CHAPTER 2 LITERATURE REVIEW

Transcription

1 19 CHAPTER 2 LITERATURE REVIEW 2.1 GENERAL This chapter summarizes the literature survey that was conducted as a part of the research reported in this thesis. It covers pertinent established concepts and techniques related to multiobjective cascade control system design for servo and regulatory processes. Simulation and real time implementation results for servo and regulatory process using conventional and intelligent control techniques employed in various literature are analyzed and discussed. Literature survey was conducted on various optimization techniques involved in optimizing the gains of different controllers with single and multiple conflicting objectives. A detailed survey was conducted on various evolutionary multiobjective optimization techniques and the performances of the algorithms in optimizing the controller structure in both simple feedback and cascade modes of operation are discussed. 2.2 LIQUID LEVEL CONTROL SYSTEMS Liquid level loops are commonly encountered in process industries. Since the desired production rates and inventories are achieved through the proper control of flows and levels, level control is quite important for the successful operation of many chemical plants as proposed by Marlin (1995). The industrial importance of level loops has led to extensive research interest to achieve the enhanced control performance of the level loop.

2 Simple Feedback Control A lot of literature focuses on control of liquid level as a simple feedback loop. The research activities done in these areas reflect the interest in improving the operation and control of highly nonlinear process in single loop mode. Literature based on control of liquid level with a single objective function is presented and discussed. Both conventional and intelligent tuning techniques proposed in various literature are discussed in this section Conventional methods Luyben and Buckley (1977) have proposed a Proportional-Lag (PL) control as a potentially good solution to the problem of liquid level control systems with feed forward compensation from the inlet flow. Proportional- Lag control is a potentially good solution for liquid level control systems with feed forward compensation. In PL control, the proportional control based feedback loop provides flow smoothing while the feed forward compensation eliminates the steady state offset in liquid level. The limitation of such feed forward control schemes is that they require an additional measurement which may be unavailable. Cheung and Luyben (1979) have studied the liquid level control system with P-only and PI feedback controllers. They adopted a tuning procedure of the Proportional-Lag (PL) controller with a design chart using specifications for the maximum level deviation and maximum rate of change of the manipulated flow to achieve optimal filtering for inlet flow step changes. It is concluded that PL controller does not offer much of an improvement compared to a standard PI Controller.

3 21 Rivera et al (1986) have proposed the P-only controller using the Internal Model Control (IMC) principle for the critically damped closed loop response of a liquid level control system to determine the PID parameters to ensure a desired closed loop response to the step change in set point. IMC method employs the desired closed loop response as first order lag system with dead time without overshooting, while providing adequate suppression of output disturbances, do a poor job of suppressing the load disturbances when the process dynamics are significantly slower than the desired closed loop dynamics. McDonald et al (1986) have proposed an interesting method of deriving an averaging level control algorithm to minimize the maximum rate of change of the manipulated flow. The continuous time optimal level controller, which minimizes the maximum rate of change of the outlet flow, was derived and is found to achieve good results even in the presence of disturbances. A limitation is that they are hard to tune, and also this method allows excessive volume changes. Wu et al (2001) have proposed a two degree of freedom control structure to address both the regulatory and servo problems in level control. A two degree-of-freedom scheme is proposed for the level control systems. It is shown by the author that the scheme gives satisfactory responses for both the systems without dead time and with small dead time and also suggested that the controllers should be designed according to operational specifications namely maximum rate of change in outflow and maximum peak height for a given inlet flow variation. In spite of its industrial and economic importance, the constrained optimal control strategy has rarely been employed in practical level loops. One of the main reasons is that most industrial level loops use simple P-only or PI controllers, which are generally accepted as being too simple to implement any sophisticated control strategy. The main obstacle to

4 22 achieving optimal control is that it normally requires an optimization package to find the optimal control action, and many practitioners are not familiar with the use of these complicated packages. Limitation is that, for large dead times involved in the process, performance of the system becomes unsatisfactory. Using a sophisticated advanced controller such as a model predictive controller might be a solution for the constraint control of the level loop, but it cannot be considered as a practical approach Intelligent techniques Conventional control approaches are not convenient to solve the complexities. Fuzzy logic and neural networks control have emerged over the years and become one of the most active and fruitful areas of the research in the intelligent control applications. Fuzzy logic, neural networks and genetic algorithms are three popular artificial intelligence techniques that are widely used in many applications. Due to their distinct properties and advantages, they are currently being investigated and integrated to form new models or strategies in the areas of system control. Seng et al (1998) have proposed a neuro-fuzzy controller to a coupled-tank liquid-level laboratory process based on the radial basis function neural network tuned automatically using genetic algorithms (GA). They have used a linear mapping method to encode the GA chromosome, which consists of the width and centre of the membership functions, and also the weights of the controller. Dynamic crossover and mutation probabilistic rates are also applied for faster convergence of the GA evolution. Compared to a manually tuned conventional fuzzy logic controller and a Proportional-Integral- Derivative (PID) controller which are applied to the same process, the proposed controller shows considerable robustness and advantages. Human expert intelligence in framing the rule base of fuzzy logic controller is a major limitation.

5 23 Naman et al (2000) have presented an adaptive model reference fuzzy controller (AMRFC) for controlling the water level in a water tank. Performance is compared with conventional methods of proportional-integral (PI) control and Model Reference Adaptive Control (MRAC). Unlike most of the literature reviewed which use the error and error change as inputs to the fuzzy system, this method uses the theoretical background developed for MRAC in choosing these inputs. Although the controller uses many inference rules (441 rules), it is shown that the required mathematical calculations are not much, making implementation on a low-end microcontroller feasible. The control algorithms are implemented in simulation and real-time on an 8-bit microcontroller. It was found that the MRAC proved to be better compared to the PI controller. Limitation is the similarity in performance due to the linearity of the plant. Han (2006) have proposed an adaptive neural network control strategy based on fuzzy self-tuning to control the drum water level of coalfired power plant. Fuzzy Inference Engine (FIE) is used to train neural network online. The control strategy possesses feed forward compensation ability for steam flow disturbance by introducing the steam flow signal to neural network controller. Robust controller is constructed to guarantee good regulating performance while dynamic behaviour of the controlled plant changes or external steam flow disturbances exists. In contrast to conventional cascade PID control, simulation results show the efficiency and superiority of the proposed strategy. Yazdizadeh et al (2009) have proposed two novel adaptive PIDlike controllers such as Neural network PID and Neural network PID with internal dynamic feedbacks for controlling multivariable, nonlinear Multiple- Input Multiple-Output (MIMO) systems. Comparative analyses of the novel adaptive controllers are tested with conventional methods and results confirm

6 24 that the algorithm exhibits excellent performance. It has been applied to control the water level of tanks in water refinement process, which is highly non linear and good performance and stability was achieved. Limitation is the learning rate and the system disturbances are not taken into consideration for the design. Hasan et al (2011) have investigated and found a solution by designing the intelligent controller such as neural network for controlling the water level system. The controller also can be specifically run under the circumstance of system disturbances. To achieve these objectives, a prototype of water level control system has been built and implementations of both PID and neural network control algorithms are performed. In PID control, Ziegler Nichols tuning method is used to control the system. In neural network control, the approach of Model Reference Adaptive Neural Network (MRANN) control based on the back propagation algorithm is applied on training the system. Both control algorithms are developed to embed into a standalone DSP-based micro-controller and their performances are compared. Limitation is that the non linear process characteristics are not preserved. Jun et al (2011) has introduced a Modified Particle Swarm Optimization (MPSO) for tuning the parameters of PID controller on boiler drum water level. Compared with Basic Particle Swarm Optimization (BPSO), MPSO enhances the searching ability, prevents particles falling into the local minimum and makes the searching of global optimal solution more efficiently by adopting solution sharing and searching range sharing. A new error criterion was proposed to validate the performance of PID controller based on MPSO (MPSO-PID). The experimental results show that the MPSO- PID achieves better time response than the PID controller based on BPSO (BPSO-PID). A major drawback is the increase in execution time.

7 Cascade Control Literature survey on liquid level control system with level as primary process and flow as secondary process has been carried out and the outcomes of the various literature are presented and discussed. Literature on cascade control of liquid level process, based on single objective optimization using conventional and intelligent techniques are presented in this section Conventional methods Sadasivarao and Chidambaram (2006) have applied a simple genetic algorithm for tuning PID controllers of the cascade control systems. A methodology for selecting the search region is proposed using Ziegler Nichols tuning method. Stability and robustness criteria are ensured in the selection of the search region, enabling the method to be applicable to online tuning. The inner and outer loops are tuned simultaneously, making the method applicable without disturbing the control strategy and ensuring overall optimal solution. Integral Absolute Error (IAE) values of the regulatory response is used as the objective function. The results show the superiority of genetic algorithm (GA) over the other methods. However, this algorithm fails to yield better results in terms of time domain specifications. Chen et al (2008) have proposed a cascade integral sliding mode control for a water tank level control system to realize level position regulating and tracking control. The key feature of this control scheme is the use of cascade back-step design method for the cascade nonlinear water tank level system to improve the control performance. The validity of the proposed control scheme was verified through practical testing on an experimental tank level system device. In the cases of step, multi step, and sinusoidal level position command inputs, the test results show that the proposed control scheme is capable of improving the tracking precision.

8 26 The limitation (Sadasivarao and Chidambaram 2006 and Chen et al 2008) of the above literature is that the control is based on single objective, even if the process has multiple objectives Intelligent techniques Tunyasrirut and Wangnipparnto (2007) have developed a cascade control scheme with fuzzy logic controller in primary (level) loop and conventional PID in secondary (flow) loop in a liquid level control system to control the level of horizontal tank that has diameter 300 mm and 480 mm long. Interface card module in computer and LabVIEW software program is used for building the cascade controller. The inner loop uses a PID controller for regulating the flow rate of the system and outer loop uses a fuzzy logic controller to control the level. The response time, steady state error, load disturbance and control valve action of cascade control system are tested and compared with the simple controller. The experimental results shows that for the same water level of 50% set point, the rising time noticed with the cascade controller was less than the simple controller about 1750 ms, and has a steady state error less than simple controller of about ±1%. The load disturbance on the plant has no affect when using the cascade controller. The cascade controller that comprises of the PID and the fuzzy logic control improves the dynamic characteristics of the liquid level control system. Limitation is that an increase in settling time of 100 sec is noticed with this method. Kumar et al (2008) have proposed a cascade control strategy with combined Fuzzy PI and Fuzzy PD in primary loop and conventional PI in secondary loop in a liquid level control system and achieved a settling time of 194 sec and overshoot of 4.5%. A comparative study was carried out to evaluate the real time performance of Fuzzy Proportional-Integral plus Fuzzy Proportional-Derivative (Fuzzy PI + Fuzzy PD) controller with the real time performance of conventional PI controller for a liquid level process. The

9 27 process considered for this experiment shows highly nonlinear behaviour due to equal percentage pneumatic control valve. National Instruments based hardware and software tools (LabVIEW) were used for precise and accurate acquisition, measurement and control. The real time implementation of the Fuzzy PI + Fuzzy PD controller was carried out in two configurations namely, feedback and cascade. In cascade control configuration, Fuzzy PI + Fuzzy PD controller was implemented in the primary loop. The secondary loop was tuned using the conventional PI controller and the results illustrate that that fuzzy controller perform better in comparison with conventional controller in both the feedback and cascade control configurations. Limitation is that the tuning of controllers becomes a difficult task when fuzzy logic controllers are combined with PI and PD controllers. Sangeetha et al (2012) have proposed a PID based cascade control scheme for the regulation of level in level control systems through Supervisory Control and Data Acquisition System (SCADA) Programmable Logic Controller (PLC) Open Process Control (OPC) interface and a network architecture with no overshoot. The cascade control system was the combination of level (primary process) and flow (secondary) processes. The SCADA was developed with PID controller. The characteristics of the cascade control system have been analyzed through the performance indices, such as peak time, rise time and settling time. The error values such as Integral Square Error (ISE) and Integral Absolute Error (IAE) are also calculated. The performances of cascade control system are validated through number of architectures, such as SCADA, PLC, OPC and internet. The introduction of PLC and National Instruments NI-OPC server has significantly improved the performance of conventional processes such as level, flow and cascade control systems. However a huge settling time of 107seconds is noticed with this method.

10 DC SERVO MOTOR CONTROL SYSTEM DC servo motors play a vital role in all the process industries. Development of accurate controllers for such processes becomes a difficult task. Literature that focus on DC servo motor control systems are discussed in this section Simple Feedback Control A lot of literature focuses on speed control of dc servo motor as simple feedback loop. The research activities done in these areas reflect the interest in improving the capability of control for servo process. Literature based on speed control with single objective function in single feedback loop are presented and discussed. Both conventional and intelligent tuning techniques proposed in various literature are discussed in this section Conventional methods Weber (1965) have proposed a Pulse Width Modulation (PWM) based DC motor control to control the speed of a DC series field motor at different required torque levels by adjusting the voltage applied to the motor. For any particular constant voltage, the motor speed was determined solely by the torque requirements and top speed is reached under minimum torque conditions. A series motor was used as a traction drive for vehicles. Voltage to the motor is controlled to fit the various torque requirements of grades, speed and load. The common method of varying the speed of the motor is by inserting resistance R in series with motor to reduce the supplied power. This type of motor speed control is very inefficient due to the wasteful of battery power due to the I 2 R loss, especially under high current and high torque conditions.

11 29 Chevrel et al (1996) have presented a methodical approach based on switched quadratic regulators for designing an efficient DC motor speed controller. This control strategy is compared with two cascade control design methods in terms of performances, robustness and complexity. Switched LQ (Linear Quadratic) regulators were particularly well suited to manage the performance-robustness trade off. The performance robustness has been proved for the strategies, in spite of large parametric uncertainties. Stability analysis of the system was not carried out to demonstrate whether the system is stable or not. Chevrel and Siala (1997) have proposed the speed control of a DC servo motor when the current is the only one measurement. A LQG (Linear Quadratic Gaussian) controller is designed by assembling an integral LQ state feedback and a Kalman observer to estimate the speed and the load torque. Its robustness and performances are discussed and experimentally tested. The current is limited by switching from the speed regulator to a current one when necessary. A control design was presented that performs well the compromise between speed control performances, robustness and current limitation of a dc-motor as no mechanical sensor is available. Especially good results were experimentally obtained when the motor parameters were identified at the starting of the motor. It is also shown that a variation on the resistance produces a non zero steady state error and demonstrated that no linear regulator can solve this problem. Limitation is that, accurate speed control without mechanical sensor includes on-line identification of the resistance which is slowly varying and also the use of adaptive control. Praesomboon et al (2009) have proposed a speed sensorless DC motor control using Kalman filter. Kalman filter considers the DC motor mathematical model. The model will start from a continuous state space into a discrete state space form. Inputs of the system are the armature voltage and

12 30 the armature current with noises. The output of the system is the estimated speed. Kalman filter was used to estimate the speed without noise interference. The result from the estimated speed was compared with the reference speed and the speed error will be used by the controller to control the linear amplifier. The final output was the constant speed of the DC motor. Even though the load of the DC motor changed, the speed of the DC motor remains constant under the control. The results illustrate that the Kalman filter can reject the noise from the system and estimate the speed of the DC motor with a high accuracy. The estimated speed will be feedback to control the system for constant speed. Sampling rate is a major limitation while converting continuous signal to discrete signal. Bindu and Namboothiripad (2012) have proposed the position control of DC servo motors since they are extensively deployed in various servomechanisms. Normally PID controllers are used to improve the transient response of DC servo motors. At present, most tuning methods are designed to provide workable initial values, which are then further manually optimized for a specific requirement. They have presented a flexible and fast tuning method based on Genetic Algorithm (GA) to determine the optimal parameters of the PID controller for the desired system specifications. Simulation results show that a wide range of requirements with respect to PID parameters are satisfied with the proposed tuning method. Limitation is on the range of requirements and can be widened by increasing the range of initial population but the number of generations required to converge to optimal value may increase Intelligent techniques Lin (1994) employed an inference method to develop the control law and applied it to the speed control of a DC servo motor system. Stability, robustness and other behaviours of the fuzzy controller are compared with the

13 31 classical PI controller. A mathematical model is developed, and used in the computer simulation to aid the selection of the control parameters. Also, the feasibility of the fuzzy controller is validated by laboratory experiments. Domain expert s knowledge on framing the rule base of fuzzy logic controller is a major limitation. Allaoua et al (2001) have proposed an Adaptive Neuro-Fuzzy Inference System (ANFIS) for the speed control of DC servo motor optimized with swarm collective intelligence. Initially, controller is designed based on fuzzy rules and then an adaptive neuro-fuzzy mechanism is adopted and ANFIS is optimized by Swarm Intelligence. ANFIS has the advantage of expert knowledge of the Fuzzy inference system and the learning capability of neural networks. Simulation results demonstrate that the deigned ANFIS- Swarm speed controller realize a good dynamic behaviour of the DC motor, a perfect speed tracking with no overshoot, gives better performance and high robustness than those obtained by the ANFIS alone. Adequate knowledge of experts required to design fuzzy logic rules for servo motor control is a major drawback. Kang and Kim (2001) have designed a neuro-fuzzy controller to improve some problems that occur when the non linear system is controlled by a fuzzy logic controller. Their model obtains fast time response, maximized learning effect and shortened settling time. To prove the capability of the designed neuro-fuzzy controller, the neuro-fuzzy model is applied to a DC servomotor. As a result, this controller does not produce overshoot, which occurs in the PID controller, and also does not produce the steady state error of FLC. Also, it shortens the settling time by about 10%. The model has only about 60% of the value of current peak of the PID controller. Limitation is that the training algorithm and the learning rate of neural networks have to be properly selected.

14 32 Liu et al (2003) have proposed a non linear Multiple-Input Multiple- Output (MIMO) feedback linearization technique to a separately excited DC motor system that is operated in high speed field-weakening regime. Load adaptive and sensorless control techniques are adopted to improve dynamic speed performance. Based on non linear MIMO feedback linearization, several non linear control techniques have been investigated for speed tracking performances. Under unknown load torque disturbance, the design with the linear control law obtained yields the steady-state errors as well as the degraded system responses as a result of the incomplete linearization. This problem cannot be effectively tackled by introducing an integrator. To overcome this limitation, load adaptive control design has been developed to improve the dynamic control performance. The main advantage of the scheme is that it uses only current measurements, eliminating costly speed sensors. Because of the decoupling and linearization, control implementation of the motor speed and back emf is achieved independently and the designs are able to assure speed tracking at any desired field point. High dynamic tracking performance is achieved even when the system with unknown load torque is operated in wide dynamic regimes of field weakening. Design of reference model for unknown load torque disturbance requires adequate knowledge is a major limitation. Thepsatorn et al (2006) have implemented the speed control of a separately excited DC motor using fuzzy logic control (FLC) based on LabVIEW. The results of experiment on the real plant demonstrate that the fuzzy logic controller is sensitive to variations of the reference speed attention. It was shown that the controller gains optimal performance and overcomes the disadvantage of the conventional control of sensitiveness to inertia variation and variation of the speed with drive system of DC motor. A limitation is that the load disturbances are not taken into account while designing the controller.

15 33 Gui et al (2006) have implemented a three-phase full control rectification circuit controller which is widely used in the system of speed regulation of DC motor. Conventional PID controller with double closedloops has been used in speed control of separately excited DC Motor. Under normal operation, it is difficult to achieve high performance, because of the low robustness of conventional PID controller. In order to overcome these problems, fuzzy self adaptive PID control method under fuzzy control theory and DSP (Digital Signal Processing) based cascade control system is designed to regulate the speed of DC motor. The practicality and validity of the control method are confirmed from the simulation results. The robustness of the system is improved and fuzzy self-adaptive PID controller has better tracking and anti-jamming performance than conventional PID controller. Limitation is the lack of domain expert s knowledge to develop fuzzy rule base. Shaker and Khashab (2010) have designed and tested an integrated electronic system that utilizes an interface card through the parallel port in addition to some auxiliary circuits to perform fuzzy control operations for DC motor speed control with load and no load. Software is constructed and the parameters for fuzzy logic controller (membership function, rules and scaling factor) and evaluation of the gain factors (K p, K i and K d ) by Ziegler-Nichols method are performed. The design dramatically reduced the hardware to the least possible. It is capable of enhancing the system performance by altering the Membership Functions (MF s) and the fuzzy rules in order to obtain the optimal result by monitoring the speed response of the motor. In Fuzzy Logic Controller (FLC), the corresponding values of all parameters (overshoot, settling time, rise time, steady state) are very less in comparison with the parameters of the conventional control and has higher stability, reaching the desired speed in a shorter time, but requires more tuning than conventional control, because FLC requires more computation parameters (MF s, rules and additional scaling factor in the inputs and output) unlike conventional

16 34 control (gain factors for inputs only). To achieve better performance and less utilization of available hardware resources, the work had been built to reduce the total number of used rules and to eliminate those which are not effective on the system. From experience, accurate results can be achieved if the relationship between the FLC & conventional control parameters were computed. Limitation is that the system software was more complex. Verma and Jain (2011) have proposed a new approach to control the speed of linear brushless DC motor and presented an overview of Performance Dependant Particle Swarm Optimization (PDPSO), as an alternative to evolutionary algorithm. Performance Dependent Particle Swarm Optimization(PDPSO) is used to determine the optimal gains of Proportional- Integral- Derivative controller (PID). Robustness of the system under critical conditions is improved when conventional optimization methods fail. PDPSO method is used to obtain optimum gains of PID controller for DC motor. PDPSO is new variant of PSO with faster speed because of strong selection principle. In simple PSO, after certain iterations, the populations set are almost identical and no further improvement is observed. PDPSO utilizes the global and local best value to search optimal setting of the state variable by considering security constraints. Like any other algorithms, the positive aspect of this method is its reliability and the number of required generation for convergence decreases with increase of population size and the performance is found to better than simple PSO-PID controller. Limitation of this method is that it produces sluggish response Multiobjective approach Literature related to multiobjective approaches to cascade control system design and its advantages and limitations are discussed and presented.

17 35 Gammel and Samahy (2009) have presented a control scheme based on MultiObjective Particle Swarm optimization (MOPSO), which is able to tune the PID controller parameters simultaneously in order to obtain the set of trade off optimal solutions called pareto set optimization solution for the conflicting objective functions of DC motor drive system. Multi Objective Particle Swarm Optimization (MOPSO) is implemented to tackle a number of conflicting goals that define the optimality problem. Five conflicting objective functions considered are minimization of maximum overshoot, rise time, speed tracking error, steady state error and settling time. Limitation is a good trade off between overshoot and settling time is not obtained due to the lack of non dominated sorting procedure in MOPSO Cascade Control Literature survey on DC servo motor control system with speed in the primary loop and armature current in the secondary loop has been carried out and the outcomes of the various literature are presented and discussed. Most of the literature on cascade control of DC servo motor are based on single objective optimization techniques. Conventional and intelligent techniques applied to the servo process in various literature are presented in this section Conventional methods Stephan et al (1988) have proposed a cascade speed control strategy for a thyristor driven DC motor subject to parameter variations. A dual mode adaptive inner current loop is cascaded with a model reference adaptive speed control loop to achieve better performance and this approach guarantees a stable and predefined performance at all the operating conditions. Comparisons with conventional analog speed control in continuous and

18 36 discontinuous current mode are made. Variations of the moment of inertia, field excitation and load torque were investigated. Model Reference Adaptive Control uses a reference model and an expertise in the field is required for design which becomes a major limitation. Alfaro et al (2008) have proposed a design approach for Two- Degree-of-Freedom (2-DOF) PID controllers within a cascade control configuration that guarantees smooth control. The rationale of operation associated to both the inner and outer controllers, determines the need of good performance for disturbance attenuation (regulation) as well as set-point following (tracking). It provides the complete set of tuning parameters for the inner (2-DOF PI) controller and the outer (2-DOF PID) controller and the design equations are formulated in such a way that a non oscillatory response is specified for both the inner and outer loop. The advantage of providing the complete set of parameters is that it avoids the need for the usual identification experiment for the tuning of the outer controller. However the use of 2-DOF controllers introduces additional parameters that need to be tuned appropriately which becomes a major limitation Intelligent techniques Pisano et al (2008) have presented a novel scheme for the speed and position control of Permanent Magnet (PM) DC motor drives. A cascade control scheme, based on multiple instances of a second order sliding mode control (2-SMC) algorithm is suggested, which provides accurate tracking performance under large uncertainty about the motor and load parameters. The overall control scheme is composed of three main blocks, a 2-SMC based velocity observer which uses only position measurements, a 2-SMC based velocity control loop that provides a reference command current and a 2-SMC-based current control loop generating the reference voltage. The scheme has been implemented and tested experimentally on a commercial

19 37 PMDC (Permanent Magnet Direct Current) motor drive. The experimental results confirm the precise and robust performance and the ease of tuning and implementation. A major limitation is the oscillations about the set point Multiobjective approach Literature related to cascade control of DC servo motor involving multiple objective functions are discussed and presented. Wang et al (2004) have proposed a genetic algorithm (GA) based multiobjective optimization evolutionary algorithm (MOEA) approach to take care of load disturbances. In traditional PI-type control strategies, the current controller is tuned first and then the speed controller is tuned. This often leads to a set of non optimal solutions. This control algorithm simultaneously tunes both the controllers in order to obtain the globally optimum control systems. Using the genetic algorithm based MOEA, the Pareto-set optimization solutions are evolved successfully. Simulation results verify the effectiveness of the proposed controller design approach. This control scheme can also be applied to ac drives such as vector controlled asynchronous drives without significant modifications. Multiple conflicting objectives and time domain specifications were not taken into consideration and hence the design approach failed to produce better results in terms of overshoot and settling time. Bottura and Serra (2004) have proposed a neural gain scheduling multiobjective genetic fuzzy PI control for DC servo motor systems. A discrete time version of a fuzzy PI controller is developed and the data bases as well as the constant PI control gains are optimally designed by using a genetic algorithm for simultaneously satisfying the following specifications, overshoot and settling time minimizations and output response smoothing. Hence, the optimization problem is a multiobjective one, from which results

20 38 an optimal fuzzy PI controller. A neural gain scheduler is designed, by the hack propagation algorithm, to tune the optimal parameters of the fuzzy PI controller at some operating points. Simulation results are shown to demonstrate the efficiency of the proposed structure for a DC servomotor adaptive speed control system used as an actuator of robotic manipulators. In spite of the advantages of the proposed approach, the limitation is that, since all the intelligent techniques are incorporated in the design, the control task becomes more complicated. 2.4 MULTIOBJECTIVE EVOLUTIONARY ALGORITHMS A detailed survey was conducted on various multiobjective evolutionary algorithms used for multiobjective optimization and some of the literature are discussed and presented. Kuhn and Tucker (1951) have developed the first technique for the generation of non inferior solutions for multiobjective optimization. The main strength of this method is its efficiency and its suitability to generate a strongly non dominated solution that can be used as an initial solution for other techniques. Its main weakness is the difficulty to determine the appropriate weights that can appropriately scale the objectives when enough information about the problem is not available, particularly if it is considered that any optimal point obtained will be a function of such weights. Still more important is the fact that this approach does not generate proper Pareto optimal solutions in the presence of non-convex search spaces regardless of the weights used. Schaffer (1985) have developed an approach to use an extension of the Simple Genetic Algorithm (SGA) called the Vector Evaluated Genetic Algorithm (VEGA) that differed from SGA only in the way in which selection was performed. This operator was modified so that at each

21 39 generation a number of sub populations were generated by performing proportional selection according to each objective function in turn. Limitation is that the generated solutions are non dominated in a local sense but their non-dominance was limited to the current population. Also, problem of genetics known as speciation i.e. the evolution of species within the population which excel on different aspects of performance arises because this technique selects individuals who excel in one dimension of performance, without looking at the other dimensions. Fonseca and Fleming (1993) have proposed a scheme in which the rank of a certain individual corresponds to the number of chromosomes in the current population by which it is dominated. The main advantage is that population sorting is done according to their rank. Fitness assignment is likely to produce a large selection pressure that might produce premature convergence was the major drawback. Srinivas and Deb (1995) have proposed the Non dominated Sorting Genetic Algorithm (NSGA) based on several layers of classifications of the individuals. Before selection is performed, the population is ranked on the basis of non domination and all non dominated individuals are classified into one category with a dummy fitness value, which is proportional to the population size, to provide an equal reproductive potential for these individuals. To maintain the diversity of the population, these classified individuals are shared with their dummy fitness values. This group of classified individuals is ignored and another layer of non-dominated individuals is considered. The process continues until all individuals in the population are classified. A stochastic remainder proportionate selection was used for this approach. Since individuals in the first front have the maximum fitness value, they always get more copies than the rest of the population. This allows searching for non dominated regions, and results in quick convergence

22 40 of the population toward such regions. The efficiency of NSGA lies in the way in which multiple objectives are reduced to a dummy fitness function using a non dominated sorting procedure. Both maximization and minimization problems can be handled using this approach. High computational complexity, non elitism approach and the need for specifying a sharing parameter was the major limitation. Deb et al (2002) have suggested a non dominated sorting based MultiObjective EA (MOEA), called Non dominated Sorting Genetic Algorithm-II (NSGA-II), a fast non-dominated sorting approach with low computational complexity which alleviates all the difficulties in NSGA. It uses a selection operator that creates a mating pool by combining the parent and offspring populations and selecting the best (with respect to fitness and spread) solutions. Simulation results on difficult test problems show that the proposed NSGA-II, in most problems is able to find much better spread of solutions and better convergence near the true Pareto-optimal front compared to Pareto-archived evolution strategy and strength-pareto EA, two other elitist MOEAs that pay special attention to creating a diverse Pareto-optimal front. Simulation results of the constrained NSGA-II on a number of test problems including a five objective seven constraint nonlinear problem are compared with another constrained multiobjective optimizer and much better performance of NSGA-II is observed. Coello and Lechuga (2002) have introduced a proposal to extend the heuristic called "Particle Swarm Optimization" (PSO) to deal with multiobjective optimization problems. This approach uses the concept of Pareto dominance to determine the flight direction of a particle and it maintains previously found non dominated vectors in a global repository that is later used by other particles to guide their own flight. The approach is validated using several standard test functions from the specialized literature.

23 41 The results indicate that this approach is highly competitive with current evolutionary multiobjective optimization techniques as NSGA-II and other techniques. Liu (2008) developed a new non-dominated sorting particle swarm optimization (NSPSO) that combines the operations (fast ranking of nondominated solutions, crowding distance ranking and elitist strategy of combining parent population and offspring population together) of a known MOGA, NSGA-II and the other advanced operations (selection and mutation operations) with a single Particle Swarm Optimizer (PSO). The efficiency of this algorithm is demonstrated on two test functions and the comparison is made with the NSGA-II and MultiObjective Particle Swarm Optimization (MOPSO). The simulation results suggest that the proposed optimisation framework is able to achieve good solutions as well diversity compared to NSGA-II and MOPSO optimisation framework. 2.5 SUMMARY Various literature on conventional and intelligent control of liquid level control system in case of simple feedback and cascade control system with single and multiple objectives are discussed. Literature that focus on speed control of DC servo motor using conventional and intelligent approaches with single and multiple objectives are presented. A brief survey of the state of art multiobjective evolutionary algorithms are reviewed and discussed. The review of the literature shows that there is still considerable challenges in the application of multiobjective evolutionary algorithms for servo and regulatory processes and remains an interesting research activity, and hence the research work. This work focuses on applying the evolutionary multiobjective algorithms such as NSGA-II and NSPSO to cascade control of servo and regulatory processes to provide better response in terms of various time domain specifications in comparison with the existing techniques.