Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

A Design Method of Bang-Bang and PID Integrated Controller

Based on Rough Set

Shaomian Chen, Jun Zhao, Jixin Qian


Institute of Industrial Control, Zhejiang University, Hangzhou, China
smchen@iipc.zju.edu.cn

Abstract

An efficient design method to improve the control perfor-


mance of rough controller which integrates the Bang-Bang f

control rules with PID control rules is presented in this pa-


per. The two parts of rules extracted from Bang-Bang con-
troller and PID controller by rough set, respectively, are
combined to a uniform decision making system under a cer- f
tain selection mechanism. This rough controller reserved
all the advances of Bang-Bang control and PID control en-
ables to deal with different magnitudes of set point variation
and large disturbance. Simulations show the effectiveness
of this method.

Figure 1. Process of Bang-Bang control


1. Introduction

Rough set theory [5, 9, 10] provides a methodology for is used to construct Bang-Bang and PID knowledge bases
generating decision tables defined as sets of decision rules comprised of corresponding control strategies emulating
coupled with rule selection mechanisms from abundant pro- separately from Bang-Bang controller and PID controller.
cess data and has been applied to industry control process And the inference engine is used to combine both bases
successfully [3, 6, 7, 11, 12]. These works suggest the idea to be a uniform decision-making controller, which not only
of an automatic system based on rough set that is capable of guarantees fast dynamic response but also ensures the con-
emulating the existing controllers or the human operators to trol accuracy.
control complex plants. Such a processor of decision tables
is referred as a ”rough controller” [8]. 2. Bang-Bang control and PID control
In industry process, the operators or experts prefer
to manually settle the tracking control problem for load Bang-Bang control, which can be assumed as the plant
change and the rejection of large disturbance, which are operator’s behavior, is one of the most popular methods of
based on their experience or knowledge of the processes; time optimal control [2]. It aims at the shortest system tran-
and the automatic controller such as PID controller is used sition time when set point changes from a value to another.
in normal operation to eliminate the small error deviated Suppose that the set point changed from r0 to rf , the corre-
from set point. Bang-Bang control is a special type of con- sponding response process of Bang-Bang control is shown
trol method where control move is only allowed to be dis- in Figure 1. Where, the top chart is the variational process
crete value, which is very similar to operators’ control ac- of the manipulated value u changed from u0 to uf as the
tions. Therefore it is natural to set up a rough set controller form of u0 → K → k → uf , the bottom chart is the vari-
combined with Bang-Bang and PID control rules as a refer- ational process of the controlled value y changed from r0
ence to simulate the whole industry process. to rf . It can be seen that the time t0 ∼ t1 and t2 ∼ t3
A design approach to Bang-Bang and PID integrated are caused by the time delay τ of the process. t2 and t4 are
rough controller is proposed in this paper. The rough set the time when the control action happens to change direc-

Fourth International Conference on Fuzzy Systems and Knowledge Discovery (FSKD 2007)
0-7695-2874-0/07 $25.00 © 2007
Bang-Bang
Input Knowledge Base
Signals ¨r Output
¨r
Rough Rough Signals
Data e e Inference ¨u u
A/D D/A
Preprocess ¨e ¨e Engine
Converter Converter

PID
Knowledge Base

Feedback Signals
Process
Disturbance Signals

Figure 2. Integrated rough control system

tion, which are often called initial switch time and terminal 3. The integrated rough controller
switch time, respectively. t5 is the time when the process
achieves the new steady value rf . 3.1. Rough controller synthesis design
Proportional-integral-derivative (PID) control is cer-
tainly the most widely used control strategy today. It is es-
The structure of Bang-Bang and PID integrated rough
timated that over 90% of control loops employ PID control;
control system is shown in Figure 2. It consists of the fol-
most loops are actually PI control [1]. The PID algorithm
lowing subsystems:
usually can be described by:
 • Rough A/D Converter: performing the quantization
 1 t de(t)  and classification of the input signals.
u(t) = K e(t) + e(τ )dτ + Td
Ti 0 dt • Knowledge Base: control rules storage, including a
Bang-Bang knowledge base comprised of control rules
Where y is the measured process variable, ysp is the refer-
emulating a Bang-Bang controller and a PID knowl-
ence variable or called set point, u is the control action and
edge base comprised of control rules emulating a PID
e is the control error (e = ysp − y). The control action is
controller.
thus a linear combination of three terms: the P-term (which
• Inference Engine: decision-making center, which se-
is proportional to the error), the I-term (which is propor-
lects rules whose conditions are satisfied from the
tional to the integral of the error), and the D-term (which is
knowledge bases and performs actions specified by
proportional to the derivative of the error). The controller
their decisions.
parameters are proportional gain K, integral time Ti , and
• Rough D/A Converter: obtains decision attributes from
derivative time Td .
the inference engine and converts them to output sig-
In process control, Bang-Bang control holds more emi-
nals.
nent dynamic performance than PID control, especially in
Knowledge bases and Inference engine are the central
hurdling the change of set point value and the influence of
part that is in charge of storing and exporting the rules de-
large disturbance. But it is inclined to be vibratory in static
rived from rough set.
state. When the deviation enters into small scope, however,
the control quality is not better than PID control. By con-
traries, the situation of PID control is quite different. It’s 3.2. Rough knowledge base
usually not easy for PID control to guarantee the control
performance in the case of large deviation or large distur- The control strategies of Bang-Bang controller and PID
bance. But it has better static performance and more power- controller for a specific process can be respectively emu-
ful capability of eliminating the steady-state error when it’s lated by two rough knowledge bases comprised of control
in the scope of small deviation. As a result, it’s necessary to rules, which are constructed by the design method of rough
combine both advantages of above-mentioned control meth- controller [14]. The approach to extract rules including:
ods. The idea of Bang-Bang and PID integrated control is 1. Select condition attributes and decision attributes ac-
generating under this background, which means selecting cording to working principle or practical relationship,
Bang-Bang control in large deviation and delivering to PID and discretize these attributes by manual discretization
control for eliminating the steady-state error. method.

Fourth International Conference on Fuzzy Systems and Knowledge Discovery (FSKD 2007)
0-7695-2874-0/07 $25.00 © 2007
Table 1. Decision table of Bang-Bang control

Rule Conditions Decisions


Number e ∆r ∆e t ∆u flag
1 5 5 5 0 8.175
2 5 5 -0.01 0.1 0
3 4.5 5 -0.023 0.4 0
e k <= a
4 4.0 5 -0.04 0.8 0
... ... ... ... ... ...
272 0 -5 0 100 0

2. Take extra testing signals to excite system for obtain-


u t
ing sufficient data.
3. Check and remove redundant data. t T
4. Check contradictory data and process them with
weighted average method.
5. Generate minimal decision rules from data. u
flag = flag =
6. Use newly batches of data to update and enrich deci-
sion rules until the knowledge base is adequate.
In rough set, rules are organized as decision table, which
specifies what decisions (actions) should be undertaken
when certain conditions are satisfied. Take that emulat-
ing Bang-Bang controller as example. Because the control
move increment ∆u and control sustaining time t applied Figure 3. Flow chart of inference engine
by the rough controller are the functions of control error e,
error increment ∆e and set point increment ∆r [4], it’s nat-
ural to choose them as the attributes for decision table. The first Bang-Bang control action to now, T is the reference
decision rules have the form shown in Table 1. time, which is sum of the terminal switch time and the tran-
Notice that, Bang-Bang control is a type of time opti- sition time which is proportional to the main time constant
mal control, which is comprised of three successive discrete of the process. The constraint t <= T is used to ensure the
control actions that are calculated by the model of the con- unbroken execution of Bang-Bang control strategies. The
trolled process. The control actions have to be performed in working mechanism of inference engine is shown in Fig-
the predefined time. So, it is necessary to attach a time at- ure 3.
tribute to decision table, which also means that a time logic The inference engine can combine two part rules through
module for inference engine is needed to guarantee the un- the following operations:
broken Bang-Bang control strategies. 1. If ∀ x1 ∈ U1 , V[∆r,e,∆e] = V(x1 ,[e,∆e]) , and
∃ x2 ∈ U2 , V[∆r,e,∆e] = V(x2 ,[∆r,e,∆e]) ; then Vu =
3.3. Inference engine V(x2 ,[∆r,e,∆e]) .
2. If ∀ x2 ∈ U2 , V[∆r,e,∆e] = V(x2 ,[∆r,e,∆e]) , and
The inference engine is the core of the controller. Ac- ∃ x1 ∈ U1 , V[∆r,e,∆e] = V(x1 ,[∆r,e,∆e]) ; then Vu =
cording to the principle of Bang-Bang and PID integrated V(x1 ,[∆r,e,∆e]) .
control, the controller should calculate the absolute value of 3. If ∀ x1 ∈ U1 , V[∆r,e,∆e] = V(x1 ,[∆r,e,∆e]) , and ∃ x2 ∈
deviation e primarily, and then deliver it to the inference en- U2 , V[∆r,e,∆e] = V(x2 ,[∆r,e,∆e]) ; if ∃ V(x1 ,t) <= T ,
gine for a proper response. It works as follows: if the devia- then Vu = V(x1 ,[∆r,e,∆e]) , else Vu = V(x2 ,[∆r,e,∆e]) .
tion e is smaller than the reference value a and the controller Where, suppose that U1 represents the rule set in the Bang-
is not in the interim of Bang-Bang control, the inference Bang knowledge base, and x1 is a rule of the set as x1 ∈ U1 ,
engine will just select the PID rough control strategies, or U2 represents the rule set in PID knowledge base, and x2 is
else it will execute the complete Bang-Bang rough control a rule of the set as x2 ∈ U2 . V(x,[∆r,e,∆e]) is a condition
strategies firstly, which can make the response of process attribute value of control rule, and V(x,[t,u]) is a decision
close to the set point, and then apply the PID rough control attribute value of Bang-Bang control rule, V(x2 ,u) is a deci-
strategies to eliminating the steady-state error. Where, t is sion attribute value of PID control rule,. V[∆r,e,∆e] is any
the control sustaining time from the beginning time of the input value, and Vu is the output value.

Fourth International Conference on Fuzzy Systems and Knowledge Discovery (FSKD 2007)
0-7695-2874-0/07 $25.00 © 2007
4. Simulations provides an idea for designing a control system that is ca-
pable of emulating complex process controlled by human
Using the following process as example: operator and automatic controller.

1
6. Acknowledgement
50s2 + 15s + 1
The control error e, error increment ∆e and set point in- We gratefully acknowledge the financial support of Na-
crement ∆r are chosen as condition attributes and control tional Natural Science Foundation of China under Grant
move increment ∆u is chosen as a decision attribute while (No.60504004) and the National Key Basic Research and
the control sustaining time t of Bang-Bang control is added Development Program (No.2002CB312200).
as another accessorial decision attribute in Bang-Bang con-
trol. Variables ∆r, e, ∆e, ∆u and t are artificially classified
References
as 21, 41, 41, 61 and 61 sections, respectively. Pseudo-
random signal is used as the testing signal in PID control
[1] K. J. Astrom and T. Hagglund. PID Controllers: Theory,
with the boundary [-0.4, 0.4], and step signal evenly dis- Design, and Tuning. Instrument Society of America, Re-
tributed is used in Bang-Bang control with the boundary search Triangle Park, NC, 1995.
[-5, 5]. [2] D. E. Kirk. Optimal Control Theory. Prentice Hall, 1970.
Data of different batches are used to extract rules. For ex- [3] G. Lamber-Torres. Application of rough sets in power sys-
ample, in the example of emulating PID controller, total 16 tem control center data mining. Power Engineering Society
tests are performed successively. The numbers of the rules Winter Meeting, 1:627–631, 2002.
[4] Y. M. Li and W. Q. Weng. Nonlinear Control System. Chem-
after being updated and enriched are shown in Table 2. It
ical Industrial Press, 1985. (In Chinese).
can be seen that the rules are not sufficient only for the first [5] A. Mrózek. Rough sets and dependency analysis among at-
several tests, and the rule number increases with successive tributes in computer implementations of expert’s inference
tests. The test can be terminated when the rule numbers do models. International Journal of Man-Machine Studies,
not change. In this way, the final rule number for the PID 30:457–473, 1989.
knowledge base is 115, and for Bang-Bang knowledge base [6] A. Mrózek. Rough Sets in Computer Implementation of
is 272. Both the rough knowledge bases are combined to- Rule-based Control of Industrial Process, Intelligent Deci-
gether by the inference engine to form the new Bang-Bang sion Support: Handbook of Applications and Advances of
and PID integrated rough controller. the rough sets Theory. Kluwer Academic Publishers, The
Neatherlands, 1992.
The Bang-Bang and PID integrated rough controller is
[7] A. Mrózek and L. Planka. Rough Sets in Industrial Applica-
expected to response to any input or disturbance distributed tions, Rough Sets in Knowledge Discovery 2: Applications,
in [-5, 5]. The responses of the integrated rough con- Case Studies and Software Systems. Physica-Verlag, Heidel-
troller and the PID controller when they are tracking step berg, 1998.
input signal from 0 to 3.1 are shown in Figure 4. The [8] A. Mrózek, L. Planka and J. Kedziera. The methodology
responses to step disturbance from 0 to 4.6 are shown in of rough controller synthesis. Proceeding of the Fifth IEEE
Figure 5. In these simulations, the parameters of PID are International Conference on Fuzzy Systems, 1996.
tuned by internal model control (IMC) method [13], where [9] Z. Pawlak. Rough sets. International Journal of Information
and Computer Science, 11:341–356, 1982.
gain Kc = 1.67, integral time Ti = 15, differential time
[10] Z. Pawlak. Rough Sets: Theoretical Aspects of Reasoning
Td = 3.33, filter parameter λ = 9. It can be seen that, com- about Data. Kluwer Academic Publishers, Dordrecht, The
pared with PID controller, the rough controller can achieve Neatherlands, 1991.
faster dynamic response with almost the same steady time. [11] J. F. Peters, A. Skowron and Z. Suraj. An application of
It uses Bang-Bang control strategies to track large set point rough set methods in control design. Fundamental Informa-
variation or reject large disturbance at first, and then switch tion, 43:269–290, 2000.
to PID control strategies for eliminating small deviation at [12] L. Planka and A. Mrózek. Rule-based stabilization of the in-
the time of T . verted pendulum. Computational Intelligence, 11:348–356,
1995.
[13] D. E. Rivera, M. Morari and S. Skogestad. Internal model
5. Conclusions control. 4. pid controller design. Industrial and Engineering
Chemistry Research, 25:252–265, 1986.
[14] G. Wang, J. Zhao and J. X. Qian. A successive design
The paper presents a design method for rough controller
method of rough controller using extra excitation. Pro-
synthesis regarding hybrid system contained Bang-Bang
ceedings of the Second International Conference on Fuzzy
control (discrete component) and PID control (continuous Systems and Knowledge Discovery (FSKD’05), 3613:1266–
component). Two parts of independent rules are combined 1270, 2005.
to be a uniform system by the inference engine. The method

Fourth International Conference on Fuzzy Systems and Knowledge Discovery (FSKD 2007)
0-7695-2874-0/07 $25.00 © 2007
Table 2. Decision table of Bang-Bang control

Test 1 2 3 4 5 6 7 8
Rule Number 73 97 106 110 110 112 112 112
Test 9 10 11 12 13 14 15 16
Rule Number 13 113 114 115 115 115 115 115

3.5 5.5
Integrated Bang−Bang and PID
Rough Controller
3 PID Controller
5

2.5
4.5

2
Output

Input
4
1.5

3.5
1

3
0.5 Integrated Bang−Bang and PID
Rough Controller
PID Controller
0 2.5
0 20 40 60 80 100 0 20 40 60 80 100
t(s) t(s)

Figure 4. Comparison between PID controller and rough controller with step input from 0 to 3.1
(Left: system output, Right: control input)

5 −4
Bang−Bang and PID
4.5 Integrated Rough Controller
PID Controller −4.5
4
−5
3.5

3 −5.5

2.5
Output

Input

−6
2

1.5 −6.5

1
−7
0.5
Bang−Bang and PID
−7.5
0 Integrated Rough Controller
PID Controller
−0.5 −8
0 20 40 60 80 100 0 20 40 60 80 100
t(s) t(s)

Figure 5. Comparison between PID controller and rough controller with step disturbance from 0 to 4.6
(Left: system output, Right: control input)

Fourth International Conference on Fuzzy Systems and Knowledge Discovery (FSKD 2007)
0-7695-2874-0/07 $25.00 © 2007

You might also like