set Kp, Ki, and Kd to 0. Slowly increase Kp until oscillation. Once it reaches oscillation, increase Kd until the oscillation goes away. Once the oscillation goes away, increase Kp until it re-oscillates, then increase Kd until it goes away. Repeat that cycle until no amount of Kd will take away the oscillation. Then increase Ki if necessary to take away steady state error.

[josh_siegel]

Trying the first method and getting ok results, but still quite a bit of error.

[Ashwin_Gupta]

Here is a copy/paste from one of my team’s log entries on tuning a turn PID using a gyroscope. The method may work on other plants (systems) as well. Its a very quick method, but will not yield perfect results. Actually, the second method you yourself mentioned is usually effective.

1. Increase P so that the turn slightly overshoots the target (5-10*) and then stops, but cannot return due to P output being too low for motors
2. Now add the I term in order to help bring the robot back to the target after over shooting. This does make it overshoot more (due to error accumlating during the reaching target phase of the turn) but then helps bring the robot to the intended value. This causes the robot to oscilate back and forth around the target tremndously for a few seconds. If it oscilates for more then a few seconds, decrease. If it does not reach the desired value and stops oscilating, increase.
3. The integral oscilations are helpful for reaching the target, but not so nice because they eat up valuable time and tend to shake the robot which is bad for the motors (rapid direction change), causes gyroscope drift to accumulate, and makes the robot drift horizontally if on omni-directional wheels. This can be fixed with derrivative which serves as a dampener. Add in the D term.
4. Increase D such that the robot applies “breaks” as it nears the target and slows tremndously. If the robot stops in the middle of the turn. Decrease D because stopping probably means the D has become too high and is counteracting the proportionality too strongly. If the robot is still oscilating at the integral, increase D.
5. Continue to modify D so that integral oscilations do not exceed 1-3 and the robot still is within 1-2 degrees of the target. If the robot is not quite reaching the target with an increased D, consider increasing P. If the oscilations are overwhelming and are not being brought down, decrease I (very slightly).
6. Once accuracy and repeatiblity have been achieved. Switch desired value. Repeat steps 4 and 5 until turns on 45, 90, and 180 are all excellent.

I will warn you, depending on what this PID is for you might need another 2-8 hours to start tuning from scratch. I hope you are prepared for an all nighter tomorrow or tonight. Otherwise, continue with the method you are already using if that is working for you even in part, because you are probably closer to reasonably accurate results. Best of luck.

[josh_siegel]

@Ashwun Gupta Thank you! I think am going to try the second method I mentioned since I have a good 10 hours at least to work…

[nenik]

@josh_siegel At least part of your Method 1 doesn’t make sense. Increasing Kd won’t speed your PID up, rather to the contrary - Kd is slowing down the approach more close to the target value
Kd helps avoid oscillations by braking earlier if you’re getting to the target too fast.
So your Method 1 would be more like increase Kp until oscillation, then increase Kd until it stops oscillation.
Throw in some Ki for the steady state, then do one or two more rounds of increasing Kp and Kd to make your PID faster.

[josh_siegel]

@nenik what you described wrong with method 1 is what method 2 is. Thank you for letting me know about the derivative! I was way off lol.

[Ashwin_Gupta]

@nenik

@josh_siegel At least part of your Method 1 doesn’t make sense. Increasing Kd won’t speed your PID up, rather to the contrary - Kd is slowing down the approach more close to the target value
Kd helps avoid oscillations by braking earlier if you’re getting to the target too fast.
So your Method 1 would be more like increase Kp until oscillation, then increase Kd until it stops oscillation.
Throw in some Ki for the steady state, then do one or two more rounds of increasing Kp and Kd to make your PID faster.

yes good catch. kD will slow you down. @josh_siegel Remember, D is the instantaneous rate of change in the error so as you begin to accelerate and approach the target, the current error is much less than the previous error. This results in a negative value for kD* (currentError - prevError) and therefore reducing motor speeds and applying a “break”.

[Team80_Giraffes]

@josh_siegel

This is my first year using PID and I have found that most people use two different methods of manual tuning and I am just wondering what are benefits/drawbacks or if 1 is better than the other or whatever there is to know. These are the 2 methods I found, with my understanding of them.
Method 1: Set Kp, Ki, and Kd, to 0. Then slowly increase Kp until oscillation. Once it reaches oscillation set Kp to half that value. Then increase Ki slowly to get rid of any steady state error. Then increase Kd to speed it up.
Method 2: Set Kp, Ki, and Kd to 0. Slowly increase Kp until oscillation. Once it reaches oscillation, increase Kd until the oscillation goes away. Once the oscillation goes away, increase Kp until it re-oscillates, then increase Kd until it goes away. Repeat that cycle until no amount of Kd will take away the oscillation. Then increase Ki if necessary to take away steady state error.

Zeigler nichols is not that far from Method 1 and is faster. You should wait until critical oscillation. Meaning when it stops settling down.

Then you can apply zeigler nichols formulas. You need to log data to get the time period of oscillation for the i and d constants.

Should not take you more than an hour. Maybe half an hour. Start with the arm down at the floor, and set the target to the height you want. Start the program. Log the data. Keep going up until oscillation gets out of control. Measure the milliseconds from peak to peak.

Do repeated executions or get a potentiometer to vary the kp value for you (more advanced). Buttons to send the arm to the floor help trimming time.

Manually tune after zeigler nichols formulas are applied. Should be close.

Optional: repeat process going from up position down to the floor. More of a tuning down kp and increasing kd because gravity assists in that momentum down to the floor. Whacking the floor makes oscillation near impossible.