You are on page 1of 10

Sensors and Actuators A 107 (2003) 109118

Fuzzy logic based closed-loop strapdown attitude system for unmanned aerial vehicle (UAV)
Sung Kyung Hong
Department of Aerospace Engineering, Sejong University, 98 Kunja-Dong, Kwangjin-Ku, Seoul 143 747, South Korea Received 11 October 2002; received in revised form 9 May 2003; accepted 4 June 2003

Abstract This paper describes the development of a fuzzy logic based closed-loop strapdown attitude reference system (SARS) algorithm, integrated ltering estimator for determining attitude reference, for unmanned aerial vehicles (UAVs) using low-cost solid-state inertial sensors. The SARS for this research consists of three single-axis rate gyros in conjunction with two single-axis accelerometers. For the solution scheme fuzzy modules (rules and reasoning) are utilized for online scheduling of the parameters for the ltering estimator. Implementation using experimental ight test data of SURV-1 Sejong UAV has been performed in order to verify the estimation. The proposed fuzzy logic aided estimation results demonstrate that more accurate performance can be achieved in comparison with conventional xed parameter ltering estimators. The estimation results were compared with the on-board vertical gyro used as the reference standard or truth model for this analysis. 2003 Elsevier B.V. All rights reserved.
Keywords: Strapdown attitude reference system (SARS); Filtering estimator; Fuzzy logic; UAV

1. Introduction Research and development in the technologies related to unmanned aerial vehicles (UAVs) have been interesting topics in military and civil aviation industry for the past decade. In military eld, UAVs could be considered as a cost-effective alternative for important missions such as gathering electro-optical information, delivering munitions, deceiving and jamming radars, and as targets for missile tests. Civil applications of UAVs are providing topographical data for geographical researches, agricultural spraying, meteorological measurements at high altitude, pollution studies and so on. In the above military and civil applications, UAVs are more preferable to manned aircraft considering the expenses [13]. An attitude measuring system is indispensable for ight control of a UAV all the time from launch to landing, and it needs to be as accurate as possible. For a high altitude UAV, it is required that these measurement should be better than order 1.0 and a resolution of 0.5 [4]. A gimbaled or strapdown inertial sensor system is able to meet the requirements for attitude measurements.

Tel.: +82-23408-3772; fax: +82-23408-3333. E-mail address: skhong@sejong.ac.kr (S.K. Hong).

The current state of the art in the determination of aircraft attitude angles is still primarily by means of the vertical gyroscope (gyro) [5]. The traditional vertical gyro is a spinning wheel in a gimbaled frame, and uses torquers at each gimbal to adjust and correct attitude. Pendulum devices are also used in initialization to erect the gyro and, during operation, to remove accumulated errors [6]. Such mechanical based systems are complex, expensive, bulky, and heavy, in addition to being prone to wear and failure. To reduce cost and increase reliability this paper proposes to replace traditional vertical gyro systems with a solid-state, strapdown attitude reference system (SARS), that is, by nature, more electronic than mechanical [7]. In the strapdown system, the outputs of a triad of body-mounted rate gyros are integrated by the Euler equations, or quaternions, to provide aircraft attitude angles. Because this strapdown inertial measurement has unbounded drift without external aiding, costly inertial-grade rate gyros are required to achieve satisfactory precision over periods of several hours. Therefore, many researchers [810] have been interested in looking at ways to replace costly inertial-grade rate gyros of SARS with low-cost solid-state miniature inertial sensors to minimize drift or bias with an estimating lter. To observe and compensate for sensor drift, a process called augmentation is used, where other system states compensate for gyro errors. For example, assuming that inertial

0924-4247/$ see front matter 2003 Elsevier B.V. All rights reserved. doi:10.1016/S0924-4247(03)00353-4

110

S.K. Hong / Sensors and Actuators A 107 (2003) 109118

systems are generally operated around an equilibrium state, linear acceleration will integrate to zero over time. Then attitude can be processed via gravitational acceleration, thus giving true angular measurements to compensate gyro drift. This concept is so called the accelerometer aided mixing algorithm of SARS. Because the orientation is augmented, this system has no restrictions on measuring time [4,11,12]. However, it should be noted that the above method of compensating for drift is not error-free; these are other inherent system error sources. One good example of augmentation errors is again provided by aviation [13]. If an aircraft and its SARS are circling for an extended period, the accelerometer will not only detect gravitational acceleration, but also centrifugal forces, resulting in incorrect attitude information. In addition, the change in transient forward acceleration is another possible error source. The result is that gyro angle measurements are augmented with incorrect accelerometer data resulting from aircraft dynamic motion. To control these effects, it is necessary to expand the current SARS algorithm to have an adaptive function under varying dynamic conditions. In this paper, a new closed-loop SARS algorithm based on fuzzy logic for the UAV using low-cost solid-state inertial sensors will be derived. By recognizing the situation of dynamic condition via fuzzy logic inference process, each parameter of the estimation lter of the existing SARS algorithm is determined online adaptively under varying aircraft dynamics. For this solution scheme, fuzzy rules and reasoning are based on the error signal of the gyro and accelerometer and the magnitude of dynamic motion [1417]. Measurements of inertial sensors outputs obtained during the experimental ight test of the SURV-1 Sejong UAV are used for implementation of the estimation scheme. The estimation results were compared with the measurements from the on-board high-precision vertical gyro used as the reference standard or truth model for this analysis. The comparison explicitly demonstrates that more accurate SARS performance can be achieved with the proposed fuzzy logic based integrated ltering estimator even though low-cost and low-performance inertial sensors are used.

Fig. 1. Sejong unmanned research vehicle (SURV-1).

direct open-loop products of attitude angles from each inertial sensor are introduced rst. 2.1. Experimental ight data During all the scheme investigation process in this paper, a set of ight test data from SURV-1 (Fig. 1) is used. The following gures show parameter time histories of the ight SURV-1. All parameters of approximately 600 s ight duration were sent telemetrically with 10 Hz from launching. During the test ight, the UAV climbed to cruise altitude at a relatively high rate of climb while maneuvering about the roll axis. The solid-state low-cost rate gyro that we adopted is RRS-75 of Inertial Science Inc. (ISI) and is shown in Fig. 2. Also, we used CXL04LP3 of Crossbow as an accelerometer. The measurement of angular rates (p, q, and r) and accelerations (ax , ay , az ) from rate gyros and accelerometers are shown in Figs. 3 and 4, respectively. Fig. 5 shows attitude angles obtained from the on-board vertical gyro used as the reference standard or truth model. 2.2. Attitude determination from rate gyros A transformation from one coordinate frame to another can be carried out as three successive rotations about different axes. The three rotations may be expressed mathematically as three separate direction cosine matrices Cz , Cy

2. Attitude determination from inertial sensors In this work, the navigation frame and the body frame are used as the reference and moving frames, respectively. Both frames have the same origin at some point xed in the aircraft where the measurement system is located. The navigation frame has axes aligned with the directions of north, east and the local vertical (down), and the orthogonal axis set of the body frame is aligned with the roll, pitch and yaw axes of the vehicle. The Euler angle representation is utilized to determine the attitude of the moving coordinate with respect to the other reference because it is one of the simplest techniques despite its drawback such as gimbal lock (singularities). In order to describe the strapdown scheme,

Fig. 2. Solid-state low-cost rate gyros and DSP processor.

S.K. Hong / Sensors and Actuators A 107 (2003) 109118


5 p (d e g / s )

111

-5 5 q (d e g / s )

100

200

300

400

500

600

-5 20 r(d e g / s ) 10 0 -1 0

100

200

300

400

500

600

100

200

300 t im e (s e c )

400

500

600

Fig. 3. Flight test rate gyro measurements.

and Cx which mean rotation angle about z-axis, angle about y-axis and angle about x-axis, respectively, and are denoted as follows: cos sin 0 Cz = sin cos 0 , 1 cos 0 sin Cy = 0 1 0 , sin 0 cos 1 0 0 Cx = 0 cos sin , 0 sin cos 0 0

The roll, pitch and yaw rate (p, q, and r, respectively) of the aircraft are measured using rate gyros with respect to its body axis system. The relationship between rate gyro output and time rate of Euler angles can be described using the direction cosine matrices (2.1), as follows: p 0 0 (2.2) q = 0 + C x + C x Cy 0 r 0 0 Starting with the matrix form of differential equations expressed in terms of Euler angles: 1 sin tan cos tan p 0 cos sin q (2.3) = sin cos r 0 cos cos

(2.1)

0 .5 a x (g )

-0 . 5 0 .2 a y (g )

100

200

300

400

500

600

-0 . 2 0 .5 a z (g )

100

200

300

400

500

600

-0 . 5 0 100 200 300 time (sec) 400 500 600

Fig. 4. Flight test accelerometer measurements.

112

S.K. Hong / Sensors and Actuators A 107 (2003) 109118


Roll & Pitch (deg) 10 0 -1 0 -2 0 -3 0 0 20 15 10 5 0 -5 0 100 200 300 400 500 600 100 200 300 400 500 600

Fig. 5. Attitude angles from the on-board vertical gyro (reference standard).

it is apparent that solving this equation for a strapdown system using numerical integration will provide attitude angles in terms of Euler angles , , and with respect to the chosen reference frame. As a physical instrument, rate gyros also carry some errors such as axis misalignment, xed bias, drift bias, xed scale factor errors, asymmetric scale factor error, and so on. The bias drift would be the most serious and deteriorate the accuracy of a SARS [10]. The bias drift, which normally shows nonlinear characteristics, causes the integration result to drift-off from the true attitude as a function of time and rapidly renders the calculation useless.

Euler angles obtained from open-loop integration of Eq. (2.3) using rst order Euler method of integration are illustrated in Fig. 6 and compared with the reference from the vertical gyros. This result shows: (1) how the bias errors cause the attitude angles to deviate, and (2) the instability of the integration which drifts as a function of time unless corrected. 2.3. Attitude determination from accelerometers Attitude angles and can also be obtained form aircraft acceleration. The following equation denes the equation of motion in terms of specic force (fx , fy and fz ), the

Fig. 6. The Euler angles calculated from rate gyro measurement.

S.K. Hong / Sensors and Actuators A 107 (2003) 109118

113

accelerometer readings in the body axes [18]: u 0 w v p fx = v + w 0 u q fy w v u 0 r fz r 2 q2 pq r 2 px pr + q pq + r p2 r 2 rq p py pz pr q rq + p q2 p2 sin + g cos sin (2.4) cos cos where (u, v, w) are linear velocity components and (px , py , pz ) are accelerometer coordinates along each axis in the body frame with its origin at the center of gravity of aircraft. If all of the parameters of Eq. (2.4), (fx , fy , fz ), (u, v, w), (u, v, w), and (p, q, r), can be measured in ight, near-true attitude angles can be determined. Normally, however, it is hard for a low-cost UAV to measure (v, w), and their time derivatives, and it is not easy to obtain true attitude measurements from accelerometers. However, if we assume that the UAV is ying level in a steady state, linear acceleration will integrates to zero over time and (p, q, r) can be neglected. With this simplication, Eq. (2.4) is reduces to the following form: fy fx = sin1 , = sin1 (2.5) g g cos It should be noted that the attitude determination using Eq. (2.5) is true only for specic conditions such as steady level ight. If an aircraft is circling for an extended period, the accelerometer will not only detect gravitational acceleration, but also centrifugal forces, resulting in incorrect attitude determination. In addition, the change in transient forward acceleration is another possible error source. This

means that the attitude calculation from accelerometers is not valid under all dynamic conditions. Accelerometer measurements during the ight test are shown in Fig. 4, and the attitudes obtained from simple calculation using Eq. (2.5) are illustrated in Fig. 7 where it is compared with the reference from the vertical gyros. These results show that the attitude from accelerometers tend to follow the vertical gyro output approximately in steady state ight, but large errors and poor reliability result in transient and high dynamic ight.

3. Conventional attitude estimator on gyros and accelerometers To observe and compensate for gyro drift, a process called augmentation is used, utilizing other system states to compensate for gyro errors. One of the conventional approaches is called the accelerometer aided mixing algorithm of SARS. This scheme involves a set of 3-axis rate gyros that provide the required attitude information from integration in combination with a 2-axis accelerometer. The basic concept of this scheme is that proper combination (ltering) of gyro and accelerometer measurements could make precise attitude information available. The gyros are responsible for measuring the transient dynamic part of aircraft motion, while the accelerometers give precise absolute measurement of attitude over a longer period (Fig. 8). In this section, the theoretical approach to conventional attitude estimation is provided and reveals some typical problems and drawbacks. 3.1. Filtering structure via blending gyros and accelerometers Fig. 8 shows a conventional ltering estimator block diagram for a roll axis channel. The role of the estimator is to

Fig. 7. The Euler angles calculated from accelerometer measurements.

114

S.K. Hong / Sensors and Actuators A 107 (2003) 109118

Fig. 8. Conventional closed-loop SARS block diagram for roll axis channel.

compare attitude angles resulting from the integration of the gyros with the attitude angle products from the accelerometers. The error between m and a is fed-back through a proportional and integral controller with a pair of estimator

gains Kp and Ki . A similar block diagram could also apply to the pitch channel. The mathematical relation of the estimated attitude (m ), the attitude products from the accelerometers (a ), and the

Fig. 9. Effects of estimation lter cut-off frequency on estimated attitude angles.

S.K. Hong / Sensors and Actuators A 107 (2003) 109118

115

attitude rate products from the gyros (g ) in Laplace form is as follows: Kp 1 Ki m = g + (a m ) + 2 (a m ) (3.1) s s s The error equation of the estimated attitude yields: m = sg + (Kp s + Ki )a s 2 + Kp s + K i (3.2)

Applying the nal value theorem to Eq. (3.2), we can see that the estimated attitude error converges to the error of the angle calculated from the accelerometer:
t

the lter improve, but the low frequency characteristics become worse. Therefore, the cut-off frequency needs to be optimized to improve the performance of the estimation lter over a wide range of dynamic conditions. However, it is difcult to achieve acceptable performance in all dynamic conditions using this xed gains ltering structure, especially when using low-cost (high bias drift) solid-state sensors. To overcome these problems, it is necessary to expand the current SARS algorithm to have an adaptive function under varying ight dynamics.

lim m = a

(3.3)

4. Fuzzy attitude estimator from inertial sensors In this section, a new closed-loop SARS algorithm based on fuzzy logic using low-cost solid-state inertial sensors is derived. By using fuzzy logic inference, each parameter of the ltering estimator of the existing SARS algorithm is determined adaptively under varying aircraft dynamics. For this solution, fuzzy rules and reasoning are based on the error signal of the gyro and accelerometer, and the magnitude of dynamic motion. 4.1. Tuning of parameters based on fuzzy logic Fig. 10 shows a fuzzy logic based attitude ltering estimator. The approach taken here is to exploit fuzzy rules and reasoning to generate parameters of the ltering estimator. As seen earlier in Eqs. (3.4) and (3.5), the parameters are determined only by the cut-off frequency. Using simulation study results, the nominal cut-off frequency (0 ) is set at 0.02 (rad/s). In the proposed scheme, the cut-off frequency is adaptively determined based on the magnitude of the current error e(k) and the current dynamic motion m(k) as shown in Fig. 10. Here the magnitude of the dynamic motion m(k) is dened as m(k) = p(k)2 + q(k)2 + r(k)2 (4.1)

The controller gains, Kp and Ki , are chosen by relating them to the cut-off frequency, , and damping ratio, , of the estimator as Ki = 2 , Kp = 2 (3.4)

During this investigation, the damping ratio is xed to a suitable value of 0.707 to provide a good transient response. Hence (3.5) Kp = 2 Therefore, the system is only characterized by the cut-off frequency, which is chosen appropriately to optimize the process. 3.2. Cut-off frequency effect on estimation Variation in the estimators cut-off frequency (natural frequency) had been exercised at 0.01 and 0.001 rad/s. The estimation results are compared with the vertical gyro outputs in Fig. 9. When cut-off frequency is 0.01 rad/s (a relatively high frequency), the estimation angles converge well at the steady state, but not during high dynamic ight motion. Reducing the cut-off frequency is set to 0.001 rad/s does not improve the situation. This is because as a cut-off frequency goes lower, the high frequency dynamic characteristics of

Fig. 10. Fuzzy based closed-loop SARS block diagram for roll axis channel.

116

S.K. Hong / Sensors and Actuators A 107 (2003) 109118

Fig. 11. Membership functions for e(k) and m(k).

The parameters of the estimation lter is thus obtained by Ki = (0 )2 and Kp = 20 (4.2) where is determined by a set of fuzzy rules of the form: if e(k) is Ai and m(k) is Bi , then = i (4.3)

The truth value of the ith rule in (4.3) i is obtained by the product of the MF values in the antecedent part of the rule: i = Ai [e(k)] Bi [m(k)] (4.4)

Here, Ai and Bi are fuzzy sets of the corresponding supporting sets; i is a constant. The membership functions (MF) of the fuzzy sets for e(k) and m(k) are shown in Fig. 11. In this gure, S represents small, M medium, B big. Note that may also be considered as a fuzzy number which has a singleton MF as shown in Fig. 12. In this gure, NO represents nominal, XS very small, XXS extra small. The fuzzy rule in (4.3) was extracted experimentally based on the characteristics of gyros and accelerometers. The either case when the magnitudes of the current error e(k) or the current dynamic motion m(k) is relatively big, the fuzzy inference engine judges that the attitude calculation from accelerometer is inaccurate. To assign more weighting to gyros, the crossover frequency should be decreased. Thus the constant can be represented by a fuzzy set XXS. Thus a set of rules, as shown in Table 1, may be used to adapt for the parameters of estimator, Kp and Ki .
Table 1 Fuzzy tuning rules for e(k) B m(k) B M S XXS XXS XS M XXS XS S S XS S NO

where Ai is the MF value of the fuzzy set Ai given a value of e(k), and Bi the MF value of the fuzzy set Bi given a value of m(k). Based on mi , the value of for each rule is determined from their corresponding MFs. By using the membership functions in Fig. 10, we have the following condition [16]:
m

i = 1
i=1

(4.5)

Then, the defuzzication yields the following:


m

=
i=1

i i

(4.6)

Here i is the value of corresponding to the grade i for the ith rule. Once is obtained, the parameters of the estimator, Kp and Ki , are calculated from the (4.2). 4.2. Experimental results The proposed fuzzy SARS ltering estimator has been tested and compared with a conventional estimator in Section 2. Outputs from the low-cost solid-state inertial sensor during an experimental ight test of SURV-1 Sejong UAV were used to implement the estimation scheme. The time responses for roll and pitch attitudes are plotted in Fig. 13. The results obtained from the on-board high-precision vertical gyro are also presented for comparison. It is obvious that these results are much better than the conventional closed-loop SARS results shown in Fig. 9. Table 2 shows the experimental results of integral of the

Table 2 Experimental results of integral of the absolute error (IAE) Conventional = 0.01 IAE for roll angle IAE for pitch angle 12645 11237 = 0.001 9578 7964 = 0.0001 13155 9669 Fuzzy based 1935 1799

Fig. 12. Singleton membership functions for .

S.K. Hong / Sensors and Actuators A 107 (2003) 109118

117

Fig. 13. Response of the fuzzy based closed-loop SARS.

Fig. 14. Variation of cut-off frequencies.

Fig. 15. Adjusted rate gyro bias error.

118

S.K. Hong / Sensors and Actuators A 107 (2003) 109118

absolute error (IAE) for both the conventional SARS and fuzzy based SARS. In order to verify the effectiveness of the estimation, the levels of agreement between truth model and the scheme results are calculated. The calculated errors based upon a simple root mean square (RMS) method are 0.32 and 0.29 for roll and pitch attitude, respectively. This level of performance is very encouraging for low-cost and low-performance inertial sensors. Figs. 14 and 15 show how the cut-off frequencies are adapted and the bias errors in rate gyros adjusted under dynamic conditions.

References
[1] C.A. Jones, Unmanned aerial vehicles: an assessment of historical operations and future possibilities, Thesis Research Paper, Air Command and Staff College, 1997. [2] T. Ehrhard, The US Air Force and unmanned aerial vehicles, Ph.D. Dissertation, Johns Hopkins University, 1999. [3] New demonstrator spurs Navy UCAV development, Aviation Week and Space Technology, February 2001. [4] S.A. Whitmore, M. Fife, L. Brasher, Development of strapdown attitude system for an ultrahigh altitude ight experiment, NASA TM-4775, January 1997. [5] G.M. Siouris, Aerospace Avionics Systems: A Modern Synthesis, Academic Press, New York, 1993. [6] B. Wie, Space Vehicle Dynamics and Control, AIAA Education Series, AIAA Inc., USA, 1998. [7] S.C. Garg, L.D. Morrow, R. Mamen, Strapdown navigation technology: a literature survey, J. Guid. Contr. Dyn. 1 (3) (1978) 287291. [8] J.P. Gilmore, Modular strapdown guidance unit with embedded micro processor, J. Guid. Contr. 3 (1) (1980) 560565. [9] M. Koifman, S.J. Merhav, Autonomously aided strapdown attitude reference system, J. Guid. Contr. Dyn. 14 (6) (1991) 11641172. [10] S.K. Hong, Compensation of nonlinear thermal bias drift of resonant rate sensor (RRS) using fuzzy logic, Sens. Actuators A 73 (23) (1999) 143148. [11] J.R. Catford, Application of strapdown inertial system with particular reference to underwater vehicle, Strapdown Inertial System, AGARD-LS-95, 1995. [12] I. Humphrey, Inertial navigation system for a micro unmanned aircraft, in: Proceedings of the AIAA-97-3567 Paper, AIAA Guidance, Navigation, and Control Conference, New Orleans, 1997. [13] I. Madani, An investigation into a reduced sensor t for unmanned aircraft, M.Sc. Thesis, Craneld University, 1998. [14] A. Fujimori, H. Tsunetomo, Z.-Y. Wu, Gain-scheduled control using fuzzy logic and its application to ight control, J. Guid. Contr. Dyn. 22 (1) (1999) 419435. [15] Z.Y. Zhao, M. Tomizuka, S. Isaka, Fuzzy gain scheduling of PID controllers, IEEE Trans. Syst. Man Cybernet. 23 (5) (1993) 1392 1398. [16] S.K. Hong, R. Langari, Robust fuzzy control of a magnetic bearing system subject to harmonic disturbances, IEEE Trans. Contr. Syst. Technol. 8 (2) (2000) 366371. [17] S.K. Hong, R. Langari, An LMI-based H-innitive fuzzy control system design with TS framework, Inform. Sci.: An Int. J. 123 (1) (2000) 163179.

5. Conclusions As an alternative to expensive and heavy devices such as vertical gyros on unmanned aerial vehicle (UAV), a solution scheme for a low-cost strapdown attitude reference system (SARS) architecture has been investigated to estimate closed-loop attitude using ight test experimental data from the SURV-1 Sejong UAV. The SARS consists of three single-axis rate gyros in conjunction with two single-axis accelerometers. The proposed closed-loop SARS scheme uses fuzzy rules and reasoning to determine the ltering estimator parameters by adjusting cut-off frequency. Although a rule of thumb for choosing the ranges for e(k), m(k), and was obtained experimentally, it is still possible to make further performance improvements by ne-tuning the ranges as well as by modifying the tuning rules. Also, a supervisory level may also be included to monitor the stability of the SARS. The proposed scheme has shown a good damping characteristic for drift errors form gyro measurement as well as accelerometer measurement noise corruption in all ight dynamic conditions. The scheme gave on accuracy of better than 0.32 in roll and 0.29 in pitch compared to a truth model vertical gyro reading. This performance is very encouraging and indicates that high accuracy SARS should be possible with low-cost, low-performance inertial.

Biography
Sung Kyung Hong received the BS degree and MS degree from Yonsei University, Korea, in 1987 and 1989, and PhD degree in mechanical engineering from Texas A&M University, College Station, in 1998. From 1989 to 2000, he was with the Unmanned Aerial Vehicle System Division and Flight Dynamics and Control Laboratory of the Agency for Defense Development (ADD) in Korea. He is currently Assistant Professor of School of Mechanical and Aerospace Engineering, Sejong University, Korea. His current interests include fuzzy logic controls, inertial sensor applications, and ight control system for UAVs.

Acknowledgements This work was supported by Korea Research Foundation Grant (KRF-000-Y00199). Also, I would like to thank Dr. Don Kim, who is the president of Inertial Science, Inc., USA, for his excellent instruction and support for the experiment.

You might also like