You are on page 1of 4

2010 IEEE 26-th Convention of Electrical and Electronics Engineers in Israel

Non Linear PID and its application in Process


Control
Guy Zaidner, Sammy Korotkin, Eli Shteimberg, Amir Ellenbogen, Meir Arad and Yosef Cohen
Nuclear Research Center - Negev, P.O.B 9001, Beer - Sheva, Israel

Abstract –In the last four decades, the proportional, integral mathematical model is not easy; therefore using a Model
and derivative (PID) controller remains by far the most widely Based Control (MBC) is almost not possible.
used in industry and continues to be an important method in
control engineering, along with high control performance
Commonly, when applying in industry, the proportional,
requirements, advanced control techniques are needed. Since 90 integral and derivative gains are usually scheduled according
percent of applied control technique among industry is related to to the process value (PV) and the magnitude of the error in
PID, any improvement of this technique may result a big impact order to overcome the nonlinearity of the plant. This method
on all related industrial processes [1].
Furthermore, it is well known that most of the industrial
called gain-scheduling.
processes are nonlinear and inherit long time delay i.e. thermal, The disadvantage of this method is that the finding of the
pressure, P.H and flow. Since the conventional PID is a linear optimal set of gains based on many empirical tests which are
controller it is efficient only for a limited operating range when time consuming. Moreover, the conventional PID has a
applying in nonlinear processes. During the last two decades, a
nonlinear PID forms has been developed. This paper reviews one limited robustness to uncertainties in the model parameters i.e.
of the nonlinear PID methods, suggests an optional tuning rules gain and time delay.
accompanied with simulation results and the implementation in During the past two decades several types of enhanced PID
Programmable Logic Controller (PLC).
controllers have been introduced for industrial control. One is
Index Terms–PID, NPID, Gain-Scheduling, PLC. the nonlinear PID (NPID) as introduced by HAN (1994) [2].
The main idea was to create continuous dynamic nonlinear
function instead of gain-scheduling by creating a nonlinear
I. INTRODUCTION 
gain function with combination of , ∫  and ̇ to achieve a


better tracking and better noise rejection. It is achieved by


The conventional PID is the most common control scheme creating a small linear area in the nonlinear function when  is
used in industry and considered as simple to implement and near zero.
affective in broad range of linear process. The conventional This paper describes one of the main algorithms for NPID
PID has two types of representation: controller, its application in industrial control [3] and basic
tuning methods of NPID accompanied by experimental
 
 =   + ∫  + ̇  (1) results. Furthermore, this paper describes the implementation
 
of NPID algorithm in Programmable Logic Controller (PLC)

 =   +  ∫  +  ̇ (2) which is commonly used in industry.


Where  represent the error between the desired output and II. MATHEMATICAL APPROACH
the process value. The integral of the error functions as
accumulator to achieve better tracking. The derivative of the As previously discussed, the main algorithm of the NPID
controller is based on a nonlinear function as inherent part of
error reveals the rate of change in the error helping to prevent
the controller. The main goal is to achieve a desired response
overshoot. Nevertheless, the differential control is carefully in the output of the plant when conventional PID could not
used in order to prevent noise amplification by the derivative achieve it. Therefore, the PID has been reconstructed using a
part. However, since most of the industrial processes are nonlinear function as follows:
nonlinear, and inherit long time delay the conventional PID is
efficient only for a limited operating range. Processes such as  =   , ∝ ,   +  ( , ∝ ,  ) +  ( , ∝ ,  ) (3)
thermal, pressure P.H and flow has nonlinear characteristic
and long time delay making it difficult to control based on
classic methods. For such processes, achieving the

000574
978-1-4244-8682-3/10/$26.00 ©2010 IEEE
Where (, !, ) is the nonlinear function: by the parameter . This parameter defines the linear area in
the nonlinear function where the controller acts like a
||# ∙ %&'*(), -ℎ* || >  conventional PID. A proper selection of ∝ will also affect the
(, !, ) = " (4)
 #/0 ∙ , -ℎ* || ≤  behavior of the controller. The parameter ∝ needs to be
chosen under consideration of the desired response of the
 ,  ,  are the controller gains and they having the same proportional, integral and derivative parts according to
meaning as the PID gains. The error expressions are: disadvantage of the conventional PID.

 =  TABLE I

 = ∫  (5) NPID PARAMETERS EFFECT

 = ̇ Parameter Value Effect
∝ <1 Smaller gain as error is large,
Obviously, this controller has much more degrees of freedom not sensitive to small error.
(DOF) making it much more designable but still more
complex to tune. >1 Higher gain when error is large,
higher gain when error is small
and by that more sensitive to
small changes.

∝ −1 <∝ < 0 Solving integral windup problem


by reducing the integral action
when error is large.

∝ ∝ > 1 Make the differential gain small


when the error is small which
results in less sensitivity to
noise.

The main considerations that should be taken:


Fig. 1- Illustration of NPID function [4] In order to achieve more sensitivity to small errors of the
proportional term ∝ should be larger than 1. In order to
overcome the integral windup problem ∝ should be −1 <
The method described below is the general form of NPID ∝ < 0. This will reduce the integral action when the error is
scheme. Yet, simpler forms can be achieved by selecting some large, better performance could be achieved even if the model
parameters properly and by that decreasing the DOF making it includes dead time. In order to overcome noise perturbation
easier to tune. over steady state ∝ should be larger than one. It will make
the differential gain small when the error is small. Further
rules of how might each parameter affect the output are shown
in Table I.

III. SIMULATION RESULTS

A. PID and NPID Controllers tuning


The NPID controller has been implemented in PLC and tested
on the following 2nd order system model:

̇ 0 1 0 0
Fig. 2- Basic NPID implementation 6 08 = 6 8 A B + A /D B  (6)
̇ 7 −1/1432 −0.127 7 6
It can be shown that a selection of ∝= 1 and  = 0 will
lead to the conventional PID form. Selection of different ∝ of Its performance compared to the performance of the
each gain, will lead to different behavior of the controller. The conventional PID controller. First an optimal tuning of PID
has been made based on simplex search optimization and
∝ parameters are conventionally the error weighting. Since the
according to step response. Then, the parameters of the NPID
motivation for designing such a controller is the demand for a
controller have been set by trial and error.
better performance in transient period and over steady state,
the design of the parameters need to be focused on the desired
response of the controller when error is around zero and when
the error is larger than normal error. This behavior is achieved

000575
TABLE II In order to overcome the dispersion of PID response, new
PID AND NPID TUNED PARAMETERS
tuning has to be done. The problem is that the new tuning does
Kp Ki Ki ∝ ∝ ∝ E not longer feet the non-delayed model.

19 0.01515 0.1676 1.1 1.3 1.6 0

Where  ,  ,  are the PID and NPID gains. The


∝ , ∝ , ∝ and E are the NPID controller parameters. Notice
that there was a difference between the recommended ∝ and
the calculated one. The reason is that the basic form of NPID
with  = 0 has been used. The basic form seems to be
sufficient enough. As shown in Fig.3 the NPID has lower
overshoot and shorter settling time, without significant
increasing of rise time. Fig. 5- System with delay

D. Robustness to uncertainties
Since the model is not always accurate and there are
possibilities of model parameters changes during the system
lifetime, robustness to uncertainties in model parameters has
been tested.

Fig. 3- Step response tuning

B. Desired trajectory
The controllers have been tested for the desired trajectory.

Fig. 6- Uncertainty of 25% in the system gain

The NPID controller shows much more robustness to


uncertainties and almost has no effect on its performance over
25% changes in parameters. The performance of the PID
controller has been dramatically deteriorated (Fig.6).

Fig. 4- Desired trajectory response

The simulation shows that the NPID controller tracks the


reference signal much better than the optimal tuned PID
controller (Fig.4).
C. Robustness with Dead time
In order to represent the real plant, a various delay time has Fig. 7- Controllers manipulated variable
been added (Fig.5). The results show that the NPID controller
has much more robustness to delay. The tolerable range of In general, achieving a better performance could lead to non-
time delay in PID is 0-4 seconds compared to 0-10 second physically controller output therefore in this simulation the
range in the NPID. Beyond six seconds delay the PID appropriate constraints has been considered. The simulation
response has been dispersed. shows (Fig. 7) that there are no dramatic changes in the
controller output by using NPID.

000576
E. Stability and control of noisy model REFERENCES
In order to compare the stability of the NPID controller and its [1] B.G. HU, "NPID-PCA: A Simulation Toolkit of Nonlinear PID
noise rejection a white Gaussian noise with variance of eight control on Scilab/Scicos(c) INRIA-ENPC", 2004.
has been added to the system model. The simulation shows [2] J. Han, "From PID to Active Disturbance Rejection Control",
that NPID is much stable with the presence of noise such as IEEE Transaction on Industrial Electronics, Vol.56, no.3 march
PID. The simulation shows almost no improvements when 2009 (translation from Chinese).
[3] D.B Ender, "Special nonlinear PID Controllers", Application
using the NPID compared to the PID (Fig.8). The reason for Manual, ch.6, 2001.
that might be that the model in used as natural LPF and thus [4] F. Jiang, Z. Gao, "An Application of Nonlinear PID Control to a
filters the high frequencies noises. Class of Truck ABS Problem", Proceedings of the 40th IEEE
Conference on Decision and Control, On page(s): 516 - 521
vol.1 2001.
[5] J. Han and W. Wang, Nonlinear tracking-differentiator, Syst.
Sci. Math.
[6] vol. 14, no. 2, pp. 177–183, 1994, (in Chinese)Y. Huang and J.
Han, A new synthesis method for uncertain systems- the self
stable region approach, Int. J. Systems Sci., 1999, 30 (1):33-39.

Fig. 8- Controlling noisy model

IV. CONCLUSIONS

The proposed method of the NPID controller suggests an


alternative nonlinear implementation of PID which is also
based on error instead of model based control. The NPID
controller is less depended on the accuracy of the estimated
mathematical model, what makes it more robust to
uncertainties and delay. This method was discovered to be
easy to implement in PLC with relation to other techniques i.e.
gain-scheduling, due to its simple mathematics.
Basic rules for the controller tuning have been shown
where the optimal NPID's parameters were easily found by
trial and error.
The simulations show that the control signal (controller
output signal) has not dramatically changed and does not have
non-physically demands. Finally, it can be said that by proper
tuning and prior simulations NPID controller seems to be a
good alternative to PID (when it is needed) and can have
impact on many of industrial controlled processes.

V. FURTHER WORK

Still, further work needs to be done in order to increase


noise rejection. According to simulation results, the current
NPID controller seems to be as noise sensitive as PID. To
achieve better noise rejection, the combination of NPID with
TD (Tracking Differentiator) called ADRC [5] is suggested as
future work. Stability analysis is also needed to be done as can
be found in [6].

000577

You might also like