This action might not be possible to undo. Are you sure you want to continue?

Background

PID Controller Optimisation for the Translational Motion of a Remotely Operated Vehicle using Metamodeling Technique

• Currently, the autonomous navigation of underwater vehicle is a growing research and very interesting for application field. • Underwater activities such as environmental and industrial monitoring or geological surveying, which is required data acquisition at precise location (Liddle, 1986). (off-shore structure or pipelines, exploration of the sea bottom).

BY : AMRUL FARUQ

2013 | Universiti Teknologi Malaysia

background

• These applications usually for monitoring and maintenance, uses of Unmanned Underwater Vehicle, Remotely Operated Vehicle or Autoomous Underwater Vehicle. • Therefore, underwater vehicle are supposed to be completely autonomous, thus relaying to onboard power system and intelligence.

background

• In the last years, Remotely Operated Vehicles (ROVs) received increasing attention due to their significant impact in several underwater operations. • ROV is controllable to get the desired operations, hence a main problem in underawater vehicle is to maintain its position for allowing it to perform its desired operations.

1

• only the 3DoF (three degrees of freedoms) dynamics in the translational motions will be considered. The control software will be implemented in MATLAB/SIMULINK. Scopes of Study • The emphasis of this study will be on the aspect of controlling an underactuated ROV. To optimize PID controllers of the translational motion of the MIMO ROV system. 1994). scopes of study • Development of mathematical modeling in this study is based on the previous study by (Fossen. 2 . Significant of Study • This study is potential to produce an applicable control design for a MIMO ROV system. To compare between the Metamodeling approach and genetic algorithms optimization in terms of execution time and the accuracy of results. To design a Multi Input Multi Output (MIMO) control system for a Remotely Operated Vehicle (ROV) written in the MATLAB/SIMULINK environment. • Proportional-Integral-Derivative (PID) Controller will be designed.11/07/2013 Objectives 1. sway and heave). by using offline tuning optimization utilizing Metamodel approach. It expects to contribute a significant impact in the development of a new optimization approach for computer experiments. which is to investigate the problem of translation motions (surge. 3. 2. Offline tuning optimization by using Metamodel will be investigated in this study.

2004: Extensively approaches System Identification for modeling AUV for various type of underwater vehicles. • Gabriella Dellino et al. They have successfully optimized the propeller for the AUV sprint speed and velocity. • Zanoli and Conte. simulation and experiment. Current Studies of Genetic Algorithm Applied on Optimization • Martz and Neu. Current Studies of Metamodel Applied on Optimization • Sultan et al. 2010: Studied a multi-objective GA applied for an AUV.. 2008: Developed a multiple objective genetic optimization (MOGO) algorithm for an AUV.. 1993) and (Cheng & Titterington.. 3 . Response Surface Methodology (Myers & Montgomery. • Chen et al.. 2003: Proposed a smoothes PID control and Discrete Fuzzy Smoother (DFS) to reduce overshoot depth control of the ROV.. 2009: Use a RBF Metamodel technique to tune the PID controller parameters for a non-linear process mixing system. 2009: Described a GA based PID for AUV motion control. 2002 and 2006: Identified the underactuated ROV and its parameters by using modeling. • Moura et al. They use MOGA to solve route planning problem of AUV.. • Koh et al. Current Studies of Metamodel Applied on Optimization • A variety of Metamodeling techniques exist. 1994) are two well-known approaches for constructing simple and fast approximations of complex computer codes. 1995) and Artificial Neural Network (ANN) methods (Smith. 2009: Present a methodology that solves robustness issues in simulation optimization by combining Taguchi’s approach with either Response Surface Methodology (RSM) or Kriging Metamodel.11/07/2013 Current Studies of ROV • Ridao et al.

forces and moments Figure 1.. from eq... 1.11/07/2013 Hydrodynamic model of ROV system Hydrodynamic model of ROV system • The velocity vector is represented as: v = [u v w p q r] T .. so they will be: and The Electro mechanical model of motor: . a simplified steady state model can be obtained: Where T is thruster output.. Thruster configurtion – front of ROV From three eq above. 4 . 1..3.. KM is modified motor constant and L is motor inductance.. Table 1... R is motor resistance.. ROV The body fixed-frames has a component of motion modeled by six velocities components: [ u v w p q r ] that represent translation of motion (surge.. underwater vehicle motion.. Ω is propeller rotational speed KT is parameters of various constants. France.1 The ROV has been described by Koh et al. 1.2 and 1. 2006. which is modified from Super Safir from Hytech Hydro – technologies.4 Figure 2. sway. Thruster Model of ROV The hydrodynamics propeller model: ... yaw) that is represented rotation of motion. V is motor voltage. 1. 1.3 . heave) and (roll.. the thruster transience is ignored. pitch..2 Thruster Model of ROV In this study.

This tools developed by Chin and Lau at NTU (Chin et al.ROV Controlllers 5 . can be written as: Controller Desing of ROV • This section illustrates the perfomance of the proposed study using the ROV Design and Analysis (RDA) package. 2006).1 and directions. control and system design analysis. RDA Packages Figure 4. RDA Packages . T3-T4 With . • RDA provides the necessary resources for a rapid and systematic implementation for mathematical models of ROV which focus on ROV modeling. < 0. T3+T4 . There are four different combinations of control input: T1+T2 . they can be used to control ROV Desing and Analysis Package ROV Controller Figure 3. T1-T2 .11/07/2013 Thruster Model of ROV The vector of forces acting on the ROV and moments about Centre of Gravity. < 1.

6 . Where is the mass. are the gain matrices for the proportional. Smaller error means the respond is better. respectively. In this study. the control input vector was assumed related to the thruster and force moments. 2002). In multi – objective optimization for controller system when we move one point to another in Pareto – front. in moving the vehicle from A to B. at least one ISE value decrease (lower error) and at least one other ISE increase (higher error). it subtracts this measurements from the set points to determine the error. integral and derivative segments of the controller. hence the coupled motions in roll and pitch were no controlled Hence.11/07/2013 ROV Models PID Controller for ROV System • Since the vehicle is underactuated. RDA Packages . Figure 5. while is the output. not all of motions can be controlled simultaneously (Koh. As the control read a sensor (translational motion). is the gravitation vector. Since the ISE is a vector. centrifugal and hydrodynamic damping matrices. 1994: Objective Function • The objective function used in this study was the Integral Square Error (ISE) given by: Where is the desired output (set point). et al. four directions given by the vector will be controlled.ROV Models PID Controller Design • A control law is chosen defined by Fossen. the concept of non-dominated (Pareto Optimality) is used to characterize the objectives.

is a basis function. Architecture of RBF NN RBF .unperturbed (6DoF RRCROV 1) Euler Transformation 6 DoF u 2 1 s Integrator1 3 Out3 Z square1 eout1 output earth pos eout2 output earth pos1 u2 square2 input display 1 s Integrator2 4 Out4 Yaw Double-click here to load parameters Clock1 t time Plot 3D Response Figure 6.11/07/2013 PID Controller Design 1 u 2 1 s Out1 bout1 X output body vel u2 PID sq 1 s Int 2 PID Demux Mux Body Velocity Scope inp thrust thrust Metamodeling Technique • Radial Basis Function Neaural Network The Radial Basis Function was first used to design Artificial Neural Network in 1988 by (Broomhead & Lowe. ROV Controller Simulink Design Figure 7. If the number of output. incorporates elitism and no sharing parameter need to be chosen (Deb et al. Q = 1. the output of the RBF ANN is calculated according to: MultiObjective Optimisation Using NSGA • Non-Dominated Sorting Genetic Algorithms (NSGA) is a popular non domination based genetic algorithm for multi-objective optimization. square3 Integrator PID Y Scope Body Velocity Scope Out2 ROV reference input PID pos pos vel v el v el eout Earth Position Scope Earth Position Scope Mux Scope1 -CConstant Nonlinear RRC ROV. 2002). a hidden layer and an output layer. denotes an Euclidean norm.NN The network consists of three layers: an input layer. YES End Searching Best Individuals Satisfactory Criteria NO Mutation 7 . Initialize the populations Select individuals Select fitness function Crossover/ Reproduction Where is an input vector. S1 is the number of neurons (and centers) in the hidden layer and are the RBF centers in the input vector space. 1988). are the weights in the output layer. It is a very effective algorithm which has a better sorting algorithm.

680} {640. 700} {860. …. …. 900} {640. Kp4 Kd4) for the PD controller. 875. 200} {640. The performance measure that was used in this study was the ISE value. 680} {860. Kp2 Kd2. 680} {660. 680. 655.11/07/2013 Result and Analysis • Metamodeling Development Here. The initial data sets are used to obtain the ISE by simulation. 680. T from simulation of the model based on input training data Define input training space. 650. 680} {660. 650. 700} {860. 670. 900} {640. 880. the measuring of each ISE has reached not longer than 1000 Result and Analysis Metamodel Illustration Initialize the parameters Obtain output target data. 180. the PD controller was chosen since the value all of I (integral) controller was selected to 0. …. …. These are (Kp1 Kd1. 170. 875. 680} {640.001. 660. 675. there are 8 parameters need to be tuned in order to obtain the best performance. Kp3 Kd3. Since using a sequence design space. 700} 6561 Kp1 Kd1 Kp2 Kd2 Kp3 Kd3 Kp4 Kd4 {160. 880. …. 200} {640. 900} {660. PD controller was selected due to its simplicity in implementation and its wide used in control applications. 680} {860. …. 660. D Define input testing data Choose RBFNN design Train RBF based on input data Simulate trained RBF based on test data Error estimated YES Define a Pareto-front Pareto optimal Satisfactory result? NO End Result and Analysis • Input training data of the RBF Metamodel Training Input Total number of data configurations Result and Analysis • Testing data Testing Input Total number of data configurations Kp1 Kd1 Kp2 Kd2 Kp3 Kd3 Kp4 Kd4 {160. …. 660. …. 900} {660. 700} 50625 8 .

• After the training stage (initial data sets). Radial Basis Function Neural Network is used to evaluate 50625 data sets (large test data).11/07/2013 Result and Analysis • Variable Constraint of ISE Variable ISE-X ISE-Y ISE-Z ISE-Yaw Result and Analysis • The identified initial data sets (input training) are used to obtain the Integral Square Error (ISE) for translational measure by simulation • 6561 initial data set (input training) configuration used to obtain output target data from simulation of the model based in train data. 9 . Upper Saturation Limit Lower Saturation Limit Spread Number of Centre 100 0 100 0 10 0 100 600 1000 0 Simulation Result (input train data) Result and Analysis • The ISE initial data sets is used to train the Radial Basis Function Neural Network which then will be used as the Metamodel of the Remotely Operated Vehicle (ROV) system to evaluate the ISE for corresponding large data sets (test data) of the controller parameters. This value can be adjusted to achieve a better response. • The ISE large data sets is used to estimate ISE by Pareto Optimal to predict the minimal ISE as simulated by Pareto front • The variable spread value can be used in the training process.

0728 minutes) Kp1 Kd1 Kp2 Kd2 Kp3 Kd3 Kp4 Kd4 200 680 680 700 900 680 900 700 6.25 6.95 85 X: 87.11/07/2013 Simulation Result (Metamodel) Pareto Front The ISE large data sets is used to estimate ISE by Pareto Optimal to predict the minimal ISE as simulated by Pareto front Figure 8.27 Y: 89.1 6.2 z-directionerror Best Gain .Metamodel The ISE estimated by pareto optimal comes from ISE large data sets by RBF-NN Best gain based on RBF-NN Metamodel (92. ISE estimated by Pareto optimal 10 .12 Z: 6.05 6 5. ISE estimated by Metamodel Pareto Optimal 6.15 6.015 90 105 95 100 100 105 95 110 90 115 85 y-directionerror x-directionerror Figure 9.

crossover probability and estimation recombination remain default Simulation Result . number of generation and population was used in this study is 1600 and 120 respectively.32 0 -1 -2 -3 -4 -5 5 10 15 20 25 30 35 40 45 50 55 60 time (sec) 11 . • Number of population will be determined as total number of Pareto – front while generation will determine the optimal value of Pareto – front.Metamodel x (surge) .5 Y: 111.11/07/2013 Simulation Result .0326 minutes) Kp1 Kd1 Kp2 Kd2 Kp3 Kd3 Kp4 Kd4 196.NSGA-II set point 0 -1 5 10 15 20 25 30 35 40 45 50 55 60 time Y (sway) Respons 2 1 y (sway) Metamodel y (sway) .5 7 100 120 140 160 180 f(x 1)-ISEX 105 110 115 120 125 130 135 f(x 2)-ISEY Figure 10.56 864.NSGA MOP using NSGA-II 8.NSGA-II set point y (sway) desired (m) 650. The other run – time parameters used by NSGA such as mutation parameter.desired Best gain based on NSGA-II (260.45 676. ISE estimated by NSGA Best Gain .NSGA 6 5 4 Overall Comparison X (surge) Respons x .5 f(x3)-ISEZ 8 X: 108.347 7.5 Z: 7.2 679.8 670.35 663.19 877. • By understanding these basic.NSGA • NSGA uses random population initialy and new candidates developed using genetic algorithm in every generation. The new candidates have probability to produce better result thus improve the final Pareto – front.5 3 2 1 x (surge) .

0 7.0 0.5 1 0.19 877.2 0.805 0.0 8.5 RBF-NN Metamodel (92.49 111.0 14.8 670.001 92.56 864.0 0.031 14.001 260.NSGA-II set point Yaw Angle Respons 35 30 25 20 15 10 5 0 yaw desired (degree) Kd4 60 5 10 15 20 25 30 35 40 45 50 55 time (sec) Overall Comparison Performance result of Surge motion Comparison Time Delay (s) Rise Time (s) Overshoot (%) Settling Time (s) Steady-state error (cm) Time taken (s) ISE value Metamodeling 7.Metamodel z (heav e) .5 NSGA-II 7.Metamodel yaw angle .11/07/2013 Overall Comparison Z (heav e) Respons 2 Overall Comparison Gains z (heav e) .0326 minutes 108.0326 minutes) 196.0728 minutes 78.198 34.45 650.001 92.5 Kp1 Kd1 60 0 5 10 15 20 25 30 35 40 45 50 55 time (sec) Kp2 Kd2 Kp3 Kd3 Kp4 yaw angle .0326 minutes Time Delay (s) Rise Time (s) Overshoot (%) Settling Time (s) Steady-state error (cm) Time taken (s) 12 .034 15.32 676.61 0.0 12.5 ISE value Comparison Overall Comparison Performance result of Sway motion Metamodeling 6.NSGA-II set point z (heave) desired (m) 1.0 0.0 0.0728 minutes 89.002 260.447 36.871 0.35 663.28 NSGA-II 9.0728 minutes) 200 680 680 700 900 680 900 700 NSGA-II (260.2 679.

Using Metamodeling.0728 minutes 6. the results obtained by using Metamodel approach are almost equal to the result evaluated by using genetic algorithm. Wassalam.0 0 92..5 hours compared to 4 hours. Using Genetic Algorithm. we managed to get better ISE but the time taken is too long. Thank you! Amrul Faruq 13 . • From the simulation results. 1.347 Summary • RBF-ANN has proven its effectiveness as a method of controller optimization in this case.0 0 12 0 260.. • However.0326 minutes 7. there is a difference in the simulation time and ISE value. we managed to obtain a reasonably good ISE in a much shorter time. i.11/07/2013 Overall Comparison Performance result of Heave motion Comparison Time Delay (s) Rise Time (s) Overshoot (%) Settling Time (s) Steady-state error (cm) Time taken (s) ISE value Metamodeling 6.e.0 0 9. It is able to give a good estimate of the controller parameters in a short time.0 6.0 5.045 NSGA-II 7.

- Installation Guide CBS
- aa.txt
- cek.txt
- aa
- Sim Param
- erorcolor.txt
- erorcolor
- Check Sum
- 12-integral-garis.pdf
- MAKALAHRADIOSTREAM_radiopendidikan_10827
- Signalling Diagram - Simple
- Testbed IP Plan v1.1
- prak9.pdf
- InternetRadio.pdf
- Getting Started With OpenBTS
- Load Tes
- Load Tes
- Membangun Multi OpenBTS
- Download 0010
- Book2
- Laser 1
- SAP Elektronika Optik
- Storage
- Engineering Village Detailed 1-25-2016 83538481
- Cover Title

Sign up to vote on this title

UsefulNot usefuloptimisation

optimisation

- Settles 2005 - An Introduction to Particle Swarm Optimizationby egonfish
- Heuristic Optimization Applied on Dynamic Matrix Control and Cascade PI Control Tunning for Speed Control of DC Motorby Hanzo Hasashi
- Optimización multi-objetivoby Angel Mariano
- PSO based optimization of a PI controller for a Real time Pressure processby International Journal for Scientific Research and Development

- Optimization of PID Parameter In Control System Tuning With Multi-Objective Genetic Algorithmby Anonymous 7VPPkWS8O
- A Heuristic Approach for optimization of Non Linear process using Firefly Algorithm and Bacterial Foraging Algorithmby Anonymous 7VPPkWS8O
- ICST 1018by International Jpurnal Of Technical Research And Applications
- Study of the Implementation of a Robust MPC in a Propylene-propane Splitter Using Rigorous Dynamic Simulation - Hinojosa2014by Hesam Ahmadian

- Settles 2005 - An Introduction to Particle Swarm Optimization
- Heuristic Optimization Applied on Dynamic Matrix Control and Cascade PI Control Tunning for Speed Control of DC Motor
- Optimización multi-objetivo
- PSO based optimization of a PI controller for a Real time Pressure process
- Px c 3876790
- Research on PID Parameter Tuning Based on Intelligent Fusion Optimization Algorithm in Control System
- 119-667-3-PB
- Paulo Oliveira.pdf
- Intelligent PID Controller
- 1204.0885
- Design and Optimization of Pid Controller Using Genetic Algorithm
- ABC
- Optimization of PID Parameter In Control System Tuning With Multi-Objective Genetic Algorithm
- A Heuristic Approach for optimization of Non Linear process using Firefly Algorithm and Bacterial Foraging Algorithm
- ICST 1018
- Study of the Implementation of a Robust MPC in a Propylene-propane Splitter Using Rigorous Dynamic Simulation - Hinojosa2014
- 12 Article Azojete Vol 8 125-131
- Optimal PID Controller Desing for AVR System
- P_696
- soft.pptxsoftcomputing_Genetic algorithm_introduction
- 2007 03 TOSCA NX DesignProcess
- Trade Space Exploration
- 011
- IEM-101
- Book Abstrats EngOpt2010
- Local Search
- 105
- PROJECT SYNOPSIS
- Optimization Basic Concepts
- Benchmarking of Heuristic Optimization Methods
- Amrul Faruq ROV 2013