INTRODUCTION
DC motors have long been the primary means of electrical traction. DC motor
has at torque/speed characteristics compatible with most mechanical loads. The
speed control methods of a dc motor are simpler and less expensive than those
of A.C Motors and speed control over a large range both below and above rated
speed can be easily achieved (Yu and Hwang, 2004).
In a typical electric drive controller, there are usually several nested control
loops for the control of current/torque, speed and position each of which may
use a separate Proportional Integral Derivative (PID) controller. Although many
alternative control structures have been proposed for this application, the
PI controller continues to be the most popular controllers used in industrial
processes (Visioli, 1999).
The most advantage of this kind of controller is its simplicity to implement.
In spite of the fact that the effect acquired as a result of disturbances and
environmental conditions on the structure of the system (Montiel
et al., 2007) adding complexity to the controller’s design,
it is not easy to find another controller with such a simple structure to be
comparable in performance.
A very important step in the use of controllers is the controller parameters
tuning process. In a PID controller, each mode (proportional, integral and derivative
mode) has a gain to be tuned giving as a result three variables involved in
the tuning process. There have been a lot of approaches to search the parameters
of PID controllers including time response tuning (Hang et
al., 1991), time domain optimization (Zhuang and
Atherton, 1994), frequency domain shaping (Voda and
Landaq, 1995) and genetic algorithms (ChunLiang et
al., 2003).
The speed response of the drive with PID controllers designed with the above techniques may be satisfactory but not necessarily be the best, since they do not pose any constraint on settling time, overshoot/undershoot etc.
In any classical PID control problem, the required controller parameters should be optimally designed. Despite the method of ZeiglerNichols (ZN) ultimate cycle tuning scheme, these parameters can be optimally obtained via Particle Swarm Optimization (PSO) and Genetic Algorithm (GA).
The optimal design of the PID controller is a complex and challenging problem
since it often involves various conflicting objectives and goals. So the selection
of an optimal solution can be stated as a nonlinear programming problem. This
problem may be solved using suitable numerical optimization technique (Wang,
2004; Cupertino et al., 2002).
Many real world control problems track several objectives simultaneously. The
objectives under consideration conflict with each other and optimizing a particular
solution with respect to single objective will result in unacceptable results
with respect to other objective. Competing goals of real world problems gives
rise to a set of compromise solutions called as pareto ptimal solutions. A reasonable
solution to a multiobjective problem is to investigate a set of solutions each
of which satisfies all the objectives at an acceptable level without being dominated
by any other solution. Thus the solution to any Multiobjective problem is a
family of points known as nondominated solutions or pareto optimal points and
the curve that joins all pareto optimal points are called Pareto optimal front.
If all the objective functions of a solution cannot be improved simultaneously,
then that solution is said to have a non dominated character.
Classical Proportional and Integral (PI) or Proportional and Derivative (PD)
is used not only for their simplicities but also due to its success in a large
number of industrial applications. These controllers are tuned based on trialerror
approaches, there for have large frequency deviations. A number of state feedback
controllers based on linear optimal control theory have been proposed to achieve
better performance (Astrom and Wittenmark, 1996).
In this study multiobjective NonDominated Sorting Genetic Algorithm (NSGAII) is usedinter for tuning of nonlinear PID controller parameters for speed control of dc motor drives. There is no constraint in the searching space of the optimal PID parameters. The new PID tuning algorithm is applied to the speed control of DC motors. The performance measure to be minimized contains the objectives of the PID controller.
Unlike classical methods such as ZieglerNichols and CohenCoon (Montiel
et al., 2007) and single objective optimization methods such as binary
coded GA, multiobjective optimization can minimize some important aspect of
a system such as overshoot/undershoot and settling time simultaneously so that
various solutions with different overshoot/undershoot and settling time are
obtained. From these different PID Parameters, one can select a single solution
based on system constraints, reliability and etc. For example in such cases
overshoot/undershoot has more importance than setting time and vice versa.
DC MOTOR MODELING
The motor torque, T is related to the armature current, i by a constant factor K_{t}: T = K_{t}I_{a}. For the separately excited DC motor, the back emf, e is related to the rotational velocity by: e = K_{b}u_{T}, In SI units K_{t} (armature constant) is equal to K_{b} (motor constant) (Fig. 1).
V = input voltage (V), R = nominal resistance (Ω), L = nominal inductance
(H), J = Inertial load (kg*m^2/s^2), V_{emf} = back emf voltage (V),
b = damping constant (Nm.S), τ = motor output torque (Nm), θ = motor
shaft angle (rad).
 Fig. 1: 
Electromechanical mode of a DC motor 

The DC motor equations based on Newton's law combined with Kirchhoff's law:
Where:
J 
= 
The moment of inertia 
B 
= 
Damping ratio of the mechanical system 
R_{a} 
= 
The electrical resistance of the armature circuit 
L_{a} 
= 
The electrical inductance of the armature circuit 

In the statespace form, the equations above can be expressed by choosing the rotational speed and electric current as the state variables and the voltage as an input. The output is chosen to be the rotational speed.
MULTIOBJECTIVE OPTIMIZATION OF PID< BASED ON NSGAII ALGORITHM
Multiobjective genetic algorithm belongs to an evolutionary algorithm for solving multiobjective optimization problem. Its core is to coordinate the relationship between the objective functions to find the optimal solution forcing them to approach to maximum or minimum.
NSGA is a new type of multiobjective genetic algorithm. High efficiency of
NSGA algorithm lies in using a nondominant classification procedure tosimplify
the multiobjective into a way of fitness function by which the method can solve
any number of objectives and to seek its maximum and minimum. In Deb
et al. (2002) presented improved NSGA, i.e., NSGAII, a quick noninferiority
sorting method, the key techniques of NSGAII algorithm as the following:
Fast nondominated sorting: Before selecting computation, classify all individuals of noninferior solution in current population into Level 1.
Then remove the individuals out from the population and find a new noninferior solution in the rest of individuals. Set it into Level 2. Repeat above process until all individuals in the population have been set into corresponding levels.
Virtual fitness: In order to maintain the diversity of individuals to prevent local accumulation, NSGAII algorithm first proposed the concept of virtual fitness. It means the local crowding distance between every point and another adjacent to it in the same objective space. For example, the crowding distance of point i in objective space is equal to the sum of two side lengths in a rectangular composed of adjacent points i1 and i+1 as shown in Fig. 2. This can be adjusted so that calculation results in the objective space are spread more evenly and with better robustness.
The selection for calculations: Let the selection process towards Pareto optimal solution direction and spread the solutions evenly. After sorting and crowding distance calculation, each individual in same group gets two attributes: nondomination order i_{rank} and crowding distance i_{d}. When i_{rank}<j_{rank} or i_{rank} = j_{rank} and _{id}>j_{d}, it refers to that individual i is superior to individual j. It means that if the nondominated orders of two individuals being different, select the individual with lower order; if two individuals lie in the same level, select the individual with less crowded around it.
The elite strategy: The elite strategy in the study implies that superior individual in parent generation will be retained and transfer into children generation directly. Combine all individuals from to parent generation Pt and children generation Q_{t} into as a population R_{t} = P_{t} Q_{t}, with individual number 2N; Make the population R_{t} rapid nondominated sorting and calculate the local crowded distance, select the individuals in turn according to their orders from high to low, until the individual number equals to N, form a new parent population P_{t+1}.
 Fig. 2: 
Crowding distance calculation 

FITNESS FUNCTIONS EMPLOYED IN MULTIOBJECTIVE DESIGN
In the general control problem, the optimization of different number of systems
performances is desired. The following simultaneous performance specifications
(the objectives) are adopted:
• 
Overshoot/Undershoot minimization: 

• 
Settling time minimization: 

• 
Rise time minimization: 

Where OU denotes overshoot, T_{N} denotes total settling time and T_{R} denotes rise time. These objectives are being simultaneously optimized and results are obtained.
SIMULATION AND RESULTS
The specifications of the DC motor are given below: Armature circuit Resistance
(R_{a}) = 7.56 Ω, Armature circuit inductance (L_{a}) =
0.055 H, Moment of inertia (J) = 0.068 Kg m^{2}, Coefficient of friction
(B) = 0.03475 N.m.sec rad^{1}, Torque constant (K_{T}) 3.475V.
Sec rad^{1}, BackEmf constant (K_{b}) = 3.475 V.sec rad^{1}.
 Fig. 3: 
Pareto optimal front with settling time and overshoot as objectives 

 Fig. 4: 
Speed track response based on settling time minimization 

The multiobjective PID controller can coordinate various performance indices
of the system and provide an effective tool for tradeoff analysis among speediness,
stability and robustness.
The decisionmaker can choose the PID parameter needed from the Pareto solutions according to the requirement of actual system.
The multiobjective PID controller using the design method presented has good control quality which provides an effective approach for optimal design of PID controller and can be widely used for practical PID control.
Figure 3 shows the Pareto optimal front with settling time
and overshoot as objective functions. From the Pareto optimal front, three samples
are taken. From these samples, corresponding Kp, Ki and Kd values are obtained
as per the requirements of the user. The following results are shown based on
a selected point from the optimal Pareto front of the controller parameters
(the set of acceptable (tradeoff) optimal solutions) for compromising of different
objective functions: settling time, maximum overshoot, rise time, steady state
error and speed track error.
 Fig. 5: 
Armature voltage based on settling time minimization 

 Fig. 6: 
Armature voltage based on settling time minimization for sine
input 

The selected values of the PID controller are: K_{p} = 29.1; K_{i}
= 8.23; K_{d} = 15.26.
The values of K_{p}, K_{d} and K_{i} are selected from the Pareto front and used in the block diagram shown in reference speed as a first track and the second track is assumed as a sinusoidal track.
The response of the first speed track (step change) is shown in Fig.
46 shows the system response of the second speed track
(sinusoidal Track).
Figure 7 shows the response of the control system for load torque disturbance based on settling time minimization.
 Fig. 7: 
Shows the response of the control system for load torque disturbance 

CONCLUSION
This study has explained application of Multiobjective NSGAII for the optimal PID controller design of an electromechanical DC motor drive. The main objective functions to be minimized are maximum overshoot, rise time and settling time. The optimization solution results are a set of near optimal tradeoff values which are called the Pareto front or optimality surfaces. Pareto front enables the operator to choose the best compromise or near optimal solution that reflects a tradeoff between key objectives. The iterative simulation results show the effectiveness of the multiobjective Non Dominated Sorting Approach NSGAII since it allows the operator to find a near optimal good compromise among the proposed goals which is the best tradeoff low cost PID controller design. The computer simulation results show that an optimized speed response is obtained always with load torque disturbance and change reference speed and demonstrates the excellent performance of PID controller.