422

IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 36, NO. 3, JULY 2011

Simultaneous Determination of Drag Coefcient and Added Mass

Wai Leung Chan and Taesam Kang
AbstractThe drag coefcient and added mass (hydrodynamic mass) are the essential parameters for the dynamics analysis of submerged objects with mobility such as bio-mimicking sh robots or underwater vehicles. The shape dependence of these parameters makes them difcult to have good theoretical approximations and the parameters should be determined either numerically or experimentally. Different experiments have been proposed to obtain either the drag coefcient or added mass. This paper presents a new method to simultaneously determine the drag coefcient and added mass from a simple and economic experiment and a numerical identication procedure. An experiment was carried out to demonstrate the method and the identication error was studied analytically and numerically for some experimental uncertainties. Index TermsAdded mass, drag coefcient, hydrodynamic mass.

I. INTRODUCTION HE drag coefcient and added mass (hydrodynamic mass or virtual mass) are the essential parameters in the dynamic analysis of submerged objects with mobility such as sh robots or underwater vehicles. The knowledge of these parameters helps engineers to model the dynamics, determine the energy efciency, improve existing designs, etc. When an object moves in a liquid media with speed , it experiences a drag force and the drag coefcient is dened as (1.1) where is the water density and is the frontal cross-sectional area. If the object is accelerated, the surrounding uid is also accelerated and the object mass is virtually heavier than it should be. The virtual increase of mass is called the added mass or the hydrodynamic mass . Both the drag coefcient and the added mass are the functions of shape and direction of object motion. In addition, the drag coefcient also depends on the Reynolds number. For simple geometries, the drag coefcient can be found in [1] and [2] and the added mass in [3] and [11], for example. For arbitrary shape objects, literature rarely provides enough information for estimation; the drag coefcient
Manuscript received August 01, 2009; revised March 09, 2011; accepted April 28, 2011. Date of publication June 16, 2011; date of current version July 01, 2011. This work was supported by Konkuk University and the Korea Research Foundation Grant (KRFJ03303). Associate Editor: L. Whitcomb. The authors are with the Department of Aerospace Information Engineering, Konkuk University, Seoul 143-701, Korea (e-mail: [email protected]; e-mail: [email protected]). Color versions of one or more of the gures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identier 10.1109/JOE.2011.2151370

and the added mass must be determined numerically [4], [12] or experimentally in these cases. Different experiments have been designed to determine the drag coefcient of underwater objects. In constant speed method, an object is towed at different constant speeds, by measuring the towing force, and the drag coefcient is determined from the plot of force versus the speed squared (e.g., [6], [7], [13], [15], and [17]). This method requires a motion control unit to pull the object moving at constant speed and force sensor to measure the towing force. In the deceleration method, the motion of a decelerated object is lmed to determine the speed. By plotting the inverse of speed versus time, the drag-coefcient-to-mass ratio (object mass added mass) can be determined from the slope (e.g., [8] and [5]). This method is applicable if the added mass is known or estimated. In the drop tank method, it is possible to exclude the added mass effect when the object is falling, and it reaches the terminal speed in a drop tank apparatus, if the tank is deep enough. In this case, the drag force is balanced by the object weight in water, and the drag coefcient can be obtained from (1.1) (e.g., [9] and [10]). In the free decay, the decaying oscillatory motion of an object in water is recorded and the drag coefcient is determined with a numerical minimization scheme [18]. Compared with drag coefcient measurements, there are relatively fewer reports about the experimental measurements of added mass. For an object that has a plane of symmetry and the plane is perpendicular to the transverse motion, the added mass can be found by comparing the oscillation frequency in air to the frequency in liquid media, with the oscillation along the transverse motion [11]. Obviously, the method is inadequate if the symmetry is broken. Lin [7] and Fernandes [13] measured the force that tows an object at constant acceleration and computed the added mass with the knowledge of drag coefcient. Smallwold [14] used least square method to numerically identify the added mass from experimental data. Rosss [16] and Morrison [18] measured the motion data of an object in free decay tests and determined the added mass with system identication. This paper presents a new method to integrate the determination of the drag coefcient and the added mass of objects with mobility in a liquid media. It is assumed that the drag coefcient is constant over the range of force applied in the experiment. The method involves some experimental measurements followed by a numerical identication procedure. The principle of the experiments is to obtain the displacement-time relationship of an underwater object and to compare it with a dynamic model. The experimental setups for data acquisition are simple and economic to be constructed, as they do not require expensive instruments such as high-speed camera or force sensor, but a

0364-9059/$26.00 2011 IEEE

CHAN AND KANG: SIMULTANEOUS DETERMINATION OF DRAG COEFFICIENT AND ADDED MASS

423

Fig. 1. Horizontal setup.

low-cost optical encoder and electronics for timing. The identication procedure does not require the objects to reach the terminal speed, and therefore, the setup can be carried out indoor with limited pool or tank size. In addition to the theoretical explanation, a boxsh robot was employed to demonstrate the experimental method. The accuracy of the numerical identication for some experimental uncertainties was studied analytically and numerically. It will be shown that timing accuracy can seriously affect the identication. II. THEORETICAL BACKGROUND The experiment presented in this paper for the drag coefcient and the added mass determination is suitable for small-scale submerged objects. The experimental setup is economic and compact, which makes it feasible for most of the laboratories. For large objects, it is still possible to determine the parameters, if a small-scaled model is applied. The principle of the experiment is to record the displacement-time data of an immersed object, which is towed by a known external force. The data are then compared with a theoretical model. External weights are employed for towing and hence a pulley system is required to transmit the force in an appropriate direction. Three experimental setups are introduced depending on the availability of facilities. A. Experimental Setups Fig. 1 shows the horizontal setup, which is carried out in a water pool. An object is connected to an external weight of mass with a string routed through a pulley system. Orientation and buoyancy adjustments on the object are required so that the object can immerse at a constant depth below the water surface and yield a stable horizontal motion upon external pulling. The object, and hence the external weight, is held xed until water is calm and the data acquisition is started. The string will pull the object to cruise straight toward the bottom pulley due to the falling of the external weight. Fig. 2 shows the vertical setups, which are similar to the drop tank method. Experiments are carried out in a water tank or a pool. The setups can be classied as D-vertical and U-vertical for the object moving downward and upward, respectively. Similar to the horizontal setup, an object is connected to an external weight through a pulley system. The object weight distribution

Fig. 2. D-vertical (left) and U-vertical (right) setups.

should be adjusted so that oating is prohibited and the body axis aligns with the vertical direction for motion. The external weight is held xed until water is calm and the data acquisition is started. The object will cruise straight downward in D-vertical setup or upward in U-vertical setup when the external weight is released. In all setups, an optical encoder is mounted to a pulley. By timing the events of encoder signal detection, the displacement time data of a motion trajectory can be obtained for the numerical identication. As will be shown in Section V, the accuracy of results (drag coefcient and added mass) highly depends on the timing precision and thus, high data sampling rate is recommended. B. External Weight Selection The function of the external weight is to pull the object for motion with a known external force. The weight should be selected so that the string force acting on the object yields the terminal speed, which is within the object swimming or propulsion capability. If the object of length can travel at a terminal speed , where is the terminal body length speed, drag force acting on the object should be (2.1) For most sh robot or underwater vehicle designs, the mass can be approxidensity is close to that of water, and mass mated as , thus (2.2) Assuming that the objects terminal speed is larger than 0.5 body and the drag coefcient is 0.5, length per second then drag force is approximated as 0.125 mL (2.3)

If the pool or the tank in the experiment is large enough such that the object can move at the terminal speed, the string force should balance the drag force, i.e., (2.4)

424

IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 36, NO. 3, JULY 2011

From the setups in Figs. 1 and 2, assuming that the string mass is negligible and the total pulley friction is constant, the string force can be expressed as horizontal D-vertical U-vertical (2.5)

is the shortest distance where is the rotational inertia and between the string and the center of the th pulley. Solving (2.8) with the initial conditions , the solutions for the displacement and speed are (2.12) (2.13) (2.14) (2.15) Note that, from (2.12) and (2.13), time and speed written as the explicit functions of displacement as can be

for

where is the buoyancy of the object. Therefore, if the pulley friction is negligible, the minimum/maximum external weight in the experiment can be obtained by substituting (2.3) and (2.5) into (2.4) as 0.125 mL 0.125 mL 0.125 mL for horizontal D-vertical U-vertical (2.6)

(2.16) (2.17) and these two expressions will be more useful than (2.12) and (2.13) in the numerical identication. D. Numerical Identication Procedure The numerical identication procedure utilizes models (2.14)(2.17) to generate two equations; one equation is nonlinear while the other equation is linear. Experimental data are substituted into these two equations to extract useful information for the determination of drag coefcient , added mass , and pulley friction . Since the linear equation is a function of the additional weight different motion trajectories (due to different ) are required in the identication procedure. The nonlinear equation is obtained by considering the following quantity: (2.18) where , and with are three distinct displacement-time points within a motion trajectory. The substitution of in (2.16) into (2.18) yields (2.19) Nonlinear equation (2.19) indicates that can be solved by Newtons method with the knowledge of other parameters in the equation. Therefore, by picking three different displacement points , and hence the value of , in an experimental motion trajectory, the corresponding value of can be computed. One should note that there will be different values of determined from (2.19) for different additional weights . With the denition of the quantity , the linear equation can be obtained by taking the square of in (2.17) and using the denitions of and in (2.14) and (2.15) as (2.20) (2.21)

The expression of external weight in (2.6) is just an estimate to nd the required minimum or maximum external weight. Testing must be carried out to check whether the pulley friction is signicant and to make the ne adjustment of the weight. In D-vertical setup, either object mass or buoyancy should be adjusted so that is a positive quantity. Due to the existence of pulley friction, a single motion trajectory is not enough to determine both the drag coefcient and the added mass from the second-order Newtonian equation of motion. It is necessary to have several different motion trajectories by using different external weights. Once the minimum or maximum external weight of a setup is available, the external weight can be expressed as for horizontal D-vertical U-vertical (2.7)

where , called additional weights, are some positive weights leading to different motion trajectories. C. Equation of Motion and Solution It is assumed that the air drag on the external weight is negligible, the total pulley friction is constant, the string is inelastic, and its mass is negligible for the experimental setups as shown in Figs. 1 and 2. The equation of motion can be written in the form (2.8) where (2.9) for horizontal D-vertical U-vertical (2.10)

is the added mass and is the inertia contribution Term from the pulley system, which can be expressed as (2.11)

CHAN AND KANG: SIMULTANEOUS DETERMINATION OF DRAG COEFFICIENT AND ADDED MASS

425

Linear equation (2.20) suggests that, if a xed displacement point is selected for all experimental motion trajectories (due to different ), has a linear relationship with . In each motion trajectory, speed can be computed by the linear regression of some nearest experimental displacement-time points; and the values of in (2.20) can be computed with the value of determined from nonlinear equation (2.19). From the plot of versus , the drag coefcient can be determined from slope in (2.21), the added mass from (2.14) and the pulley friction from the -intercept in (2.21) as (2.22)

Fig. 3. The ostraciiform sh robot.

(2.23)

for

horizontal D-vertical U-vertical

(2.24)

Note that, due to measurement uncertainty and external disturbance, different motion trajectory may yield different value of the added mass. The added mass determined from the numerical identication procedure in this paper is the average value of the added mass obtained from (2.14) for different motion trajectories (i.e., different ). If the rst slit of the encoder, which is to invoke the rst encoder signal, is positioned initially at , the time data will suffer an offset, which should be accounted for the verication of model (2.12). Once the values of , the drag coefcient, the added mass, and friction are obtained, the value of can be computed through (2.15). With (2.16), the time offset due to the initial displacement is (2.25) Therefore, the identication procedure can also account for the timing offset in measurement due to the initial displacement. III. EXPERIMENTS A robotic boxsh, as shown in Fig. 3, was employed to demonstrate the drag coefcient and added mass determination using the horizontal setup as shown in Fig. 1. The boxsh weight was adjusted so that it could stay at a constant depth of 3 cm below the water surface. The experiment was carried out in a circular inatable pool with the diameter of 1.5 m and the depth of 25 cm. An optical encoder circuit was mounted to a pulley in the setup and recorded motion as shown in Fig. 4. In this experiment, the conguration of the encoder plate is illustrated in Fig. 5. When the external weight falls, the string will bring the pulley and hence the encoder plate to rotate. Light sensor is triggered when a slit passes through the horizontal dotted line and a microcontroller will record the time of the event. With the knowledge of the encoder resolution and radius, the discrete displacement data are (3.1)

Fig. 4. The horizontal setup with an encoder circuit attached to a pulley.

where is the number of signal detections since the beginning of motion; 0.7854 cm is the distance interval traveled by the string between two consecutive signal detections; and the initial displacement is the displacement of the object before the rst signal detection. As is shown in Section II-D, three distinct displacement-time points in a motion trajectory are required for numerical identication; and the displacement can be represented by integer , such that , and with . Table I lists the information of the setup. Seven external weights of 1040 g with a step of 5 g, i.e., 10 g and 0, 5, 10, , 30 g, were used and 80 data were recorded in all motion trajectories. IV. RESULTS The experimental data (down-sampled discrete points) for the displacement versus time are plotted in Fig. 6 for different external weights. Three displacement points, 19.08 cm 42.64 cm , and 58.35 cm were chosen to compute using (2.19) in every motion trajectory. To compute the values of in linear equation (2.20), 59.13 cm was selected as the displacement point in linear equation (2.20) and the corresponding speed was determined using linear regression with the nearest six points. Fig. 7 plots versus according to (2.20). From the slope, -intercept, and expressions (2.22)(2.24), the determined drag coefcient, the added mass, and the pulley friction are 0.70 g, 619 g, and 32.8 mN respectively. From these values, tting curves for the displacement-time relationships are computed

426

IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 36, NO. 3, JULY 2011

Fig. 5. Conguration of an encoder plate before the start of a motion trajectory.

TABLE I INFORMATION FOR THE HORIZONTAL SETUP

Fig. 7.

versus

at

58.35 cm.

the numerical procedure. For a timer with time unit , precision error , the magnitude of which is always less than a timing unit , is dened as (5.1) where and are the experimental time measured (with nite precision) and the actual time (with innite precision), respectively. Therefore, the precision errors in the time data are always less than 1 ms for the experiment in Section III, for example. If the timing device starts before the motion trajectory, the starting time will be different for every motion trajectory and hence, the time precision error will be different even though two motion trajectories have the same experimental parameters. The initial displacement uncertainty affects both values of in (2.19) and of in (2.20) directly, while the time precision error leads to the speed error and the uncertainty of through the error of dened in (2.18). The error of will nally affect the calculation of the drag coefcient and the added mass. A. Analytical Calculation of Errors To study the effects of initial displacement uncertainty and time precision error analytically, the error of results is computed by the rst-order error approximation. The time precision error rst leads the error of . From the denition of in (2.18), we have

Fig. 6. Experimental data (discrete points) and the curve tting (solid lines) . using model (2.12) for different external weights

using model (2.12) and plotted in Fig. 6. It can be seen that the experimental data agree well with the model. V. EFFECTS
OF INITIAL AND

(5.2) to is the time precision where error on the actual time data due to the measurement. Therefore, the rst-order change in , denoted , due to the time precision error is bounded by (5.3)

DISPLACEMENT UNCERTAINTY TIME PRECISION ERROR

In the experiment, the initial displacement is not the same and there exists an initial displacement uncertainty in every motion trajectory. Besides the displacement uncertainty, the timing precision will also affect the results as determined in

CHAN AND KANG: SIMULTANEOUS DETERMINATION OF DRAG COEFFICIENT AND ADDED MASS

427

where (5.4) The error of , together with the initial displacement uncertainty, results in the error of upon solving (2.19). Taking the rst-order approximation of (2.19), the error of can be approximated as (5.5) where

TABLE II SIMULATION PARAMETERS

(5.6) The error of , the initial displacement uncertainty, and the time precision error will nally contribute to the calculation error of in (2.20) as follows:

(5.7) is the speed error generated in the linear regression where due to the truncation of time data. With the error of for each motion trajectory due to different , the least square error of slope in (2.20) is

can be seen from (5.5) and (5.6), quantity and hence the error of will diverge as close to zero. From (5.3), points and should be chosen at a large distance from point to increase the denominators and and reduce the error in . On the other hand, points and cannot be too close since is close to 1 and leads to a singularity problem of solving (2.19). The value of in (2.20) should be evaluated at a larger displacement point to reduce the calculation error. From (5.7), the error of due to the rst two terms is reduced when a larger value of is applied. The acceleration is also smaller in this situation; hence, the error of speed, computed by linear regression, can be reduced. B. Numerical Simulations The initial displacement uncertainty and time precision error can affect the results as seen in the previous rst-order error calculations. To nd the reliability of the experimental results as computed in Section IV with respect to this uncertainty and measurement error, numerical simulations were executed to compute the error range of the results. Table II lists the parameters that simulate the experiment in Section III. It is assumed that the drag coefcient is 0.7, the added mass is 620 g, and the pulley friction is 33 mN. In addition to the horizontal setup, the same parameters are also applied in the D-vertical and U-vertical simulations with the buoyancy of 1096 g 10.74 N. Every computational experiment (for each setup) simulates seven motion trajectories, which use different external weights ( 1040 g with a step of 5 g). In each simulated motion trajectory, 80 data of theoretical displacement and time are rst generated using (5.11) (5.12) where values of and are calculated from (2.14) and (2.15), respectively; and time data use model (2.16) with the substitution of from (5.11) into . Initial displacement is assumed to be a uniformly distributed random number between 0.13 and 0.33 cm (i.e., average 0.23 cm) and is different for every simulated trajectory. Since the timer is running before the

(5.8) Therefore, the rst-order changes in the drag coefcient from (2.22) and the added mass from (2.23) are (5.9)

(5.10) The error of the drag coefcient is proportional to the ac, while tual value of the drag coefcient with a factor the error of the added mass is approximately proportional to the quantity , with a larger factor of . Therefore, the maximum percentage error of the drag coefcient is always less than that of the added mass. Given three displacement points , and with , these displacement points should be chosen carefully to reduce the error of the drag coefcient and the added mass in the numerical identication. Point should be large enough to reduce the percentage error of the displacement due to . As

428

IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 36, NO. 3, JULY 2011

TABLE III CONFIDENCE LEVELS OF COMPUTATIONAL RESULTS WITHIN SOME SPECIFIC RANGE OF ERRORS; TIMING UNIT 1 ms

TABLE V CONFIDENCE LEVELS OF COMPUTATIONAL RESULTS WITHIN SOME SPECIFIC RANGE OF ERRORS; TIMING UNIT 10 ms

TABLE IV CONFIDENCE LEVELS OF COMPUTATIONAL RESULTS WITHIN SOME SPECIFIC RANGE OF ERRORS, PERFECT TIMING

start of a motion trajectory, the time is not equal to zero when the experiment starts, but at a starting time , which is a uniformly distributed random number between 0 and 1 s. To generate experimental displacement and time , it is assumed that initial displacement is distributed uniformly between 0.13 and 0.33 cm with average 0.23 cm for all trajectories and the timing unit 1 ms. Hence (5.13) (5.14) where denotes the ooring of to an integer. From these generated experimental data and , the drag coefcient and the added mass in every computational experiment are computed using the numerical identication as described in Section II-D. Table III summarizes the simulation results of 10 000 computational experiments. It can be seen that the drag coefcient errors are almost less than 2.5% of the theoretical value in all setups. The added mass errors are larger for the same condence interval compared with the drag coefcient; nevertheless, all the errors are less than 10% of the theoretical values. From the simulation results, it can be claimed that, due to the initial displacement uncertainty and time precision error, the percentage errors of the drag coefcient and the added mass are less than 2.5% and 5.0% of the theoretical values ( and 620 g) respectively, with 99% condence interval. C. Precision of Time Data To see the effect of time precision error, two computational experiments were repeated using the parameters in Table II, except the timing accuracy. For the case of the perfect timing (no truncation error, i.e., ), Table IV lists the condence levels of the results. Comparing the results with Table III, the initial displacement uncertainty has small effects on the error of the results. If the timing unit is changed to 10 ms, then (5.15)

The condence intervals of the results are summarized in Table V. It can be seen that, with the poor timing accuracy (larger precision error) compared to the cases in Table III, the error ranges of both the drag coefcient and the added mass are increased for a given condence interval. The percentage error of the drag coefcient is within 10% of the theoretical value with 95% condence interval, while the added mass is no longer reliable. Note that the condence interval of the drag coefcient within a specic range of error is always larger than that of the added mass, which is agreed with conclusion drawn in (5.9) and (5.10). This means that the determined drag coefcient is always more reliable that the added mass. D. Nonlinear Least Square Curve Fitting The numerical identication presented in Section II-D is designed for the experiments in this paper specically. The nonlinear least square curve tting is an option to nd the values of and simultaneously. However, this method is no longer reliable when the initial displacement uncertainty exists. To compare the numerical identication presented in this paper and the nonlinear least square method, computational experiments, using Matlab nonlinear least square curve tting function lsqcurvet for nding of and , was carried out with the same parameters used in Table II. Due to the existence of the initial displacement , and noting that the time offset can be computed by (2.25), model (2.16) should be modied as (5.16) and nonlinear function can be written as (5.17) For each external weight, nonlinear function (5.17) together with the generated experimental data are applied in the nonlinear least square method to nd the values of and . Similar to the identication procedure presented in this paper, the values of and are then substituted into (2.20) to plot versus , and the drag coefcient and the added mass are calculated using (2.22) and (2.23).

CHAN AND KANG: SIMULTANEOUS DETERMINATION OF DRAG COEFFICIENT AND ADDED MASS

429

TABLE VI CONFIDENCE INTERVAL OF COMPUTATIONAL RESULTS (USING NONLINEAR LEAST SQUARE CURVE FITTING) WITHIN SOME SPECIFIC RANGE 1 ms OF ERROR; TIMING UNIT

which is undetermined. Therefore, expression (2.19) is free of the time offset problem. VII. CONCLUSION In this paper, a new method is presented to determine the drag coefcient and the added mass of some submerged objects simultaneously. Three simple and economic experimental setups are introduced for data acquisition. It is assumed that the drag coefcient is constant over the range of force applied in the experiments and a numerical identication procedure is presented to calculate the drag coefcient and the added mass from the experimental data. A boxsh was employed to demonstrate the method. From the experimental data, the drag coefcient is 0.7 and the added mass is 619 g. Simulations were executed to explore the effect of initial displacement uncertainty and time precision error; the results show that the drag coefcient and the added mass determined by the numerical identication procedure have small percentage of error compared with the theoretical values. The time precision error is critical to the reliability of the results and high timing accuracy can reduce the error of the results. REFERENCES
[1] B. R. Munson, D. F. Young, and T. H. Okiishi, Fundamentals of Fluid Mechanics, 4th ed. New York: Wiley, 2002, pp. 594595. [2] F. M. White, Fluid Mechanics, 4th ed. New York: McGraw-Hill, 2007, p. 460. [3] J. E. Slater, A review of hydrodynamic added mass inertia of vibrating submerged structures, Defense Res. Establishment Atlantic, Canada, Tech. Memorandum, 1984. [4] M. Listak, D. Pugal, and K. , CFD simulations and real world measurements of drag of biologically inspired underwater robot, in Proc. US/EU-Baltic Int. Symp., May 2729, 2008, pp. 14. [5] W. L. Chan, T. Kang, Y. J. Lee, and S. K. Sung, Swimming study on an ostraciiform sh robot, in Proc. Int. Conf. Control Autom. Syst., Seoul, Korea, Oct. 1720, 2007, pp. 700705. [6] R. L. McMasters, C. P. Grey, J. M. Sollock, R. Mukherjee, A. Benard, and A. R. Diaz, Comparing the mathematical models of lighthill to the performance of a biomimetic sh, Bioinsp. Biomim., vol. 3, 2008, 016002. [7] C. C. Lin, R. C. Chen, and T. L. Li, Experimental determination of the hydrodynamics coefcients of an underwater manipulator, J. Robot. Syst., vol. 16, no. 6, pp. 329338, 1999. [8] D. Bilo and W. Nachtigall, A simple method to determine drag coefcients in aquatic animals, J. Exp. Biol., vol. 87, pp. 357359, 1980. [9] E. G. Richardson, The physical aspects of sh locomotion, J. Exp. Biol., vol. 13, pp. 6374, 1936. [10] R. W. Blake, Fish Locomotion. Cambridge, U.K.: Cambridge Univ. Press, 1983, pp. 5456. [11] H. Chung and S. S. Chen, Hydrodynamic mass, DOE Tech. Rep. CONF-840647-9, 1984. [12] A. Tyagi and D. Sen, Calculation of transverse hydrodynamic coefcients using computational uid dynamic approach, Ocean. Eng., vol. 33, pp. 789809, 2006. [13] A. C. Fernandes and F. P. S. Mineiro, Assessment of hydrodynamic properties of bodies with complex shapes, J. Appl. Ocean Res., vol. 29, pp. 155166, 2007. [14] D. A. Smallwood and L. L. Whitcomb, Adaptive identication of dynamically positioned underwater robotic vehicles, IEEE Trans. Control Syst. Technol., vol. 11, no. 4, pp. 505515, Jul. 2003. [15] A. J. Healey and M. R. Good, The NPS AUV II autonomous underwater vehicle test-tube: Design and experimental verication, Naval Eng. J., vol. 104, pp. 191202, May 1992. [16] A. Ross, T. I. Fossen, and T. A. Johansen, Identication of underwater vehicle hydrodynamic coefcient using free decay test, in Proc. IFAC Conf. Control Appl. Mar. Syst., Ancona, Italy, Jul. 2004, pp. 363368.

Ten thousand computational experiments were carried out and the results are summarized in Table VI. Comparing with Table III, the numerical identication method presented in this paper is superior if the initial displacement uncertainty exists. VI. DISCUSSIONS A. Optical Encoder Versus High-Speed Camera High-speed camera is a popular tool to capture motion data, but it is highly recommended to use optical encoder for all experimental setups presented in this paper for the following reasons. The high-speed camera may require proper lighting for exposure. To reduce the error of displacement measurement due to a view angle, the setup may require a movable mounting to follow the object or external weight during motion. The tedious setups make high-speed camera an unfavorable option. As is shown in the error analysis and simulations, the timing accuracy can seriously affect the results. High-speed cameras with higher frame rate would be more expensive and require larger memory for picture storage. Optical encoder method, on the other hand, can record the data at high rate even with lowcost electronics, and it requires less memory for storage since a oating-point datum (time datum), instead of a picture, is saved for every captured moment. The optical encoder method does not require data postprocessing, since the time data are recorded directly and the displacement data are automatically known if the encoder radius and resolution are known. Therefore, the displacement-time data are immediately available after measurements. On the other hand, image processing is required to extract the displacement information if high-speed camera is applied, and this could be time consuming. B. Constraint Function for Finding the Value of It is possible to generate other constraint functions to nd the value of instead of using (2.19). Nevertheless, the computation of in (2.18) can effectively eliminate any time offset, which is a nite amount of time for an object to travel the initial offset,

430

IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 36, NO. 3, JULY 2011

[17] C. Aage and S. L. Wagner, Hydrodynamic manoeuvrability data of a atsh type AUV, Proc. Oceans Eng. Todays Technol. Tomorrows Preservation, vol. 3, pp. 425430, Sep. 1316, 1994. [18] A. T. Morrison and D. R. Yoerger, Determination of the hydrodynamic parameters of an underwater vehicle during small scale, nonuniform, 1-dimensional translation, Proc. IEEE/MTS OCEANS Conf., pp. 277282, Oct. 1821, 1993. Wai Leung Chan received the B.S. degree in physics from the Hong Kong University of Science and Technology, Hong Kong, in 1997 and the M.Sc. and Ph.D. degrees from the University of California San Diego, La Jolla, in 2002 and 2005, respectively. His Ph.D. research was about the minimal mass design, control, and experimental study of the deployable structures called tensegrity. After graduating from his Ph.D. program, he was a Systems Engineer in an engineering company in San Diego, CA. The projects involved the deployment and vibration isolation of a space antenna, and energy harvesting under-

water vehicle. He is currently working as a Researcher in the Department of Aerospace Information Engineering, Konkuk University, Seoul, Korea, and his research includes bio-mimicking robots, swimming study, and experiments on autonomous sh robots.

Taesam Kang received the B.S., M.S., and Ph.D. degrees from the Department of Control and Instrumentation Engineering, Seoul National University, Seoul, Korea, in 1986, 1988, and 1992, respectively. He was an Associate Professor at Hoseo University, Asan, Korea, from 1994 to 2001. Since September 2001, he has been teaching control system design and its applications as a faculty member in the Department of Aerospace Information Engineering, Konkuk University, Seoul, Korea. His research area includes sensor design and signal processing, control theory, and applications for aerospace, marine, and ground systems.