Re: Question about epid record

[Bruce Hill <bhill@slac.stanford.edu>]

Thanks Mark and Trivan for the advise and clarification!
I think I would have preferred to have them independent,
as I'm less likely to change input or output units than I
am to want to tweak them independently, but changing
it would be a problem for all existing epid records.

Best regards,
- Bruce

On 02/27/2013 05:56 AM, Mark Rivers wrote:
> Hi Bruce,
>
> Thinking about this a little more I think there are advantages of the EPID formulation versus the classical one in terms of scale-independence of the KI and KD terms.
>
> There are input units for the error (let's use degrees for example) and output units for the control variable (let's use volts for example).
>
> The dimensions of the KP, KI and KD terms in the classical formulation are:
>
> KP = volts/deg
> KI = volts/deg/sec
> KD = volts*sec/deg
>
> In the EPID formation the units are:
>
> KP = volts/deg
> KI = 1/sec = Hz
> KD = sec
>
> So KI and KD are independent of the input and output units, they are in pure units of Hz and time.  This means that if, for example, I change the input units from degrees C to degrees F, or the output units from volts to millivolts I don't need to change KI or KD, I only need to change KP.
>
> Mark
>
> ________________________________
> From: Bruce Hill [bhill@slac.stanford.edu]
> Sent: Wednesday, February 27, 2013 12:36 AM
> To: Mark Rivers
> Cc: tech-talk@aps.anl.gov
> Subject: Question about epid record
>
> Hi Mark,
> I'm in the process of replacing a genSub based PID controller with
> the epid record that you and Bob developed, and I'm puzzled by a difference
> in how the two compute the I and D terms.
>
> In the classical PID algorithm, the value of M(n), the manipulated variable is
> M(n) = KP*(E(n) + KI*SUMi(E(i)*dT(i)) + KD*(E(n) -E(n-1))/dT(n)
>
> In fact, the above is from line 33 of devEpidSoft.c, and agrees with
> other descriptions I've found of the PID calculation.
>
> However, in the do_pid() function, and in epidRecord.html, the
> calculation includes the KP gain in the integral and derivative terms.
> M(n) = KP*(E(n) + KP*KI*SUMi(E(i)*dT(i)) + KP*KD*(E(n) -E(n-1))/dT(n)
>
> I can scale my KI and KD gain factors by KP to get the equivalent result,
> but I don't see why the effective integral gain should be affected when
> the proportional gain is tweaked.  Ditto for KD.
>
> Any comments or clarification would be appreciated.
>
> Thanks,
> - Bruce
>
> --
> Bruce Hill
> Member Technical Staff
> SLAC National Accelerator Laboratory
--
Bruce Hill
Member Technical Staff
SLAC National Accelerator Lab
2575 Sand Hill Road M/S 10
Menlo Park, CA  94025