734 lines
34 KiB
JavaScript
734 lines
34 KiB
JavaScript
|
var longDescriptionObj = {
|
||
|
|
//*************************************************************************/
|
||
|
|
// PID
|
||
|
|
//*************************************************************************/
|
||
|
|
"PID":
|
||
|
|
`# ***Linear PID Controllers***
|
||
|
|
|
||
|
|
The basic controller type described here is a linear PID. The PID implementations in the
|
||
|
|
DCL include several features not commonly found in basic PID designs, and this
|
||
|
|
complexity is reflected the benchmark figures. Applications which do not require
|
||
|
|
derivative action, or are more sensitive to cycle efficiency, may be better served by the
|
||
|
|
simpler PI controller structure described in the next section.
|
||
|
|
|
||
|
|
***NOTE***: The DCL contains one implementation of a linear PID controller in double precision
|
||
|
|
floating point form. This controller may be used with the FPU32 CPU, however support
|
||
|
|
for the double precision data type currently relies on the run time support libraries which
|
||
|
|
are not cycle efficient. The structure of the controller is identical to the PID_C1 & PID_C2
|
||
|
|
controllers described earlier in this chapter. Support functions for PIDF64 do not currently
|
||
|
|
include the ability to load the controller from transfer function coefficient or ZPK3
|
||
|
|
descriptions.
|
||
|
|
|
||
|
|
PID control is widely used in systems which employ output feedback control. In such
|
||
|
|
systems, the controlled output is measured and fed back to a summing point where it is
|
||
|
|
subtracted from the reference input. The difference between the reference and feedback
|
||
|
|
corresponds to the control loop error and forms the input to the PID controller.
|
||
|
|
|
||
|
|
Conceptually, the PID controller output is the parallel sum of three paths which act
|
||
|
|
respectively on the error, error integral, and error derivative. The relative weight of each
|
||
|
|
path may be adjusted by the user to optimize transient response, or to emulate the
|
||
|
|
behavior of a specified transfer function expressed in terms of its poles and zeros.
|
||
|
|
|
||
|
|
*Figure: Parallel form PID controller*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
The diagram above shows the structure of a continuous time parallel PID controller. The
|
||
|
|
output of this controller is captured in the following equation:
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
Conceptually, the controller comprises three separate paths connected in parallel. The
|
||
|
|
upper path contains an adjustable gain term (Kp). Its effect is to fix the open loop gain of
|
||
|
|
the control system. Since loop gain is proportional to this term, Kp is known as
|
||
|
|
proportional gain.
|
||
|
|
A second path contains an integrator which accumulates error history. A separate gain
|
||
|
|
term acts on this path. The output of the integral path changes continuously as long as a
|
||
|
|
non-zero error (e) is present at the controller input. A small but persistent servo error has
|
||
|
|
the effect of driving the output of the integrator such that the loop error will eventually
|
||
|
|
disappear. The principal effect of the integral path is therefore to eliminate steady state
|
||
|
|
error. The effect of the integral gain term is to change the rate at which this happens.
|
||
|
|
Integral action is especially important in applications such as electronic power supplies,
|
||
|
|
which must maintain accurate regulation over long periods of time.
|
||
|
|
The third path contains a differentiator. The output of this path is large whenever the rate
|
||
|
|
of change of the error is large. The principal effects of the derivative action are to damp
|
||
|
|
oscillation and reduce transients.
|
||
|
|
The operation of the PID controller can be visualized in terms of the transient error
|
||
|
|
following a step change of set-point.
|
||
|
|
|
||
|
|
*Figure: PID control action*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
The figure above shows the action of the PID controller in terms of the control loop error
|
||
|
|
at time t1. The proportional term contributes a control effort which is proportional to the
|
||
|
|
instantaneous loop error. The output of the integral path is the accumulated error history:
|
||
|
|
the shaded area in the lower plot. The contribution of the derivative path is proportional to
|
||
|
|
the rate of change of the loop error. Derivative gain fixes the time interval over which a
|
||
|
|
tangential line to the error curve is projected forward in time.
|
||
|
|
Tuning the PID controller is a matter of finding the optimum combination of these three
|
||
|
|
effects. This in turn means finding the best balance of the three gain terms. For more
|
||
|
|
information on PID control & tuning, see the references in section 6.1.
|
||
|
|
The PID shown above is known as the “parallel” form because the three controller gains
|
||
|
|
appear in separate parallel paths. A slightly different PID architecture in which the
|
||
|
|
proportional gain is moved into the output path (i.e. after the summing point), so that the
|
||
|
|
proportional path becomes a direct connection between the controller input and the
|
||
|
|
summing point, is known as the “series” or “ideal” form. In the ideal form, the open loop
|
||
|
|
gain is directly influenced by the proportional controller gain, and there is less interaction
|
||
|
|
between the controller gains. However the proportional gain cannot be zero (since the
|
||
|
|
loop would be opened), and to maintain good control cannot be small. The parallel form
|
||
|
|
allows the proportional gain to be small, however there is slightly more interaction
|
||
|
|
between the controller gains, complicating the tuning process. The DCL contains both
|
||
|
|
ideal and parallel PID functions.
|
||
|
|
|
||
|
|
## ***Implementation***
|
||
|
|
|
||
|
|
The linear PID controllers in the DCL include the following features:
|
||
|
|
* Parallel and ideal forms
|
||
|
|
* Programmable output saturation
|
||
|
|
* Independent reference weighting on proportional path
|
||
|
|
* Anti-windup integrator reset
|
||
|
|
* Programmable low-pass derivative filter
|
||
|
|
* External saturation input for integrator anti-windup
|
||
|
|
* Adjustable output saturation
|
||
|
|
|
||
|
|
It is important to note that the controller sample period is not accounted for in the selection
|
||
|
|
of integral gain (Ki). This is relevant when computing the integral gain as opposed to
|
||
|
|
manual tuning against a transient response, for example. In such situations, users must
|
||
|
|
multiply the computed integral gain by the sample period before loading Ki (either directly
|
||
|
|
or through the SPS). The element T in the CSS sub-structure can be used to store the
|
||
|
|
sample period for this purpose.
|
||
|
|
|
||
|
|
All PID type controllers in the library implement integrator anti-windup reset in a similar
|
||
|
|
way. A clamp is present at the controller output which allows the user to set upper and
|
||
|
|
lower limits on the control effort. If either limit is exceeded, an internal floating-point
|
||
|
|
controller variable changes from 1.0 to 0.0. This variable is multiplied by the integrator
|
||
|
|
input, such that the integrator accumulates zero data only when the output is saturated,
|
||
|
|
thus preventing the well-known “wind-up” phenomenon.
|
||
|
|
|
||
|
|
The PID controllers in the library make provision for anti-windup reset to be triggered from
|
||
|
|
an external part of the loop. This is useful in situations where a component outside the
|
||
|
|
controller may be saturated. The floating-point variable lk is expected to be either 1.0 or
|
||
|
|
0.0 in the normal and saturated conditions respectively. If this feature is not required, the
|
||
|
|
functions should be called with the lk argument set to 1.0. Note that all the controllers
|
||
|
|
here require non-zero proportional gain to recover from loop saturation.
|
||
|
|
|
||
|
|
The derivative PID path includes a digital low-pass filter to avoid amplification of unwanted
|
||
|
|
high frequency noise. The filter implemented here is a simple first order lag filter
|
||
|
|
(with differentiator), converted into discrete form using the Tustin transform. Referring to
|
||
|
|
Figure 6, the difference equation of the filtered differentiator is
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
The temporary storage elements d2 & d3 must be preserved from the (k - 1)th interval, so
|
||
|
|
the following must be computed after the differentiator update.
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
The derivative filter coefficients are
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
Both the sample period (T) and filter time constant (tau) must be determined by the user.
|
||
|
|
The time constant is the reciprocal of the desired filter bandwidth in radians per second.
|
||
|
|
All linear PID controller functions use a common C structure to hold coefficients and data.
|
||
|
|
Refer to the header file DCLF32.h for details of the DCL_PID controller structure.
|
||
|
|
The library PID controller architectures are shown below:
|
||
|
|
|
||
|
|
*Figure: DCL_PID C1, C2, & L1 architecture*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
*Figure: DCL_PID C3, C4, & L2 architecture*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
`,
|
||
|
|
//*************************************************************************/
|
||
|
|
// PI
|
||
|
|
//*************************************************************************/
|
||
|
|
"PI":
|
||
|
|
`# ***Linear PI Controllers***
|
||
|
|
The continuous time parallel PI control equation is
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
The linear PI controllers in the DCL differ from the PID in the following respects.
|
||
|
|
* Removal of derivative path
|
||
|
|
* Removal of set-point weighting
|
||
|
|
* No provision for external saturation input
|
||
|
|
|
||
|
|
In all other respects, the PI controllers are similar to the PID controllers.
|
||
|
|
|
||
|
|
## ***Implementation***
|
||
|
|
|
||
|
|
All linear PI controller functions use a common C structure to hold coefficients and data,
|
||
|
|
defined in the header files DCLF32.h and DCLCLA.h.
|
||
|
|
|
||
|
|
The PI controller architectures are shown in the following diagrams.
|
||
|
|
|
||
|
|
*Figure: DCL_PI C1, C2, L1, & L3 architecture*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
*Figure: DCL_PI C3, C4, L2 & L4 architecture*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
*Figure: DCL_PI C5 architecture*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
Note that the C5 parallel form PI controller contains enhanced anti-windup reset logic which allows the
|
||
|
|
integral path to recover from saturation even when the proportional gain is zero.
|
||
|
|
|
||
|
|
*Figure: DCL_PI C6, C7, & L5 architecture*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
The C6, C7, & L5 controllers combine a series form PI with a Tustin integrator. This configuration
|
||
|
|
is best suited to applications in which the controller gains are selected on the basis of high
|
||
|
|
frequency loop gain and zero frequency, because Kp and Ki are effectively un-coupled in the
|
||
|
|
series form controller. Furthermore the Tustin integrator has fixed 90 degree phase lag at all
|
||
|
|
frequencies below the Nyquist limit, simplifying design of the compensator.
|
||
|
|
`,
|
||
|
|
//*************************************************************************/
|
||
|
|
// NLPID
|
||
|
|
//*************************************************************************/
|
||
|
|
"NLPID":
|
||
|
|
`# ***Non-linear PID Controller***
|
||
|
|
The DCL includes two implementations of a non-linear PID controller, denoted NLPID.
|
||
|
|
The controllers are broadly similar to the ideal PID_C1 controller, except that there is no
|
||
|
|
set-point weighting and in one case the derivative path sees the servo error instead of the
|
||
|
|
feedback. A non-linear gain block appears in series with each of the three paths.
|
||
|
|
|
||
|
|
To improve computational efficiency, the non-linear law is separated into two parts: one
|
||
|
|
part which is common to all paths, and a second part which contains terms specific to
|
||
|
|
each path. The non-linear part of the high-level controller structure is shown below:
|
||
|
|
|
||
|
|
*Figure: DCL_NLPID C1 input architecture*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
The linear part of the DCL_NLPID_C1 controller is shown below:
|
||
|
|
|
||
|
|
*Figure: DCL_NLPID C1 output architecture*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
*Figure: DCL_NLPID C2 & C3 architecture*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
The non-linear control law is based on a power function of the modulus of the servo error,
|
||
|
|
with a linearized region about the zero error point. In the equation below, x represents the
|
||
|
|
input to the control law, y, the output, and "ALPHA" is a user selectable exponent representing
|
||
|
|
the degree of non-linearity.
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
The following plot shows the x-y relationship for values of "ALPHA" between 0.2 and 2. Notice
|
||
|
|
that the curves intersect at x = 0, and x = ±1.
|
||
|
|
|
||
|
|
*Figure: Non-linear control law input-output plot*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
The gain of the control law is the slope of the x-y curve. Observe that with "ALPHA" = 1 the
|
||
|
|
control is linear with unity gain. With "ALPHA" > 1 the gain is zero when x = 0, and increases as x
|
||
|
|
increases. In the controller, a value of "ALPHA" in this range produces controller gain which
|
||
|
|
increases with increasing control error. With 0 < "ALPHA" < 1 the gain at x = 0 is infinite, and falls
|
||
|
|
as x increases. This range of "ALPHA" setting produces controller gain which decreases with
|
||
|
|
increasing servo error.
|
||
|
|
|
||
|
|
The plots below show the gain vs. control loop error curves for different values of "ALPHA".
|
||
|
|
Notice particularly the singularity at x = 0 when "ALPHA" < 1.
|
||
|
|
|
||
|
|
*Figure: Gain vs error curves for varying alpha*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
The presence of zero or infinite gain at the zero control error point leads to practical
|
||
|
|
difficulties. With "ALPHA" > 1 the response becomes sluggish for small errors; with "ALPHA" < 1 it is
|
||
|
|
common to encounter oscillation or “chattering” near steady-state. These issues can be
|
||
|
|
overcome by limiting the controller gain in a region close to x = 0. This is done by
|
||
|
|
modifying the control law to introduce a user selectable region about x = 0 with linear gain
|
||
|
|
is applied. The non-linear control law becomes
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
When the magnitude of servo error falls below δ the linear gain is applied, otherwise the
|
||
|
|
gain is determined by the non-linear law. For computational efficiency, we pre-compute
|
||
|
|
the gain in the linear region ("GAMMA") as follows.
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
A typical plot of the linearized control law is shown below. Observe that when x = δ the
|
||
|
|
linear and non-linear curves intersect, so the controller makes a smooth transition
|
||
|
|
between the linear and non-linear regions as the servo error passes through x = ±δ.
|
||
|
|
|
||
|
|
*Figure: NLPID linearized region*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
In addition to the P, I, and D gains, the user must select two additional terms in each
|
||
|
|
control path: "ALPHA" and δ. The library includes a separate function to compute and update "GAMMA"
|
||
|
|
for each path using the SPS structure. The use must initialize "GAMMA" parameters before calling
|
||
|
|
the NL controllers.
|
||
|
|
|
||
|
|
The NLPID controller has been seen to provide significantly improved control in many
|
||
|
|
cases, however it must be remembered that increased gain, even if only applied to part of
|
||
|
|
the control range, can lead to significantly increased output from the controller. In most
|
||
|
|
cases, this 'control effort' is limited by practical factors such as actuator saturation, PWM
|
||
|
|
modulation range, and so on. The corollary is that not every application benefits equally
|
||
|
|
from the use of non-linear control and some may see no benefit at all. In cases where
|
||
|
|
satisfactory performance cannot be achieved through the use of linear PID control, the
|
||
|
|
user is advised to start with all "ALPHA" = 1, and experiment by introducing non-linear terms
|
||
|
|
gradually while monitoring both control performance and the magnitude of the control
|
||
|
|
effort. In general, P and I paths benefit from increased gain at high servo error ("ALPHA" > 1),
|
||
|
|
while the D path benefits from reduced gain at high servo error ("ALPHA" < 1), but this is not
|
||
|
|
universally true. Further information on tuning the nonlinear PID controller can be found
|
||
|
|
in the Nonlinear PID Controller Tuning Guide in the \docs directory of the DCL.
|
||
|
|
|
||
|
|
Further information on this control law can be found in: “From PID to Active Disturbance
|
||
|
|
Rejection Control”, Jingqing Han, IEEE TRANSACTIONS ON INDUSTRIAL
|
||
|
|
ELECTRONICS, VOL. 56, NO. 3, MARCH 2009
|
||
|
|
|
||
|
|
## ***Implementation***
|
||
|
|
The NLPID controller uses a C structure to hold coefficients and data, defined in the
|
||
|
|
header file DCL_NLPID.h. Note that the NLPID functions make use of the pow()
|
||
|
|
function in the standard C library. For this reason the header file math.h must be
|
||
|
|
included, which is not supported by the CLA compiler. To allow different DCL functions to
|
||
|
|
be run on both the CPU and CLA in the same program, the NLPID functions are located in
|
||
|
|
a separate header file. Refer to the file DCL_NLPID.h for details of the NLPID controller
|
||
|
|
structure.
|
||
|
|
`,
|
||
|
|
//*************************************************************************/
|
||
|
|
// DF11
|
||
|
|
//*************************************************************************/
|
||
|
|
"DF11":
|
||
|
|
`# ***Direct Form 1 (First Order) Compensators***
|
||
|
|
The DCL includes one first order compensator in Direct Form 1. The DF11 compensator
|
||
|
|
implements a first order, or “simple lag”, type frequency response. The general form of
|
||
|
|
discrete time first order transfer function is
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
Denominator coefficients must be normalized accordingly. The corresponding difference
|
||
|
|
equation is
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
A diagrammatic representation of the DF11 is shown below:
|
||
|
|
|
||
|
|
*Figure: DCL_DF11 C1, C2, & L1 architecture*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
## ***Implementation***
|
||
|
|
All DF11 functions use a common C structure to hold coefficients and data, defined in the
|
||
|
|
header files DCLF32.h and DCLCLA.h.
|
||
|
|
`,
|
||
|
|
//*************************************************************************/
|
||
|
|
// DF13
|
||
|
|
//*************************************************************************/
|
||
|
|
"DF13":
|
||
|
|
`# ***Direct Form 1 (Third Order) Compensators***
|
||
|
|
The Direct Form 1 (DF1) structure is a common type of discrete time control structure
|
||
|
|
used to implement a control law or dynamical system model specified either as a polezero
|
||
|
|
set, or as a rational polynomial in z (i.e. a discrete time transfer function). The DCL
|
||
|
|
includes one third order DF1 compensator, denoted “DF13”.
|
||
|
|
|
||
|
|
In general, the Direct Form 1 structure is less numerically robust than the Direct Form 2
|
||
|
|
(see below), and for this reason users are encouraged to choose the latter type whenever
|
||
|
|
possible. However, the DCL_DF13 structure is very common in digital power supplies
|
||
|
|
and for that reason is included in the library. The same function supports a second order
|
||
|
|
control law after the superfluous coefficients (a3 & b3) have been set to zero.
|
||
|
|
|
||
|
|
The general form of third order transfer function is
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
Notice that the coefficients have been adjusted to normalize the highest power of z in the
|
||
|
|
denominator. There is no notational standard for numbering of the controller coefficients;
|
||
|
|
the notation used here has the advantage that the coefficient suffixes are the same as the
|
||
|
|
delay line elements and this helps with clarity of the assembly code, however other
|
||
|
|
notations may be found in the literature. The corresponding difference equation is
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
The DF13 controller uses two, three-element delay lines to store previous input and
|
||
|
|
output data required to compute u(k). A diagrammatic representation is shown below.
|
||
|
|
|
||
|
|
*Figure: DCL_DF13 C1, C4, & L1 architecture*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
The DF13 control law consists of seven multiplication operations which yield seven partial
|
||
|
|
products, and six addition or subtraction operations which combine the partial products to
|
||
|
|
obtain the compensator output, u(k). When implemented in this way, the control law is
|
||
|
|
referred to as the “full” DF13 form.
|
||
|
|
|
||
|
|
The DF13 control law can be re-structured to reduce control latency by pre-computing six
|
||
|
|
of the seven partial products which are already known in the previous sample interval.
|
||
|
|
The control law is then broken into two parts: the “immediate” part and the “partial” part.
|
||
|
|
|
||
|
|
The advantage of doing this is to reduce the “sample-to output” delay, or the time between
|
||
|
|
e(k) being sampled, and a corresponding u(k) becoming available. By partially precomputing
|
||
|
|
the control law, the computation delay can be reduced to one multiplication
|
||
|
|
and one addition.
|
||
|
|
|
||
|
|
In the kth interval, the immediate part is computed.
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
Next, the v(k) partial result is pre-computed for use in the (k+1)th interval.
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
Structurally, the pre-computed control law can be drawn as below:
|
||
|
|
|
||
|
|
*Figure: DCL_DF13 C2, C3, C5, C6, L2, & L3 architecture*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
The pre-computed structure allows the controller output (u(k)) to be used as soon as it is
|
||
|
|
computed. The remaining terms in the third order control law do not involve the newest
|
||
|
|
input e(k) and therefore do not affect u(k). These terms can be computed after u(k) has
|
||
|
|
been applied to the control loop and the input-output latency of the controller is therefore
|
||
|
|
reduced.
|
||
|
|
|
||
|
|
A further benefit of the pre-computed structure is that it allows the control effort to be
|
||
|
|
clamped after the immediate part. Computation of the pre-computed part can be made
|
||
|
|
dependent on the outcome of the clamp such that if u(k) matches or exceeds the clamp
|
||
|
|
limits there is no point in pre-computing the next partial control variable and the
|
||
|
|
computation can be avoided. The DCL includes three clamp functions intended for this
|
||
|
|
purpose.
|
||
|
|
|
||
|
|
## ***Implementation***
|
||
|
|
All DF13 functions use a common C structure to hold coefficients and data, defined in the
|
||
|
|
header files DCL.h and DCLCLA.h.
|
||
|
|
|
||
|
|
The assignment of coefficients and data in the DCL_DF13 structure to those in the
|
||
|
|
diagram is shown below:
|
||
|
|
|
||
|
|
*Figure: DCL_DF13 data & coefficient layout*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
To implement a full DF13 controller, the user might call:
|
||
|
|
|
||
|
|
uk = DCL_runDF13_C1(&df13, ek);
|
||
|
|
|
||
|
|
…where ek and uk are the controller input and output respectively, and df13 is the
|
||
|
|
controller structure.
|
||
|
|
|
||
|
|
To implement a pre-computed DF13 controller, the user might call:
|
||
|
|
|
||
|
|
uk = DCL_runDF13_C2(&df13, ek, vk);
|
||
|
|
vk = DCL_runDF13_C3(&df13, ek, uk);
|
||
|
|
|
||
|
|
...where vk is an intermediate float variable. This arrangement allows the user to make
|
||
|
|
use of the output immediately by placing instructions between the two lines above to use
|
||
|
|
the newest value of uk.
|
||
|
|
|
||
|
|
For applications where it is necessary to restrict the output of the compensator, a clamp
|
||
|
|
function can be used as follows.
|
||
|
|
|
||
|
|
uk = DCL_runDF13_C2(&df13, ek, vk);
|
||
|
|
s = DCL_runClamp_C1(&uk, upperLim, lowerLim);
|
||
|
|
if (0U == s)
|
||
|
|
{
|
||
|
|
vk = DCL_runDF13_C3(&df13, ek, uk);
|
||
|
|
}
|
||
|
|
|
||
|
|
The clamp function limits the controller output to lie between “lowerLim” and “upperLim”
|
||
|
|
and sets the unsigned integer “s” to 1 is that range is exceeded. The pre-computed part
|
||
|
|
of the controller will only be executed when the immediate result is in range.
|
||
|
|
`,
|
||
|
|
//*************************************************************************/
|
||
|
|
// DF22
|
||
|
|
//*************************************************************************/
|
||
|
|
"DF22":
|
||
|
|
`# ***Direct Form 2 (Second Order) Compensators***
|
||
|
|
The C2000 Digital Controller Library contains a second order implementation of the Direct
|
||
|
|
Form 2 controller structure, denoted “DCL_DF22”. This structure is sometimes referred to
|
||
|
|
as a “bi-quad” filter and is commonly used in a cascaded chain to build up digital filters of
|
||
|
|
high order.
|
||
|
|
|
||
|
|
***NOTE***: From version 3.4, the DCL contains support for the DF22 compensator in double precision
|
||
|
|
floating point. The controller architecture is identical to that described in section 3.8 so
|
||
|
|
the details need not be repeated here. One full controller and a pair of functions to
|
||
|
|
support a pre-computed controller are implemented. There are also 'reset' and 'update'
|
||
|
|
functions, and a double precision clamp function. Other single precision
|
||
|
|
functions to support coefficient calculation are not implemented at this time.
|
||
|
|
|
||
|
|
The transfer function of a second order discrete time compensator is
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
The corresponding difference equation is
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
A diagrammatic representation of the full Direct Form 2 realization is shown below:
|
||
|
|
|
||
|
|
*Figure: DCL_DF22 C1, C4,L1, & L4 architecture*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
As with the DCL_DF13 compensator, sample-to-output delay can be reduced through the
|
||
|
|
use of pre-computation. The immediate and pre-computed control laws are as follows. In
|
||
|
|
the kth interval, the immediate part is computed
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
Next, the v(k) partial result is pre-computed for use in the (k+1)th interval.
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
The pre-computed form of DCL_DF22 is shown in the following diagrams.
|
||
|
|
|
||
|
|
*Figure: DCL_DF22 C2, C5, & L2 architecture*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
*Figure: DCL_DF22 C3, C6, & L3 architecture*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
Notice that pre-computation is a little different from the Direct Form 1 case because the
|
||
|
|
intermediate value exists as one of the internal states and is therefore automatically
|
||
|
|
stored as “x1” in the DCL_DF22 structure. Therefore it is not necessary to create a
|
||
|
|
separate variable to store v(k).
|
||
|
|
|
||
|
|
## ***Implementation***
|
||
|
|
All DF22 functions use a common C structure to hold coefficients and data, defined in the
|
||
|
|
header file DCL.h and DCLCLA.h.
|
||
|
|
`,
|
||
|
|
//*************************************************************************/
|
||
|
|
// DF23
|
||
|
|
//*************************************************************************/
|
||
|
|
"DF23":
|
||
|
|
`# ***Direct Form 2 (Third Order) Compensators***
|
||
|
|
The third order Direct Form 2 compensator (DF23) is similar in all respects to the DF22
|
||
|
|
compensator. Separate full and pre-computed forms are supplied in C and assembly for
|
||
|
|
computation on the FPU32, and in assembly for computation on the CLA.
|
||
|
|
The control law is the same as the DF13 compensator.
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
A diagrammatic representation of the full third order Direct Form 2 compensator is shown
|
||
|
|
below:
|
||
|
|
|
||
|
|
*Figure: DCL_DF23 C1, C4, & L1 architecture*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
Sample-to-output delay can be reduced through the use of pre-computation, in a similar
|
||
|
|
way to the DF22 compensator. In the kth interval, the immediate part is computed.
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
Next, the v(k) partial result is pre-computed for use in the (k+1)th interval.
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
The pre-computed form of DF23 is shown in the following diagrams:
|
||
|
|
|
||
|
|
*Figure: DCL_DF23 C2, C5, & L2 architecture*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
*Figure: DCL_DF23 C3, C6, & L3 architecture*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
## ***Implementation***
|
||
|
|
All DF23 functions use a common C structure to hold coefficients and data, defined in the header
|
||
|
|
file DCL.h. and DCLCLA.h.
|
||
|
|
`,
|
||
|
|
//*************************************************************************/
|
||
|
|
// PID32
|
||
|
|
//*************************************************************************/
|
||
|
|
"PID32":
|
||
|
|
`# ***Fixed-Point PID Controllers***
|
||
|
|
The DCL contains one implementation of a parallel form fixed-point PID controller. The
|
||
|
|
structure is similar to the floating-point C1 controller.
|
||
|
|
|
||
|
|
The linear PID controller in the DCL32 includes the following features.
|
||
|
|
* Parallel form
|
||
|
|
* Programmable output saturation
|
||
|
|
* Anti-windup integrator reset
|
||
|
|
* Programmable low-pass derivative filter
|
||
|
|
* Feedback input to derivative path
|
||
|
|
|
||
|
|
Both PID and PI type controllers in the DCl32 library implement integrator anti-windup
|
||
|
|
reset in a similar way. A clamp is present at the controller output which allows the user to
|
||
|
|
set upper and lower limits on the control effort. If either limit is exceeded, an internal
|
||
|
|
floating-point controller variable changes from logical 1 to logical 0. This variable is
|
||
|
|
converted into Q24 format and multiplied by the integrator input, such that the integrator
|
||
|
|
accumulates successive zero data when the output is saturated, avoiding the “wind-up”
|
||
|
|
phenomenon.
|
||
|
|
|
||
|
|
The following equations describe the implementation of the PID32 controller. Note that
|
||
|
|
the storage of static variables {i14 i10 d2 d3} is not shown.
|
||
|
|
|
||
|
|
The servo error equation is
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
The proportional path equation is
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
The integral path equations are
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
The derivative path equations are
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
Note that the derivative coefficient c1 must be divided by two on initialization. This
|
||
|
|
element is typically much larger than c2 so we enter half its value in the code and multiply
|
||
|
|
twice. This allows greater numerical range for a given Q-format.
|
||
|
|
|
||
|
|
The output path equations are
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
The DCL_PID32 implementation is shown below:
|
||
|
|
|
||
|
|
*Figure: DCL_PID32 A1 architecture*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
The linear PID_A1 controller uses a C structure to hold coefficients and data, defined in
|
||
|
|
the header file DCLC28.h. The order of these structure elements must not be changed
|
||
|
|
by the user.
|
||
|
|
`,
|
||
|
|
//*************************************************************************/
|
||
|
|
// PI32
|
||
|
|
//*************************************************************************/
|
||
|
|
"PI32":
|
||
|
|
`# ***Fixed-Point PI Controllers***
|
||
|
|
The DCL contains one implementation of a fixed-point series form PI controller. The PI is
|
||
|
|
similar in operation to the PID controller, with the removal of the derivative path.
|
||
|
|
|
||
|
|
## ***Implementation***
|
||
|
|
|
||
|
|
The following equations describe the implementation of the PI32 controller.
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
The ideal form PI implementation is shown below:
|
||
|
|
|
||
|
|
*Figure: DCL_PI A1 architecture*
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
The linear PI controller uses a C structure to hold coefficients and data, defined in the
|
||
|
|
header file DCLC28.h. The order of these structure elements must not be changed by
|
||
|
|
the user.
|
||
|
|
`,
|
||
|
|
//*************************************************************************/
|
||
|
|
// PI2
|
||
|
|
//*************************************************************************/
|
||
|
|
"PI2":
|
||
|
|
`# ***Double Integrator PI Controller***
|
||
|
|
The DCL contains one implementation of a linear PI controller having two series
|
||
|
|
integrators. This type of controller is similar to the parallel PI controller
|
||
|
|
except that the anti-windup reset logic is more complicated.
|
||
|
|
|
||
|
|
In this controller, allowance has been made for the Kp element to be zero. This scenario
|
||
|
|
presents a problem if the controller enters saturation because without the proportional
|
||
|
|
path there is no way to recover. The PI2 resolves this by releasing the anti-windup lock
|
||
|
|
when the integrator input reverses sign. The logic has to be implemented twice, since
|
||
|
|
there are two cascaded integrators. Similar anti-windup reset logic is present in the
|
||
|
|
PI_C5 controller.
|
||
|
|
|
||
|
|
## ***Implementation***
|
||
|
|
The double integrator PI2 controller uses a C structure to hold coefficients and data,
|
||
|
|
defined in the header file DCLF32.h.
|
||
|
|
|
||
|
|
The PI2 implementation is shown below:
|
||
|
|
|
||
|
|
*Figure: DCL_PI2 C1 architecture*
|
||
|
|
|
||
|
|
|
||
|
|
`,
|
||
|
|
}
|
||
|
|
|
||
|
|
function get_longDescriptionObj()
|
||
|
|
{
|
||
|
|
return longDescriptionObj;
|
||
|
|
}
|
||
|
|
|
||
|
|
exports = {
|
||
|
|
get_longDescriptionObj:get_longDescriptionObj,
|
||
|
|
}
|