ISSN: 1992-8645 Jatit E-ISSN: ACTIVE VORTEX .

Journal of Theoretical and Applied Information Technology10th December 2014. Vol.70 No.1 2005 - 2014 JATIT & LLS. All rights reserved.ISSN: 1992-8645www.jatit.orgE-ISSN: 1817-3195ACTIVE VORTEX INDUCED VIBRATION CONTROLLERAND NEURO IDENTIFICATION FOR MARINE RISERS11MOHAMMED JAWAD MOHAMMED, 2INTAN Z. MAT DARUSFaculty of Mechanical Engineering, UTM, Skudai, Johor, Malaysia-813101Faculty of Electromechanical Engineering, University of Technology, Baghdad, Iraq2Faculty of Mechanical Engineering, UTM, Skudai, Johor, Malaysia-81310E-mail: 1msc.mohammed83@gmail.com, 2intan@fkm.utm.myABSTRACTIn this work, a vortex induced vibration controller within discrete time has been investigated on marinecylinder pipe risers which represented by using nonlinear neuron identification models namely NARX andNAR. Input-output data have been extracted from the experimental rig of vortex induced vibration marineriser. A proposed work in this paper is to create the nonlinear system identification model undergoing forvortex induced vibration of marine riser depends on Neural Network which didn’t represented before thistime in this application and using PID controller to suppress the vibration. Two nonlinear systemidentification methods used to represent the models which are: Neural Network based on Nonlinear AutoRegressive External (Exogenous) Input (NARX) and Neural Network based on Nonlinear Auto-Regressive(NAR). Also, the best model has been chosen based on the lowest value of Mean Square Error (MSE)between actual and predicted response. While, PID controller has been used to suppress the oscillation ofpipe cylinder for all models and the comparison of the controller’s performance on each model by tuningthe controller parameter (KP, KI and KD) using heuristic method. Finally, the outcomes show that theNARX model performed better than the NAR model to predict the dynamic response of the system. On theother hand, PID controller has been managed to reduce the pipe cylinder fluctuation for all models speciallythe NARX model. Using particle swarm optimization (PSO) to improve the stability for marine riser on theparameters of PID controller are planned for future work.Keywords: Neuro Identification, NARX Model, NAR Model, PID Controller, Vortex Induced henomenon of Vortex Induced Vibration (VIV) isone of the main problems in the offshoreengineering fields because of the elasticallydynamic which is mounted on the rigid cylinder toextract the oil from the depths of the sea [1-2]. VIVhappens because of the interaction between thebody and the vortices shedding behind thestructure. Whereas, the vortices behind of thestructure create the forces which Working onfluctuation the cylinder from side to side in one ortwo directions [3].The motion of the cylinder from theprevious studies can divide into movements: firstly,when the cylinder motion in perpendicular(transverse) to the air or water crossflow generatingforce called lift force. Secondly, when the cylindermotion in same direction of stream flow (inline)generating force called drag force [4-5]. However,there are two ways in this field can suppress theoscillation of a pipe cylinder under VIV which is:passive and active vibration control. The firstmethod depends on adding stiffness or mass orchanging the sectional area of the structure of thesystem to raise the natural frequency of the bodywithout the need for external power signal. [6-7].While, the other method its very used from the firstattempt by Baz and Ro (1991) up to now because ofthis method is cheaper and produce a higherperformance more than the passive method and itsdepend on using sensors and actuators as anexternal power signal to the system [8].Nowadays, there are few researchers inthis field represented the system behavior by usingsystem identification model to suppress theattenuation of the cylinder during from usingcontrolled methods such as PID and Fuzzy-PIDcontrollers [9-12]. All papers in this field usedlinear System identification methods (Least Squareand Recursive Least Square) to predict the system.153

Journal of Theoretical and Applied Information Technology10th December 2014. Vol.70 No.1 2005 - 2014 JATIT & LLS. All rights reserved.ISSN: 1992-8645www.jatit.orgThe initial goal in this work to create the modeldepends on nonlinear system identification methodsthen using discrete time PID controller.In this work, the reduction process the effect ofVIV on marine risers has been divided into fourmain sections: firstly, extracting the input-outputdata of the experimental setup from prior paper.Secondly, using NARX and NAR as a systemidentification method to predict the dynamicresponse of the system and validating the resultsduring from MSE technique. Thirdly, study theeffectiveness of PID controller discrete time onNARX and NAR models. Recently, discuss theoutcomes that obtained from this work.2.EXPERIMENTAL DATAIn this study, the input-output data extractedfrom previous paper by Shaharuddin and Mat Darus[11-12]. The researchers used the rigid cylinderwith flexibly supporting in the crossflow vibrationonly. Table 1 showing the specification and theparametervalueswhichestimatedfromexperimental test.E-ISSN: 1817-3195Figure 1. Accelerometer Positions For Detecting AndObserving Data Of Experimental Setup Diagram [11-12]Accelerometer A represents the detected inputdata for system identification. While, accelerometerB is represents the observed output for systemidentification. In Figure 2 and 3 shown therelationship between the input and outputamplitudes with the timeTable 1. Obtained Parameters inderLengthAspect RatioCylindermassMass ratioNaturalFrequency inwaterDampingratio in sionlessk265.34N/mFigure 2. Detected Amplitude With Simple Time SeriesFor Pipe CylinderInput (from accelerometer A) and output (fromaccelerometer B) data gotten from Shaharuddin andMat Darus as shown in Figure 1 which included33000 data for input and output data[11-12].Figure 3. Observed Amplitude With Simple Time SeriesFor Pipe Cylinder3.154SYSTEM IDENTIFICATION

Journal of Theoretical and Applied Information Technology10th December 2014. Vol.70 No.1 2005 - 2014 JATIT & LLS. All rights reserved.ISSN: 1992-8645www.jatit.orgE-ISSN: 1817-3195In this work, the system identification methodsNeural Network time series: Neural Network basedonNonlinearAuto-RegressiveExternal(Exogenous) Input (NARX) and Neural Networkbased on Nonlinear Auto-Regressive (NAR) usedto predict the dynamic response characteristicsundergoing vortex induced vibration on the pipecylinder.Neural Network is one of artificialmethods which can be used for the nonlineardynamic system identification. It is consists of anumber of neurons arranged in numerous layers.According to Figure 4, Neural Network modelsinclude at least three layers which are input, hiddenand output layers. The equations can be calculatedas follows:2.1 Nonlinear Auto-Regressive Model (NAR)2.3.1NAR is one of the methods used topredict the output data for the nonlinear model byusing the input or the output data only. Theequation of NAR model is [13]:v x wy t f y t1 , . , y tn NARX model is employed for differentpurposes such as control and system identification.Also, it is able to forecast the output data in realtime. The NARX algorithm can be defined asfollows [14]: షభ u t (2) షభ A z 1 a z a zB z b b z b z f ష౬భ షೡమ y w b (6)(8)(9)For output layerf v f w f w b where y wwherex , x are the actual input data of input network.y , y are the actual output of input network.w , w , w , w , w , w , w , w theweights between input and hidden layers.b , b the bias weights for hidden layerv , v the summation values for hidden layerf , f the final values for hidden layer.2.3.2 షభ Then 1 2.2 Nonlinear Auto-Regressive External Input Model(NRAX) షభ x wv x w x w y w y w b (7)The n represents the previous value of theoutput. f is represents the nonlinear function whichcan be carried out by using intelligent methods suchas neural network.y t For hidden layer ష౬య(10)(11)(3) (4)After neglecting the noise error and definingthe B z and A z :y t f y t 1 , . , y t n , . . , u tn , . . , u t nn 1 (5)wherew , w is the weights between the hidden andoutput layers.b is the bias weights for the output layer.v is the summation values for the output layer.f is the final predicted value for the output layer orneural network process. , are represent the previous value of input andoutput respectively. While, the represents theinput delay. Finally, represents the nonlinearfunction which can be carried out by usingintelligent methods such as neural network.2.3 Neural Network Time Series155

Journal of Theoretical and Applied Information Technology10th December 2014. Vol.70 No.1 2005 - 2014 JATIT & LLS. All rights reserved.ISSN: 1992-8645www.jatit.orgE-ISSN: 1817-3195Figure 4. Neural Network Architecture2.4 Validation TechniqueThe verification process of the result isconsidered one of the important ways to measurethe performance of the mathematical algorithmswhich used in this paper. Also, check the algorithmeffectiveness to ensure that which one is best fromthe other. In this paper, Mean Square Error method(MSE) used to verify the results obtained in thiswork. The equation of Mean Square Error is [15]:1ε t y t NFigure 5: Specify The Percentage Values For Training,Validating And Testing Data y t 12 wherey t is represents the actual output fromexperimental setup.y t is represents the predicted output whichobtained from system identification methods.The data which used in this paper are divided intotwo partitions. The first partition is used for testingand the other partition for validating.4.(a)SYSEM IDENTIFICATION SIMULATIONDESIGNDepends on the Neural Network times series inMATLAB toolbox, number of neurons and numberof delay considered the important variables togenerate the nonlinear model for NARX and NARmodels after specifying number of training,validating and testing data as shown in figure 5.The simulation procedure has been divided intothree parts: firstly, 70% of data used for training,15% of fata used for validating and 15% of dataused for testing. Secondly, specify the number ofneurons and delay with variant cases and finallycreate the model depends on the MSE values andfind the best model as shown in figure 6.(b)Figure 6: Specify The Number of Neurons And Delay (a)and Block Scheme For System Identification Models5.PID CONTROLLEROne of the conventional controllers used in thecontrol field of all disciplines around the world. Inthis work, PID controller employed to suppress themovement of the pipe cylinder for marine risersunder vortex emerged vibration to calculate the156