Paul & Sabrina's cheap DIY 144v motor controller

[Articus]

Yep power section is built. I used the entire board just for the little DC motor. RTD explorer said 7 amps max on the current feedback.

[jedsmd]

Originally Posted by Articus

Yep power section is built. I used the entire board just for the little DC motor. RTD explorer said 7 amps max on the current feedback.

Congrats! Photos?

(Sorry I've no help on your current limiting question)

[Articus]

Originally Posted by jedsmd

Congrats! Photos?

(Sorry I've no help on your current limiting question)

Thanks and of course . I made 2 of them, here is the first:


I'm waiting on an order from Mouser for the VR1 trim pot on the second one because I lost the ones I originally ordered

[jedsmd]

Nice, what are you putting these into?

[MPaulHolmes]

Depending on the motor, it may not be using 7 amps. Also, there would be very little resolution, since it's a LEM Hass 300-s. There are 128 ticks of the A/D for each 300 amps. So, around 2 amps per tick of the A/D. Figure +/-1 tick of resolution, and it can only resolve +/- 2 amps. My guess is that the motor is using less than 7 motor amps, and so the PWM just ramps to 100% trying to see 7 motor amps.

[Astro]

Originally Posted by MPaulHolmes

Depending on the motor, it may not be using 7 amps. Also, there would be very little resolution, since it's a LEM Hass 300-s. There are 128 ticks of the A/D for each 300 amps. So, around 2 amps per tick of the A/D. Figure +/-1 tick of resolution, and it can only resolve +/- 2 amps. My guess is that the motor is using less than 7 motor amps, and so the PWM just ramps to 100% trying to see 7 motor amps.

Since it is low current and just for testing with a small motor. How about taking the LEM 300 off of the copper bar and running a wire from the copper bar, looping through the LEM several times (not sure how many) and then to the motor? Thereby increasing the magnetic field measured by the LEM. Effectively reducing the LEM to say a LEM 20 rather than an LEM 300.

[MPaulHolmes]

That works very well to do that. Good idea!

[artc]

I'm totally confused by the "PI" loop in the Cougar11B code. Can someone please help me see where I've made a mistake:

In config you have:
Kp ( from the example in the code, this is 1 or 2)
Ki ( from the example in the code somewhere between 20 and 160)


In config_pi:

v1 =Kp<<10
pi.K1=v1 (so ~2000)
pi.k2=Ki-pi.K1 (so ~ -1800)


In the main loop:

pi.error_new = current_ref - current_fb;
pi.pwm += (pi.K1 * pi.error_new) + (pi.K2 * pi.error_old);

drop the pi. and substitute for K2 (watch the i's and 1's):

pwm += (K1 * error_new) + ((Ki-K1) * error_old);
pwm += (K1 * error_new) + (Ki * error_old) - (K1 * error_old);
pwm += (K1 * (error_new - error_old)) + (Ki * error_old);

finally :

pi.error_old = pi.error_new;

I can't see an integral term in there. Normally you'd have something like

pi.total_error += pi.error_new;

but the code is just setting new to old, not accumulating.

The first term appears to be a differential term. If so, what's running is a PD controller, but the proportional term (Ki) is acting on the last error, not the current error.

Substituting some imaginary values:

pi.pwm += (2000*10) -1800*50;
pi.pwm += 20000 - 90000 = -70000;

the final signal is pi.pwm >> 16 so the output would change by 254, not unreasonable.

Empirically, the loop is stable, so I feel I must be missing something - PD is not unstable.

[MPaulHolmes]

I didn't do the PI loop the traditional way. I did it like this:

Let P and I be the constants. error(t) is the error between the desired current, and the actual current. pwm(t) is the duty at time t.

pwm(t) = P*error(t) + I*INTEGRAL(error(TAU)), where the integral goes from 0 to t, hence the dummy variable TAU (remember that in calculus? hehe. They always use TAU! What's up with that?).


Differentiate both sides:

(2) d (pwm)/dt = P*d(error)/dt + I*error(t), by the Fundamental Theorem of Calculus. (The one where differentiating the integral leaves the function)


Now, I approximated d(error)/dt and d(pwm)/dt by finding the slope of two points for a small time interval, like 0.001, which was my throttle sampling rate back in the day, and substitute it into equation (2).



So, d(error)/dt IS REPLACED WITH (errorNew - errorOld)/deltaT. (remember the difference quotient. We just aren't letting h --> 0. Just using small h)



Also, d(pwm)/dt IS REPLACED WITH deltaPWM/deltaT.



Also, replace error(t) with errorOld. (You could use either errorOld or errorNew if you wanted I think).



So, after substituting, Equation (2) becomes:
(2b) deltaPWM/deltaT = P*(errorNew - errorOld)/deltaT + I*errorOld

Now, multiply both sides by deltaT:

(3) deltaPWM = P*(errorNew - errorOld) + I*deltaT*errorOld

Now, distribute P, and get all the errorOld stuff together:



(4) deltaPWM = P*errorNew - P*errorOld + I*deltaT*errorOld

(5) deltaPWM = P*errorNew + (I*deltaT - P)*errorOld


So, kp = P, and ki = I*deltaT - P. kp and ki are constants, because I, P, and deltaT are constant.

[artc]

Originally Posted by MPaulHolmes

I didn't do the PI loop the traditional way.

No argument from me there

You're right in that you're calculating deltaPWM rather than PWM so there's an additional differentiation that comes into play.

Originally Posted by MPaulHolmes

(5) deltaPWM = P*errorNew + (I*deltaT - P)*errorOld

(As a PI equation that seem intuitively horribly wrong, but I can' see a problem in the mathematics).

What bothers me about (5) is that somewhere I'm missing an inversion (possibly U5C?). With a "direct" drive I think that lower pwm, lower OCR1A would lead to an increased current which would imply a negative proportional term. However, I think that output of that loop increases with increasing error.