Identification Modeling Based On Analysis Of Ship Manoeuvring Tests

Ship manoeuvrability is one of the most important hydrodynamic performances ofships and is closely related to navigation safety. With the booming development ofshipbuilding and shipping, ships have been presenting some characteristics like large-sized,diversification and so on. Ship manoeuvring becomes more and more complex anddifficult, so that the risk of marine casualty is increasing. Therefore, early in1985,International Maritime Organization (IMO) proposed the preliminary guides of shipmanoeuvring performance evaluation and prescribed the basic requirements of shipmanoeuvrability. Later in1993and2002, IMO promulgated the interim standards andstandards for ship manoeuvrability respectively, and put forword explicit quanlitativerequirements of ship manoeuvrability.Accordind to the requirements of IMO standards for ship manoeuvrabiliy, to avoiddesigning and building ships which do not meet the basic requirements of shipmanoeuvrability, it is necessary to predict ship manoeuvrability at the initial ship designstage. The prediction methods include mainly the methods of database or empiricalformula, free-running model tests, computer simulation based on mathematical modelsand Computational Fluid Dynamics (CFD) based direct numerical simulation, amongthem the third method is one of the most popular and effective methods. To apply thismethod, accurately determining the hydrodynamic derivatives and manoeuvrabilityindexes in the mathematical models is vital to improve the prediction accuracy.Nowadays, four kinds of methods can be applied to determine the hydrodynamicderivatives in the mathematical models at the ship design stage, i.e., the methods ofdatabase or empirical formula, captive model tests, theoretical and numerical calculation,and system identification combined with model tests. The effectiveness of the first methodis greatly influenced by ship types, which limits its applicability. For the second method,not only special facilities are needed and implement is too laborious and time-consuming, but also there exists the so-called scale effect. The third method can calculate thehydrodynamic forces and moments acting on ships. By using this method, however, it isvery hard to get all hydrodynamic derivatives, especially the nonlinear ones, withaccuracy satisfying engineering requirements. The fourth method is an effective one formodeling of ship manoeuvring motion and has a long history of development andapplication. With the continuous development of modern measurement techniques andsystem identification techniques, this method has been more and more widely used.There are two main types of mathematical model of ship manoeuvring motion, i.e.,the hydrodynamic model and the response model. The hydrodynamic model includesAbkowitz model and MMG model. Abkowitz model is also known as whole-ship model.It treats the hydrodynamic forces acting on the hull, propeller and rudder as a whole andexpands them by Taylor series about the equilibrium point of forward motion withconstant speed. In MMG model, also known as modular model, the hydrodynamic forcesare decomposed into three components, the ones on hull, on rudder and on propeller,respectively, with the interactions between the hull, the rudder and the propeller beingtaken into account. The response model may be derived from the linear hydrodynamicmodel and reflects the yaw response of ship to rudder action and is mainly used in theautopilot design; it may also be used to predict simple ship manoeuvring motion.In this thesis, a novel system identification method, Support Vector Machines (SVM),is applied in modeling of ship manoeuvring motion by analyzing ship manoeuvring tests,including the free-running model tests and the captive model tests. SVM mainly includesLeast Square-SVM (LS-SVM), ε-SVM, ν-SVM and so on. Abkowitz model isprincipally used as study object, and to solve the nonlinear problems in the analysis ofship manoeuvring tests, ε-SVM and neural network are applied to identify the nonlinearfunctions in the model respectively. In the preprocessing of data from ship manoeuvringtests, wavelet analysis method, the so-called “mathematical microscope”, is applied for thefirst time in the denoising of ship manoeuvring test data.In analysis of the simulated free-running model zig-zag tests, simulation verificationof ε-SVM method and its application are conducted, i.e., ε-SVM is used to analyze thesimulated free-running model tests. Using Abkowitz model and response model,simulation of zig-zag tests is carried out. By analyzing the simulation tests, ε-SVM based on the linear kernel is used for the first time to identify the parameters inmathematical models of ship manoeuvring motion. Predictions of ship manoeuvringmotion are conducted by using the identified mathematical models. From the comparisonbetween the identified parameters and the parameters used for simulation test and thecomparison between the predicted results and simulation data, the feasibility of ε-SVMmethod in the analysis of free-running model tests is verified. In the identificationmodeling of response models, linear response models are adopted. To investigate theeffects of insensitive factor ε in ε-SVM on the analysis of ship manoeuvring tests, theparameters of linear response models are identified by ε-SVM with different ε. Theidentified response models are used to predict the zig-zag manoeuvres. It is shown that thelearning efficiency and prediction accuracy of ε-SVM can reach the optimum valuesimultaneously by adjusting ε. In the identification modeling of Abkowitz model,random number sequence is added into the training samples to damp the extent ofparameter drift. The results show that parameter drift is controlled effectively by thismeasure.In identifiying the nonlinear functions in the mathematical models of shipmanoeuvring motion, to overcome some inherent shortcomings of the classical BP neuralnetwork such as relatively slow convergence and existence of local minima etc., a newtype of neural networks, the Neural Networks based on Chebyshev Orthogonal basis orChebyshev Neural Networks (CNN) for short, is proposed and applied for identifying thenonlinear functions for the first time. This neural model adopts a group of Chebyshevorthogonal polynomial as the hidden layer neurons’ activation functions, while weightupdate formulas are derived by employing the standard BP training method. With the testdata of rudder angle and manoeuvring motion variables as inputs and the hydrodynamicforces as outputs, ε-SVM, classical BP neural network and CNN are applied to identifythe nonlinear functions in Abkowitz model respectively. Based on the identified nonlinearfunctions, hydrodynamic forces are predicted. The comparison of the predicted resultsshows that ε-SVM behaves best, CNN behaves better and BP behaves worst.In analysis of the captive model tests, test verification of ε-SVM method and itsapplication are conducted. The KVLCC1model recommended by the InternationalTowing Tank Conference (ITTC) Manoeuvring Committee for international comparative study is used as study object and the data of oblique towing tests by Korean Maritime&Ocean Engineering Research Institute (MOERI) and the pure sway tests by Italian ShipModel Basin (INSEAN) are utilized. By analyzing the captive model tests, ε-SVM isapplied to obtain the regression expression of hydrodynamic forces in Abkowitz model forthe first time. By using the regression expression of hydrodynamic forces, hydrodynamicforces under different operating conditions are predicted. Comparison between thepredicted results and test data shows that ε-SVM method is feasible in the analysis ofcaptive model tests. In analysis of the pure sway tests, equivalence transformation isconducted for expression of hydrodynamic forces to eliminate the remarkable linearcorrelation between the variables in the regression expression. The results indicate thatparameter drift is controlled effectively by this measure.In the data preprocessing of ship manoeuvring tests, wavelet analysis method isapplied in the denoising of test data for the first time. To verify this method, the responsemodels are used for simulation of zig-zag tests and the polluted test data are obtained byadding random number sequence into the simulated test data. Then, wavelet analysismethod is used to denoise the polluted data. By analyzing the polluted data and thedenoised data, ε-SVM is applied to identify the parameters in the response models andpredictions of zig-zag tests are conducted. From the comparison between the identifiedmodel parameters and the parameters for the simulated tests and the comparison betweenthe predicted results and simulation data, it is shown that wavelet analysis method isfeasible in the data preprocessing of ship manoeuvring tests.The main innovation of this research can be summarized as follows:1. ε-SVM is applied for the first time in the modeling of ship manoeuvring motion byanalyzing the ship manoeuvring tests including free-running model tests and captivemodel tests.2. A new type of neural networks, the neural networks based on Chebyshev Orthogonalbasis, or Chebyshev neural networks for short, is developed and applied for the firsttime to identify the nonlinear functions of Abkowitz model.3. Wavelet analysis method is applied for the first time in the data preprocessing of shipmanoeuvring tests, and its effectiveness is validated. 4. In analysis of the simulated free-running model zig-zag tests, random numbersequence is added into the training samples to weaken the linear correlation betweenthe variables of Abkowitz model so as to effectively damp the parameter drift inidentification. In analysis of the pure sway tests, equivalence transformation isconducted for the expression of hydrodynamic forces in Abkowitz model, so that theremarkable linear correlation between the variables in the expression ofhydrodynamic forces is eliminated effectively and the parameter drift is avoided.In this thesis, simulation verification and test verification are conducted for ε-SVMmethod and its application in identification modeling of ship manoeuvring motion basedon analyzing the ship manoeuvring tests. Moreover, wavelet analysis method is applied indata preprocessing of ship manoeuvring tests to improve the accuracy of identificationmodeling. This study provides a new method for identification modeling of shipmanoeuvring motion and a new way to design ship manoeuvring tests.